This includes a few nice bits of refactoring (e.g splitting out the
exclusive locking code into a common utility).
Patch by Rainer Orth!
Differential Revision: https://reviews.llvm.org/D40944
llvm-svn: 320726
Summary:
This patch implements flock for Windows, needed to make gcda writing work in a multiprocessing scenario.
Fixes https://bugs.llvm.org/show_bug.cgi?id=34923.
Reviewers: zturner
Reviewed By: zturner
Subscribers: rnk, zturner, llvm-commits
Differential Revision: https://reviews.llvm.org/D38891
llvm-svn: 317705
Summary:
Fixes https://bugs.llvm.org/show_bug.cgi?id=34922.
Apparently, the mode in **fdopen** gets simply ignored and Windows only cares about the mode of the original **open**.
I have verified this both with the simple case from bug 34922 and with a full Firefox build.
Reviewers: zturner
Reviewed By: zturner
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D38984
llvm-svn: 316048
The buildbots have shown that -Wstrict-prototypes behaves differently in GCC
and Clang so we should keep it disabled until Clang follows GCC's behaviour
llvm-svn: 312246
Clang 5 supports -Wstrict-prototypes. We should use it to catch any C
declarations that declare a non-prototype function.
rdar://33705313
Differential Revision: https://reviews.llvm.org/D36669
llvm-svn: 312240
Summary: In the current implementation, the defaul number of values per site tracked by value profiler is 8, which is too small and could introduce inaccuracies to profile. Changing it to 16 will be able to gain more accurate value profiler.
Reviewers: davidxl, tejohnson
Reviewed By: tejohnson
Subscribers: sanjoy, llvm-commits
Differential Revision: https://reviews.llvm.org/D35964
llvm-svn: 309388
Users can specify the path a raw profile is written to by passing
-fprofile-instr-generate=<path>, but this functionality broke on Darwin
after __llvm_profile_filename was made weak [1], resulting in profiles
being written to "default.profraw" even when <path> is specified.
The situation is that instrumented programs provide a weak definition of
__llvm_profile_filename, which conflicts with a weak redefinition
provided by the profiling runtime.
The linker appears to pick the 'winning' definition arbitrarily: on
Darwin, it usually prefers the larger definition, which is probably why
the instrprof-override-filename.c test has been passing.
The fix is to move the runtime's definition into a separate object file
within the archive. This means that the linker won't "see" the runtime's
definition unless the user program has not provided one. I couldn't
think of a great way to test this other than to mimic the Darwin
failure: use -fprofile-instr-generate=<some-small-path>.
Testing: check-{clang,profile}, modified instrprof-override-filename.c.
[1] [Profile] deprecate __llvm_profile_override_default_filename
https://reviews.llvm.org/D22613https://reviews.llvm.org/D22614
Differential Revision: https://reviews.llvm.org/D34797
llvm-svn: 306710
Introduces a 'owner' struct to include the overridable write
method and the write context in C.
This allows easy introdution of new member API to help reduce
profile merge time in the follow up patch.
llvm-svn: 306432
And also r295364 [PGO] remove unintended debug trace. NFC
I removed the test case: it's hard to write synchronized test b/w processes
in this framework. I will revisit the test-case later.
llvm-svn: 298113
This patch adds profile run time support to profile a range of values.
This interface will be used in profiling the size of memory intrinsic calls.
Differential Revision: http://reviews.llvm.org/D28964
llvm-svn: 297895
Summary:
We found a nondeterministic behavior when doing online profile merging
for multi-process applications. The application forks a sub-process and
sub-process sets to get SIGKILL when the parent process exits,
The first process gets the lock, and dumps the profile. The second one
will mmap the file, do the merge and write out the file. Note that before
the merged write, we truncate the profile.
Depending on the timing, the child process might be terminated
abnormally when the parent exits first. If this happens:
(1) before the truncation, we will get the profile for the main process
(2) after the truncation, and before write-out the profile, we will get
0 size profile.
(3) after the merged write, we get merged profile.
This patch temporarily suspend the SIGKILL for PR_SET_PDEATHSIG
before profile-write and restore it after the write.
This patch only applies to Linux system.
