This diff fixes the clean build of the target generate-order-file.
In llvm/tools/clang/CMakeLists.txt
add_subdirectory(utils/perf-training) should go after the block where
the value of the variable CLANG_ORDER_FILE is set - otherwise
(tested with cmake's version 3.6.2) the arguments of perf-helper.py gen-order-file
will be ill-formed (CLANG_ORDER_FILE will be empty).
Differential revision: https://reviews.llvm.org/D28153
llvm-svn: 290781
This was originally motivated by a compile time problem I've since figured out how to solve differently, but the cleanup seemed useful. We had the same logic - which essentially implemented find - in several places. By commoning them out, I can implement find and allow erase to be inlined at the call sites if profitable.
Differential Revision: https://reviews.llvm.org/D28183
llvm-svn: 290779
Summary: Update the Phabricator docs to clarify how changes are merged for contributors without commit access.
Reviewers: delcypher, aaron.ballman
Subscribers: aaron.ballman, anmol, llvm-commits
Differential Revision: https://reviews.llvm.org/D28184
llvm-svn: 290767
There was a small error in the code in the tutorial. The tutorial contains a few errors which results in code not being able to compile.
One error was described here: https://llvm.org/bugs/show_bug.cgi?id=25583 .
I found and fixed the error and one additional error.
Reviewers: aaron.ballman, malcolm.parsons
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D28180
llvm-svn: 290766
Assert that the size of the MD5 result is the same size as the signature
field being populated. Use the sizeof operator to determine the size of
the field being written rather than hardcoding it to the magic number
16. NFC.
llvm-svn: 290764
Summary:
gep 0, 0 is equivalent to bitcast. LLVM canonicalizes it
to getelementptr because it make SROA can then handle it.
Simple case like
void g(A &a) {
z(a);
if (glob)
a.foo();
}
void testG() {
A a;
g(a);
}
was not devirtualized with -fstrict-vtable-pointers because luck of
handling for gep 0 in Memory Dependence Analysis
Reviewers: dberlin, nlewycky, chandlerc
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D28126
llvm-svn: 290763
CVP doesn't care about the order of blocks visited, but by using a pre-order traversal over the graph we can a) not visit unreachable blocks and b) optimize as we go so that analysis of later blocks produce slightly more precise results.
I noticed this via inspection and don't have a concrete example which points to the issue.
llvm-svn: 290760
There were two problems with the initial fix.
1. The added tests flushed out that we misconfigured _LIBCPP_EXPLICIT with GCC.
2. Because the boolean type was a member function template it caused weird link
errors. I'm assuming due to the vague linkage rules. This time the bool type
is a non-template member function pointer. That seems to have fixed the
failing tests. Plus it will end up generating less symbols overall, since
the bool type is no longer per instantiation.
original commit message below
-----------------------------
std::basic_ios has an operator bool(). In C++11 and later
it is explicit, and only allows contextual implicit conversions.
However explicit isn't available in C++03 which causes std::istream (et al)
to have an implicit conversion to int. This can easily cause ambiguities
when calling operator<< and operator>>.
This patch uses a "bool-like" type in C++03 to work around this. The
"bool-like" type is an arbitrary pointer to member function type. It
will not convert to either int or void*, but will convert to bool.
llvm-svn: 290754
std::basic_ios has an operator bool(). In C++11 and later
it is explicit, and only allows contextual implicit conversions.
However explicit isn't available in C++03 which causes std::istream (et al)
to have an implicit conversion to int. This can easily cause ambiguities
when calling operator<< and operator>>.
This patch uses a "bool-like" type in C++03 to work around this. The
"bool-like" type is an arbitrary pointer to member function type. It
will not convert to either int or void*, but will convert to bool.
llvm-svn: 290750
The bug was introduced in r289619.
Reviewers: Mehdi Amini
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D28134
llvm-svn: 290749
Back in r240527 I added a knob to prevent thread-unsafe functions from
being exposed. mblen(), mbtowc() and wctomb() were also added to this
list, as the latest issue of POSIX doesn't require these functions to be
thread-safe.
It turns out that the only circumstance in which these functions are not
thread-safe is in case they are used in combination with state-dependent
character sets (e.g., Shift-JIS). According to Austin Group Bug 708,
these character sets "[...] are mostly a relic of the past and which
were never supported on most POSIX systems".
Though in many cases the use of these functions can be prevented by
using the reentrant counterparts, they are the only functions that allow
you to query whether the locale's character set is state-dependent. This
means that omitting these functions removes actual functionality.
Let's be a bit less pedantic and drop the guards around these functions.
Links:
http://austingroupbugs.net/view.php?id=708http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2037.htm
Reviewed by: ericwf
Differential Revision: https://reviews.llvm.org/D21436
llvm-svn: 290748
I have a created a new check for clang tidy: misc-string-compare. This will check for incorrect usage of std::string::compare when used to check equality or inequality of string instead of the string equality or inequality operators.
Example:
```
std::string str1, str2;
if (str1.compare(str2)) {
}
```
Reviewers: hokein, aaron.ballman, alexfh, malcolm.parsons
Subscribers: xazax.hun, Eugene.Zelenko, cfe-commits, malcolm.parsons, Prazek, mgorny, JDevlieghere
Differential Revision: https://reviews.llvm.org/D27210
llvm-svn: 290747
I remove one extra line, but because annoyingly llvm-lit does not
clean the output directory before running the test, it didn't fail
locally (the file was present from a previous run).
llvm-svn: 290740
This is similar to the allocfn case - if an alloca is not captured, then it's
necessarily thread-local.
Differential Revision: https://reviews.llvm.org/D28170
llvm-svn: 290738
Summary:
The current loop complete unroll algorithm checks if unrolling complete will reduce the runtime by a certain percentage. If yes, it will apply a fixed boosting factor to the threshold (by discounting cost). The problem for this approach is that the threshold abruptly. This patch makes the boosting factor a function of runtime reduction percentage, capped by a fixed threshold. In this way, the threshold changes continuously.
The patch also simplified the code by reducing one parameter in UP.
The patch only affects code-gen of two speccpu2006 benchmark:
445.gobmk binary size decreases 0.08%, no performance change.
464.h264ref binary size increases 0.24%, no performance change.
Reviewers: mzolotukhin, chandlerc
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D26989
llvm-svn: 290737
Some incoming changes in ThinLTO will break this test.
Instead of relying on the heuristic to import, we
force the importing to happen with llvm-link.
llvm-svn: 290736
"Changed" doesn't actually change within the loop, so there's
no reason to keep track of it - we always return false during
analysis and true after the transformation is made.
llvm-svn: 290735
We correctly canonicalized (add (sext x), (sext y)) to (sext (add x, y))
where possible. However, we didn't perform the same canonicalization
for zexts or for muls.
llvm-svn: 290733
Alexander Shaposhnikov