This allows their reuse across projects. The name of the module
is intentionally generic because we would like to move more platform
checks there.
Differential Revision: https://reviews.llvm.org/D115276
In D116472 we created conditionally defined variables for the tools to
unbreak the legacy build where they are in `llvm/tools`.
The runtimes are not tools, so that flexibility doesn't matter. Still,
it might be nice to define (unconditionally) and use the variable for
the runtimes simply to make the code a bit clearer and document what is
going on.
Also, consistently put project dirs at the beginning, not end of `CMAKE_MODULE_PATH`. This ensures they will properly shadow similarly named stuff that happens to be later on the path.
Reviewed By: mstorsjo, #libunwind, #libc, #libc_abi, ldionne
Differential Revision: https://reviews.llvm.org/D116477
Using the out-of-line LSE atomics helpers for AArch64 on FreeBSD also
requires adding support for initializing __aarch64_have_lse_atomics
correctly. On Linux this is done with getauxval(3), on FreeBSD with
elf_aux_info(3), which has a slightly different interface.
Differential Revision: https://reviews.llvm.org/D109330
Add `__c11_atomic_fetch_nand` builtin to language extensions and support `__atomic_fetch_nand` libcall in compiler-rt.
Reviewed By: theraven
Differential Revision: https://reviews.llvm.org/D112400
There's a lot of duplicated calls to find various compiler-rt libraries
from build of runtime libraries like libunwind, libc++, libc++abi and
compiler-rt. The compiler-rt helper module already implemented caching
for results avoid repeated Clang invocations.
This change moves the compiler-rt implementation into a shared location
and reuses it from other runtimes to reduce duplication and speed up
the build.
Differential Revision: https://reviews.llvm.org/D88458
There's a lot of duplicated calls to find various compiler-rt libraries
from build of runtime libraries like libunwind, libc++, libc++abi and
compiler-rt. The compiler-rt helper module already implemented caching
for results avoid repeated Clang invocations.
This change moves the compiler-rt implementation into a shared location
and reuses it from other runtimes to reduce duplication and speed up
the build.
Differential Revision: https://reviews.llvm.org/D88458
This avoids pulling in all of them if only one of them is needed
(if builtins are built without -ffunction-sections), and matches how
the similar aliases for AEABI are set up.
Differential Revision: https://reviews.llvm.org/D107815
Multiple copies of emulated TLS state means inconsistent results when
accessing the same thread-local variable from different shared objects
(https://github.com/android/ndk/issues/1551). Making `__emutls_get_address`
be a weak default visibility symbol should make the dynamic linker
ensure only a single copy gets used at runtime. This is best-effort, but
the more robust approach of putting emulated TLS into its own shared
object would (a) be a much bigger change, and (b) shared objects are
pretty heavyweight, and adding a new one to a space-constrained
environment isn't an easy sell. Given the expected rarity of direct
accesses to emulated TLS variables across different shared objects, the
best-effort approach should suffice.
Reviewed By: danalbert, rprichard
Differential Revision: https://reviews.llvm.org/D107127
On Apple platforms the builtins may be built for both arm64 and arm64e.
With Makefile generators separate targets are built using Make sub-invocations.
This causes a race when creating the symlink which may sometimes fail.
Work around this by using a custom target that the builtin targets depend on.
This causes any sub-invocations to depend on the symlinks having been created before.
Mailing list thread: https://lists.llvm.org/pipermail/llvm-dev/2021-July/151822.html
Reviewed By: thakis, steven_wu
Differential Revision: https://reviews.llvm.org/D106305
The dsb after instruction cache invalidation only needs to be executed
if any instruction cache invalidation did happen.
Without this change, if the CTR_EL0.DIC bit indicates that instruction
cache invalidation is not needed, __clear_cache would execute two dsb
instructions in a row; with the second one being unnecessary.
Differential Revision: https://reviews.llvm.org/D104371
Apparently __builtin_abort() is not supported when targetting Windows.
This should fix the following builder errors:
clang_rt.builtins-x86_64.lib(int_util.c.obj) : error LNK2019: unresolved
external symbol __builtin_abort referenced in function __compilerrt_abort_impl
When compiled with -ffreestanding, we should not assume that headers
declaring functions such as abort() are available. While the compiler may
still emit calls to those functions [1], we should not require the headers
to build compiler-rt since that can result in a cyclic dependency graph:
The compiler-rt functions might be required to build libc.so, but the libc
headers such as stdlib.h might only be available once libc has been built.
