Dmitri Gribenko [Sat, 15 Sep 2012 20:22:05 +0000 (20:22 +0000)]
Fix Doxygen issues: wrap code examples in \code and use \p to refer to
parameters.
llvm-svn: 163984
Dmitri Gribenko [Sat, 15 Sep 2012 20:20:27 +0000 (20:20 +0000)]
Use LLVM_DELETED_FUNCTION in place of 'DO NOT IMPLEMENT' comments.
llvm-svn: 163983
Dmitri Gribenko [Sat, 15 Sep 2012 20:16:46 +0000 (20:16 +0000)]
Make PTHManager properly noncopyable.
llvm-svn: 163982
Dmitri Gribenko [Sat, 15 Sep 2012 19:23:53 +0000 (19:23 +0000)]
Address a long-standing FIXME: remove an unused copy constructor and make
UnqualifiedId() properly noncopyable.
llvm-svn: 163981
Craig Topper [Sat, 15 Sep 2012 18:47:51 +0000 (18:47 +0000)]
Fix includes of llvm and clang files that used angle brackets.
llvm-svn: 163980
Craig Topper [Sat, 15 Sep 2012 18:45:38 +0000 (18:45 +0000)]
Fix includes of llvm files that used angle brackets.
llvm-svn: 163979
Craig Topper [Sat, 15 Sep 2012 18:41:37 +0000 (18:41 +0000)]
Fix a couple include directives that used angle brackets for llvm files.
llvm-svn: 163978
Craig Topper [Sat, 15 Sep 2012 17:09:36 +0000 (17:09 +0000)]
Use LLVM_DELETED_FUNCTION in place of 'DO NOT IMPLEMENT' comments.
llvm-svn: 163974
Craig Topper [Sat, 15 Sep 2012 17:08:51 +0000 (17:08 +0000)]
Remove unused private fields to silence -Wunused-private-field.
llvm-svn: 163973
Jakob Stoklund Olesen [Sat, 15 Sep 2012 16:32:11 +0000 (16:32 +0000)]
Don't depend on kill flags in removeCopyByCommutingDef().
Kill flags are removed more and more aggressively during the register
allocation passes, it is better to get information from LiveIntervals.
llvm-svn: 163972
Jakob Stoklund Olesen [Sat, 15 Sep 2012 16:29:49 +0000 (16:29 +0000)]
Make LiveRangeQuery work for PHIDefs as well.
If a PHI value happens to be live out from the layout predecessor of its
def block, the def slot index will be in the middle of the segment:
%vreg11 = [192r,240B:0)[352r,416B:2)[416B,496r:1) 0@192r 1@480B-phi %2@352r
A LiveRangeQuery for 480 should return NULL from valueIn() since the
PHI value is defined at the block entry, not live in to the block.
No test case, future code depends on this functionality.
llvm-svn: 163971
Craig Topper [Sat, 15 Sep 2012 16:23:52 +0000 (16:23 +0000)]
Use LLVM_DELETED_FUNCTION in place of 'DO NOT IMPLEMENT' comments.
llvm-svn: 163970
Craig Topper [Sat, 15 Sep 2012 16:22:27 +0000 (16:22 +0000)]
Use LLVM_DELETED_FUNCTION in place of 'DO NOT IMPLEMENT' comments.
llvm-svn: 163969
Benjamin Kramer [Sat, 15 Sep 2012 15:11:10 +0000 (15:11 +0000)]
Disable new sroa now that all buildbots have tested it.
What we have so far:
- Some clang test failures (these were known already)
- Perf results are mixed, some big regressions
http://llvm.org/perf/db_default/v4/nts/3844
http://llvm.org/perf/db_default/v4/nts/3845
bullet suffers a lot. matmul is interesting: slower scalar code, faster with -vectorize.
