[libcxxabi] Fix the long double demangling for PowerPC

[mstorsjo]

PowerPC uses a slightly different type of floats for their 128 bit long doubles, as "double-double", with LDBL_MANT_DIG == 106 rather than LDBL_MANT_DIG == 113 for IEEE 128 bit floats.

This fixes compiling libcxxabi for PowerPC after
3b70715.

[llvmbot]

@llvm/pr-subscribers-libcxxabi

Author: Martin Storsjö (mstorsjo)

Changes

PowerPC uses a slightly different type of floats for their 128 bit long doubles, as "double-double", with LDBL_MANT_DIG == 106 rather than LDBL_MANT_DIG == 113 for IEEE 128 bit floats.

This fixes compiling libcxxabi for PowerPC after
3b70715.

---

Full diff: https://github.com/llvm/llvm-project/pull/135332.diff

1 Files Affected:

• (modified) libcxxabi/src/demangle/ItaniumDemangle.h (+1-1)

diff --git a/libcxxabi/src/demangle/ItaniumDemangle.h b/libcxxabi/src/demangle/ItaniumDemangle.h
index b9d17060c04f9..eca9ddad66f92 100644
--- a/libcxxabi/src/demangle/ItaniumDemangle.h
+++ b/libcxxabi/src/demangle/ItaniumDemangle.h
@@ -5741,7 +5741,7 @@ struct FloatData<double>
 template <>
 struct FloatData<long double>
 {
-#if __LDBL_MANT_DIG__ == 113
+#if __LDBL_MANT_DIG__ == 113 || __LDBL_MANT_DIG__ == 106
   static const size_t mangled_size = 32;
 #elif __LDBL_MANT_DIG__ == 53 || defined(_MSC_VER)
   // MSVC doesn't define __LDBL_MANT_DIG__, but it has long double equal to

[mstorsjo]

If ok, I would like to merge this one without waiting for the full CI pipeline, as this fixes a broken buildbot at https://lab.llvm.org/buildbot/#/builders/72/builds/10061.

[philnik777]

I think this is fine to land without waiting for the CI. We don't have anything in the CI that would catch this anyways.

[mstorsjo]

I think this is fine to land without waiting for the CI. We don't have anything in the CI that would catch this anyways.

Thanks! And as long as at least some CI tasks have completed successfully, we can be kinda safe that the patch isn't entirely broken (causing compilation of libcxxabi to fail). Landing it now.