[llvm] [llvm] Preserve existing calling conv when auto-upgrading nvvm.annota… (PR #125568)

[via llvm-commits]

https://github.com/macurtis-amd created https://github.com/llvm/llvm-project/pull/125568

…tions

Kernels produced by ROCm classic flang may have both nvvm "kernel" annotation and amdgpu_kernel calling convention.

>From c46e3eef1e62db86f9675b5b4be6e1361a6c6504 Mon Sep 17 00:00:00 2001
From: Matthew Curtis <macurtis at amd.com>
Date: Mon, 3 Feb 2025 12:46:47 -0600
Subject: [PATCH] [llvm] Preserve existing calling conv when auto-upgrading
 nvvm.annotations

Kernels produced by ROCm classic flang may have both nvvm "kernel" annotation
and amdgpu_kernel calling convention.
---
 llvm/lib/IR/AutoUpgrade.cpp                        |  6 ++++--
 .../CodeGen/AMDGPU/upgrade-nvvm-annotations.ll     | 14 ++++++++++++++
 2 files changed, 18 insertions(+), 2 deletions(-)
 create mode 100644 llvm/test/CodeGen/AMDGPU/upgrade-nvvm-annotations.ll

diff --git a/llvm/lib/IR/AutoUpgrade.cpp b/llvm/lib/IR/AutoUpgrade.cpp
index e886a6012b219ae..c64aec69d9dbb1f 100644
--- a/llvm/lib/IR/AutoUpgrade.cpp
+++ b/llvm/lib/IR/AutoUpgrade.cpp
@@ -5023,8 +5023,10 @@ bool llvm::UpgradeDebugInfo(Module &M) {
 bool static upgradeSingleNVVMAnnotation(GlobalValue *GV, StringRef K,
                                         const Metadata *V) {
   if (K == "kernel") {
-    if (!mdconst::extract<ConstantInt>(V)->isZero())
-      cast<Function>(GV)->setCallingConv(CallingConv::PTX_Kernel);
+    auto *F = cast<Function>(GV);
+    if (!mdconst::extract<ConstantInt>(V)->isZero() &&
+        !F->hasKernelCallingConv())
+      F->setCallingConv(CallingConv::PTX_Kernel);
     return true;
   }
   if (K == "align") {
diff --git a/llvm/test/CodeGen/AMDGPU/upgrade-nvvm-annotations.ll b/llvm/test/CodeGen/AMDGPU/upgrade-nvvm-annotations.ll
new file mode 100644
index 000000000000000..057569f2f1fd6af
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/upgrade-nvvm-annotations.ll
@@ -0,0 +1,14 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -mtriple=amdgcn-amd-amdhsa -O0 -S | FileCheck %s
+
+define amdgpu_kernel void @foo(i32 %a, i32 %b) {
+; CHECK-LABEL: define amdgpu_kernel void @foo(
+; CHECK-SAME: i32 [[A:%.*]], i32 [[B:%.*]]) {
+; CHECK-NEXT:    ret void
+;
+  ret void
+}
+
+!nvvm.annotations = !{!0}
+!0 = !{ptr @foo, !"kernel", i32 1}
+