8 files changed, 42 insertions, 11 deletions
diff --git a/source/slang/slang-compiler.cpp b/source/slang/slang-compiler.cpp
index 8426b7eec..a5f46da3e 100644
--- a/source/slang/slang-compiler.cpp
+++ b/source/slang/slang-compiler.cpp
@@ -1548,6 +1548,8 @@ namespace Slang
     // Do emit logic for a zero or more entry points
     SlangResult CodeGenContext::emitEntryPoints(ComPtr<IArtifact>& outArtifact)
     {
+        CompileTimerRAII recordCompileTime(getSession());
+
         auto target = getTargetFormat();
 
         switch (target)
diff --git a/source/slang/slang-compiler.h b/source/slang/slang-compiler.h
index 4d3df5433..66454b2da 100755
--- a/source/slang/slang-compiler.h
+++ b/source/slang/slang-compiler.h
@@ -2947,9 +2947,10 @@ namespace Slang
 
         SLANG_NO_THROW void SLANG_MCALL setDownstreamCompilerForTransition(SlangCompileTarget source, SlangCompileTarget target, SlangPassThrough compiler) override;
         SLANG_NO_THROW SlangPassThrough SLANG_MCALL getDownstreamCompilerForTransition(SlangCompileTarget source, SlangCompileTarget target) override;
-        SLANG_NO_THROW double SLANG_MCALL getDownstreamCompilerElapsedTime() override
+        SLANG_NO_THROW void SLANG_MCALL getCompilerElapsedTime(double* outTotalTime, double* outDownstreamTime) override
         {
-            return m_downstreamCompileTime;
+            *outDownstreamTime = m_downstreamCompileTime;
+            *outTotalTime = m_totalCompileTime;
         }
         
             /// Get the downstream compiler for a transition
@@ -3028,6 +3029,7 @@ namespace Slang
         ~Session();
 
         void addDownstreamCompileTime(double time) { m_downstreamCompileTime += time; }
+        void addTotalCompileTime(double time) { m_totalCompileTime += time; }
 
         ComPtr<ISlangSharedLibraryLoader> m_sharedLibraryLoader;                    ///< The shared library loader (never null)
 
@@ -3060,6 +3062,7 @@ namespace Slang
         CodeGenTransitionMap m_codeGenTransitionMap;
 
         double m_downstreamCompileTime = 0.0;
+        double m_totalCompileTime = 0.0;
     };
 
     void checkTranslationUnit(
@@ -3164,6 +3167,24 @@ SLANG_FORCE_INLINE SlangCompileTarget asExternal(CodeGenTarget target) { return
 
 SLANG_FORCE_INLINE SlangSourceLanguage asExternal(SourceLanguage sourceLanguage) { return (SlangSourceLanguage)sourceLanguage; }
 
+// Helper class for recording compile time.
+struct CompileTimerRAII
+{
+    std::chrono::high_resolution_clock::time_point startTime;
+    Session* session;
+    CompileTimerRAII(Session* inSession)
+    {
+        startTime = std::chrono::high_resolution_clock::now();
+        session = inSession;
+    }
+    ~CompileTimerRAII()
+    {
+        double elapsedTime = std::chrono::duration_cast<std::chrono::microseconds>(
+            std::chrono::high_resolution_clock::now() - startTime)
+            .count() / 1e6;
+        session->addTotalCompileTime(elapsedTime);
+    }
+};
 }
 
 #endif
diff --git a/source/slang/slang-emit.cpp b/source/slang/slang-emit.cpp
index 071ea9639..a96cb6908 100644
--- a/source/slang/slang-emit.cpp
+++ b/source/slang/slang-emit.cpp
@@ -584,17 +584,14 @@ Result linkAndOptimizeIR(
     specializeResourceUsage(codeGenContext, irModule);
     specializeFuncsForBufferLoadArgs(codeGenContext, irModule);
 
-    //
-    simplifyIR(irModule);
-
     // For GLSL targets, we also want to specialize calls to functions that
     // takes array parameters if possible, to avoid performance issues on
     // those platforms.
     if (isKhronosTarget(targetRequest))
     {
         specializeArrayParameters(codeGenContext, irModule);
-        simplifyIR(irModule);
     }
+    simplifyIR(irModule);
 
