| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
| |
See https://reviews.llvm.org/D47106 for details.
Reviewed By: probinson
Differential Revision: https://reviews.llvm.org/D47172
llvm-svn: 336844
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Lifting from Bob Wilson's notes: The hash value that we compute and
store in PGO profile data to detect out-of-date profiles does not
include enough information. This means that many significant changes to
the source will not cause compiler warnings about the profile being out
of date, and worse, we may continue to use the outdated profile data to
make bad optimization decisions. There is some tension here because
some source changes won't affect PGO and we don't want to invalidate the
profile unnecessarily.
This patch adds a new hashing scheme which is more sensitive to loop
nesting, conditions, and out-of-order control flow. Here are examples
which show snippets which get the same hash under the current scheme,
and different hashes under the new scheme:
Loop Nesting Example
--------------------
// Snippet 1
while (foo()) {
while (bar()) {}
}
// Snippet 2
while (foo()) {}
while (bar()) {}
Condition Example
-----------------
// Snippet 1
if (foo())
bar();
baz();
// Snippet 2
if (foo())
bar();
else
baz();
Out-of-order Control Flow Example
---------------------------------
// Snippet 1
while (foo()) {
if (bar()) {}
baz();
}
// Snippet 2
while (foo()) {
if (bar())
continue;
baz();
}
In each of these cases, it's useful to differentiate between the
snippets because swapping their profiles gives bad optimization hints.
The new hashing scheme considers some logical operators in an effort to
detect more changes in conditions. This isn't a perfect scheme. E.g, it
does not produce the same hash for these equivalent snippets:
// Snippet 1
bool c = !a || b;
if (d && e) {}
// Snippet 2
bool f = d && e;
bool c = !a || b;
if (f) {}
This would require an expensive data flow analysis. Short of that, the
new hashing scheme looks reasonably complete, based on a scan over the
statements we place counters on.
Profiles which use the old version of the PGO hash remain valid and can
be used without issue (there are tests in tree which check this).
rdar://17068282
Differential Revision: https://reviews.llvm.org/D39446
llvm-svn: 318229
|
|
|
|
|
|
|
|
|
| |
The root cause of the issues reported in D32406 and D34680 is that clang
instruments functions without bodies. Make it stop doing that, and also
teach it how to use old (incorrectly generated) profiles without
crashing.
llvm-svn: 306883
|
|
|
|
| |
llvm-svn: 306882
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Clang warns that a profile is out-of-date if it can't find a profile
record for any function in a TU. This warning became noisy after llvm
started allowing dead-stripping of instrumented functions.
To fix this, this patch changes the existing profile out-of-date warning
(-Wprofile-instr-out-of-date) so that it only complains about mismatched
data. Further, it introduces a new, off-by-default warning about missing
function data (-Wprofile-instr-missing).
Differential Revision: https://reviews.llvm.org/D28867
llvm-svn: 301570
|
|
|
|
|
|
|
|
|
| |
There is a bot which doesn't use '%1' as the IR name of the first
argument to a function:
http://lab.llvm.org:8011/builders/clang-with-thin-lto-ubuntu/builds/2050/steps/test-stage1-compiler/logs/stdio
llvm-svn: 296248
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
2nd attempt: the first was in r296231, but it had a use after lifetime
bug.
Clang has logic to lower certain conditional expressions directly into llvm
select instructions. However, it does not emit the correct profile counter
increment as it does this: it emits an unconditional increment of the counter
for the 'then branch', even if the value selected is from the 'else branch'
(this is PR32019).
That means, given the following snippet, we would report that "0" is selected
twice, and that "1" is never selected:
int f1(int x) {
return x ? 0 : 1;
^2 ^0
}
f1(0);
f1(1);
Fix the problem by using the instrprof_increment_step intrinsic to do the
proper increment.
llvm-svn: 296245
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
(PR32019)"
This reverts commit r296231. It causes an assertion failure on 32-bit
machines
clang: /export/users/atombot/llvm/clang-atom-d525-fedora-rel/llvm/lib/IR/Instructions.cpp:263: void llvm::CallInst::init(llvm::FunctionType*, llvm::Value*, llvm::ArrayRef<llvm::Value*>, llvm::ArrayRef<llvm::OperandBundleDefT<llvm::Value*> >, const llvm::Twine&): Assertion `(i >= FTy->getNumParams() || FTy->getParamType(i) == Args[i]->getType()) && "Calling a function with a bad signature!"' failed.