Reviewers: davidxl
Reviewed By: davidxl
Subscribers: xur, llvm-commits
Differential Revision: https://reviews.llvm.org/D29954
llvm-svn: 295364
Summary: In profile data paths, we replace "%h" with the hostname of the machine the program is running on. On Windows, we used gethostname() to obtain the hostname. This requires linking with ws2_32. With this change, we instead get the hostname from GetComputerNameExW(), which does not require ws2_32.
Reviewers: rnk, vsk, amccarth
Subscribers: zturner, ruiu, hans
Differential Revision: https://reviews.llvm.org/D27178
llvm-svn: 288146
This makes __llvm_profile_set_filename() work across dylib boundaries on
Darwin.
This functionality was originally meant to work on all platforms, but
was moved to a Linux-only directory with r272404. The root cause of the
test failure on Darwin was that lprofCurFilename was not marked weak.
Each dylib maintained its own copy of the variable due to the two-level
namespace.
Tested with check-profile (on Darwin). I don't expect this to regress
other platforms.
Differential Revision: https://reviews.llvm.org/D25707
llvm-svn: 284440
This reverts commit r282294. It breaks a Linux bot:
http://lab.llvm.org:8011/builders/clang-cmake-aarch64-42vma/builds/12180
It looks like the test checks that __llvm_profile_set_filename() alters the raw
profile filename in both the dylib and the main program. Now that
lprofCurFilename is hidden, this can't work, and we get two profiles (one for
the call to "main" and one for "func").
Back this change out so that we don't affect external users.
llvm-svn: 282304
This patch builds on LLVM r279776.
In this patch I've done some cleanup and abstracted three common steps runtime components have in their CMakeLists files, and added a fourth.
The three steps I abstract are:
(1) Add a top-level target (i.e asan, msan, ...)
(2) Set the target properties for sorting files in IDE generators
(3) Make the compiler-rt target depend on the top-level target
The new step is to check if a command named "runtime_register_component" is defined, and to call it with the component name.
The runtime_register_component command is defined in llvm/runtimes/CMakeLists.txt, and presently just adds the component to a list of sub-components, which later gets used to generate target mappings.
With this patch a new workflow for runtimes builds is supported. The new workflow when building runtimes from the LLVM runtimes directory is:
> cmake [...]
> ninja runtimes-configure
> ninja asan
The "runtimes-configure" target builds all the dependencies for configuring the runtimes projects, and runs CMake on the runtimes projects. Running the runtimes CMake generates a list of targets to bind into the top-level CMake so subsequent build invocations will have access to some of Compiler-RT's targets through the top-level build.
Note: This patch does exclude some top-level targets from compiler-rt libraries because they either don't install files (sanitizer_common), or don't have a cooresponding `check` target (stats).
llvm-svn: 279863
The API is intended to be used by user to do fine
grained (per-region) control of profile dumping.
Differential Revision: http://reviews.llvm.org/D23106
llvm-svn: 278092
Summary:
This patch is a refactoring of the way cmake 'targets' are grouped.
It won't affect non-UI cmake-generators.
Clang/LLVM are using a structured way to group targets which ease
navigation through Visual Studio UI. The Compiler-RT projects
differ from the way Clang/LLVM are grouping targets.
This patch doesn't contain behavior changes.
Reviewers: kubabrecka, rnk
Subscribers: wang0109, llvm-commits, kubabrecka, chrisha
Differential Revision: http://reviews.llvm.org/D21952
llvm-svn: 275111
Here's the warnings and how they were fixed:
- InstrProfilingUtil.c(110): warning C4013: '_open_osfhandle' undefined; assuming extern returning int
Include io.h to get the prototype.
- warning C4005: 'FILE_MAP_EXECUTE': macro redefinition
Stop trying to support pre-XP versions of Windows, don't attempt to
define this macro.
- InstrProfilingWriter.c(271): warning C4221: nonstandard extension used: 'Data': cannot be initialized using address of automatic variable 'Header'
- InstrProfilingWriter.c(275): warning C4221: nonstandard extension used: 'Data': cannot be initialized using address of automatic variable 'Zeroes'
Turn this warning off. This is definitely legal in C++, all compilers
accept it, and I only have room for half of one language standard in my
brain.
