gador/nixpkgs
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

cuda_nvcc: fix host_defines.h __noinline__ conflict under clang

Browse Source 
host_defines.h redefines __noinline__ as __attribute__((noinline)).
When clang compiles CUDA code against libstdc++ >=12, which uses
__attribute__((__noinline__)), the macro causes the invalid expansion
__attribute__((__attribute__((noinline)))). Clang natively understands
__noinline__ as an attribute, so the macro is unnecessary — skip it
when clang is the compiler.
This commit is contained in:
Samuel Ainsworth
2026-03-10 20:33:31 -04:00
committed by Gaetan Lepage
parent 5cf92f5e27
commit 8e55aea6ee
1 changed files with 13 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
pkgs/development/cuda-modules/packages/cuda_nvcc.nix
+13
View File
@@ -213,6 +213,19 @@ buildRedist (finalAttrs: {
"__func__(float cospif(const float a))" \
"__func__(float cospif(const float a) throw())"
''
# Fix clang CUDA compilation: host_defines.h redefines __noinline__ as
# __attribute__((noinline)), which conflicts with libstdc++ >=12 using
# __attribute__((__noinline__)) — the macro expands to
# __attribute__((__attribute__((noinline)))) which is invalid.
# Clang natively understands __noinline__ as an attribute so the macro
# is unnecessary. Skip it when clang is the compiler.
+ lib.optionalString (cudaOlder "13.0") ''
nixLog "Patching host_defines.h to skip __noinline__ macro under clang"
substituteInPlace "''${!outputInclude:?}/include/crt/host_defines.h" \
--replace-fail \
'#if defined(__CUDACC__) || defined(__CUDA_ARCH__) || defined(__CUDA_LIBDEVICE__)' \
'#if (defined(__CUDACC__) || defined(__CUDA_ARCH__) || defined(__CUDA_LIBDEVICE__)) && !defined(__clang__)'
''
);
brokenAssertions = [