- Some dragonegg selfhost bots crash in SROA during selfhost now
http://lab.llvm.org:8011/builders/dragonegg-x86_64-linux-gcc-4.6-self-host-checks/builds/1632
http://lab.llvm.org:8011/builders/dragonegg-x86_64-linux-gcc-4.5-self-host/builds/1891
llvm-svn: 163968
Benjamin Kramer [Sat, 15 Sep 2012 12:44:27 +0000 (12:44 +0000)]
X86: Emitting x87 fsin/fcos for sinf/cosf is not safe without unsafe fp math.
This was only an issue if sse is disabled.
llvm-svn: 163967
Dmitri Gribenko [Sat, 15 Sep 2012 11:56:32 +0000 (11:56 +0000)]
Add bindings for clang_getCompletionBriefComment to cindex.py.
llvm-svn: 163966
Chandler Carruth [Sat, 15 Sep 2012 11:43:14 +0000 (11:43 +0000)]
Port the SSAUpdater-based promotion logic from the old SROA pass to the
new one, and add support for running the new pass in that mode and in
that slot of the pass manager. With this the new pass can completely
replace the old one within the pipeline.
The strategy for enabling or disabling the SSAUpdater logic is to do it
by making the requirement of the domtree analysis optional. By default,
it is required and we get the standard mem2reg approach. This is usually
the desired strategy when run in stand-alone situations. Within the
CGSCC pass manager, we disable requiring of the domtree analysis and
consequentially trigger fallback to the SSAUpdater promotion.
In theory this would allow the pass to re-use a domtree if one happened
to be available even when run in a mode that doesn't require it. In
practice, it lets us have a single pass rather than two which was
simpler for me to wrap my head around.
There is a hidden flag to force the use of the SSAUpdater code path for
the purpose of testing. The primary testing strategy is just to run the
existing tests through that path. One notable difference is that it has
custom code to handle lifetime markers, and one of the tests has been
enhanced to exercise that code.
This has survived a bootstrap and the test suite without serious
correctness issues, however my run of the test suite produced *very*
alarming performance numbers. I don't entirely understand or trust them
though, so more investigation is on-going.
To aid my understanding of the performance impact of the new SROA now
that it runs throughout the optimization pipeline, I'm enabling it by
default in this commit, and will disable it again once the LNT bots have
picked up one iteration with it. I want to get those bots (which are
much more stable) to evaluate the impact of the change before I jump to
any conclusions.
NOTE: Several Clang tests will fail because they run -O3 and check the
result's order of output. They'll go back to passing once I disable it
again.
llvm-svn: 163965
Richard Smith [Sat, 15 Sep 2012 06:09:58 +0000 (06:09 +0000)]
const _Atomic(T) is not an atomic type, so do not allow it as the type 'A' in
C11 7.17's atomic operations. GNU's __atomic_* builtins do allow const-qualified
atomics, though (!!) so don't restrict those.
llvm-svn: 163964
Ted Kremenek [Sat, 15 Sep 2012 05:51:10 +0000 (05:51 +0000)]
Revert "Utilize new build system support in Xcode 4.5 for easier interposition"
It's not clear if this is working yet on the buildbot. Reverting until we have
time to investigate.
llvm-svn: 163963
Jordan Rose [Sat, 15 Sep 2012 02:48:31 +0000 (02:48 +0000)]
-Warc-retain-cycles: warn at variable initialization as well as assignment.
Specifically, this should warn:
__block block_t a = ^{ a(); };
Furthermore, this case which previously warned now does not, since the value
of 'b' is captured before the assignment occurs:
block_t b; // not __block
b = ^{ b(); };
(This will of course warn under -Wuninitialized, as before.)
<rdar://problem/
11015883>
llvm-svn: 163962
Jason Molenda [Sat, 15 Sep 2012 01:59:02 +0000 (01:59 +0000)]
Update LLDB_DISABLE_PYTHON #ifdefs in FormatManager.cpp to get it
building on no-Python systems again.
llvm-svn: 163961
Akira Hatanaka [Sat, 15 Sep 2012 01:52:08 +0000 (01:52 +0000)]
Remove aligned/unaligned load/store fragments defined in MipsInstrInfo.td and
use load/store fragments defined in TargetSelectionDAG.td in place of them.
