Simple test profiling (#1062)

* First pass support for performance profiling

* Test across all elements

* Fix bug - sourceContents is not used, should use rawSource.

* * Add ability to get prelude from API.
* Allow specifying source language for render-test
* Made it possible to compile a test input file as C++
  * Special handling for reflection
* Added C++ impl to performance-profile.slang

* Remove some clang warnings.

* Output profile timings on appveyor and other TC.

* Remove passing around of StdWriters (can use global).
Small comment improvements.

7 files changed, 195 insertions, 34 deletions

diff --git a/tools/render-test/cpu-compute-util.cpp b/tools/render-test/cpu-compute-util.cpp
index 81325ce80..688e6b254 100644
--- a/tools/render-test/cpu-compute-util.cpp
+++ b/tools/render-test/cpu-compute-util.cpp
@@ -181,7 +181,7 @@ static CPUComputeUtil::Resource* _newOneTexture2D(int elemCount)
                 }
                 else
                 {
-                    throw TextFormatException("Invalid input syntax at line " + parser.NextToken().Position.Line);
+                    throw TextFormatException(String("Invalid input syntax at line ") + parser.NextToken().Position.Line);
                 }
             }
 
@@ -301,7 +301,7 @@ static CPUComputeUtil::Resource* _newOneTexture2D(int elemCount)
     return SLANG_OK;
 }
 
-/* static */SlangResult CPUComputeUtil::calcExecuteInfo(ExecuteStyle style, const uint32_t dispatchSize[3], const ShaderCompilerUtil::OutputAndLayout& compilationAndLayout, Context& context, ExecuteInfo& out)
+/* static */SlangResult CPUComputeUtil::calcExecuteInfo(ExecuteStyle style, ISlangSharedLibrary* sharedLib, const uint32_t dispatchSize[3], const ShaderCompilerUtil::OutputAndLayout& compilationAndLayout, Context& context, ExecuteInfo& out)
 {
     auto request = compilationAndLayout.output.request;
     auto reflection = (slang::ShaderReflection*) spGetReflection(request);
@@ -314,9 +314,6 @@ static CPUComputeUtil::Resource* _newOneTexture2D(int elemCount)
 
     const char* entryPointName = entryPoint->getName();
 
-    ComPtr<ISlangSharedLibrary> sharedLibrary;
-    SLANG_RETURN_ON_FAIL(spGetEntryPointHostCallable(request, 0, 0, sharedLibrary.writeRef()));
-
     // Copy dispatch size
     for (int i = 0; i < 3; ++i)
     {
@@ -334,7 +331,7 @@ static CPUComputeUtil::Resource* _newOneTexture2D(int elemCount)
             StringBuilder groupEntryPointName;
             groupEntryPointName << entryPointName << "_Group";
 
-            CPPPrelude::ComputeFunc groupFunc = (CPPPrelude::ComputeFunc)sharedLibrary->findFuncByName(groupEntryPointName.getBuffer());
+            CPPPrelude::ComputeFunc groupFunc = (CPPPrelude::ComputeFunc)sharedLib->findFuncByName(groupEntryPointName.getBuffer());
             if (!groupFunc)
             {
                 return SLANG_FAIL;
@@ -346,7 +343,7 @@ static CPUComputeUtil::Resource* _newOneTexture2D(int elemCount)
         case ExecuteStyle::GroupRange:
         {
             CPPPrelude::ComputeFunc groupRangeFunc = nullptr;
-            groupRangeFunc = (CPPPrelude::ComputeFunc)sharedLibrary->findFuncByName(entryPointName);
+            groupRangeFunc = (CPPPrelude::ComputeFunc)sharedLib->findFuncByName(entryPointName);
             if (!groupRangeFunc)
             {
                 return SLANG_FAIL;
@@ -359,7 +356,7 @@ static CPUComputeUtil::Resource* _newOneTexture2D(int elemCount)
             StringBuilder threadEntryPointName;
             threadEntryPointName << entryPointName << "_Thread";
 