- InstrProfilingValue.c(320): warning C4113: 'uint32_t (__cdecl *)()' differs in parameter lists from 'uint32_t (__cdecl *)(void)'
Fix this with an explicit (void) in the prototype.
- InstrProfilingMerge.c.obj : warning LNK4006: _VPMergeHook already defined in InstrProfilingMergeFile.c.obj; second definition ignored
Last remaining warning. This is from linking a selectany definition with
a strong definition. We need to sort out weak symbols in compiler-rt in
general, though.
llvm-svn: 273026
- lprofCurFilename was intended to have external visibility. This is
pending further discussion.
- The raw version number doesn't need to be hidden: hiding it may make
it easier to accidentally combine FE/IR profiles.
See the mailing list discussion on r272081.
llvm-svn: 272089
There are still a few external symbols visible from InstrProfData.inc.
The plan for dealing with those isn't as straightforward, so I'll try it
in a separate commit.
llvm-svn: 272081
(Part-1 merging API is in profile runtime)
This patch implements a portable file opening API
with exclusive access for the process. In-process
profile merge requires profile file update to be
atomic/fully sychronized.
llvm-svn: 271864
The max warning check was masking the "return 0" codepath.
See the thread "Warnings and compile-time failure on 458.sjeng" for more
info.
llvm-svn: 270762
After r270617 I am getting an error when building:
projects/compiler-rt/lib/profile/InstrProfilingFile.c:33:36:
error: missing field 'PidChars' initializer
[-Werror,-Wmissing-field-initializers]
lprofFilename lprofCurFilename = {0};
^
Fix the aggregate initializer.
llvm-svn: 270697
licensees actually see in the toolchain we deliver to them. This will
reduce the set of local patches we have to maintain. The triple is
not changing. (The term ORBIS is an internal code name for PS4.)
llvm-svn: 269672
This is part-3 of the effort to eliminate dependency on
libc allocator in instr profiler runtime. With this change,
the profile dumper is completely free of malloc/calloc.
Value profile instr API implementation is the only remaining
piece with calloc dependency.
llvm-svn: 269576
This reverts commit r269493 as the corresponding LLVM commit was
reverted due to lots of warnings. See the review thread for the original
LLVM commit (r269491) for details.
llvm-svn: 269550
With this change, dynamic memory allocation is only used
for testing purpose. This change is one of the many steps to
make instrument profiler dynamic allocation free.
llvm-svn: 269453
This reverts commit r268840, as it breaks Thumb2 self-hosting. There is something
unstable in the profiling for Thumb2 that needs to be sorted out before we continue
implementing these changes to the profiler. See PR27667.
llvm-svn: 268864
to function names
Summary:
Hopefully this will make it easier for the next person to figure all
this out...
Reviewers: bogner, davidxl
Subscribers: davidxl, cfe-commits
Differential Revision: http://reviews.llvm.org/D18489
llvm-svn: 264680
This is implicitly needed at least by gcc-flag-compatibility.test
The thing that needs it is the `\` preceding the "default.profraw"
appended internally by clang when doing `-fprofile-use=`.
Clang uses `\` because is uses sys::path::append which will use `\` on a
Windows host. This is wrong, but I don't think there's an easy way to
solve it (maybe just always using `/` since places that accept `\` also
tend to accept `/`, but not the other way around).
llvm-svn: 264665
Summary:
Hopefully this will make it easier for the next person to figure all
this out...
Reviewers: bogner, davidxl
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D18490
llvm-svn: 264612
The GCDA writer closed the arc file before unlocking it. This causes an
EBADF while unlocking the file, and opens us up to racy behavior.
Fixes PR26847.
llvm-svn: 262779
Summary:
Hi David, SCE folks,
What is implemented in this patch is enough for the upstream libprofile to
work for PGO with the PS4 game codebase I tested ("game7" for you SCE
folks; this is with a standalone build of compiler-rt).
The first change, which is simple, is to stub out gethostname. PS4
doesn't have a simple analog for this that doesn't bring in extra
OS libraries, so for now we do not support `%h` expansion.
This is consistent with internal B#136272.
The second change implies future work, but is a simple change at present.
PS4 does not have `getenv`, so for now we will introduce a shim.
This obviously makes it impossible for many of the tests to be run since
they require setting `LLVM_PROFILE_FILE=`.