Unaligned loads/stores are either expanded or lowered to target-specific nodes,
so instruction selection should see only aligned load/store nodes.
No changes in functionality.
llvm-svn: 163960
Craig Topper [Sat, 15 Sep 2012 01:22:42 +0000 (01:22 +0000)]
Revert r163878 as it breaks on targets with alternate register names. Such targets do not exist in the main tree so this was not noticed.
llvm-svn: 163959
Argyrios Kyrtzidis [Sat, 15 Sep 2012 01:10:20 +0000 (01:10 +0000)]
Allow disabling PCH validation when loading an AST file by checking
an environment variable for debugging purposes.
llvm-svn: 163958
Sean Callanan [Sat, 15 Sep 2012 01:05:12 +0000 (01:05 +0000)]
More runtime work. We now successfully traverse
the dynamic and static runtime class tables to
construct our isa table. This is putting the runtime
in contact with unrealized classes, which we need
to deal with in order to get accurate information.
That's the next piece of work.
<rdar://problem/
10986023>
llvm-svn: 163957
Akira Hatanaka [Sat, 15 Sep 2012 01:02:03 +0000 (01:02 +0000)]
Handled unaligned load/stores properly in Mips16
Patch by Reed Kotler.
llvm-svn: 163956
Manman Ren [Sat, 15 Sep 2012 00:39:57 +0000 (00:39 +0000)]
PGO: preserve branch-weight metadata when simplifying two branches with a common
destination.
Updated previous implementation to fix a case not covered:
// PBI: br i1 %x, TrueDest, BB
// BI: br i1 %y, TrueDest, FalseDest
The other case was handled correctly.
// PBI: br i1 %x, BB, FalseDest
// BI: br i1 %y, TrueDest, FalseDest
Also tried to use 64-bit arithmetic instead of APInt with scale to simplify the
computation. Let me know if you have other opinions about this.
llvm-svn: 163954
Andrew Trick [Sat, 15 Sep 2012 00:20:02 +0000 (00:20 +0000)]
TableGen subtarget parser. Handle new machine model.
Collect processor resources from the subtarget defs.
llvm-svn: 163953
Andrew Trick [Sat, 15 Sep 2012 00:19:59 +0000 (00:19 +0000)]
TableGen subtarget parser. Handle new machine model.
Infer SchedClasses from variants defined by the target or subtarget.
llvm-svn: 163952
Andrew Trick [Sat, 15 Sep 2012 00:19:57 +0000 (00:19 +0000)]
TableGen subtarget parser. Handle new machine model.
Collect SchedClasses and SchedRW types from the subtarget defs.
llvm-svn: 163951
Daniel Dunbar [Fri, 14 Sep 2012 23:36:56 +0000 (23:36 +0000)]
cmake: Fix file path.
llvm-svn: 163950
Howard Hinnant [Fri, 14 Sep 2012 23:28:54 +0000 (23:28 +0000)]
Update CREDITS.TXT
llvm-svn: 163949
Daniel Dunbar [Fri, 14 Sep 2012 23:15:56 +0000 (23:15 +0000)]
formatted_raw_ostream: Fix a serious bug in tell().
- The current_pos function is supposed to return all the written bytes, not the
current position of the underlying stream.
- This caused tell() to be broken whenever the underlying stream had buffered
content.
llvm-svn: 163948
Bill Wendling [Fri, 14 Sep 2012 23:05:52 +0000 (23:05 +0000)]
Some small reorganization to get read for Attributes overhaul.
llvm-svn: 163947
Enrico Granata [Fri, 14 Sep 2012 22:41:44 +0000 (22:41 +0000)]
Fixing a potential crasher where the new C++ synthetic children can return a NULL FrontEnd and cause LLDB to crash. This patch introduces a dummy front-end which the ValueObjectSynthetic can use lacking a real FrontEnd
llvm-svn: 163946
Bill Wendling [Fri, 14 Sep 2012 22:35:49 +0000 (22:35 +0000)]
Remove comment.
llvm-svn: 163945
David Blaikie [Fri, 14 Sep 2012 22:26:11 +0000 (22:26 +0000)]
Fix up erroneous alignas usage while making this portable to GCC 4.7
Review by Chandler Carruth.
llvm-svn: 163944
Jim Ingham [Fri, 14 Sep 2012 22:16:10 +0000 (22:16 +0000)]
Remove a duplicate frame_sp local that was shadowing the one we copied the incoming SBFrame into.
