CVS Repository/ src/ gnu/ lib/ libcompiler_rt/ Makefile

amd64 needs the quad-precision support functions as well (for __float128
support).  From Brad (with a small comment tweak).

ok rsadowski@

Remove obsolete ppc/atomic_lock_free.c

Back when compiler-rt was missing __atomic_is_lock_free, we added it
for 64-bit atomic ops on powerpc.  A later version of compiler-rt
added __atomic_is_lock_free to builtins/atomic.c; the linker has been
finding atomic.c and ignoring atomic_lock_free.c.

ok kettenis@

Clean up the handling of the "TF" (quad-precision) helper functions.
Create a single arch-independent list for them and allow for replacement
of the generic C implementation with arch-dependent assembly code.

ok rsadowski@

Remove x86-specific stanza that adds Float16 conversion functions that
are already part of the generic list of files.

ok tb@

Hook up compiler-rt cpu_model detection on i386

Lack of support spotted by sthen@ in bulk build failures
(archivers/blosc2, archivers/unrar, etc).

ok kettenis@ tb@

aarch64 supports Float16; also add two missing files for BFloat16 support.

ok rsadowski@, miod@

update build infrastructure for compiler-rt-19.1.7

- Adds SANITIZER_COMMON_NO_REDEFINE_BUILTINS define for ubsan_minimal
- Enable Float16 and BFloat16 support
  - Includes Float16 conversion functions (extendhfsf2.c, etc.)
  - Adds BFloat16 support for x86_64 and aarch64
- Adds CPU model detection (x86.c)
- Adds 80-bit extended precision support (divxc3.c, fixunsxfdi.c, etc.)
- Removes generic cpu_model from common sources
- Adds arch-specific CPU model detection (aarch64, x86_64)

Quad-Precision Support:

- Restructures TF (128-bit float) functions into separate TF_SRCS variable
- Enables quad-precision for aarch64, mips64, riscv, and sparc64
- Adds arch-specific TF functions (addtf3.c, divtf3.c, etc.)

TODO:
- Comments out SME (Scalable Matrix Extension) support.

Special thanks to kettenis@, who put all the magic in the right place.

ok jca@ kettenis@

Add risc-v support code for clang -msave-restore

ok kettenis@

Fix RTARCH in our Makefile and lets us access riscv-specific implementations

This brings us fp_mode.c used in softfloat code, and muldi3.S used when
the target ISA doesn't have the Multiply extension.

ok kettenis@

Compile out-of-line helpers for atomic operations which can be enabled
through -moutline-atomics.  These are included by default in this updated
version of compiler-rt, we just haven't enabled them yet.  Some ports start
to make use of that option, so it makes sense to provide these helpers.

The helpers would make use of the ARMv8.1 LSE instructions, if we flagged
that the running system supports those.  As we do not yet have a mechanism
to show support for LSE, the code will always fall back to regular atomics.

Issue raised by jca@
Tested by phessler@
Input from jsg@
ok kettenis@

Build libcompiler_rt for riscv64
ok kettenis@

Don't build double-double functions since long double is the same as double
on OpenBSD.

ok gkoehler@

Remove fixtfti.c from ppc-if to unbreak macppc build.  I guess kettenis@
was right that this isn't really needed.  Now the list is the same as what
we had for the previous compiler-rt version.

fix path for testing if a .S exists

Add build infrastructure for compiler-rt.

ok kettenis@