This adds driver support for building DLLs (the /LD and /LDd flags).
It basically does two things: runtime selection and passing -dll and
-implib to the linker.
llvm-svn: 190428
Apparently folks run into this (PR17097). The flag is not supported by
MSVC either, but we should parse it so we don't get confused when it occurs.
This changes the clang-cl output for "clang-cl /c /o foo.obj" from:
clang-cl.exe: error: no such file or directory: '/o'
clang-cl.exe: error: no such file or directory: 'foo.obj'
to:
clang-cl.exe: warning: argument unused during compilation: '/o bajs.obj'
llvm-svn: 190323
I don't think Clang intends to implement this functionality.
ASan should be used instead. Since /RTC is often passed by default
from MSBuild, ignore the option to avoid bloating the output.
llvm-svn: 190202
We already use .obj as extension when the user provides a stem file
name (via /Fo), but were failing in the most basic case when the file
name is based on the input file.
llvm-svn: 190071
* In C, as before, if the "warning flag" is enabled, warnings are produced by
forcing string literals to have const-qualified types (the produced warnings
are *not* -Wwrite-strings warnings). However, more recent GCCs (at least 4.4
onwards) now take -w into account here, so we now do the same.
* In C++, this flag is entirely sane: it behaves just like any other warning
flag. Stop triggering -fconst-strings here. This is a bit cleaner, but there's
no real functionality change except in the case where -Xclang -fno-const-strings
is also specified.
llvm-svn: 190006
Passing inconsistent munaligned-access / mno-unaligned-access
flags, intentionally resulted in a warning and the flag
no-unaligned-access being used.
Gcc does, at least in practice, use the last flag in such a
case. This patch updates clang behaviour accordingly; use the
last flag or base alignment behaviour on the target (which
llvm will do if no flag is explicitly passed)
Patch by Jeroen Hofstee.
llvm-svn: 189542
As Chandler pointed out, we should not be using -backend-option because this
will cause crashes for users of the tooling interface, etc. A better way to fix
this will be to provide the unrolling pass-manager flag to the loop vectorizer
directly.
Original commit message:
Disable loop vectorizer unrolling when no unrolling requested
In addition to the regular loop unrolling transformation, the loop vectorizer
can also unroll loops. If no unrolling has specifically been requested (by
-fno-unroll-loops), and the loop vectorizer will be used, then add the backend
option to (also) prevent the loop vectorizer from unrolling loops.
I confirmed with Nadav (off list) that disabling vectorizer loop unrolling when
-fno-unroll-loops is provided is the desired behavior.
llvm-svn: 189441
In addition to the regular loop unrolling transformation, the loop vectorizer
can also unroll loops. If no unrolling has specifically been requested (by
-fno-unroll-loops), and the loop vectorizer will be used, then add the backend
option to (also) prevent the loop vectorizer from unrolling loops.
I confirmed with Nadav (off list) that disabling vectorizer loop unrolling when
-fno-unroll-loops is provided is the desired behavior.
llvm-svn: 189440
This exposes the -fsanitize=address option and adds the runtime library
to the link command.
Differential Revision: http://llvm-reviews.chandlerc.com/D1526
llvm-svn: 189389
We error on -O5 and higher. While it is tempting to do the same for -O4, I
agree with Jordan Rose: we should warn for a release at least first.
llvm-svn: 189369
which add another wrinkle to the installation of the libstdc++ headers.
Add at least some basic testing of the weirdnesses of Gentoo's layout.
llvm-svn: 189212
The original idea was to implement it all on the driver, but to do that the
driver needs to know the sse level and to do that it has to know the default
features of a cpu.
Benjamin Kramer pointed out that if one day we decide to implement support for
' __attribute__ ((__target__ ("arch=core2")))', then the frontend needs to
keep its knowledge of default features of a cpu.
To avoid duplicating which part of clang handles default cpu features,
it is probably better to handle -mfpmath in the frontend.
For ARM this patch is just a small improvement. Instead of a cpu list, we
check if neon is enabled, which allows us to reject things like
-mcpu=cortex-a9 -mfpu=vfp -mfpmath=neon
For X86, since LLVM doesn't support an independent ssefp feature, we just
make sure the selected -mfpmath matches the sse level.
llvm-svn: 188939
This moves the logic for handling -mfoo -mno-foo from the driver to -cc1. It
also changes -cc1 to apply the options in order, fixing pr16943.