<rdar://problem/
12304255>
llvm-svn: 163943
Manman Ren [Fri, 14 Sep 2012 21:53:06 +0000 (21:53 +0000)]
PGO: preserve branch-weight metadata when simplifying a switch with a single
case to a conditional branch and when removing dead cases.
llvm-svn: 163942
Daniel Dunbar [Fri, 14 Sep 2012 21:30:17 +0000 (21:30 +0000)]
compiler-rt/darwin: Install asan_osx_dynamic with the other runtime libraries.
llvm-svn: 163941
Evan Cheng [Fri, 14 Sep 2012 21:25:34 +0000 (21:25 +0000)]
Stylistic and 80-col fixes
llvm-svn: 163940
Argyrios Kyrtzidis [Fri, 14 Sep 2012 21:17:41 +0000 (21:17 +0000)]
In StringLiteral::setString make sure that we copy the number of
bytes of the buffer and not the size of the string, otherwise we
may overwrite the buffer if there is a mismatch between the size
of the string and the CharByteWidth, and assertions are disabled.
The bug where this could occur was fixed in r163931.
Related to rdar://
12069503
llvm-svn: 163939
Jim Ingham [Fri, 14 Sep 2012 21:07:14 +0000 (21:07 +0000)]
SBThread::StepOut and SBThread::StepOutOfFrame should both run all threads.
llvm-svn: 163938
Jim Ingham [Fri, 14 Sep 2012 21:04:15 +0000 (21:04 +0000)]
"thread step-out" should run all threads by default.
llvm-svn: 163937
Jim Ingham [Fri, 14 Sep 2012 20:48:14 +0000 (20:48 +0000)]
Use the frame index passed into "thread until" rather than using the selected frame.
llvm-svn: 163936
Andrew Trick [Fri, 14 Sep 2012 20:27:25 +0000 (20:27 +0000)]
comment typo
llvm-svn: 163935
Andrew Trick [Fri, 14 Sep 2012 20:26:46 +0000 (20:26 +0000)]
TargetSchedModel interface. To be implemented...
llvm-svn: 163934
Andrew Trick [Fri, 14 Sep 2012 20:26:41 +0000 (20:26 +0000)]
Define MC data tables for the new scheduling machine model.
llvm-svn: 163933
Andrew Trick [Fri, 14 Sep 2012 20:26:39 +0000 (20:26 +0000)]
whitespace
llvm-svn: 163932
Argyrios Kyrtzidis [Fri, 14 Sep 2012 20:24:53 +0000 (20:24 +0000)]
[libclang] When loading an AST file, make sure to apply the language options
to the target info.
Related to rdar://
12069503.
llvm-svn: 163931
Bill Wendling [Fri, 14 Sep 2012 20:23:31 +0000 (20:23 +0000)]
Remove unused variable.
llvm-svn: 163930
Sid Manning [Fri, 14 Sep 2012 20:04:36 +0000 (20:04 +0000)]
Fix environment path problem
llvm-svn: 163929
Alex Rosenberg [Fri, 14 Sep 2012 19:19:57 +0000 (19:19 +0000)]
Review feedback from Duncan Sands. Alphabetize includes and simplify
lit config.
llvm-svn: 163928
Ted Kremenek [Fri, 14 Sep 2012 19:06:33 +0000 (19:06 +0000)]
Utilize new build system support in Xcode 4.5 for easier interposition
of a build without needing to use ccc-analyzer.
llvm-svn: 163927
Manman Ren [Fri, 14 Sep 2012 19:05:19 +0000 (19:05 +0000)]
Try to fix the bots by detecting inconsistant branch-weight metadata.
llvm-svn: 163926
Jordan Rose [Fri, 14 Sep 2012 18:59:03 +0000 (18:59 +0000)]
scan-build: Determine the location of the clang binary before printing help.