I see two paths forward:
1. In the tests we are already wrapping execution with `%run` and so by
setting a PS4-specific expansion for `%run` we can pass the information
in another way We can adapt the getenv shim as appropriate.
We will need to experiment with this internally.
Maggie, Phillip, Filipe? Any ideas? Maybe ping me internally since we
may need to get into some PS4 vagaries. I'm thinking a fake getenv
library that uses some side channel for communication.
2. Another possibility which is more verbose is to use a separate clang
invocation with `-profile-generate=<filename>` to set the filename in
each test.
This might require redundant clang invocations though which may be
undesirable for upstream. David, thoughts?
Also, this is a fairly libprofile-specific workaround, so it e.g.
doesn't help Filipe's ASan work.
Overall, this approach sounds like a bit of a hack to me.
Small detail:
InstrProfilingPort.h seems like the natural place for the getenv shim,
but GCDAProfiling.c needs it as well. InstrProfilingUtil.h is currently
the only header common between InstrProfilingFile.c and GCDAProfiling.c.
I can move the shim to InstrProfilingPort.h and add an include to
GCDAProfiling.c as per your preference David.
Reviewers: davidxl, MaggieYi, phillip.power, filcab
Subscribers: simon.f.whittaker, slingn, probinson, llvm-commits
Differential Revision: http://reviews.llvm.org/D17676
llvm-svn: 262527
Compiler-rt miscalculates the number of entries in the __llvm_prf_data section
on i386 Darwin. This results in a number of test failures (which we started
catching after r261344).
The fix we attempted earlier is insufficient (r261683). It caused some tests to
start passing again, but that hid the fact that we drop some data entries.
This patch should fix the real problem. It fixes the way we compute DataSize by
taking into account the way the Darwin linker lays out __llvm_prf_data.
Differential Revision: http://reviews.llvm.org/D17623
llvm-svn: 261957
Fix a crash when gathering value profile data on i386 Darwin.
The Darwin linker shrinks sections containing aligned structures when
padding is not explicitly added to the end of the structure. When
iterating over these structures, be sure to not walk past the end of the
section.
No tests added, since running `ninja check-profile` on i386 Darwin is
enough to reproduce the original crash.
llvm-svn: 261683
It thinks that these functions don't match the function pointer type
that they are passed with:
GCDAProfiling.c(578) : warning C4113: 'void (__cdecl *)()' differs in parameter lists from 'void (__cdecl *)(void)'
GCDAProfiling.c(579) : warning C4113: 'void (__cdecl *)()' differs in parameter lists from 'void (__cdecl *)(void)'
GCDAProfiling.c(580) : warning C4113: 'void (__cdecl *)()' differs in parameter lists from 'void (__cdecl *)(void)'
llvm-svn: 260475
This reduces sizes of instrumented object files, final binaries,
process images, and raw profile data.
The format of the indexed profile data remain the same.
Differential Revision: http://reviews.llvm.org/D16388
llvm-svn: 260118
This patch adds support for expanding "%h" out to the machine hostname
in the LLVM_PROFILE_FILE environment variable.
Patch by Daniel Waters!
Differential Revision: http://reviews.llvm.org/D16371
llvm-svn: 259272
Summary:
This patch is provided in preparation for removing autoconf on 1/26. The proposal to remove autoconf on 1/26 was discussed on the llvm-dev thread here: http://lists.llvm.org/pipermail/llvm-dev/2016-January/093875.html
"I am the punishment of God... If [autoconf] had not committed great sins, God would not have sent a punishment like me upon [it]."
-Genghis Khan
Reviewers: chandlerc, grosbach, bob.wilson, zaks.anna, kubabrecka, samsonov, echristo
Subscribers: iains, llvm-commits
Differential Revision: http://reviews.llvm.org/D16473
llvm-svn: 258863
IR level instrumentation needs to override version with variant bits.
No change for FE instrumentation is needed. Test case is added to
detect version mismatch.
llvm-svn: 257230
Summary: This change configures Windows builds to build the complier-rt profile support library (clang_rt.profile-i386.lib). Windows API incompatibilities in the compiler-rt profile lib are also fixed.