[1] From https://gcc.gnu.org/onlinedocs/gcc/Standards.html:
GCC requires the freestanding environment provide memcpy, memmove,
memset and memcmp. Finally, if __builtin_trap is used, and the target
does not implement the trap pattern, then GCC emits a call to abort.
Reviewed By: phosek
Differential Revision: https://reviews.llvm.org/D103876
cmake-3.16+ for AIX changes the default behavior of building a `SHARED` library which breaks AIX's build of libatomic, i.e., cmake-3.16+ builds `SHARED` as an archive of dynamic libraries. To fix it, we have to build `libatomic.so.1` as `MODULE` which keeps `libatomic.so.1` as an normal dynamic library.
Reviewed By: jsji
Differential Revision: https://reviews.llvm.org/D103786
When toolchain can supports all of arm, armhf and armv6m architectures compiler-rt
libraries won't compile because architecture specific flags are appended to single
BUILTIN_CFLAGS variable.
Differential revision: https://reviews.llvm.org/D103363
On AIX, we have to ship `libatomic.a` for compatibility. First, a new `clang_rt.atomic` is added. Second, use added cmake modules for AIX, we are able to build a compatible libatomic.a for AIX. The second step can't be perfectly implemented with cmake now since AIX's archive approach is kinda unique, i.e., archiving shared libraries into a static archive file.
Reviewed By: jsji
Differential Revision: https://reviews.llvm.org/D102155
While implementing support for the float128 routines on x86_64, I noticed
that __builtin_isinf() was returning true for 128-bit floating point
values that are not infinite when compiling with GCC and using the
compiler-rt implementation of the soft-float comparison functions.
After stepping through the assembly, I discovered that this was caused by
GCC assuming a sign-extended 64-bit -1 result, but our implementation
returns an enum (which then has zeroes in the upper bits) and therefore
causes the comparison with -1 to fail.
Fix this by using a CMP_RESULT typedef and add a static_assert that it
matches the GCC soft-float comparison return type when compiling with GCC
(GCC has a __libgcc_cmp_return__ mode that can be used for this purpose).
Also move the 3 copies of the same code to a shared .inc file.
Reviewed By: compnerd
Differential Revision: https://reviews.llvm.org/D98205
After a follow-up change (D98332) this header can be included the same time
as fenv.h when running the tests. To avoid enum members conflicting with
the macros/enums defined in the host fenv.h, prefix them with CRT_.
Reviewed By: peter.smith
Differential Revision: https://reviews.llvm.org/D98333
Trying to build the builtins code fails because `arm64_32_SOURCES` is
missing. Setting it to the same list used for `aarch64_SOURCES` solves
that problem and allow the builtins to compile for that architecture.
Additionally, arm64_32 is added as a possible architecture for watchos
platforms.
Reviewed By: compnerd
Differential Revision: https://reviews.llvm.org/D99690
This fixes detection when linking isn't supported (i.e. while building
builtins the first time).
Since 8368e4d54c, after setting
CMAKE_TRY_COMPILE_TARGET_TYPE to STATIC_LIBRARY, this isn't strictly
needed, but is good for correctness anyway (and in case that commit
ends up reverted).
Differential Revision: https://reviews.llvm.org/D98737
Previously, that configuration only used the generic sources, in
addition to the couple specifically chosen arm/mingw files.
Differential Revision: https://reviews.llvm.org/D98547
The existing value of 0x1000 sets the IXE bit (Inexact floating-point exception
trap enable), but we really want to be setting IXC, bit 4:
Inexact cumulative floating-point exception bit. This bit is set to 1 to
indicate that the Inexact floating-point exception has occurred since 0 was
last written to this bit.
Reviewed By: kongyi, peter.smith
Differential Revision: https://reviews.llvm.org/D98353
When building builtins, the toolchain might not yet be at a stage
when linking a test application works yet, as builtins aren't
available. Therefore set CMAKE_TRY_COMPILE_TARGET_TYPE to STATIC_LIBRARY,
to avoid failing the compiler sanity check.
