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
Xinliang David Li