Reviewers: davidxl, dnovillo
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D15830
llvm-svn: 256848
Extract the buffered filer writer code used by value profile
writer and turn it into common/sharable buffered fileIO
interfaces. Added a test case for the buffered file writer and
rewrite the VP dumping using the new APIs.
llvm-svn: 256604
The profile reader no longer depends on this field to be updated and point
to owning func's vp data. The VP data also no longer needs to be allocated
in a contiguous memory space.
Differential Revision: http://reviews.llvm.org/D15258
llvm-svn: 256543
Patch by: Johan Engelen
On windows, opening in text mode will result in
line ending chars to be appended leading to
profile corruption.
llvm-svn: 255684
Value profile runtime depends on libc which breaks
buffer API implemenation with current file organization.
Test case is also updated to check more symbols.
llvm-svn: 255294
InstrProfiling.h file declares profile runtime public APIs.
It has become a dumping place for many different things, which
needs cleanups. In this change, core type declarations and
portability macros are moved to a new file InstrProfilingPort.h.
llvm-svn: 255270
This allows the profile runtime to pick the right impl
for cmp&swap for a given target.
Differential Revision: http://reviews.llvm.org/D15248
llvm-svn: 255173
cmp&swap is not well supported -- the new test
case triggers some assembler error.
This is a partial fix to the general problem (lack
of atomics operation support for certain targets).
llvm-svn: 254701
With the latest refactoring and code sharing patches landed,
it is possible to unify the value profile implementation between
raw and indexed profile. This is part in prfofile runtime.
Differential Revision: http://reviews.llvm.org/D15057
llvm-svn: 254678
(patch suggested by silvas)
With this patch, the IO information is wrapped in struct
ProfDataIOVec, and interface of writerCallback takes a vector
of IOVec and a pointer to writer context pointer.
Differential Revision: http://reviews.llvm.org/D14859
llvm-svn: 253764
There seems to be a problem in system header (stdint.h) of FreeBSD
where uint8_t nor uint16_t are defined. Explicitly define the key
types as done for FreeBSD i386.
llvm-svn: 253703
Value profile enumerator change to match LLVM code
ProfData new member field name change to match LLVM code
ProfData member type change to match LLVM code
Do not use lower case for types that are internal to implementation (not exposed to APIs)
There is no functional change. This is a preparation patch to enable more code sharing
in follow up patches
Differential Revision: http://reviews.llvm.org/D14841
llvm-svn: 253700
MIPS build bots failed due to lack of 64bit atomic operations.
The fix is to disable VP for MIPS target until a better solution
is found.
llvm-svn: 253687
1. Added missing public API decl in InstrProfiling.h
2. Clang formatting fix
3. Added more comments for new VP code
4. refactor the VP allocation code to make it more readable.
llvm-svn: 253508
This change adds extends the data structures and adds in the routines
for handling runtime calls for value profiling. The profile data format
is modified and the version number is incremented.
llvm-svn: 253483
- This is to handle a corner case where profile lib is linked
in but non of the modules are instrumented (On linux, since
we avoided the overhead to emit runtime hook use functions so
this is the side effect of that size optimization).
- Added a profile runtime test case to cover all scenarios of
shared library builds.
Differential Revision: http://reviews.llvm.org/D14468
llvm-svn: 253098
Runtime code implicitly depends on the defintions and const
values defined in LLVM proper, but currently such dependency
is made implicitly by duplicating code across two dirs. As
part of the PGO cleanup effort, there will be changes to share
common sources. This is a preparation patch to enable it (NFC).
Differential Revision: http://reviews.llvm.org/D14487
llvm-svn: 252570
To make them the same as the common def in InstrProfData.inc.
This is a preparation to make the runtime code to use the
template as well. NFC
llvm-svn: 252421
This patch introduces a well defined header struct
to represent raw profile header instead of using raw array.
Previously the raw array is used in two different files and
is very error prone when header structure is re-organized.
This is a small cleanup with NFC.
llvm-svn: 250561
On Linux, the profile runtime can use __start_SECTNAME and __stop_SECTNAME
symbols defined by the linker to locate the start and end location of
a named section (with C name). This eliminates the need for instrumented
binary to call __llvm_profile_register_function during start-up time.
llvm-svn: 250200
Summary: This refactoring moves much of the Apple-specific behavior into a function in AddCompilerRT. The next cleanup patch will remove more of the if(APPLE) checks in the outlying CMakeLists.