-            CPPPrelude::ComputeThreadFunc threadFunc = (CPPPrelude::ComputeThreadFunc)sharedLibrary->findFuncByName(threadEntryPointName.getBuffer());
+            CPPPrelude::ComputeThreadFunc threadFunc = (CPPPrelude::ComputeThreadFunc)sharedLib->findFuncByName(threadEntryPointName.getBuffer());
             if (!threadFunc)
             {
                 return SLANG_FAIL;
@@ -470,7 +467,7 @@ static CPUComputeUtil::Resource* _newOneTexture2D(int elemCount)
 }
 
 
-/* static */ SlangResult CPUComputeUtil::checkStyleConsistency(const uint32_t dispatchSize[3], const ShaderCompilerUtil::OutputAndLayout& compilationAndLayout)
+/* static */ SlangResult CPUComputeUtil::checkStyleConsistency(ISlangSharedLibrary* sharedLib, const uint32_t dispatchSize[3], const ShaderCompilerUtil::OutputAndLayout& compilationAndLayout)
 {
     Context context;
     SLANG_RETURN_ON_FAIL(CPUComputeUtil::calcBindings(compilationAndLayout, context));
@@ -478,7 +475,7 @@ static CPUComputeUtil::Resource* _newOneTexture2D(int elemCount)
     // Run the thread style to test against
     {
         ExecuteInfo info;
-        SLANG_RETURN_ON_FAIL(calcExecuteInfo(ExecuteStyle::Thread, dispatchSize, compilationAndLayout, context, info));
+        SLANG_RETURN_ON_FAIL(calcExecuteInfo(ExecuteStyle::Thread, sharedLib, dispatchSize, compilationAndLayout, context, info));
         SLANG_RETURN_ON_FAIL(execute(info));
     }
 
@@ -489,7 +486,7 @@ static CPUComputeUtil::Resource* _newOneTexture2D(int elemCount)
         SLANG_RETURN_ON_FAIL(CPUComputeUtil::calcBindings(compilationAndLayout, checkContext));
 
         ExecuteInfo info;
-        SLANG_RETURN_ON_FAIL(calcExecuteInfo(style, dispatchSize, compilationAndLayout, checkContext, info));
+        SLANG_RETURN_ON_FAIL(calcExecuteInfo(style, sharedLib, dispatchSize, compilationAndLayout, checkContext, info));
         SLANG_RETURN_ON_FAIL(execute(info));
 
         // Make sure the out buffers are all the same
diff --git a/tools/render-test/cpu-compute-util.h b/tools/render-test/cpu-compute-util.h
index 1284735c0..9430eb841 100644
--- a/tools/render-test/cpu-compute-util.h
+++ b/tools/render-test/cpu-compute-util.h
@@ -50,11 +50,11 @@ struct CPUComputeUtil
 
     
         /// Runs code across run styles and makes sure output buffers match
-    static SlangResult checkStyleConsistency(const uint32_t dispatchSize[3], const ShaderCompilerUtil::OutputAndLayout& compilationAndLayout);
+    static SlangResult checkStyleConsistency(ISlangSharedLibrary* sharedLib, const uint32_t dispatchSize[3], const ShaderCompilerUtil::OutputAndLayout& compilationAndLayout);
 
     static SlangResult calcBindings(const ShaderCompilerUtil::OutputAndLayout& compilationAndLayout, Context& outContext);
 
-    static SlangResult calcExecuteInfo(ExecuteStyle style, const uint32_t dispatchSize[3], const ShaderCompilerUtil::OutputAndLayout& compilationAndLayout, Context& context, ExecuteInfo& out);
+    static SlangResult calcExecuteInfo(ExecuteStyle style, ISlangSharedLibrary* sharedLib, const uint32_t dispatchSize[3], const ShaderCompilerUtil::OutputAndLayout& compilationAndLayout, Context& context, ExecuteInfo& out);
 
     static SlangResult execute(const ExecuteInfo& info);
 
diff --git a/tools/render-test/options.cpp b/tools/render-test/options.cpp
index e13a2b88f..3d5df6f62 100644
--- a/tools/render-test/options.cpp
+++ b/tools/render-test/options.cpp
@@ -44,6 +44,32 @@ static SlangResult _setRendererType(RendererType type, const char* arg, Slang::W
     return SLANG_OK;
 }
 
