[libcxx-commits] [libcxx] [libc] Adjust inline assembly constraints for the AMDGPU target (PR #101747)

[Joseph Huber via libcxx-commits]

https://github.com/jhuber6 created https://github.com/llvm/llvm-project/pull/101747

Summary:
These assembly constraints are illegal / invalid on the AMDGPU target.
The `r` constraint is only valid on inputs and the `m` constraint isn't
accepted at all. The NVPTX target can handle them because it uses a more
permissive virtual machine (PTX is an IR). Simply add exceptions on the
target to make these work.


>From 248c12fbcb979a9a4be07e7faf5841793e458c15 Mon Sep 17 00:00:00 2001
From: Joseph Huber <huberjn at outlook.com>
Date: Fri, 2 Aug 2024 14:00:42 -0500
Subject: [PATCH] [libc] Adjust inline assembly constraints for the AMDGPU
 target

Summary:
These assembly constraints are illegal / invalid on the AMDGPU target.
The `r` constraint is only valid on inputs and the `m` constraint isn't
accepted at all. The NVPTX target can handle them because it uses a more
permissive virtual machine (PTX is an IR). Simply add exceptions on the
target to make these work.
---
 libcxx/test/support/test_macros.h | 18 +++++++++++++-----
 1 file changed, 13 insertions(+), 5 deletions(-)

diff --git a/libcxx/test/support/test_macros.h b/libcxx/test/support/test_macros.h
index 15fc5b69b5207..65037d03b9567 100644
--- a/libcxx/test/support/test_macros.h
+++ b/libcxx/test/support/test_macros.h
@@ -291,17 +291,25 @@ struct is_same<T, T> { enum {value = 1}; };
 // when optimizations are enabled.
 template <class Tp>
 inline Tp const& DoNotOptimize(Tp const& value) {
-    asm volatile("" : : "r,m"(value) : "memory");
-    return value;
+  // The `m` constraint is invalid in the AMDGPU backend.
+#  if defined(__AMDGPU__) || defined(__NVPTX__)
+  asm volatile("" : : "r"(value) : "memory");
+#  else
+  asm volatile("" : : "r,m"(value) : "memory");
+#  endif
+  return value;
 }
 
 template <class Tp>
 inline Tp& DoNotOptimize(Tp& value) {
-#if defined(__clang__)
+  // The `m` and `r` output constraint is invalid in the AMDGPU backend.
+#  if defined(__AMDGPU__)
+  asm volatile("" : "+v"(value) : : "memory");
+#  elif defined(__clang__)
   asm volatile("" : "+r,m"(value) : : "memory");
-#else
+#  else
   asm volatile("" : "+m,r"(value) : : "memory");
-#endif
+#  endif
   return value;
 }
 #else