Setting CMAKE_TRY_COMPILE_TARGET_TYPE to STATIC_LIBRARY has the risk
of making checks for library availability succeed falsely (e.g.
indicating that libs would be available that really aren't, as the
tests don't do any linking), but the builtins library doesn't try to
link against any external libraries (and only produces static libraries
anyway), so it should be safe here.
This avoids having to set CMAKE_C_COMPILER_WORKS when bootstrapping a
cross toolchain, when building the builtins.
Differential Revision: https://reviews.llvm.org/D91334
As reported in D93278 post-review symlinking requires privilege escalation on Windows.
Copying is functionally same, so fallback to it for systems that aren't Unix-like.
This is similar to the solution in AddLLVM.cmake.
Reviewed By: ikudrin
Differential Revision: https://reviews.llvm.org/D98111
This reverts commit 680f836c2f.
Disable the non-default-rounding-mode scalbn[f] tests when we're using
the MSVC libraries.
Differential Revision: https://reviews.llvm.org/D91841
Define inline versions of __compiler_rt_fmax* and __compiler_rt_scalbn*
rather than depend on the versions in libm. As with
__compiler_rt_logbn*, these functions are only defined for single,
double, and quad precision (binary128).
Fixes PR32279 for targets using only these FP formats (e.g. Android
on arm/arm64/x86/x86_64).
For single and double precision, on AArch64, use __builtin_fmax[f]
instead of the new inline function, because the builtin expands to the
AArch64 fmaxnm instruction.
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D91841
This fixes an apparent oversight in D91156, where the symbol was defined
without the leading underscore, then the visibility was later declared with it.
rdar://73364185
Differential revision: https://reviews.llvm.org/D95639
This fixes the implementation for architectures like CHERI with strong
pointer provenance (pointers, and thus uintptr_t, are represented as
hardware capabilities). Specifically, adding two uintptr_t's together
(as is done for `start + length` and `funcStart + landingPad`) has
ambiguous provenance, whereas using a plain integer (such as size_t) for
the offset operand does not. Also, readULEB128 is creating a plain
integer, not a pointer.
On all currently-supported architectures this should be an NFC, as
size_t and uintptr_t end up being the same underlying plain integer
type.
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D95537
FreeBSD uses -Ddouble=jagged-little-pill -Dfloat=floaty-mcfloatface to
poison uses of floating point in its standalone environment. It also
deprecates machine/limits.h in favour of sys/limits.h and does not even
provide the former on newer architectures.
This is a cleaner reimplementation of equivalent patches in FreeBSD's
vendored copy of compiler-rt.
Reviewed By: dim
Differential Revision: https://reviews.llvm.org/D95264
This function is called by the __atomic_is_lock_free() builtin if the value
cannot be resolved to true at compile time. Lack of this function is
causing the non-lockfree atomics tests in libc++ to not be run (see D91911)
This function is also added in D85044, but that review also adds support
for using lock-free atomics in more cases, whereas this is a minimal change
that just adds __atomic_is_lock_free() for the implementation of atomic.c.
Reviewed By: ldionne
Differential Revision: https://reviews.llvm.org/D92302
On Android, when the builtins are linked into a binary, they are
typically linked using -Wl,--exclude-libs so that the symbols aren't
reexported. For the NDK, compiler-rt's default behavior (build the
builtins archive with -fvisibility=hidden) is better so that builtins
are hidden even without -Wl,--exclude-libs.
Android needs the builtins with non-hidden symbols only for a special
case: for backwards compatibility with old binaries, the libc.so and
libm.so DSOs in the platform need to export some builtins for arm32 and
32-bit x86. See D56977.
Control the behavior with a new flag,
`COMPILER_RT_BUILTINS_HIDE_SYMBOLS`, that behaves similarly to the
`*_HERMETIC_STATIC_LIBRARY` in libunwind/libcxx/libcxxabi, so that
Android can build a special builtins variant for libc.so/libm.so.
Unlike the hermetic flags for other projects, this new flag is enabled
by default.
Reviewed By: compnerd, MaskRay
Differential Revision: https://reviews.llvm.org/D93431
...where the name of that variable defined in
compiler-rt/lib/builtins/cpu_model.c is decorated with a leading underscore
Differential Revision: https://reviews.llvm.org/D93390
`;` is the default comment marker, which is also used by powerpc*-*-elf target triples.
