4 files changed, 55 insertions, 12 deletions

diff --git a/source/slang/slang-emit.cpp b/source/slang/slang-emit.cpp
index 38f066c6c..0f53f74cd 100644
--- a/source/slang/slang-emit.cpp
+++ b/source/slang/slang-emit.cpp
@@ -557,6 +557,17 @@ Result linkAndOptimizeIR(
 
     switch (target)
     {
+    case CodeGenTarget::CUDASource:
+    case CodeGenTarget::PyTorchCppBinding:
+    break;
+
+    default:
+        removeTorchAndCUDAEntryPoints(irModule);
+        break;
+    }
+
+    switch (target)
+    {
     case CodeGenTarget::CPPSource:
     case CodeGenTarget::HostCPPSource:
     {
@@ -605,10 +616,19 @@ Result linkAndOptimizeIR(
     if (!targetProgram->getOptionSet().shouldPerformMinimumOptimizations())
         fuseCallsToSaturatedCooperation(irModule);
 
-    // Generate any requested derivative wrappers
-    if (requiredLoweringPassSet.derivativePyBindWrapper)
-        generateDerivativeWrappers(irModule, sink);
-
+    switch (target)
+    {   
+    case CodeGenTarget::CUDASource:
+    case CodeGenTarget::PyTorchCppBinding:
+    {
+        // Generate any requested derivative wrappers
+        if (requiredLoweringPassSet.derivativePyBindWrapper)
+            generateDerivativeWrappers(irModule, sink);
+        break;
+    }
+    default:
+        break;
+    }
     // Next, we need to ensure that the code we emit for
     // the target doesn't contain any operations that would
     // be illegal on the target platform. For example,
diff --git a/source/slang/slang-ir-pytorch-cpp-binding.cpp b/source/slang/slang-ir-pytorch-cpp-binding.cpp
index 5105b2e81..6922984d6 100644
--- a/source/slang/slang-ir-pytorch-cpp-binding.cpp
+++ b/source/slang/slang-ir-pytorch-cpp-binding.cpp
@@ -632,7 +632,6 @@ void generateReflectionFunc(IRBuilder* builder, IRFunc* kernelFunc, IRFunc* host
 
     builder->addExternCppDecoration(reflectionFunc, reflFuncExportName.getUnownedSlice());
     builder->addTorchEntryPointDecoration(reflectionFunc, reflFuncExportName.getUnownedSlice());
-    builder->addHLSLExportDecoration(reflectionFunc);
     builder->addKeepAliveDecoration(reflectionFunc);
 }
 
@@ -817,7 +816,6 @@ void generateReflectionForType(IRType* type, DiagnosticSink* sink)
  
     builder.addTorchEntryPointDecoration(reflFunc, reflFuncExportName.getUnownedSlice());
     builder.addExternCppDecoration(reflFunc, reflFuncExportName.getUnownedSlice());
-    builder.addHLSLExportDecoration(reflFunc);
     builder.addKeepAliveDecoration(reflFunc);
 }
 
@@ -899,7 +897,6 @@ IRFunc* generateCUDAWrapperForFunc(IRFunc* func, DiagnosticSink* sink)
         // Mark for host-side emit logic.
         builder.addCudaHostDecoration(hostFunc);
         // Keep alive. This method will be accessed externally.
-        builder.addHLSLExportDecoration(hostFunc);
         builder.addKeepAliveDecoration(hostFunc);
     }
 