We need the clang executable to print the list of available checkers.
llvm-svn: 163925
Jim Ingham [Fri, 14 Sep 2012 18:57:14 +0000 (18:57 +0000)]
SBThread::StepOverUntil should run all threads. It is running to breakpoints, so running one thread is likely to cause the target to stall.
llvm-svn: 163924
Bill Wendling [Fri, 14 Sep 2012 18:55:32 +0000 (18:55 +0000)]
Add support for reading the GCDA file and merging the arc information.
With the advent of the __llvm_gcov_flush function, we need to be able to merge
counts into the .gcda files in an intelligent manner. This involves reading the
file if it exists, adding the counts together, and then writing the results.
<rdar://problem/
12185886>
llvm-svn: 163923
Andrew Trick [Fri, 14 Sep 2012 18:48:46 +0000 (18:48 +0000)]
Implement getNumLDMAddresses and expose through ARMBaseInstrInfo.
llvm-svn: 163922
Andrew Trick [Fri, 14 Sep 2012 18:31:58 +0000 (18:31 +0000)]
Cortex-A9 instruction-level scheduling machine model.
This models the A9 processor at the level of instruction operands, as
opposed to the itinerary, which models each operation at the level of
pipeline stages.
The two primary motivations are:
1) Allow MachineScheduler to model A9 as an out-of-order processor. It
can now distinguish between hazards that force interlocking vs.
buffered resources.
2) Reduce long-term maintenance by allowing the itinerary and target
hooks to eventually be removed. Note that almost all of the complexity
in the new model exists to model instruction variants, which the
itinerary cannot handle. Instead the scheduler previously relied on
processor-specific target hooks which are incomplete and buggy.
llvm-svn: 163921
Richard Smith [Fri, 14 Sep 2012 18:27:01 +0000 (18:27 +0000)]
Fix some dead stores which the static analyzer warned about. No functionality
change (the problematic cases in ParseDecl.cpp are currently impossible).
llvm-svn: 163920
Richard Smith [Fri, 14 Sep 2012 18:21:10 +0000 (18:21 +0000)]
When diagnosing multiple mem-initializers in a delegating ctor, point to the delegating initializer, not to the first initializer. For good measure, also highlight the other initializer.
llvm-svn: 163919
Fariborz Jahanian [Fri, 14 Sep 2012 17:51:34 +0000 (17:51 +0000)]
Test removed.
llvm-svn: 163918
Fariborz Jahanian [Fri, 14 Sep 2012 17:48:35 +0000 (17:48 +0000)]
objective-C arc: remove -Warc-abi in its entirety.
// rdar://
10554025
llvm-svn: 163917
Manman Ren [Fri, 14 Sep 2012 17:29:56 +0000 (17:29 +0000)]
PGO: preserve branch-weight metadata when merging two switches where
the default target of the first switch is not the basic block the second switch
is in (PredDefault != BB).
llvm-svn: 163916
Andrew Trick [Fri, 14 Sep 2012 17:22:42 +0000 (17:22 +0000)]
misched: add a hook for custom DAG postprocessing.
llvm-svn: 163915
Ted Kremenek [Fri, 14 Sep 2012 17:20:56 +0000 (17:20 +0000)]
Fix copy-paste error in manpage. Should be -stdlib=library.
llvm-svn: 163914
Sean Callanan [Fri, 14 Sep 2012 17:20:18 +0000 (17:20 +0000)]
Extended the "watchpoint set variable" code to
support watchpoints on globals.
<rdar://problem/
12297238>
llvm-svn: 163913
Filipe Cabecinhas [Fri, 14 Sep 2012 17:09:15 +0000 (17:09 +0000)]
Fixed some problems with SWIG bindings.
This may (but shouldn't) break Linux (but I tested and it still worked on FreeBSD).