The handling of -mno-mmx -msse is now an explicit special case.
llvm-svn: 188817
Clang doesn't have a table mapping cl.exe to clang warnings. While some
warnings like -Wsign-compare exist in both compilers, the majority do
not correspond and should usually be ignored.
llvm-svn: 188732
AFAIK, there are no -W options for gcc-as and gcc-ld.
It caused failure to build clang with gcc-4.7 on cygwin.
FIXME: Could we recategorize Options for gcc-as and gcc-ld?
llvm-svn: 188668
Summary:
This change turns SanitizerArgs into high-level options
stored in the Driver, which are parsed lazily. This fixes an issue of multiple copies of the same diagnostic message produced by sanitizer arguments parser.
Reviewers: rsmith
Reviewed By: rsmith
CC: chandlerc, eugenis, cfe-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D1341
llvm-svn: 188660
This adds support for the /link option, which forwards
subsequent arguments to the linker.
The test for this will only work when targetting win32.
Since that's the only target where clang-cl makes sense,
use that target by default.
Differential Revision: http://llvm-reviews.chandlerc.com/D1388
llvm-svn: 188331
We used to decide whether to really vectorize depending on the optimization
level in PassManagerBuilder.
This patch moves this decision to the clang driver. We look at the optimization
level and whether the f(no-)vectorize is set and decide whether to vectorize.
This allows us to simplify the logic in PassManagerBuilder to just a check for
whether the vectorizer should run or not.
We now do the right thing for:
$ clang -O1 -fvectorize
$ clang -fno-vectorize -O3
llvm-svn: 188280
This patch adds -mmsa and -mno-msa to the options supported by
clang to enable and disable support for MSA.
When MSA is enabled, a predefined macro '__mips_msa' is defined to 1.
Patch by Daniel Sanders
llvm-svn: 188184
Various tests had sprung up over the years which had --check-prefix=ABC on the
RUN line, but "CHECK-ABC:" later on. This happened to work before, but was
strictly incorrect. FileCheck is getting stricter soon though.
Patch by Ron Ofir.
llvm-svn: 188174
This option prints information about #included files to stderr. Clang could
already do it, this patch just teaches the existing code about the /showIncludes
style and adds the flag.
Differential Revision: http://llvm-reviews.chandlerc.com/D1333
llvm-svn: 188037
This reverts commit r187991 and adjusts the comment. /Za is much more
involved, and we don't want to give anyone the impression we actually
support it.
llvm-svn: 187998
'-fno-unroll-loops'. The option to the backend is even called
'DisableUnrollLoops'. This is precisely the form that Clang *didn't*
support. We didn't recognize the flag, we didn't pass it to the CC1
layer, and even if we did we wouldn't use it. Clang only inspected the
positive form of the flag, and only did so to enable loop unrolling when
the optimization level wasn't high enough. This only occurs for an
optimization level that even has a chance of running the loop unroller
when optimizing for size.
This commit wires up the 'no' variant, and switches the code to actually
follow the standard flag pattern of using the last flag and allowing
a flag in either direction to override the default.
I think this is still wrong. I don't know why we disable the loop
unroller entirely *from Clang* when optimizing for size, as the loop
unrolling pass *already has special logic* for the case where the
function is attributed as optimized for size! We should really be
trusting that. Maybe in a follow-up patch, I don't really want to change
behavior here.
llvm-svn: 187969
These flags set some preprocessor macros and injects a dependency
on the runtime library into the object file, which later is picked up
by the linker.
This also adds a new CC1 flag for adding a dependent library.
Differential Revision: http://llvm-reviews.chandlerc.com/D1315
llvm-svn: 187945
This implements support for the /Fo option, which is used
to set the filename or output dir for object files.
Differential Revision: http://llvm-reviews.chandlerc.com/D1302
llvm-svn: 187820
These are used to specify source files, and whether to treat source
files as C or C++.
Differential Revision: http://llvm-reviews.chandlerc.com/D1290
llvm-svn: 187760
> This adds a bunch of options to clang-cl. Notably, this includes
> all the options that get passed when doing a default build of a
> command-line project with msbuild.exe in Debug and Release modes,
> and I believe all flags from Reid's original patch.
The original commit was reverted in r187640 after it broke the Mac build.
This should now be fixed, by Clang r187668, LLVM r187675, and putting
a -- before %s in the test.
llvm-svn: 187679
It broke the "phase1 - sanity" buildbot. Reverting until
we can figure out what's going on.
