Previously, we tried to avoid creating new temporary object regions if
the value to be materialized itself came from a temporary object region.
However, once we became more strict about lvalues vs. rvalues (months
ago), this optimization became dead code, because the input to this
function will always be an rvalue (i.e. a symbolic value or compound
value rather than a region, at least for structs).
This would be a nice optimization to keep, but removing it makes it
simpler to reason about temporary regions.
llvm-svn: 187160
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
Back in r140220 we removed the autoconf code that would set LLVMCC_OPTION
since it was only used by the test-suite. This patch now removes code
that would only be used if LLVMCC_OPTION was set.
llvm-svn: 187154
Recent failures on a freebsd buildbot indicated a weakness in the
Reformatting.cpp lit test. Tweaking the test to avoid false negatives and
hopefully make the buildbot happy.
llvm-svn: 187153
Also move the logic to shorten thread names from linux/Host.cpp to a new
SetShortThreadName as both FreeBSD and Linux need the functionality.
llvm-svn: 187149
- ReadLocker constructors that take a lock
- Unconditional Lock::ReadLock and ReadLocker::Lock
(all consumers use TryLock)
- Make Unlock protected, as it has no external consumers
llvm-svn: 187147
FreeBSD's Host class doesn't yet return a list of running processes,
so 'platform process list' fails and attach by process name does not
work.
llvm-svn: 187142
The previous change to local live range allocation also suppressed
eviction of local ranges. In rare cases, this could result in more
expensive register choices. This commit actually revives a feature
that I added long ago: check if live ranges can be reassigned before
eviction. But now it only happens in rare cases of evicting a local
live range because another local live range wants a cheaper register.
The benefit is improved code size for some benchmarks on x86 and armv7.
I measured no significant compile time increase and performance
changes are noise.
llvm-svn: 187140
Also avoid locals evicting locals just because they want a cheaper register.
Problem: MI Sched knows exactly how many registers we have and assumes
they can be colored. In cases where we have large blocks, usually from
unrolled loops, greedy coloring fails. This is a source of
"regressions" from the MI Scheduler on x86. I noticed this issue on
x86 where we have long chains of two-address defs in the same live
range. It's easy to see this in matrix multiplication benchmarks like
IRSmk and even the unit test misched-matmul.ll.
A fundamental difference between the LLVM register allocator and
conventional graph coloring is that in our model a live range can't
discover its neighbors, it can only verify its neighbors. That's why
we initially went for greedy coloring and added eviction to deal with
the hard cases. However, for singly defined and two-address live
ranges, we can optimally color without visiting neighbors simply by
processing the live ranges in instruction order.
Other beneficial side effects:
It is much easier to understand and debug regalloc for large blocks
when the live ranges are allocated in order. Yes, global allocation is
still very confusing, but it's nice to be able to comprehend what
happened locally.
Heuristics could be added to bias register assignment based on
instruction locality (think late register pairing, banks...).
Intuituvely this will make some test cases that are on the threshold
of register pressure more stable.
llvm-svn: 187139
PR title updated to indicate that FreeBSD is also affected: Backtrace
command does not display c++ member function names on Linux or FreeBSD
llvm-svn: 187127
The reason this clang invocation was failing is that it had two %s. We would
close stdout after the first one and report a fatal error when trying to
print the second.
llvm-svn: 187122
The tests use a plugin based on the name from sys.platform.
Unfortunately that string includes the major version number in
Python 2.7, so the tests would look for builder_freebsd9.py,
builder_freebsd10.py, etc.
The issue doesn't affect Linux as Python returns 'linux2' also
on Linux 3.x -- see http://bugs.python.org/issue12326 for details.
It seems later versions of Python will drop the major version
number, so adopt this convention now for FreeBSD.
llvm-svn: 187121
For two intrinsics 'llvm.nvvm.texsurf.handle' and 'llvm.nvvm.texsurf.handle.internal',
TableGen was emitting matching code like:
if (Name.startswith("llvm.nvvm.texsurf.handle")) ...
if (Name.startswith("llvm.nvvm.texsurf.handle.internal")) ...
We can never match "llvm.nvvm.texsurf.handle.internal" here because it will
always be erroneously matched by the first condition.
The fix is to sort the intrinsic names and emit them in reverse order.
llvm-svn: 187119
Before the patch we took advantage of the fact that the compare and
branch are glued together in the selection DAG and fused them together
(where possible) while emitting them. This seemed to work well in practice.
However, fusing the compare so early makes it harder to remove redundant
compares in cases where CC already has a suitable value. This patch
therefore uses the peephole analyzeCompare/optimizeCompareInstr pair of
functions instead.
No behavioral change intended, but it paves the way for a later patch.
llvm-svn: 187116
In TUs with large classes, a matcher like
methodDecl(ofClass(recordDecl(has(varDecl()))))
(finding all member functions of classes with static variables)
becomes unbearably slow otherwise.
llvm-svn: 187115
Bill Wendling