llvm::sys::PrintStackTrace(llvm::raw_ostream&) (/export/users/atombot/llvm/clang-atom-d525-fedora-rel/stage1/./bin/clang+0x1c5fbfa)
llvm::sys::RunSignalHandlers() (/export/users/atombot/llvm/clang-atom-d525-fedora-rel/stage1/./bin/clang+0x1c5dc7e)
SignalHandler(int) (/export/users/atombot/llvm/clang-atom-d525-fedora-rel/stage1/./bin/clang+0x1c5dde2)
__restore_rt (/lib64/libpthread.so.0+0x3f1d00efa0)
__GI_raise /home/glibctest/rpmbuild/BUILD/glibc-2.17-c758a686/signal/../nptl/sysdeps/unix/sysv/linux/raise.c:56:0
__GI_abort /home/glibctest/rpmbuild/BUILD/glibc-2.17-c758a686/stdlib/abort.c:92:0
__assert_fail_base /home/glibctest/rpmbuild/BUILD/glibc-2.17-c758a686/assert/assert.c:92:0
(/lib64/libc.so.6+0x3f1c82e622)
llvm::CallInst::init(llvm::FunctionType*, llvm::Value*, llvm::ArrayRef<llvm::Value*>, llvm::ArrayRef<llvm::OperandBundleDefT<llvm::Value*> >, llvm::Twine const&) (/export/users/atombot/llvm/clang-atom-d525-fedora-rel/stage1/./bin/clang+0x1804e3a)
clang::CodeGen::CodeGenPGO::emitCounterIncrement(clang::CodeGen::CGBuilderTy&, clang::Stmt const*, llvm::Value*) (/export/users/atombot/llvm/clang-atom-d525-fedora-rel/stage1/./bin/clang+0x1ec7891)
llvm-svn: 296234
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Clang has logic to lower certain conditional expressions directly into
llvm select instructions. However, it does not emit the correct profile
counter increment as it does this: it emits an unconditional increment
of the counter for the 'then branch', even if the value selected is from
the 'else branch' (this is PR32019).
That means, given the following snippet, we would report that "0" is
selected twice, and that "1" is never selected:
int f1(int x) {
return x ? 0 : 1;
^2 ^0
}
f1(0);
f1(1);
Fix the problem by using the instrprof_increment_step intrinsic to do
the proper increment.
llvm-svn: 296231
|
|
|
|
|
|
|
|
|
| |
Fix the fact that we don't assign profile counters to constructors in
classes with virtual bases, or constructors with variadic parameters.
Differential Revision: https://reviews.llvm.org/D30131
llvm-svn: 296062
|
|
|
|
|
|
|
|
| |
The cxx-structors.cpp test checks that some instrumentation doesn't
appear, but it should be more explicit about which instrumentation it
actually expects to appear.
llvm-svn: 295532
|
|
|
|
|
|
|
|
| |
The frontend can't see "__profn" profile name variables after IRGen
because llvm throws these away now. Tighten up some test cases which
checked for the non-existence of those variables.
llvm-svn: 295528
|
|
|
|
|
|
|
|
|
| |
This is a re-try of r295085: fix up some test cases that assume that
profile name variables are preserved by the instrprof pass.
This catches one additional case in test/CoverageMapping/unused_names.c.
llvm-svn: 295101
|
|
|
|
|
|
|
| |
This reverts commit r295085, because the corresponding llvm change was
reverted.
llvm-svn: 295100
|
|
|
|
|
|
|
| |
The 'profn' name variables shouldn't show up after we run the instrprof
pass, see https://reviews.llvm.org/D29921 for more details.
llvm-svn: 295085
|
|
|
|
| |
llvm-svn: 291853
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Much to my surprise, '-disable-llvm-optzns' which I thought was the
magical flag I wanted to get at the raw LLVM IR coming out of Clang
deosn't do that. It still runs some passes over the IR. I don't want
that, I really want the *raw* IR coming out of Clang and I strongly
suspect everyone else using it is in the same camp.
There is actually a flag that does what I want that I didn't know about
called '-disable-llvm-passes'. I suspect many others don't know about it
either. It both does what I want and is much simpler.
This removes the confusing version and makes that spelling of the flag
an alias for '-disable-llvm-passes'. I've also moved everything in Clang
to use the 'passes' spelling as it seems both more accurate (*all* LLVM
passes are disabled, not just optimizations) and much easier to remember
and spell correctly.
This is part of simplifying how Clang drives LLVM to make it cleaner to
wire up to the new pass manager.
Differential Revision: https://reviews.llvm.org/D28047
llvm-svn: 290392
|
|
|
|
|
|
| |
Differential Revision: http://reviews.llvm.org/D21741
llvm-svn: 287689
|
|
|
|
|
|
| |
Differential Revision: https://reviews.llvm.org/D25572
llvm-svn: 284293
|
|
|
|
|
|
| |
Differential Revision: https://reviews.llvm.org/D25539
llvm-svn: 284292
|
|
|
|
|
|
|
|
|
|
|
| |
Old: "__DATA,__llvm_prf_data"
New: "__DATA,__llvm_prf_data,regular,live_support"
This should fix the following bot failure:
http://bb.pgr.jp/builders/cmake-clang-x86_64-linux/builds/55158
llvm-svn: 283949
|
|
|
|
|
|
|
| |
This patch enables raw profile merging for this option which is the
new intended behavior.
llvm-svn: 276484
|
|
|
|
|
|
| |
See http://reviews.llvm.org/D22613, http://reviews.llvm.org/D22614
llvm-svn: 276356
|
|
|
|
|
|
|
|
|
|
|
|
| |
The flags:
Enable IR-level instrumentation -fprofile-generate or -fprofile-generate=
When applying profile data: -fprofile-use=/path/to/profdata
Patch by Jake VanAdrighem!