`@` is the comment marker of powerpc*-*-darwin but the Darwin support has been deleted for PowerPC (D72063).
`%%` is the statement separator used by aarch64-*-darwin (see AArch64MCAsmInfoDarwin, it uses `;` as the comment marker, which is different from most other targets)
Reviewed By: tambre
Differential Revision: https://reviews.llvm.org/D93378
aa772fc85e (D92530) has landed fixing relocations on Darwin.
3000c19df6 (D93236) has landed working around an assembly parser bug on Darwin.
Previous quick-fix d9697c2e6b (D93198) included in this commit.
Invoking the preprocessor ourselves is fragile and would require us to replicate CMake's handling of definitions, compiler flags, etc for proper compatibility.
In my toolchain builds this notably resulted in a bunch of warnings from unused flags as my CMAKE_C_FLAGS includes CPU-specific optimization options.
Notably this part was already duplicating the logic for VISIBILITY_HIDDEN define.
Instead, symlink the files and set the proper set of defines on each.
This should also be faster as we avoid invoking the compiler multiple times.
Fixes https://llvm.org/PR48494
Differential Revision: https://reviews.llvm.org/D93278
Put .cfi_startproc on a new line to avoid hitting the assembly parser bug in MasmParser::parseDirectiveCFIStartProc().
Reviewed By: tambre
Differential Revision: https://reviews.llvm.org/D93236
aa772fc85e (D92530) has landed fixing Apple builds.
Previous quick-fix d9697c2e6b (D93198) included in this commit.
Invoking the preprocessor ourselves is fragile and would require us to replicate CMake's handling of definitions, compiler flags, etc for proper compatibility.
In my toolchain builds this notably resulted in a bunch of warnings from unused flags as my CMAKE_C_FLAGS includes CPU-specific optimization options.
Notably this part was already duplicating the logic for VISIBILITY_HIDDEN define.
Instead, symlink the files and set the proper set of defines on each.
This should also be faster as we avoid invoking the compiler multiple times.
Fixes https://llvm.org/PR48494
Differential Revision: https://reviews.llvm.org/D93211
Invoking the preprocessor ourselves is fragile and would require us to replicate CMake's handling of definitions, compiler flags, etc for proper compatibility.
In my toolchain builds this notably resulted in a bunch of warnings from unused flags as my CMAKE_C_FLAGS includes CPU-specific optimization options.
Notably this part was already duplicating the logic for VISIBILITY_HIDDEN define.
Instead, symlink the files and set the proper set of defines on each.
This should also be faster as we avoid invoking the compiler multiple times.
Fixes https://llvm.org/PR48494
Reviewed By: ilinpv
Differential Revision: https://reviews.llvm.org/D93178
This patch adds both extendhftf2 and trunctfhf2 to support
conversion between half-precision and quad-precision floating-point
values. They are built iff the compiler supports _Float16.
Some notes on ARM plaforms: while fp16 is supported on all
architectures, _Float16 is supported only for 32-bit ARM, 64-bit ARM,
and SPIR (as indicated by clang/docs/LanguageExtensions.rst). Also,
fp16 is a storage format and 64-bit ARM supports floating-point
convert precision to half as base armv8-a instruction.
This patch does not change the ABI for 32-bit ARM, it will continue
to pass _Float16 as uint16.
This re-enabled revert done by https://reviews.llvm.org/rGb534beabeed3ba1777cd0ff9ce552d077e496726
Differential Revision: https://reviews.llvm.org/D92242
Support SX Aurora VE by __clear_cache() function. This modification
allows VE to run written data, e.g. clear_cache_test.c under compiler-rt
test. We still have code alignment problem in enable_execute_stack_test.c,
though.
Reviewed By: simoll
Differential Revision: https://reviews.llvm.org/D92703
This patch fixes builtins' CMakeLists.txt and their VFP tests to check
the standard macro defined in the ACLE for VFP support. It also enables
the tests to be built and run for single-precision-only targets while
builtins were built with double-precision support.