+static SlangSourceLanguage _findSourceLanguage(const UnownedStringSlice& text)
+{
+    if (text == "c" || text == "C")
+    {
+        return SLANG_SOURCE_LANGUAGE_C;
+    }
+    else if (text == "cpp" || text == "c++" || text == "C++" || text == "cxx")
+    {
+        return SLANG_SOURCE_LANGUAGE_CPP;
+    }
+    else if (text == "slang")
+    {
+        return SLANG_SOURCE_LANGUAGE_SLANG;
+    }
+    else if (text == "glsl")
+    {
+        return SLANG_SOURCE_LANGUAGE_GLSL;
+    }
+    else if (text == "hlsl")
+    {
+        return SLANG_SOURCE_LANGUAGE_HLSL;
+    }
+    return SLANG_SOURCE_LANGUAGE_UNKNOWN;
+}
+
+
 SlangResult parseOptions(int argc, const char*const* argv, Slang::WriterHelper stdError)
 {
     using namespace Slang;
@@ -169,6 +195,10 @@ SlangResult parseOptions(int argc, const char*const* argv, Slang::WriterHelper s
             arg.value = *argCursor++;
             gOptions.compileArgs.add(arg);
         }
+        else if (strcmp(arg, "-performance-profile") == 0)
+        {
+            gOptions.performanceProfile = true;
+        }
         else if (strcmp(arg, "-adapter") == 0)
         {
             if (argCursor == argEnd)
@@ -207,6 +237,24 @@ SlangResult parseOptions(int argc, const char*const* argv, Slang::WriterHelper s
                 gOptions.computeDispatchSize[i] = v;
             }
         }
+        else if (strcmp(arg, "-source-language") == 0)
+        {
+            if (argCursor == argEnd)
+            {
+                stdError.print("error: expecting a source language name for '%s'\n", arg);
+                return SLANG_FAIL;
+            }
+            UnownedStringSlice sourceLanguageText(*argCursor++);
+
+            SlangSourceLanguage sourceLanguage = _findSourceLanguage(sourceLanguageText);
+            if (sourceLanguage == SLANG_SOURCE_LANGUAGE_UNKNOWN)
+            {
+                stdError.print("error: expecting unknown source language name '%s' for '%s'\n", String(sourceLanguageText).getBuffer(), arg);
+                return SLANG_FAIL;
+            }
+
+            gOptions.sourceLanguage = sourceLanguage;
+        }
         else
         {
             // Lookup
@@ -228,7 +276,7 @@ SlangResult parseOptions(int argc, const char*const* argv, Slang::WriterHelper s
                 if (languageRenderType != RendererType::Unknown)
                 {
                     gOptions.targetLanguageRendererType = languageRenderType;
-                    gOptions.inputLanguageID = (argName == "hlsl" || argName == "glsl") ?  InputLanguageID::Native : InputLanguageID::Slang;
+                    gOptions.inputLanguageID = (argName == "hlsl" || argName == "glsl" || argName == "cpp" || argName == "cxx" || argName == "c") ?  InputLanguageID::Native : InputLanguageID::Slang;
                     continue;
                 }
             }
diff --git a/tools/render-test/options.h b/tools/render-test/options.h
index 67eae6603..1bb4af74c 100644
--- a/tools/render-test/options.h
+++ b/tools/render-test/options.h
@@ -49,6 +49,7 @@ struct Options
         /// The set render type
     RendererType rendererType = RendererType::Unknown;
     InputLanguageID inputLanguageID = InputLanguageID::Slang;
+    SlangSourceLanguage sourceLanguage = SLANG_SOURCE_LANGUAGE_UNKNOWN;
 
         /// Can be used for overriding the profile
     const char* profileName = nullptr;
@@ -59,6 +60,8 @@ struct Options
     bool useDXIL = false;
     bool onlyStartup = false;
 
+    bool performanceProfile = false;
+
     Slang::List<Slang::String> renderFeatures;          /// Required render features for this test to run
 
     Slang::List<Slang::CommandLine::Arg> compileArgs;
diff --git a/tools/render-test/render-test-main.cpp b/tools/render-test/render-test-main.cpp
index 3a8871618..13af422a9 100644
--- a/tools/render-test/render-test-main.cpp
+++ b/tools/render-test/render-test-main.cpp
@@ -55,6 +55,13 @@ static const int kVertexCount = SLANG_COUNT_OF(kVertexData);
 
 using namespace Slang;
 