Differential Revision: https://reviews.llvm.org/D21823
llvm-svn: 275668
|
|
|
|
|
|
| |
Differential revision: http://reviews.llvm.org/D19184
llvm-svn: 273198
|
|
|
|
|
|
| |
Follow-on to r267262.
llvm-svn: 267263
|
|
|
|
|
|
|
| |
I ran into this when seeing what tests would break if we make a
driver-level decision about whether FEPGO or IRPGO is the default.
llvm-svn: 267262
|
|
|
|
|
|
|
|
|
|
| |
Value profiling should not profile constants and/or constant
expressions when they appear as callees in call instructions.
Constant expressions form when a direct callee has bitcasts or
inttoptr(ptrtint (callee)) nests surrounding it. Value profiling
should avoid instrumenting such cases. Mostly NFC.
llvm-svn: 265037
|
|
|
|
| |
llvm-svn: 264795
|
|
|
|
|
|
|
|
|
| |
For terminator instructions, the value profiling instrumentation
happens in a basic block other than where the value site resides.
This CR moves the instrumentation point prior to the value site.
Mostly NFC.
llvm-svn: 264783
|
|
|
|
|
|
| |
Differential Revision: http://reviews.llvm.org/D18289
llvm-svn: 264342
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch changes cc1 option for PGO profile use from
-fprofile-instr-use=<path> to -fprofile-instrument-use-path=<path>.
-fprofile-instr-use=<path> is now a driver only option.
In addition to decouple the cc1 option from the driver level option, this patch
also enables IR level profile use. cc1 option handling now reads the profile
header and sets CodeGenOpt ProfileUse (valid values are {None, Clang, LLVM}
-- this is a common enum for -fprofile-instrument={}, for the profile
instrumentation), and invoke the pipeline to enable the respective PGO use pass.
Reviewers: silvas, davidxl
Differential Revision: http://reviews.llvm.org/D17737
llvm-svn: 262515
|
|
|
|
| |
llvm-svn: 261047
|
|
|
|
|
|
|
| |
Differential Revision: http://reviews.llvm.org/D16947
llvm-svn: 260270
|
|
|
|
| |
llvm-svn: 260161
|
|
|
|
| |
llvm-svn: 260126
|
|
|
|
| |
llvm-svn: 260022
|
|
|
|
| |
llvm-svn: 260021
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch changes cc1 option -fprofile-instr-generate to an enum option
-fprofile-instrument={clang|none}. It also changes cc1 options
-fprofile-instr-generate= to -fprofile-instrument-path=.
The driver level option -fprofile-instr-generate and -fprofile-instr-generate=
remain intact. This change will pave the way to integrate new PGO
instrumentation in IR level.
Review: http://reviews.llvm.org/D16730
llvm-svn: 259811
|
|
|
|
|
|
|
| |
1. Make test case more focused and robust by focusing on what to be tested (linkage, icall) -- make it easier to validate
2. Testing linkages of data and counter variables instead of names. Counters and data are more relavant to be tested.
llvm-svn: 259067
|
|
|
|
|
|
| |
Differential Revision: http://reviews.llvm.org/D8940
llvm-svn: 258650
|
|
|
|
|
|
|
|
|
|
|
| |
NFC. These hints are only used for inlining and the inliner now uses
the same criteria to identify hot and cold callees and set appropriate
thresholds without relying on these hints. Hence this removed code is
superfluous.
Differential Revision: http://reviews.llvm.org/D15726
llvm-svn: 256793
|
|
|
|
|
|
|
|
| |
This sets the maximum entry count among all functions in the program to the module using module flags. This allows the optimizer to use this information.
Differential Revision: http://reviews.llvm.org/D15163
llvm-svn: 255918
|
|
|
|
| |
llvm-svn: 255587
|
|
|
|
|
|
|
|
|
|
| |
(test case update)
Profile symbols have long prefixes which waste space and creating pressure for linker.
This patch shortens the prefixes to minimal length without losing verbosity.
Differential Revision: http://reviews.llvm.org/D15503
llvm-svn: 255576
|
|
|
|
| |
llvm-svn: 255447
|
|
|
|
| |
llvm-svn: 255445
|
|
|
|
| |
llvm-svn: 255437
|
|
|
|
| |
llvm-svn: 255436
|
|
|
|
|
|
|
| |
(This is part-2 of the patch of r255434 --
fixing test cases, second try)
llvm-svn: 255435
|