This changes the handling of -fno-unsafe-fp-math to stop having that
option imply -ftrapping-math. In gcc, -fno-unsafe-math-optimizations
sets -ftrapping-math, but that dependency is based on the fact the
-ftrapping-math is enabled by default in gcc. Because clang does not
enable -ftrapping-math by default, there is no reason for
-fno-unsafe-math-optimizations to set it.
On the other hand, -funsafe-math-optimizations continues to imply
-fno-trapping-math because this option necessarily disables strict
exception semantics.
This fixes https://github.com/llvm/llvm-project/issues/87523
This change replaces most of the cc1 option checks in the driver test for fast-math option handling. These changes rely on the assumption that the order in which the driver emits floating-point options is stable.
The changes also rely on the assumption that the order of prefixes listed on the FileCheck command line is unimportant and that all prefixed checks will be combined and checked as if they were a single prefix. At the time of the change, that worked.
Pick up a few more codegen parametres from downstream. The DWARF version matchesthe GCC config.
DWARF version 4, Math Errno handling, ObjC ABI handling, and debug handling.
Reviewed By: nielx
Differential Revision: https://reviews.llvm.org/D159409
The exit code for -### is inconsistent. Unrecognized options lead to
exit code 1, as expected. However, most others errors (including invalid
option value) lead to exit code 0, differing from GCC and most utilities.
This is a longstanding quirk of -###, and we didn't fix it because many
driver tests need adjustment.
Change -### to be similar to -fdriver-only -v and exit with code 1.
This requires fixing many driver tests, but the end result gives us
stronger tests.
* Existing `RUN: %clang -### ...` tests usually don't use `CHECK-NOT: error:` or `--implicit-check-not=error:`.
If a change introduces an error, such a change usually cannot be detected.
* Many folks contributing new tests don't know `-fdriver-only -v`.
To test no driver error/warning for new tests, they can use the familiar `-### -Werror`
instead of `-fdriver-only -v -Werror`.
An incomplete list of prerequisite test improvement:
* 2f79bb10461d114783a1548201928549ace09755: add -nogpulib to some AMDGPU tests
* 9155e517e6e1cda474d0d0fa82f71696c325bc10: add --cuda-path= (test w/ and w/o /usr/local/cuda)
* 80765ede5bbcca1364c2d4ae06127011eaba6389: -mcpu=native may return either 0 or 1, depending on whether `--target=` specifies a native target
* abae53f43f0d1da8d8e421f4a628d7ec64d6e365: fix -fuse-ld=lld misuses (test w/o and w/o /usr/local/bin/ld.lld)
Reviewed By: jhuber6, yaxunl, dblaikie
Differential Revision: https://reviews.llvm.org/D156363
This reverts commit e39bf32b3bc2f0cc21d783ba789bd82553493875.
Some tests have different behaviors depent on whether certain directories/files are present on the host.
An incomplete list from https://lab.llvm.org/buildbot/#/builders/109/builds/70149
csky-toolchain.c
riscv*-toolchain.c
fuchsia.*
hip-*
ohos.c
The exit code for -### is inconsistent. Unrecognized options lead to
exit code 1, as expected. However, most others errors (including invalid
option value) lead to exit code 0, differing from GCC and most utilities.
This is a longstanding quirk of -###, and we didn't fix it because many
driver tests need adjustment.
Change -### to be similar to -fdriver-only -v and exit with code 1.
This requires fixing many driver tests, but the end result gives us
stronger tests.
* Existing `RUN: %clang -### ...` tests usually don't use `CHECK-NOT: error:` or `--implicit-check-not=error:`.
If a change introduces an error, such a change usually cannot be detected.
* Many folks contributing new tests don't know `-fdriver-only -v`.
To test no driver error/warning for new tests, they can use the familiar `-### -Werror`
instead of `-fdriver-only -v -Werror`.
Reviewed By: jhuber6, yaxunl, dblaikie
Differential Revision: https://reviews.llvm.org/D156363
`-fapprox-func` should be disabled by `-fp-model={strict|precise}`,
as well as other fast-math flags. See the last changes in
`clang/test/Driver/fp-model.c`.
Probably this route (`case options::OPT_ffp_model_EQ`) was forgot
to update in D106191 and D114564. There is no appropriate reason not
to disable the flag.
This commit also updates other regression tests, which are not directly
related to this bug, for consistency with other fast-math flags.
Differential Revision: https://reviews.llvm.org/D138109
There are currently two options that are used to tell the compiler to perform
unsafe floating-point optimizations:
'-ffast-math' and '-funsafe-math-optimizations'.
'-ffast-math' is enabled by default. It automatically enables the driver option
'-menable-unsafe-fp-math'.
Below is a table illustrating the special operations enabled automatically by
'-ffast-math', '-funsafe-math-optimizations' and '-menable-unsafe-fp-math'
respectively.
