[clang] [clang][OpenMP] Don't declare HIP's ::min and ::max in an OpenMP TU (PR #225162)

[Larry Meadows via cfe-commits]

https://github.com/lfmeadow updated https://github.com/llvm/llvm-project/pull/225162

>From f79903201822ec0d465ee819eca4ccffba53ddb9 Mon Sep 17 00:00:00 2001
From: Larry Meadows <Lawrence.Meadows at amd.com>
Date: Mon, 21 Sep 2026 13:40:33 -0500
Subject: [PATCH] [clang][OpenMP] Don't put HIP's ::min and ::max in an OpenMP
 TU

__clang_hip_math.h defines templated ::min and ::max for CUDA source
compatibility. The OpenMP math wrapper includes that header for the
amdgcn device pass, so an OpenMP translation unit -- ordinary C++ that
never opted into HIP -- gets them too, and an unqualified min(a, b)
after `using namespace std` becomes ambiguous against std::min.

Guard them with __OPENMP_AMDGCN__, matching the host min/max block
below, which is already excluded from OpenMP for the same reason.
---
 clang/lib/Headers/__clang_hip_math.h            |  2 ++
 .../amdgcn_openmp_device_math_minmax.cpp        | 17 +++++++++++++++++
 2 files changed, 19 insertions(+)
 create mode 100644 clang/test/Headers/amdgcn_openmp_device_math_minmax.cpp

diff --git a/clang/lib/Headers/__clang_hip_math.h b/clang/lib/Headers/__clang_hip_math.h
index f0e6d46ed0af4a..c1fc112c95a4c1 100644
--- a/clang/lib/Headers/__clang_hip_math.h
+++ b/clang/lib/Headers/__clang_hip_math.h
@@ -1284,6 +1284,7 @@ double __fma_rn(double __x, double __y, double __z) {
 #endif // !defined(__cplusplus) && __STDC_VERSION__ >= 201112L
 
 #if defined(__cplusplus)
+#ifndef __OPENMP_AMDGCN__
 template <class T> __DEVICE__ T min(T __arg1, T __arg2) {
   return (__arg1 < __arg2) ? __arg1 : __arg2;
 }
@@ -1291,6 +1292,7 @@ template <class T> __DEVICE__ T min(T __arg1, T __arg2) {
 template <class T> __DEVICE__ T max(T __arg1, T __arg2) {
   return (__arg1 > __arg2) ? __arg1 : __arg2;
 }
+#endif
 
 __DEVICE__ int min(int __arg1, int __arg2) {
   return (__arg1 < __arg2) ? __arg1 : __arg2;
diff --git a/clang/test/Headers/amdgcn_openmp_device_math_minmax.cpp b/clang/test/Headers/amdgcn_openmp_device_math_minmax.cpp
new file mode 100644
index 00000000000000..b84d76eaaa3de4
--- /dev/null
+++ b/clang/test/Headers/amdgcn_openmp_device_math_minmax.cpp
@@ -0,0 +1,17 @@
+// RUN: %clang_cc1 -internal-isystem %S/Inputs/include -x c++ -fopenmp -triple x86_64-unknown-unknown -fopenmp-targets=amdgpu-amd-amdhsa -emit-llvm-bc %s -o %t-host.bc
+// RUN: %clang_cc1 -internal-isystem %S/../../lib/Headers/openmp_wrappers -include __clang_openmp_device_functions.h -internal-isystem %S/../../lib/Headers/openmp_wrappers -internal-isystem %S/Inputs/include -x c++ -fopenmp -triple amdgpu-amd-amdhsa -aux-triple x86_64-unknown-unknown -fopenmp-targets=amdgpu-amd-amdhsa -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-host.bc -verify -o -
+
+// The OpenMP math wrapper pulls in __clang_hip_math.h, whose ::min and ::max
+// are a CUDA compatibility extension for HIP. In an OpenMP translation unit
+// these calls are ordinary unqualified C++ lookups and must resolve to std.
+
+// expected-no-diagnostics
+
+#include <algorithm>
+#include <cmath>
+
+using namespace std;
+
+long test_min(long __a, long __b) { return min(__a, __b); }
+
+long test_max(long __a, long __b) { return max(__a, __b); }