Differential revision: https://reviews.llvm.org/D92497
On AArch64 it allows use the native FP16 ABI (although libcalls are
not emitted for fptrunc/fpext lowering), while on other architectures
the expected current semantic is preserved (arm for instance).
For testing the _Float16 usage is enabled by architecture base,
currently only for arm, aarch64, and arm64.
This re-enabled revert done by https://reviews.llvm.org/rGb534beabeed3ba1777cd0ff9ce552d077e496726
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D92241
Move the two different definitions of FUNC_ALIGN out of the ELF
specific block. Add the missing CFI_END in
END_COMPILERRT_OUTLINE_FUNCTION, to go with the corresponding CFI_START
in DEFINE_COMPILERRT_OUTLINE_FUNCTION_UNMANGLED.
Differential Revision: https://reviews.llvm.org/D92549
The MSVC specific pragmas disable this warning, but the pragmas themselves
(when not guarded by any _MSC_VER ifdef) cause warnings for other targets,
e.g. when targeting mingw.
Instead silence the MSVC warnings about unused parameters by casting
the parameters to void.
Differential Revision: https://reviews.llvm.org/D91851
Revert "[compiler-rt] [builtins] Support conversion between fp16 and fp128" & dependency
Revert "[compiler-rt] [builtins] Use _Float16 on extendhfsf2, truncdfhf2 __truncsfhf2 if available"
This reverts commit 7a94829881.
This reverts commit 1fb91fcf9c.
On AArch64 it allows use the native FP16 ABI (although libcalls are
not emitted for fptrunc/fpext lowering), while on other architectures
the expected current semantic is preserved (arm for instance).
Differential Revision: https://reviews.llvm.org/D91733
This patch adds both extendhftf2 and trunctfhf2 to support
conversion between half-precision and quad-precision floating-point
values. They are enabled iff the compiler supports _Float16.
Some notes on ARM plaforms: while __fp16 is supported on all
architectures, _Float16 is supported only for 32-bit ARM, 64-bit ARM,
and SPIR (as indicated by clang/docs/LanguageExtensions.rst). Also,
__fp16 is a storage format and promoted to 'float' for argument passing
and 64-bit ARM supports floating-point convert precision to half as
base armv8-a instruction.
It means that although extendhfsf2, truncdfhf2 __truncsfhf2 will be
built for 64-bit ARM, they will be never used in practice (compiler
won't emit libcall to them). This patch does not change the ABI for
32-bit ARM, it will continue to pass _Float16 as uint16.
Differential Revision: https://reviews.llvm.org/D91732
This patch enables building compiler-rt builtins for ARM targets that
only support single-precision floating point instructions (e.g., those
with -mfpu=fpv4-sp-d16).
This fixes PR42838
Differential Revision: https://reviews.llvm.org/D90698
Adjustment to integer division in int_div_impl.inc to avoid undefined behaviour that can occur as a result of having INT_MIN as one of the parameters.
Reviewed By: sepavloff
Differential Revision: https://reviews.llvm.org/D90218
The __isPlatformVersionAtLeast routine is an implementation of `if (@available)` check
that uses the _availability_version_check API on Darwin that's supported on
macOS 10.15, iOS 13, tvOS 13 and watchOS 6.
Differential Revision: https://reviews.llvm.org/D90367
The RISC-V implementations of the `__mulsi3`, `__muldi3` builtins were
conditionally compiling the actual function definitions depending on whether
the M extension was present or not. This caused Compiler-RT testing failures
for RISC-V targets with the M extension, as when these sources were included
the `librt_has_mul*i3` features were still being defined. These `librt_has_*`
definitions are used to conditionally run the respective tests. Since the
actual functions were not being compiled-in, the generic test for `__muldi3`
would fail. This patch makes these implementations follow the normal
Compiler-RT convention of always including the definition, and conditionally
running the respective tests by using the lit conditional
`REQUIRES: librt_has_*`.
Since the `mulsi3_test.c` wasn't actually RISC-V-specific, this patch also
moves it out of the `riscv` directory. It now only depends on
`librt_has_mulsi3` to run.
Differential Revision: https://reviews.llvm.org/D86457
This patch enables support for building compiler-rt builtins for 32-bit
Power arch on AIX. For now, we leave out the specialized ppc builtin
implementations for 128-bit long double and friends since those will
need some special handling for AIX.