The same shell scripts are now used on Xcode and Makefiles, for generating
the SWIG bindings.
Some compatibility fixes were applied, too (python path, bash-isms, etc).
llvm-svn: 163912
Filipe Cabecinhas [Fri, 14 Sep 2012 17:09:06 +0000 (17:09 +0000)]
Don't force libc++ and c++11 on everyone.
Make the flags part of configure and make invocations, for now.
llvm-svn: 163911
Douglas Gregor [Fri, 14 Sep 2012 16:38:59 +0000 (16:38 +0000)]
Add missing test from r163874.
llvm-svn: 163910
Dmitri Gribenko [Fri, 14 Sep 2012 16:35:35 +0000 (16:35 +0000)]
Comment parsing: don't parse comment marker followed by a digit as a command
since no Doxygen command starts with a digit.
llvm-svn: 163909
Hemant Kulkarni [Fri, 14 Sep 2012 16:11:34 +0000 (16:11 +0000)]
ELF support for LLD writer. The writer at present emits ELF header and section
table header. Skeleton code for ReferenceKinds.
Credits:
Doxygen by Michael Spencer.
Origianl implementation from Macho by Sidney Manning.
Templatization, implementation of section header chunks, string table, ELF header by Hemant Kulkarni.
llvm-svn: 163906
Dmitri Gribenko [Fri, 14 Sep 2012 15:37:29 +0000 (15:37 +0000)]
Comment parsing: support the "\invariant" command.
llvm-svn: 163905
Micah Villmow [Fri, 14 Sep 2012 15:36:50 +0000 (15:36 +0000)]
Add in comments that explain what the indexing and the size of the arrays is about.
llvm-svn: 163904
Sergei Larin [Fri, 14 Sep 2012 15:07:59 +0000 (15:07 +0000)]
DAG post-process for Hexagon MI scheduler
This patch introduces a possibility for Hexagon MI scheduler
to perform some target specific post- processing on the scheduling
DAG prior to scheduling.
llvm-svn: 163903
Dmitri Gribenko [Fri, 14 Sep 2012 14:57:36 +0000 (14:57 +0000)]
Fix Doxygen issues:
* wrap code blocks in \code ... \endcode;
* refer to parameter names in paragraphs correctly (\arg is not what most
people want -- it starts a new paragraph);
* use \param instead of \arg to document parameters in order to be consistent
with the rest of the codebase.
llvm-svn: 163902
Benjamin Kramer [Fri, 14 Sep 2012 13:08:09 +0000 (13:08 +0000)]
SROA: Silence unused variable warnings in Release builds.
The NDEBUG hack is ugly, but I see no better solution.
llvm-svn: 163900
Benjamin Kramer [Fri, 14 Sep 2012 12:19:58 +0000 (12:19 +0000)]
Remove redundant private field.
clang warned about this being unused in Release builds.
llvm-svn: 163899
Simon Atanasyan [Fri, 14 Sep 2012 11:27:24 +0000 (11:27 +0000)]
Do not use "lib32" directory to create a library/object files
paths when target is MIPS 32-bit.
The patch reviewed by Chandler Carruth.
llvm-svn: 163898
Chandler Carruth [Fri, 14 Sep 2012 11:08:31 +0000 (11:08 +0000)]
Rework the computation of a sub-structure natural type. There were
pointless checks in here, bad asserts, and just confusing code. I've
also added a bit more to the comment to clarify what this function is
really trying to do as it was not obvious to Duncan when studying it.
Thanks to Duncan for helping me dig through the issue.