+static void _outputProfileTime(uint64_t startTicks, uint64_t endTicks)
+{
+    WriterHelper out = StdWriters::getOut();
+    double time = double(endTicks - startTicks) / ProcessUtil::getClockFrequency();
+    out.print("profile-time=%g\n", time);
+}
+
 class RenderTestApp : public WindowListener
 {
 	public:
@@ -79,6 +86,8 @@ class RenderTestApp : public WindowListener
 		/// Called in initialize
 	Result _initializeShaders(SlangSession* session, Renderer* renderer, Options::ShaderProgramType shaderType, const ShaderCompilerUtil::Input& input);
 
+    uint64_t m_startTicks;
+
 	// variables for state to be used for rendering...
 	uintptr_t m_constantBufferSize, m_computeResultBufferSize;
 
@@ -232,6 +241,9 @@ void RenderTestApp::runCompute()
     auto pipelineType = PipelineType::Compute;
     m_renderer->setPipelineState(pipelineType, m_pipelineState);
     m_bindingState->apply(m_renderer, pipelineType);
+
+    m_startTicks = ProcessUtil::getClockTick();
+
 	m_renderer->dispatchCompute(m_options.computeDispatchSize[0], m_options.computeDispatchSize[1], m_options.computeDispatchSize[2]);
 }
 
@@ -314,24 +326,60 @@ Result RenderTestApp::update(Window* window)
     }
 
     // If we are in a mode where output is requested, we need to snapshot the back buffer here
-    if (m_options.outputPath)
+    if (m_options.outputPath || m_options.performanceProfile)
     {
         // Submit the work
         m_renderer->submitGpuWork();
         // Wait until everything is complete
         m_renderer->waitForGpu();
 
-        if (gOptions.shaderType == Options::ShaderProgramType::Compute || gOptions.shaderType == Options::ShaderProgramType::GraphicsCompute)
+        if (m_options.performanceProfile)
         {
-            SLANG_RETURN_ON_FAIL(writeBindingOutput(gOptions.outputPath));
+            // It might not be enough on some APIs to 'waitForGpu' to mean the computation has completed. Let's lock an output
+            // buffer to be sure
+            if (m_bindingState->outputBindings.getCount() > 0)
+            {
+                const auto& binding = m_bindingState->outputBindings[0];
+                auto i = binding.entryIndex;
+                const auto& layoutBinding = m_shaderInputLayout.entries[i];
+
+                assert(layoutBinding.isOutput);
+                
+                if (binding.resource && binding.resource->isBuffer())
+                {
+                    BufferResource* bufferResource = static_cast<BufferResource*>(binding.resource.Ptr());
+                    const size_t bufferSize = bufferResource->getDesc().sizeInBytes;
+                    unsigned int* ptr = (unsigned int*)m_renderer->map(bufferResource, MapFlavor::HostRead);
+                    if (!ptr)
+                    {                            
+                        return SLANG_FAIL;
+                    }
+                    m_renderer->unmap(bufferResource);
+                }
+            }
+
+            // Note we don't do the same with screen rendering -> as that will do a lot of work, which may swamp any computation
+            // so can only really profile compute shaders at the moment
+
+            const uint64_t endTicks = ProcessUtil::getClockTick();
+
+            _outputProfileTime(m_startTicks, endTicks);
         }
-        else
+
+        if (gOptions.outputPath)
         {
-            SlangResult res = writeScreen(gOptions.outputPath);
-            if (SLANG_FAILED(res))
+            if (gOptions.shaderType == Options::ShaderProgramType::Compute || gOptions.shaderType == Options::ShaderProgramType::GraphicsCompute)
             {
-                fprintf(stderr, "ERROR: failed to write screen capture to file\n");
-                return res;
+                SLANG_RETURN_ON_FAIL(writeBindingOutput(gOptions.outputPath));
+            }
+            else
+            {
+                SlangResult res = writeScreen(gOptions.outputPath);
+                if (SLANG_FAILED(res))
+                {
+                    fprintf(stderr, "ERROR: failed to write screen capture to file\n");
+                    return res;
+                }
             }
         }
         // We are done
@@ -343,6 +391,8 @@ Result RenderTestApp::update(Window* window)
     return SLANG_OK;
 }
 