     // Rewrite functions that return arrays to return them via `out` parameter,
     // since our target languages doesn't allow returning arrays.
diff --git a/source/slang/slang-ir-defunctionalization.cpp b/source/slang/slang-ir-defunctionalization.cpp
index ac4258753..15e5a3ff7 100644
--- a/source/slang/slang-ir-defunctionalization.cpp
+++ b/source/slang/slang-ir-defunctionalization.cpp
@@ -30,7 +30,6 @@ bool specializeHigherOrderParameters(
     while (changed)
     {
         changed = specializeFunctionCalls(codeGenContext, module, &condition);
-        simplifyIR(module);
         result |= changed;
     }
     return result;
diff --git a/source/slang/slang-ir-specialize-function-call.cpp b/source/slang/slang-ir-specialize-function-call.cpp
index e0a757e11..2fa93087f 100644
--- a/source/slang/slang-ir-specialize-function-call.cpp
+++ b/source/slang/slang-ir-specialize-function-call.cpp
@@ -4,6 +4,7 @@
 #include "slang-ir.h"
 #include "slang-ir-clone.h"
 #include "slang-ir-insts.h"
+#include "slang-ir-ssa-simplification.h"
 
 namespace Slang
 {
@@ -888,6 +889,8 @@ struct FunctionParameterSpecializationContext
         //
         addCallsToWorkListRec(newFunc);
 
+        simplifyFunc(newFunc);
+
         return newFunc;
     }
 };
diff --git a/source/slang/slang-ir-specialize-resources.cpp b/source/slang/slang-ir-specialize-resources.cpp
index 83f280d9b..3d28e0bb0 100644
--- a/source/slang/slang-ir-specialize-resources.cpp
+++ b/source/slang/slang-ir-specialize-resources.cpp
@@ -95,7 +95,6 @@ bool specializeResourceParameters(
     while (changed)
     {
         changed = specializeFunctionCalls(codeGenContext, module, &condition);
-        simplifyIR(module);
         result |= changed;
     }
     return result;
diff --git a/source/slang/slang-ir-specialize.cpp b/source/slang/slang-ir-specialize.cpp
index 5962f6dbb..80dcec82b 100644
--- a/source/slang/slang-ir-specialize.cpp
+++ b/source/slang/slang-ir-specialize.cpp
@@ -6,6 +6,7 @@
 #include "slang-ir-insts.h"
 #include "slang-ir-ssa-simplification.h"
 #include "slang-ir-lower-witness-lookup.h"
+#include "slang-ir-dce.h"
 
 namespace Slang
 {
@@ -929,8 +930,8 @@ struct SpecializationContext
 
             if (iterChanged)
             {
-                simplifyIR(module);
                 this->changed = true;
+                eliminateDeadCode(module->getModuleInst(), IRDeadCodeEliminationOptions());
             }
             else
             {
@@ -1617,6 +1618,8 @@ struct SpecializationContext
         //
         addToWorkList(newFunc);
 
+        simplifyFunc(newFunc);
+
         return newFunc;
     }
 
diff --git a/source/slang/slang.cpp b/source/slang/slang.cpp
index ead1e338f..d651f0737 100644
--- a/source/slang/slang.cpp
+++ b/source/slang/slang.cpp
@@ -2997,6 +2997,8 @@ RefPtr<Module> Linkage::loadModule(
     DiagnosticSink*     sink,
     const LoadedModuleDictionary* additionalLoadedModules)
 {
+    CompileTimerRAII recordCompileTime(static_cast<Session*>(getGlobalSession()));
+
     RefPtr<FrontEndCompileRequest> frontEndReq = new FrontEndCompileRequest(this, nullptr, sink);
 
     frontEndReq->additionalLoadedModules = additionalLoadedModules;
@@ -5118,9 +5120,11 @@ SlangResult EndToEndCompileRequest::EndToEndCompileRequest::compile()
 {
     SlangResult res = SLANG_FAIL;
     double downstreamStartTime = 0.0;
+    double totalStartTime = 0.0;
+
     if (m_reportDownstreamCompileTime)
     {
-        downstreamStartTime = getSession()->getDownstreamCompilerElapsedTime();
+        getSession()->getCompilerElapsedTime(&totalStartTime, &downstreamStartTime);
     }
 #if !defined(SLANG_DEBUG_INTERNAL_ERROR)
     // By default we'd like to catch as many internal errors as possible,
@@ -5171,7 +5175,10 @@ SlangResult EndToEndCompileRequest::EndToEndCompileRequest::compile()
 
     if (m_reportDownstreamCompileTime)
     {
-        double downstreamTime = getSession()->getDownstreamCompilerElapsedTime() - downstreamStartTime;
+        double downstreamEndTime = 0;
+        double totalEndTime = 0;
+        getSession()->getCompilerElapsedTime(&totalEndTime, &downstreamEndTime);
+        double downstreamTime = downstreamEndTime - downstreamStartTime;
         String downstreamTimeStr = String(downstreamTime, "%.2f");
         getSink()->diagnose(SourceLoc(), Diagnostics::downstreamCompileTime, downstreamTimeStr);