No real functionality changed here in practical cases, and certainly no
test case. This is just cleanup spotted by inspection.
llvm-svn: 163897
Chandler Carruth [Fri, 14 Sep 2012 10:30:44 +0000 (10:30 +0000)]
Rely on the recursive check for pointer types rather than adding an
explicit check before recursing. A simplification requested by Duncan
during review.
llvm-svn: 163896
Chandler Carruth [Fri, 14 Sep 2012 10:30:42 +0000 (10:30 +0000)]
Be a bit more aggressive in bailing out of this routine. Spotted by
inspection by Duncan during review. My suspicion is that we would still
have returned 0 anyways in this case, but doing it sooner is better.
llvm-svn: 163895
Chandler Carruth [Fri, 14 Sep 2012 10:30:40 +0000 (10:30 +0000)]
Add some comments clarifying that the GEP analysis for vector GEPs is
deeply suspicious and likely to go away eventually. Also fix a bogus
comment about one of the checks in the vector GEP analysis. Based on
review from Duncan.
llvm-svn: 163894
Chandler Carruth [Fri, 14 Sep 2012 10:26:38 +0000 (10:26 +0000)]
Move an instance variable to a local variable based on review by Duncan.
Originally I had anticipated needing to thread this through more bits of
the SROA pass itself, but that ended up not happening. In the end, this
is a much simpler way to manange the variable.
llvm-svn: 163893
Chandler Carruth [Fri, 14 Sep 2012 10:26:36 +0000 (10:26 +0000)]
Add a comment about debug intrinsics that I *really* don't want to
forget from Duncan's review as a FIXME.
llvm-svn: 163892
Chandler Carruth [Fri, 14 Sep 2012 10:26:34 +0000 (10:26 +0000)]
Add two asserts that Duncan thought would help ensure things don't rot
unexpectedly in the future. More fixes from his code review.
llvm-svn: 163891
Chandler Carruth [Fri, 14 Sep 2012 10:18:54 +0000 (10:18 +0000)]
Actually keep the flag default-off for now. =/ That's what I get for
being busy testing this...
llvm-svn: 163890
Chandler Carruth [Fri, 14 Sep 2012 10:18:53 +0000 (10:18 +0000)]
Remove some dead, commented out code Duncan spotted in review.
llvm-svn: 163889
Chandler Carruth [Fri, 14 Sep 2012 10:18:51 +0000 (10:18 +0000)]
Wrap the dumping and printing routines in NDEBUG and LLVM_ENABLE_DUMP macros.
llvm-svn: 163888
Chandler Carruth [Fri, 14 Sep 2012 10:18:49 +0000 (10:18 +0000)]
Lots of comment fixes and cleanups from Duncan's review.
llvm-svn: 163887
NAKAMURA Takumi [Fri, 14 Sep 2012 10:06:10 +0000 (10:06 +0000)]
SROA.cpp: Unbreak gcc, sorry!
llvm-svn: 163886
NAKAMURA Takumi [Fri, 14 Sep 2012 09:55:22 +0000 (09:55 +0000)]
SROA.cpp: Appease msvc. LLVM_ATTRIBUTE(s) should come front of "const".
llvm-svn: 163885
Chandler Carruth [Fri, 14 Sep 2012 09:30:33 +0000 (09:30 +0000)]
Speculative change to try to fix older GCC versions that can't handle
the injected class name of a dependent base class here.
llvm-svn: 163884
Chandler Carruth [Fri, 14 Sep 2012 09:22:59 +0000 (09:22 +0000)]
Introduce a new SROA implementation.
This is essentially a ground up re-think of the SROA pass in LLVM. It
was initially inspired by a few problems with the existing pass:
- It is subject to the bane of my existence in optimizations: arbitrary
thresholds.
- It is overly conservative about which constructs can be split and
promoted.
- The vector value replacement aspect is separated from the splitting
logic, missing many opportunities where splitting and vector value
formation can work together.
- The splitting is entirely based around the underlying type of the
alloca, despite this type often having little to do with the reality
of how that memory is used. This is especially prevelant with unions
and base classes where we tail-pack derived members.
- When splitting fails (often due to the thresholds), the vector value
replacement (again because it is separate) can kick in for
preposterous cases where we simply should have split the value. This
results in forming i1024 and i2048 integer "bit vectors" that
tremendously slow down subsequnet IR optimizations (due to large
APInts) and impede the backend's lowering.
