libgcc_s: 80-bit long double function are x86-only
ClosedPublic
Actions

Authored by brooks on Apr 30 2024, 6:05 PM.

Details

Reviewers

dim
arichardson

Commits

rG98c8caafffb6: libgcc_s: 80-bit long double function are x86-only

Summary

Don't try to expose them elsewhere.

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Passed

Unit

No Test Coverage

Build Status

Buildable 57479
Build 54367: arc lint + arc unit

Event Timeline

brooks created this revision.Apr 30 2024, 6:05 PM

Herald added a subscriber: imp. · View Herald TranscriptApr 30 2024, 6:05 PM

brooks requested review of this revision.Apr 30 2024, 6:05 PM

Harbormaster completed remote builds in B57469: Diff 137903.Apr 30 2024, 6:05 PM

The condition in both extendxftf2.c and trunctfxf2.c is:

#if defined(CRT_HAS_TF_MODE) && __LDBL_MANT_DIG__ == 64 && defined(__x86_64__)

In int_types.h there is:

#if defined(CRT_HAS_128BIT) && defined(CRT_HAS_F128)
#define CRT_HAS_TF_MODE
#endif

and:

#if defined(__LP64__) || defined(__wasm__) || defined(__mips64) ||             \
    defined(__SIZEOF_INT128__) || defined(_WIN64)
#define CRT_HAS_128BIT
#endif

but also:

#elif __LDBL_MANT_DIG__ == 113 ||                                              \
    (__FLT_RADIX__ == 16 && __LDBL_MANT_DIG__ == 28)
// Use long double instead of __float128 if it matches the IEEE 128-bit format
// or the IBM hexadecimal format.
#define CRT_LDBL_128BIT
#define CRT_HAS_F128

So either __LDBL_MANT_DIG__ is 113 or 28 for this to be true, and this can never occur simultaneously with the first condition... I don't really grok the tests here.

Note that the symbols *can* be compiled for PowerPC though! I think that's where they came from? @pkubaj any ideas?

@arichardson made a big change upstream that impacted the upstream conditional for extendxftf2 and trunctfxf2 prior to llvm18's release. I might want to make that change a separate commit.