[clang] abbdc13 - [CUDA][SPIRV] Use OpenCLKernel CC for CUDA -> SPIRV

[Artem Belevich via cfe-commits]

Author: Daniele Castagna
Date: 2021-12-06T15:06:57-08:00
New Revision: abbdc13e6803562bce4e42866a9ebf7f2a04a89b

URL: https://github.com/llvm/llvm-project/commit/abbdc13e6803562bce4e42866a9ebf7f2a04a89b
DIFF: https://github.com/llvm/llvm-project/commit/abbdc13e6803562bce4e42866a9ebf7f2a04a89b.diff

LOG: [CUDA][SPIRV] Use OpenCLKernel CC for CUDA -> SPIRV

Select the OpenCLKernel calling convention for kernels when compiling
CUDA targeting SPIR-V.

In this way the generated LLVM IR will have a spir_kernel calling
convention that will be translated to an OpEntryPoint when converting
to SPIRV.

Reviewed By: jlebar, tra

Differential Revision: https://reviews.llvm.org/D114407

Added: 
    clang/test/CodeGenCUDASPIRV/kernel-cc.cu

Modified: 
    clang/lib/Sema/SemaType.cpp

Removed: 
    


################################################################################
diff  --git a/clang/lib/Sema/SemaType.cpp b/clang/lib/Sema/SemaType.cpp
index b786c9f050ec9..0607d3a774aad 100644
--- a/clang/lib/Sema/SemaType.cpp
+++ b/clang/lib/Sema/SemaType.cpp
@@ -3940,6 +3940,20 @@ static CallingConv getCCForDeclaratorChunk(
         break;
       }
     }
+  } else if (S.getLangOpts().CUDA) {
+    // If we're compiling CUDA/HIP code and targeting SPIR-V we need to make
+    // sure the kernels will be marked with the right calling convention so that
+    // they will be visible by the APIs that ingest SPIR-V.
+    llvm::Triple Triple = S.Context.getTargetInfo().getTriple();
+    if (Triple.getArch() == llvm::Triple::spirv32 ||
+        Triple.getArch() == llvm::Triple::spirv64) {
+      for (const ParsedAttr &AL : D.getDeclSpec().getAttributes()) {
+        if (AL.getKind() == ParsedAttr::AT_CUDAGlobal) {
+          CC = CC_OpenCLKernel;
+          break;
+        }
+      }
+    }
   }
 
   return CC;

diff  --git a/clang/test/CodeGenCUDASPIRV/kernel-cc.cu b/clang/test/CodeGenCUDASPIRV/kernel-cc.cu
new file mode 100644
index 0000000000000..1ba906ebc90d7
--- /dev/null
+++ b/clang/test/CodeGenCUDASPIRV/kernel-cc.cu
@@ -0,0 +1,9 @@
+// RUN: %clang_cc1 -fcuda-is-device -triple spirv32 -o - -emit-llvm -x cuda %s  | FileCheck %s
+// RUN: %clang_cc1 -fcuda-is-device -triple spirv64 -o - -emit-llvm -x cuda %s  | FileCheck %s
+
+// Verifies that building CUDA targeting SPIR-V {32,64} generates LLVM IR with
+// spir_kernel attributes for kernel functions.
+
+// CHECK: define spir_kernel void @_Z6kernelv()
+
+__attribute__((global)) void kernel() { return; }