The new design takes an approach that fundamentally is not susceptible
to many of these problems. It is the result of a discusison between
myself and Duncan Sands over IRC about how to premptively avoid these
types of problems and how to do SROA in a more principled way. Since
then, it has evolved and grown, but this remains an important aspect: it
fixes real world problems with the SROA process today.
First, the transform of SROA actually has little to do with replacement.
It has more to do with splitting. The goal is to take an aggregate
alloca and form a composition of scalar allocas which can replace it and
will be most suitable to the eventual replacement by scalar SSA values.
The actual replacement is performed by mem2reg (and in the future
SSAUpdater).
The splitting is divided into four phases. The first phase is an
analysis of the uses of the alloca. This phase recursively walks uses,
building up a dense datastructure representing the ranges of the
alloca's memory actually used and checking for uses which inhibit any
aspects of the transform such as the escape of a pointer.
Once we have a mapping of the ranges of the alloca used by individual
operations, we compute a partitioning of the used ranges. Some uses are
inherently splittable (such as memcpy and memset), while scalar uses are
not splittable. The goal is to build a partitioning that has the minimum
number of splits while placing each unsplittable use in its own
partition. Overlapping unsplittable uses belong to the same partition.
This is the target split of the aggregate alloca, and it maximizes the
number of scalar accesses which become accesses to their own alloca and
candidates for promotion.
Third, we re-walk the uses of the alloca and assign each specific memory
access to all the partitions touched so that we have dense use-lists for
each partition.
Finally, we build a new, smaller alloca for each partition and rewrite
each use of that partition to use the new alloca. During this phase the
pass will also work very hard to transform uses of an alloca into a form
suitable for promotion, including forming vector operations, speculating
loads throguh PHI nodes and selects, etc.
After splitting is complete, each newly refined alloca that is
a candidate for promotion to a scalar SSA value is run through mem2reg.
There are lots of reasonably detailed comments in the source code about
the design and algorithms, and I'm going to be trying to improve them in
subsequent commits to ensure this is well documented, as the new pass is
in many ways more complex than the old one.
Some of this is still a WIP, but the current state is reasonbly stable.
It has passed bootstrap, the nightly test suite, and Duncan has run it
successfully through the ACATS and DragonEgg test suites. That said, it
remains behind a default-off flag until the last few pieces are in
place, and full testing can be done.
Specific areas I'm looking at next:
- Improved comments and some code cleanup from reviews.
- SSAUpdater and enabling this pass inside the CGSCC pass manager.
- Some datastructure tuning and compile-time measurements.
- More aggressive FCA splitting and vector formation.
Many thanks to Duncan Sands for the thorough final review, as well as
Benjamin Kramer for lots of review during the process of writing this
pass, and Daniel Berlin for reviewing the data structures and algorithms
and general theory of the pass. Also, several other people on IRC, over
lunch tables, etc for lots of feedback and advice.
llvm-svn: 163883
Duncan Sands [Fri, 14 Sep 2012 09:00:11 +0000 (09:00 +0000)]
Remove silly dead store. Patch by Ettl Martin.
llvm-svn: 163882
Evgeniy Stepanov [Fri, 14 Sep 2012 08:57:50 +0000 (08:57 +0000)]
Remove some debugging code.
llvm-svn: 163881
Craig Topper [Fri, 14 Sep 2012 08:33:11 +0000 (08:33 +0000)]
Allow the second opcode info table to be 8, 16, or 32-bits as needed to represent additional fragments. This recovers some space on ATT X86 syntax and PowerPC which only need 40-bits instead of 48-bits. This also increases ARM to 64-bits to fully encode all of its operands.
llvm-svn: 163880
Kostya Serebryany [Fri, 14 Sep 2012 06:52:46 +0000 (06:52 +0000)]
[asan] fix the va_arg usage
llvm-svn: 163879
Craig Topper [Fri, 14 Sep 2012 06:37:49 +0000 (06:37 +0000)]
Reduce size of register name index tables by using uint16_t for all in tree targets. If more than 16-bits are needed for any out of tree targets, code will detect and use uint32_t instead.
llvm-svn: 163878