+
+
 } //  namespace renderer_test
 
 SLANG_TEST_TOOL_API SlangResult innerMain(Slang::StdWriters* stdWriters, SlangSession* session, int argcIn, const char*const* argvIn)
@@ -432,6 +482,16 @@ SLANG_TEST_TOOL_API SlangResult innerMain(Slang::StdWriters* stdWriters, SlangSe
             break;
     }
 
+    if (gOptions.sourceLanguage != SLANG_SOURCE_LANGUAGE_UNKNOWN)
+    {
+        input.sourceLanguage = gOptions.sourceLanguage;
+
+        if (input.sourceLanguage == SLANG_SOURCE_LANGUAGE_C || input.sourceLanguage == SLANG_SOURCE_LANGUAGE_CPP)
+        {
+            input.passThrough = SLANG_PASS_THROUGH_GENERIC_C_CPP;
+        }
+    }
+
     // Use the profile name set on options if set
     input.profile = gOptions.profileName ? gOptions.profileName : input.profile;
 
@@ -459,24 +519,52 @@ SLANG_TEST_TOOL_API SlangResult innerMain(Slang::StdWriters* stdWriters, SlangSe
         ShaderCompilerUtil::OutputAndLayout compilationAndLayout;
         SLANG_RETURN_ON_FAIL(ShaderCompilerUtil::compileWithLayout(session, gOptions.sourcePath, gOptions.compileArgs, gOptions.shaderType, input, compilationAndLayout));
 
-       
         {
+            // Get the shared library -> it contains the executable code, we need to keep around if we recompile
+            ComPtr<ISlangSharedLibrary> sharedLibrary;
+            SLANG_RETURN_ON_FAIL(spGetEntryPointHostCallable(compilationAndLayout.output.request, 0, 0, sharedLibrary.writeRef()));
+
+            // If we are running c/c++ we still need binding information, so compile again as slang source
+            if (gOptions.sourceLanguage == SLANG_SOURCE_LANGUAGE_C || input.sourceLanguage == SLANG_SOURCE_LANGUAGE_CPP)
+            {
+                ShaderCompilerUtil::Input slangInput = input;
+                slangInput.sourceLanguage = SLANG_SOURCE_LANGUAGE_SLANG;
+                slangInput.passThrough = SLANG_PASS_THROUGH_NONE;
+                // We just want CPP, so we get suitable reflection
+                slangInput.target = SLANG_CPP_SOURCE;
+
+                SLANG_RETURN_ON_FAIL(ShaderCompilerUtil::compileWithLayout(session, gOptions.sourcePath, gOptions.compileArgs, gOptions.shaderType, slangInput, compilationAndLayout));
+            }
+
+            // calculate binding
             CPUComputeUtil::Context context;
             SLANG_RETURN_ON_FAIL(CPUComputeUtil::calcBindings(compilationAndLayout, context));
 
+            // Get the execution info from the lib
             CPUComputeUtil::ExecuteInfo info;
-            SLANG_RETURN_ON_FAIL(CPUComputeUtil::calcExecuteInfo(CPUComputeUtil::ExecuteStyle::GroupRange, gOptions.computeDispatchSize, compilationAndLayout, context, info));
+            SLANG_RETURN_ON_FAIL(CPUComputeUtil::calcExecuteInfo(CPUComputeUtil::ExecuteStyle::GroupRange, sharedLibrary, gOptions.computeDispatchSize, compilationAndLayout, context, info));
+
+            const uint64_t startTicks = ProcessUtil::getClockTick();
+
             SLANG_RETURN_ON_FAIL(CPUComputeUtil::execute(info));
-        
-            // Dump everything out that was written
-            SLANG_RETURN_ON_FAIL(CPUComputeUtil::writeBindings(compilationAndLayout.layout, context.buffers, gOptions.outputPath));
-        }
 