This patch adds a bunch of new functionality to add_compiler_rt_runtime so that the target names don't need to be reconstructed outside the call. It also updates some of the call sites to exercise the new functionality, but does not update all uses fully. Subsequent patches will further update call sites and move to using the new features.
Reviewers: filcab, bogner, kubabrecka, zaks.anna, glider, samsonov
Subscribers: beanz, rengolin, llvm-commits
Differential Revision: http://reviews.llvm.org/D12292
llvm-svn: 245970
Summary: This patch consolidates add_compiler_rt_osx_static_runtime and add_compiler_rt_darwin_dynamic_runtime into a single new function add_compiler_rt_darwin_runtime.
Reviewers: filcab, samsonov, bogner
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D12106
llvm-svn: 245317
When the file is initialized, this patch checks whether the path
specifies a directory. If so, it creates the directory tree before
truncating the file.
Use default.profdata instead of pgo-data for default indexed profile name.
llvm-svn: 241824
Summary:
Use CMake's cmake_parse_arguments() instead.
It's called in a slightly different way, but supports all our use cases.
It's in CMake 2.8.8, which is our minimum supported version.
CMake 3.0 doc (roughly the same. No direct link to 2.8.8 doc):
http://www.cmake.org/cmake/help/v3.0/module/CMakeParseArguments.html?highlight=cmake_parse_arguments
Since I was already changing these calls, I changed ARCH and LIB into
ARCHS and LIBS to make it more clear that they're lists of arguments.
Reviewers: eugenis, samsonov, beanz
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D10529
llvm-svn: 240120
the profile output from the command line via -fprofile-instr-generate=<path>,
where the specified output path/file will be overridden by the
LLVM_PROFILE_FILE environment variable.
Several changes are made to the runtime to support this:
Add a new interface __llvm_profile_override_default_filename that will
set the profile output filename, but allows LLVM_PROFILE_FILE to override.
This is the interface used by the new option.
Refactor the pid-expansion done for LLVM_PROFILE_FILE into a separate
routine that can be shared by the various filename setting routines
(so that the filename from the option can also use the "%p" syntax).
Move the truncation into setFilename, and only truncate if there is a
new filename specified (to maintain support for appending to the same
profile file in the case of multiple shared objects built with profiling).
Move the handling for a NULL filename passed to __llvm_profile_set_filename and
__llvm_profile_override_default_filename into the new setFilenamePossiblyWithPid
routine. This now correctly resets the output file to default.profraw
instead of NULL.
The handling for a null LLVM_PROFILE_FILE (which should not reset) is done
by caller setFilenameFromEnvironment.
Patch by Teresa Johnson.
llvm-svn: 236055
This avoids crashing or corrupting data if multiple concurrent
processes write to the same .gcda file. This is hard to test, since
the previous behaviour was a data race that often worked out, and it
ignores errors in flock to fall back to the old racy behaviour so that
it won't degrade the behaviour on filesystems that don't support
flock.
llvm-svn: 234036
Debugging a missing profile is a bit painful right now. We can make
people's lives a bit easier by adding a knob to enable printing a
helpful error message for such failures.
llvm-svn: 226312
On Darwin, compiler_rt uses magic linker symbols to find the profile
counters in the __DATA segment. This is a reasonable method for
normal, hosted, userspace programs. However programs with custom
memory layouts, such as the kernel, will need to tell compiler_rt
explicitly where to find these sections.
Patch by Lawrence D'Anna. Thanks!
llvm-svn: 223840
This change removes libclang_rt.profile-pic-<arch>.a version of
profile runtime. Instead, it's sufficient to always build
libclang_rt.profile-<arch>.a with -fPIC, as it can be linked into
both executables and shared objects.
llvm-svn: 221952
The naming scheme we're using for counters in profile data shares a
prefix with some fixed names we use for the runtime, notably
__llvm_profile_data_begin and _end. Embarrassingly, this means a
function called begin() can't be instrumented.
This modifies the runtime names so as not to collide with the
instrumentation.
llvm-svn: 217166
Vedant Kumar