[clang] [compiler-rt] [llvm] [SPARC] Properly handle CC for long double on sparc32 (PR #162226)

Rainer Orth via llvm-commits llvm-commits at lists.llvm.org
Wed Nov 26 04:44:18 PST 2025


================
@@ -1009,9 +1009,9 @@ else ()
         list(APPEND BUILTIN_CFLAGS_${arch} -fomit-frame-pointer -DCOMPILER_RT_ARMHF_TARGET)
       endif()
 
-      # For RISCV32, we must force enable int128 for compiling long
+      # For RISCV32 and 32-bit SPARC, we must force enable int128 for compiling long
       # double routines.
-      if(COMPILER_RT_ENABLE_SOFTWARE_INT128 OR "${arch}" STREQUAL "riscv32")
+      if(COMPILER_RT_ENABLE_SOFTWARE_INT128 OR "${arch}" STREQUAL "riscv32" OR ("${arch}" STREQUAL "sparc" AND NOT CMAKE_COMPILER_IS_GNUCC))
----------------
rorth wrote:

> > When compiled with gcc, how is the library supposed to link if int128 routines are not compiled in?
> 
> I suppose in that case the library will lack `long double` routines? @rorth probably know better about this.

I've no idea right now (have tried to forget all of this ;-).  Besides, I'm so busy with the upcoming GCC 16 release that I've no time left for LLVM.

What I've done in the past is do build with both `clang` and `gcc` as build compiler, 2-stage builds with `clang`, 1-stage ones with `gcc`.  And note that the `compiler-rt` tests still aren't run in runtime builds (which is the default these days), not even a target to do so manually, so I've always used
```
 -DLLVM_ENABLE_PROJECTS=compiler-rt
```
 to avoid that.

https://github.com/llvm/llvm-project/pull/162226


More information about the llvm-commits mailing list