-        {
-            // Check all execution styles produce the same result
-            SLANG_RETURN_ON_FAIL(CPUComputeUtil::checkStyleConsistency(gOptions.computeDispatchSize, compilationAndLayout));
+            if (gOptions.performanceProfile)
+            {
+                const uint64_t endTicks = ProcessUtil::getClockTick();
+                _outputProfileTime(startTicks, endTicks);
+            }
+
+            if (gOptions.outputPath)
+            {
+                // Dump everything out that was written
+                SLANG_RETURN_ON_FAIL(CPUComputeUtil::writeBindings(compilationAndLayout.layout, context.buffers, gOptions.outputPath));
+
+                // Check all execution styles produce the same result
+                SLANG_RETURN_ON_FAIL(CPUComputeUtil::checkStyleConsistency(sharedLibrary, gOptions.computeDispatchSize, compilationAndLayout));
+            }
         }
 
+        
         return SLANG_OK;
     }
 
diff --git a/tools/render-test/shader-input-layout.cpp b/tools/render-test/shader-input-layout.cpp
index 0d13e5ca1..b5b65f52d 100644
--- a/tools/render-test/shader-input-layout.cpp
+++ b/tools/render-test/shader-input-layout.cpp
@@ -506,7 +506,7 @@ namespace renderer_test
                                     Token nameToken = parser.ReadToken();
                                     if (nameToken.Type != TokenType::Identifier)
                                     {
-                                        throw TextFormatException("Invalid input syntax at line " + parser.NextToken().Position.Line);
+                                        throw TextFormatException(String("Invalid input syntax at line ") + parser.NextToken().Position.Line);
                                     }
                                     builder << nameToken.Content;
 
@@ -537,7 +537,7 @@ namespace renderer_test
                                         }
                                         else
                                         {
-                                            throw TextFormatException("Invalid input syntax at line " + parser.NextToken().Position.Line);
+                                            throw TextFormatException(String("Invalid input syntax at line ") + parser.NextToken().Position.Line);
                                         }
                                     }
 
@@ -553,7 +553,7 @@ namespace renderer_test
                 }
                 catch (TextFormatException)
                 {
-                    throw TextFormatException("Invalid input syntax at line " + parser.NextToken().Position.Line);
+                    throw TextFormatException(String("Invalid input syntax at line ") + parser.NextToken().Position.Line);
                 }
             }
         }
diff --git a/tools/render-test/slang-support.cpp b/tools/render-test/slang-support.cpp
index 515481d4f..9afddc09b 100644
--- a/tools/render-test/slang-support.cpp
+++ b/tools/render-test/slang-support.cpp
@@ -9,6 +9,8 @@
 #include <assert.h>
 #include <stdio.h>
 
+#include "../../source/core/slang-string-util.h"
+
 namespace renderer_test {
 using namespace Slang;
 
@@ -54,6 +56,12 @@ static const char computeEntryPointName[] = "computeMain";
     case SLANG_SOURCE_LANGUAGE_HLSL:
         spAddPreprocessorDefine(slangRequest, "__HLSL__", "1");
         break;
+    case SLANG_SOURCE_LANGUAGE_C:
+        spAddPreprocessorDefine(slangRequest, "__C__", "1");
+        break;
+    case SLANG_SOURCE_LANGUAGE_CPP:
+        spAddPreprocessorDefine(slangRequest, "__CPP__", "1");
+        break;
 
     default:
         assert(!"unexpected");
@@ -240,6 +248,23 @@ static const char computeEntryPointName[] = "computeMain";
     List<char> sourceText;
     SLANG_RETURN_ON_FAIL(readSource(sourcePath, sourceText));
 
+    if (input.sourceLanguage == SLANG_SOURCE_LANGUAGE_CPP || input.sourceLanguage == SLANG_SOURCE_LANGUAGE_C)
+    {
+        // Add an include of the prelude
+        ComPtr<ISlangBlob> prelude;
+        session->getDownstreamCompilerPrelude(SLANG_PASS_THROUGH_GENERIC_C_CPP, prelude.writeRef());
+
+        String preludeString = StringUtil::getString(prelude);
+
+        // Add the prelude 
+        StringBuilder builder;
+        builder << preludeString << "\n";
+        builder << UnownedStringSlice(sourceText.getBuffer(), sourceText.getCount());
+
+        sourceText.setCount(builder.getLength());
+        memcpy(sourceText.getBuffer(), builder.getBuffer(), builder.getLength());
+    }
+
     output.sourcePath = sourcePath;
 
     auto& layout = output.layout;