[flang-commits] [flang] [flang][cuda] Implicitly set bind(c) called kernel as global (PR #191050)

[via flang-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-flang-fir-hlfir

Author: Valentin Clement (バレンタイン クレメン) (clementval)

<details>
<summary>Changes</summary>



---
Full diff: https://github.com/llvm/llvm-project/pull/191050.diff


2 Files Affected:

- (modified) flang/lib/Semantics/expression.cpp (+6) 
- (modified) flang/test/Lower/CUDA/cuda-device-proc.cuf (+16) 


``````````diff
diff --git a/flang/lib/Semantics/expression.cpp b/flang/lib/Semantics/expression.cpp
index bd4e2a44df8ae..756bfd551a90a 100644
--- a/flang/lib/Semantics/expression.cpp
+++ b/flang/lib/Semantics/expression.cpp
@@ -3551,6 +3551,12 @@ void ExpressionAnalyzer::Analyze(const parser::CallStmt &callStmt) {
             isKernel = *attrs == common::CUDASubprogramAttrs::Global ||
                 *attrs == common::CUDASubprogramAttrs::Grid_Global;
           }
+          // Implicitly set the CUDA subprogram attributes to GLOBAL for bind(c)
+          // subprograms that are not marked as kernel.
+          if (!isKernel && isBindC && !chevrons->empty()) {
+            const_cast<semantics::SubprogramDetails *>(subpDetails)
+                ->set_cudaSubprogramAttrs(common::CUDASubprogramAttrs::Global);
+          }
         } else if (const auto *procDetails{
                        ultimate.detailsIf<semantics::ProcEntityDetails>()}) {
           isKernel = procDetails->isCUDAKernel();
diff --git a/flang/test/Lower/CUDA/cuda-device-proc.cuf b/flang/test/Lower/CUDA/cuda-device-proc.cuf
index f5ea2c5d414ea..95be49157b907 100644
--- a/flang/test/Lower/CUDA/cuda-device-proc.cuf
+++ b/flang/test/Lower/CUDA/cuda-device-proc.cuf
@@ -2,6 +2,8 @@
 
 ! Test CUDA Fortran procedures available in cudadevice module
 
+
+
 attributes(global) subroutine devsub()
   implicit none
   integer :: ret
@@ -766,3 +768,17 @@ end subroutine
 ! CHECK: nvvm.cp.async.bulk.global.shared.cta %{{.*}}, %{{.*}}, %{{.*}} : <1>, <3>
 ! CHECK: nvvm.inline_ptx "cp.async.bulk.commit_group;"
 ! CHECK: nvvm.cp.async.bulk.wait_group 0
+
+
+subroutine call_c_kernel()
+  interface
+    subroutine c_kernel() bind(c,name='c_kernel')
+    end subroutine
+  end interface
+
+  call c_kernel<<<1,1>>>()
+end subroutine
+
+! CHECK-LABEL: func.func @_QPcall_c_kernel()
+! CHECK: cuf.kernel_launch @c_kernel<<<{{.*}}>>>()
+! CHECK: func.func private @c_kernel() attributes {cuf.proc_attr = #cuf.cuda_proc<global>, fir.bindc_name = "c_kernel", fir.proc_attrs = #fir.proc_attrs<bind_c>}

``````````

</details>


https://github.com/llvm/llvm-project/pull/191050