@@ -1163,6 +1160,27 @@ void handleAutoBindNames(IRModule* module)
     }
 }
 
+void removeTorchAndCUDAEntryPoints(IRModule* module)
+{
+    // Go through global insts, find cuda & torch related entry points and remove the keep-alive decoration.
+    IRBuilder builder(module);
+    for (auto globalInst : module->getGlobalInsts())
+    {
+        if (auto func = as<IRFunc>(globalInst))
+        {
+            if (func->findDecoration<IRAutoPyBindCudaDecoration>() ||
+                func->findDecoration<IRTorchEntryPointDecoration>() ||
+                func->findDecoration<IRCudaKernelDecoration>())
+            {
+                if (auto keepAlive = func->findDecoration<IRKeepAliveDecoration>())
+                    keepAlive->removeAndDeallocate();
+                if (auto hlslExport = func->findDecoration<IRHLSLExportDecoration>())
+                    hlslExport->removeAndDeallocate();
+            }
+        }
+    }
+}
+
 void generateDerivativeWrappers(IRModule* module, DiagnosticSink* sink)
 {
     SLANG_UNUSED(sink);
@@ -1237,7 +1255,6 @@ void generateDerivativeWrappers(IRModule* module, DiagnosticSink* sink)
                     builder.addExternCppDecoration(wrapperFunc, nameBuilder.getUnownedSlice());
                 }
 
-                builder.addHLSLExportDecoration(wrapperFunc);
                 builder.addKeepAliveDecoration(wrapperFunc);
 
                 builder.addCudaKernelForwardDerivativeDecoration(func, wrapperFunc);
@@ -1296,7 +1313,6 @@ void generateDerivativeWrappers(IRModule* module, DiagnosticSink* sink)
                     builder.addExternCppDecoration(wrapperFunc, nameBuilder.getUnownedSlice());
                 }
 
-                builder.addHLSLExportDecoration(wrapperFunc);
                 builder.addKeepAliveDecoration(wrapperFunc);
 
                 builder.addCudaKernelBackwardDerivativeDecoration(func, wrapperFunc);
diff --git a/source/slang/slang-ir-pytorch-cpp-binding.h b/source/slang/slang-ir-pytorch-cpp-binding.h
index a761dbc03..6d022db7b 100644
--- a/source/slang/slang-ir-pytorch-cpp-binding.h
+++ b/source/slang/slang-ir-pytorch-cpp-binding.h
@@ -11,6 +11,7 @@ void removeTorchKernels(IRModule* module);
 void handleAutoBindNames(IRModule* module);
 void generateDerivativeWrappers(IRModule* module, DiagnosticSink* sink);
 void lowerBuiltinTypesForKernelEntryPoints(IRModule* module, DiagnosticSink* sink);
+void removeTorchAndCUDAEntryPoints(IRModule* module);
 
 }
 
diff --git a/source/slang/slang-lower-to-ir.cpp b/source/slang/slang-lower-to-ir.cpp
index f8faf7c07..cc4704fe9 100644
--- a/source/slang/slang-lower-to-ir.cpp
+++ b/source/slang/slang-lower-to-ir.cpp
@@ -1385,21 +1385,27 @@ static void addLinkageDecoration(
         {
             builder->addCudaKernelDecoration(inst);
             builder->addExternCppDecoration(inst, decl->getName()->text.getUnownedSlice());
-            builder->addHLSLExportDecoration(inst);
+
+            // Temp decorations to get this function through the linker.
             builder->addKeepAliveDecoration(inst);
+            builder->addHLSLExportDecoration(inst);
         }
         else if (as<TorchEntryPointAttribute>(modifier))
         {
             builder->addTorchEntryPointDecoration(inst, decl->getName()->text.getUnownedSlice());
             builder->addCudaHostDecoration(inst);
-            builder->addHLSLExportDecoration(inst);
-            builder->addKeepAliveDecoration(inst);
             builder->addExternCppDecoration(inst, decl->getName()->text.getUnownedSlice());
+
+            // Temp decorations to get this function through the linker.
+            builder->addKeepAliveDecoration(inst);
+            builder->addHLSLExportDecoration(inst);
         }
         else if (as<AutoPyBindCudaAttribute>(modifier))
         {
             builder->addAutoPyBindCudaDecoration(inst, decl->getName()->text.getUnownedSlice());
             builder->addAutoPyBindExportInfoDecoration(inst);
+
+            // Temp decorations to get this function through the linker.
             builder->addKeepAliveDecoration(inst);
             builder->addHLSLExportDecoration(inst);
         }