Reviewed By: hubert.reinterpretcast
Differential Revision: https://reviews.llvm.org/D87383
since we will be building both 32-bit and 64-bit compiler-rt builtins
from a single configuration.
Reviewed By: hubert.reinterpretcast
Differential Revision: https://reviews.llvm.org/D87113
The code currently uses __c11_atomic_is_lock_free() to detect whether an
atomic operation is natively supported. However, this can result in a
runtime function call to determine whether the given operation is lock-free
and clang generating a call to e.g. __atomic_load_8 since the branch is
not a constant zero. Since we are implementing those runtime functions, we
must avoid those calls. This patch replaces __c11_atomic_is_lock_free()
with __atomic_always_lock_free() which always results in a compile-time
constant value. This problem was found while compiling atomic.c for MIPS32
since the -Watomic-alignment warning was being triggered and objdump showed
an undefined reference to _atomic_is_lock_free.
In addition to fixing 32-bit platforms this also enables the 16-byte case
that was disabled in r153779 (185f2edd70).
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D86510
Add the implementation of __isOSVersionAtLeast for Android. Currently,
only the major version is checked against the API level of the platform
which is an integer. The API level is retrieved by reading the system
property ro.build.version.sdk (and optionally ro.build.version.codename
to see if the platform is released or not).
Patch by jiyong@google.com
Bug: 150860940
Bug: 134795810
Test: m
Reviewed By: srhines
Differential Revision: https://reviews.llvm.org/D86596
Previously we calculating the remainder by multiplying the
quotient and divisor and subtracting from the dividend.
__udivmod can calculate the remainder while calculating the
quotient. We just need to correct the sign afterward.
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D87433
No point in making __parityti2 go through 2 calls to get to
__paritysi2.
Reviewed By: MaskRay, efriedma
Differential Revision: https://reviews.llvm.org/D87218
This patch introduces denormal result support to soft-float division
implementation unified by D85031.
Reviewed By: sepavloff
Differential Revision: https://reviews.llvm.org/D85032
This patch replaces three different pre-existing implementations of
__div[sdt]f3 LibCalls with a generic one - like it is already done for
many other LibCalls.
Reviewed By: sepavloff
Differential Revision: https://reviews.llvm.org/D85031
The existing implementations are almost identical except for width of the
integer type.
Factor them out to int_mulo_impl.inc for better maintainability.
This patch is almost identical to D86277.
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D86289
The existing implementations are almost identical except for width of the
integer type.
Factor them out to int_mulv_impl.inc for better maintainability.
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D86277
Support -march=sapphirerapids for x86.
Compare with Icelake Server, it includes 14 more new features. They are
amxtile, amxint8, amxbf16, avx512bf16, avx512vp2intersect, cldemote,
enqcmd, movdir64b, movdiri, ptwrite, serialize, shstk, tsxldtrk, waitpkg.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D86503
It's full featured now and we can use it for the runtimes build instead
of relying on an external libtool, which means the CMAKE_HOST_APPLE
restriction serves no purpose either now. Restrict llvm-lipo to Darwin
targets while I'm here, since it's only needed there.
Reviewed By: phosek
Differential Revision: https://reviews.llvm.org/D86367
D80465 added an assembly implementation of muldi3 for RISC-V but it didn't
add it to the cmake `*_SOURCES` list, so the C implementation was being used
instead. This patch fixes that.
Differential Revision: https://reviews.llvm.org/D86036
Instead of ANDing with a one hot mask representing the bit to
be tested, we were ANDing with just the bit number. This tests
multiple bits none of them the correct one.
This caused skylake-avx512, cascadelake and cooperlake to all
be misdetected. Based on experiments with the Intel SDE, it seems
that all of these CPUs are being detected as being cooperlake.
This is bad since its the newest CPU of the 3.
long double is a 64-bit double-precision type on:
- MSVC (32- and 64-bit x86)
- Android (32-bit x86)
long double is a 128-bit quad-precision type on x86_64 Android.
The assembly variants of the 80-bit builtins are correct, but some of
the builtins are implemented in C and require that long double be the
80-bit type passed via an x87 register.
Reviewed By: compnerd
Differential Revision: https://reviews.llvm.org/D82153
Qiu Chaofan