And Eric says it broke all current Mac builds actually.
llvm-svn: 187640
This adds a bunch of options to clang-cl. Notably, this includes
all the options that get passed when doing a default build of a
command-line project with msbuild.exe in Debug and Release modes,
and I believe all flags from Reid's original patch.
Differential Revision: http://llvm-reviews.chandlerc.com/D1264
llvm-svn: 187637
This adds a few more clang-cl options. It also exposes two core clang
options to the clang-cl mode: we need to be able to claim --driver_mode
so it doesn't show up as unused in cl mode, and we need -### for tests.
Differential Revision: http://llvm-reviews.chandlerc.com/D1232
llvm-svn: 187527
is that the command is quoted differently from the arguments. The
command has '\' and the argument has '\\'. This is made unclear because
FileCheck escapes the single matched '\' when it prints the contents of
the variable, thus fooling me into thinking it had matched '\\' as
intended. The solution is to bind the gcc_install variable in the
argument list rather than out of the command. To do so we also have to
be a bit more careful so that we don't get stray other things into the
'.*' regex.
Also, because of the argument difference, '\\\\' is the correct
formulation before crtbegin, go back to that.
llvm-svn: 187489
Clang when linking and using a GCC installation from a GCC
cross-compiler.
This was desired already by two special case platforms (Android and
Mips), and turns out to be generally (if frustratingly) true. I've added
a substantial comment to the code clarifying the underlying assumptions
of doing actual cross compiles with Clang (or GCC for that matter!) and
help avoid further confusion here.
The end result is to realize that fully general form of PR12478 cannot
be resolved while we support existing cross-compiling GCC toolchains,
and linking with them (namely, linking against their libgcc and
libstdc++ installs). GCC installs these target libraries under
a target-specific prefix but one that may not be available within the
actual sysroot in use. When linking in this world, GCC works and Clang
should as well, but caveat emptor: DSOs from this tree must be
replicated and rpath-fixed to be found at runtime within the sysroot.
I've extended the cross compile test cases to cover these issues by
pointing them at a sysroot and actually checking the library search
paths.
llvm-svn: 187466
The quotes (from r187330) didn't really help here, the trick was to disable
the test on MSYS builds. This removes those quotes, changes back the comment
to explain why /? has to be quoted specifically, and moves the REQUIRES
line to the top of the file because that's important.
llvm-svn: 187366
This test would fail in weird ways on systems with a one-letter filename
in the root directory, because the shell would helpfully expand /? to e.g. /n.
Make sure this doesn't happen by adding quotes.
llvm-svn: 187295
This establishes a new Flag in Options.td, which can be assigned to
options that should be made available in clang's cl.exe compatible
mode, and updates the Driver to make use of the flag.
(The whitespace change to CMakeLists forces the build to re-run CMake
and pick up the include dependency on the new .td file. This makes the
build work if someone moves backwards in commit history after this change.)
Differential Revision: http://llvm-reviews.chandlerc.com/D1215
llvm-svn: 187280
This patch provides basic support for powerpc64le as an LLVM target.
However, use of this target will not actually generate little-endian
code. Instead, use of the target will cause the correct little-endian
built-in defines to be generated, so that code that tests for
__LITTLE_ENDIAN__, for example, will be correctly parsed for
syntax-only testing. Code generation will otherwise be the same as
powerpc64 (big-endian), for now.
The patch leaves open the possibility of creating a little-endian
PowerPC64 back end, but there is no immediate intent to create such a
thing.
The new test case variant ensures that correct built-in defines for
little-endian code are generated.
llvm-svn: 187180
Use the same filtering for assembly arguments to -cc1as as we do for
-cc1, this allows a consistent (& more useful) diagnostic experience for
users (rather than getting an error from -cc1as (which a user shouldn't
really be thinking about) about --foo, they get an error from clang
about --foo in -Wa,)
I'm sort of surprised by the separation of -cc1as & the separate
argument handling, etc, but at least this removes a little bit of the
duplication.
llvm-svn: 187156
This is still a fairly odd test. Clang wants to run gcc for assembling. At
least with -### it only prints that instead of actually trying to run it with
-ccc-echo.
llvm-svn: 186945
and add a new option --driver-mode= to control it explicitly.
The CCCIsCXX and CCCIsCPP flags were non-overlapping, i.e. there
are currently really three modes that Clang can run in: gcc, g++
or cpp, so it makes sense to represent them as an enum.