Special Operations -ffast-math -funsafe-math-optimizations -menable-unsafe-fp-math
MathErrno 0 1 1
FiniteMathOnly 1 0 0
AllowFPReassoc 1 1 1
NoSignedZero 1 1 1
AllowRecip 1 1 1
ApproxFunc 1 1 1
RoundingMath 0 0 0
UnsafeFPMath 1 0 1
FPContract fast on on
'-ffast-math' enables '-fno-math-errno', '-ffinite-math-only',
'-funsafe-math-optimzations' and sets 'FpContract' to 'fast'. The driver option
'-menable-unsafe-fp-math' enables the same special options than
'-funsafe-math-optimizations'. This is redundant.
We propose to remove the driver option '-menable-unsafe-fp-math' and use
instead, the setting of the special operations to set the function attribute
'unsafe-fp-math'. This attribute will be enabled only if those special
operations are enabled and if 'FPContract' is either 'fast' or set to the
default value.
Differential Revision: https://reviews.llvm.org/D135097
It (introduced by 556d713c70bfaf58ac18d089883f9c34c581633a) appears to be
related to the removed dragonegg project. In addition, the feature was a bit
misnamed and may lur users to unnecessarily use it.
This patch removes the -f[no-]trapping-math flags from the -cc1 command line. These flags are ignored in the command line parser and their semantics is fully handled by -ffp-exception-mode.
This patch does not remove -f[no-]trapping-math from the driver command line. The driver flags are being used and do affect compilation.
Reviewed By: dexonsmith, SjoerdMeijer
Differential Revision: https://reviews.llvm.org/D93395
Patch was reverted because https://bugs.llvm.org/show_bug.cgi?id=44048
The original patch is modified to set the strictfp IR attribute
explicitly in CodeGen instead of as a side effect of IRBuilder.
In the 2nd attempt to reapply there was a windows lit test fail, the
tests were fixed to use wildcard matching.
Differential Revision: https://reviews.llvm.org/D62731
and a follow-up NFC rearrangement as it's causing a crash on valid. Testcase is on the original review thread.
This reverts commits af57dbf12e54f3a8ff48534bf1078f4de104c1cd and e6584b2b7b2de06f1e59aac41971760cac1e1b79
Add options to control floating point behavior: trapping and
exception behavior, rounding, and control of optimizations that affect
floating point calculations. More details in UsersManual.rst.
Reviewers: rjmccall
Differential Revision: https://reviews.llvm.org/D62731
As suggested in the post-commit thread for rL331056, we should match these
clang options with the established vocabulary of the corresponding sanitizer
option. Also, the use of 'strict' is well-known for these kinds of knobs,
and we can improve the descriptive text in the docs.
So this intends to match the logic of D46135 but only change the words.
Matching LLVM commit to match this spelling of the attribute to follow shortly.
Differential Revision: https://reviews.llvm.org/D46236
llvm-svn: 331209
As discussed in the post-commit thread for:
rL330437 ( http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20180423/545906.html )
We need a way to opt-out of a float-to-int-to-float cast optimization because too much
existing code relies on the platform-specific undefined result of those casts when the
float-to-int overflows.
The LLVM changes associated with adding this function attribute are here:
rL330947
rL330950
rL330951
Also as suggested, I changed the LLVM doc to mention the specific sanitizer flag that
catches this problem:
rL330958
Differential Revision: https://reviews.llvm.org/D46135
llvm-svn: 331041
There are 2 parts to getting the -fassociative-math command-line flag translated to LLVM FMF:
1. In the driver/frontend, we accept the flag and its 'no' inverse and deal with the
interactions with other flags like -ffast-math -fno-signed-zeros -fno-trapping-math.
This was mostly already done - we just need to translate the flag as a codegen option.
The test file is complicated because there are many potential combinations of flags here.
Note that we are matching gcc's behavior that requires 'nsz' and no-trapping-math.
2. In codegen, we map the codegen option to FMF in the IR builder. This is simple code and
corresponding test.
For the motivating example from PR27372:
float foo(float a, float x) { return ((a + x) - x); }
$ ./clang -O2 27372.c -S -o - -ffast-math -fno-associative-math -emit-llvm | egrep 'fadd|fsub'
%add = fadd nnan ninf nsz arcp contract float %0, %1
%sub = fsub nnan ninf nsz arcp contract float %add, %2
So 'reassoc' is off as expected (and so is the new 'afn' but that's a different patch).
