[HIP] Compile device code

llvm · Feb 7, 2025 · 52049fc · 52049fc
1 parent 1783f4c
commit 52049fc
Show file tree

Hide file tree

Showing 4 changed files with 10 additions and 5 deletions.
diff --git a/clang/lib/CIR/CodeGen/CIRGenCall.cpp b/clang/lib/CIR/CodeGen/CIRGenCall.cpp
@@ -1641,7 +1641,8 @@ static void getTrivialDefaultFunctionAttributes(
   // AFAIK, neither of them support exceptions in device code.
   if (langOpts.SYCLIsDevice)
     llvm_unreachable("NYI");
-  if (langOpts.OpenCL || (langOpts.CUDA && langOpts.CUDAIsDevice)) {
+  if (langOpts.OpenCL ||
+      ((langOpts.CUDA || langOpts.HIP) && langOpts.CUDAIsDevice)) {
     auto noThrow = cir::NoThrowAttr::get(CGM.getBuilder().getContext());
     funcAttrs.set(noThrow.getMnemonic(), noThrow);
   }

diff --git a/clang/lib/CIR/CodeGen/CIRGenModule.cpp b/clang/lib/CIR/CodeGen/CIRGenModule.cpp
@@ -3114,7 +3114,7 @@ void CIRGenModule::emitDeferred(unsigned recursionLimit) {
   // Emit CUDA/HIP static device variables referenced by host code only. Note we
   // should not clear CUDADeviceVarODRUsedByHost since it is still needed for
   // further handling.
-  if (getLangOpts().CUDA && getLangOpts().CUDAIsDevice &&
+  if ((getLangOpts().CUDA || getLangOpts().HIP) && getLangOpts().CUDAIsDevice &&
       !getASTContext().CUDADeviceVarODRUsedByHost.empty()) {
     llvm_unreachable("NYI");
   }

diff --git a/clang/lib/CIR/CodeGen/CIRGenTypes.cpp b/clang/lib/CIR/CodeGen/CIRGenTypes.cpp
@@ -348,7 +348,11 @@ mlir::Type CIRGenTypes::convertType(QualType T) {
 
   // For the device-side compilation, CUDA device builtin surface/texture types
   // may be represented in different types.
-  if (astContext.getLangOpts().CUDAIsDevice) {
+  // NOTE: CUDAIsDevice is true when building also HIP code.
+  //  1. There is no SurfaceType on HIP,
+  //  2. There is Texture memory on HIP but accessing the memory goes through
+  //  calls to the runtime. e.g. for a 2D: `tex2D<float>(tex, x, y);`
+  if (astContext.getLangOpts().CUDA && astContext.getLangOpts().CUDAIsDevice) {
     if (Ty->isCUDADeviceBuiltinSurfaceType() ||
         Ty->isCUDADeviceBuiltinTextureType())
       llvm_unreachable("NYI");

diff --git a/clang/test/CIR/CodeGen/HIP/simple-device.cpp b/clang/test/CIR/CodeGen/HIP/simple-device.cpp
@@ -1,7 +1,7 @@
 #include "../Inputs/cuda.h"
 
-// RUN: %clang_cc1 -triple=amdgcn-amd-amdhsa -fcuda-is-device \
-// RUN:            -x hip -fclangir -emit-cir -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple=amdgcn-amd-amdhsa -x hip -fcuda-is-device \
+// RUN:            -fclangir -emit-cir -o - %s | FileCheck %s
 
 // This shouldn't emit.
 __host__ void host_fn(int *a, int *b, int *c) {}