Having a command line flag to control it helps testing.
llvm-svn: 186605
MSBuild writes response files as UTF-16 little endian with a byte order
mark. With this change, clang will be able to read them, although we
still can't parse any of their flags.
Adds a UTF-16-LE response file with a BOM for testing.
Differential Revision: http://llvm-reviews.chandlerc.com/D1137
llvm-svn: 186603
Some versions of python will expand the glob used in the test, others wont,
causing the test to fail when run with LIT_USE_INTERNAL_SHELL=1.
llvm-svn: 185653
Darwin systems currently do not support dwarf version 3 or above. When we are
ready, we can bump the default to gdwarf-4 for Darwin.
For other systems, the default is dwarf version 3, if everything goes smoothly,
we can bump the version to 4.
rdar://13591116
llvm-svn: 185483
when specifying --coverage (or related) flags.
The system for doing this was based on the old LLVM-hosted profile_rt
library, and hadn't been updated for Linux to use the new compiler-rt
library. Also, it couldn't possibly work on multiarch or biarch systems
in many cases. The whole thing now works much the same as the sanitizer
libraries that are built and used out of the compiler-rt repo.
Note that other target OSes haven't been updated because I don't know if
they're doing anything special with the installation path of profile_rt.
I suspect however that *all* of these are wrong and would encourage
maintainers of each target to take a hard look at how compiler-rt
runtime libraries are linked on their platforms.
llvm-svn: 184666
verifies that we run the assembler and linker in the correct mode, and
that we can successfully use a bi-arch variant of a GCC installation in
a generic cross compilation invocation of Clang.
llvm-svn: 184662
directory for programs used by the driver is actually the standard
behavior we want to be compatible with GCC cross compilers -- it isn't
specific to SUSE or any other distro.
Also start fleshing out testing of the different cross compilation
patterns, both with a new very bare-bones tree of cross compilers and by
extending the multilib trees. Currently, we don't correctly model doing
a cross compile using the non-triple target of a bi-arch GCC install,
but I'll add support for that (and tests) next.
llvm-svn: 184499
-gcc-toolchain foo -> --gcc-toolchain=foo
-target foo -> --target=foo
I've added legacy aliases for the original spellings. I've updated the
canonical tests to check both spellings, and switched all of the
-gcc-toolchain usages elsewhere in the test suite to use the new one.
I've updated some of the usages of -target to the new syntax, but will
finish that in a separate entirely mechanical change once I'm sure this
won't get rolled back for some reason (It touches a *huge* number of RUN
lines in the test suite unsurprisingly).
A nice result is that the three most common flags I end up using when
doing cross compiles are all now consistent: --target=, --sysroot=, and
--gcc-toolchain=.
llvm-svn: 184408
gcc's inputs are already added by the InputInfoList passed to
Action::ConstructJob.
Fixes a regression from r183989. This was manifesting when targetting
mingw as an extra input argument to gcc when assembling. It presumably
affects other situations where clang calls gcc.
Prior to r183989, forwardToGCC() was returning false because the INPUT
option defined in OptParser.td had the DriverOption flag set on it.
LLVM's Option library does not set this flag for INPUT.
Reviewers: espindola
Differential Revision: http://llvm-reviews.chandlerc.com/D999
llvm-svn: 184308
These options will add a module flag with name "Dwarf Version".
The behavior flag is currently set to Warning, so when two values disagree,
a warning will be emitted.
llvm-svn: 184276
integrated assembler then go ahead and still split the dwarf anyhow.
Add two tests, one to exercise existing behavior of not splitting
when we're just emitting assembly files and the other to test
that we split when we're not in integrated as mode.
llvm-svn: 183355
This option is used to select a dynamic loader prefix to be used
at runtime. Currently this is implemented for the Linux toolchain.
Differential Revision: http://llvm-reviews.chandlerc.com/D851
llvm-svn: 182744
If -fsanitize=leak is specified, link the program with the
LeakSanitizer runtime. Ignore this option when -fsanitize=address is specified,
because AddressSanitizer has this functionality built in.
llvm-svn: 182729
Sanitizer runtime intercepts functions from librt. Not doing this will fail
if the librt dependency is not present at program startup (ex. comes from a
dlopen()ed library).
llvm-svn: 182645
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
We've added the RS880 variant in the LLVM backend to represent an R600
GPU with no vertex cache, so we need to update the GPU mappings for
-mcpu.
llvm-svn: 181202
Hans Wennborg