This case now works as expected end-to-end although the underlying logic is still wrong:
$ ./clang -O2 27372.c -S -o - -ffast-math -fno-associative-math | grep xmm
addss %xmm1, %xmm0
subss %xmm1, %xmm0
We're not done because the case where 'reassoc' is set is ignored by optimizer passes. Example:
$ ./clang -O2 27372.c -S -o - -fassociative-math -fno-signed-zeros -fno-trapping-math -emit-llvm | grep fadd
%add = fadd reassoc float %0, %1
$ ./clang -O2 27372.c -S -o - -fassociative-math -fno-signed-zeros -fno-trapping-math | grep xmm
addss %xmm1, %xmm0
subss %xmm1, %xmm0
Differential Revision: https://reviews.llvm.org/D39812
llvm-svn: 320920
The way -ffast-math and the various related options to tweak floating-point
handling are handled is inflexible and rather confusing. This patch restructures
things so that we go through the options adjusting our idea of what's enabled as
we go, instead of trying to figure each individual thing out by working
backwards from the end, as this makes the behaviour of each individual option
more clear.
Doing it this way also means we get gcc-compatible behaviour for when the
__FAST_MATH__ and __FINITE_MATH_ONLY__ macros are defined, as they should depend
on the final set of features that are enabled and not just on -ffast-math and
-ffinite-math-only specifically.
Differential Revision: http://reviews.llvm.org/D30582
llvm-svn: 297837
-ffast-math to CC1, but it included a wrong llvm regression tests which was
removed in r280065. Although regression test noexceptionsfpmath.c makes sure
-fno-trapping-math ends up as a function attribute, this adds a test that
explicitly checks the driver output for -fno-trapping-math.
llvm-svn: 280227
to CC1, which are translated to function attributes and can e.g. be mapped on
build attributes FP_exceptions and FP_denormal. Setting these build attributes
allows better selection of floating point libraries.
Differential Revision: https://reviews.llvm.org/D23840
llvm-svn: 280064
The driver currently accepts but ignores the -freciprocal-math flag.
This patch passes the flag through and enables 'arcp' fast-math-flag
generation in IR.
Note that this change does not actually enable the optimization for
any target. The reassociation optimization that this flag specifies
was implemented by http://reviews.llvm.org/D6334 :
http://llvm.org/viewvc/llvm-project?view=revision&revision=222510
Because the optimization is done in the backend rather than IR,
the backend must be modified to understand instruction-level
fast-math-flags or a new function-level attribute must be created.
Also note that -freciprocal-math is independent of any target-specific
usage of reciprocal estimate hardware instructions. That requires
its own flag ('-mrecip').
https://llvm.org/bugs/show_bug.cgi?id=20912
llvm-svn: 234493
These names weren't referred to anywhere in the source so don't need a written
name.
Depends on the TableGen fix for anonymous records in LLVM r197869.
llvm-svn: 197896
imply -fno-math-errno if the user passed -fno-fast-math OR -ffast-math,
regardless of in which order and regardless of the tool chain default.
I've fixed this to follow the logic:
1) If the last dominating flag is -fno-math-errno, -ffast-math, or
-Ofast, then do not use math-errno.
2) If the last dominating flag is an explicit -fmath-errno, do use
math-errno.
3) Otherwise, use the toolchain default.
This, for example, allows the flag sequence
'-ffast-math ... -fno-fast-math' with no mention of '-fmath-errno' or
'-fno-math-errno' to preserve the toolchain default. Most notably, this
should prevent users trying to disable fast-math optimizations on Darwin
and BSD platforms from simultaneously enabling (pointless) -fmath-errno.
I've enhanced the tests (after more reorganization) to cover this and
other weird permutations of flags and targets.
llvm-svn: 182203
For now -fno-math-errno is the default on BSD-derived platforms (Darwin,
DragonFlyBSD, FreeBSD, NetBSD, OpenBSD). If the default is not right for
your platform, please yell. I only verified the result with the default
compilers on Darwin and FreeBSD.
llvm-svn: 155990
Linux and other (non-Darwin) platforms and have it use -fmath-errno by
default (for better or worse).
Darwin has seen the light here and uses -fno-math-errno by default, this
patch preserves that.
If any maintainers for a non-Linux platform would also like to opt-in to
-fno-math-errno by default, I'm happy to add folks, but we're currently
getting buts and misleading comparisons with GCC due to this difference
in behavior on Linux at least.
llvm-svn: 155607
Clang driver. This involves a bunch of silly option parsing code to try
to carefully emulate GCC's options. Currently, this takes a conservative
approach, and unless all of the unsafe optimizations are enabled, none
of them are. The fine grained control doesn't seem particularly useful.
If it ever becomes useful, we can add that to LLVM first, and then
expose it here.
This also fixes a few tiny bugs in the flag management around
-fhonor-infinities and -fhonor-nans; the flags now form proper sets both
for enabling and disabling, with the last flag winning.
I've also implemented a moderately terrifying GCC feature where
a language change is also provided by the '-ffast-math' flag by defining
the __FAST_MATH__ preprocessor macro. This feature is tracked and
serialized in the frontend but it isn't used yet. A subsequent patch
will add the preprocessor macro and tests for it.
I've manually tested that codegen appears to respect this, but I've not
dug in enough to see if there is an easy way to test codegen options w/o
relying on the particulars of LLVM's optimizations.
llvm-svn: 147434
