oota-llvm.git
14 years agoUse MOVLHPS and MOVHLPS x86 nodes whenever possible. Also remove some useless nodes
Bruno Cardoso Lopes [Tue, 31 Aug 2010 21:15:21 +0000 (21:15 +0000)]
Use MOVLHPS and MOVHLPS x86 nodes whenever possible. Also remove some useless nodes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112642 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRevert r112432. It appears to be exposing a problem in the emacs build.
Dan Gohman [Tue, 31 Aug 2010 20:58:44 +0000 (20:58 +0000)]
Revert r112432. It appears to be exposing a problem in the emacs build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112638 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMerge 2010-08-31-InfiniteRecursion.ll into crash.ll.
Owen Anderson [Tue, 31 Aug 2010 20:27:17 +0000 (20:27 +0000)]
Merge 2010-08-31-InfiniteRecursion.ll into crash.ll.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112635 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMore cleanups of my JumpThreading transforms, including extracting some duplicated...
Owen Anderson [Tue, 31 Aug 2010 20:26:04 +0000 (20:26 +0000)]
More cleanups of my JumpThreading transforms, including extracting some duplicated code into a helper function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112634 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoIgnore unallocatable registers in RegAllocFast.
Jakob Stoklund Olesen [Tue, 31 Aug 2010 19:54:25 +0000 (19:54 +0000)]
Ignore unallocatable registers in RegAllocFast.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112632 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRevert r112623. It is causing self host build failures.
Devang Patel [Tue, 31 Aug 2010 19:41:03 +0000 (19:41 +0000)]
Revert r112623. It is causing self host build failures.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112631 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoUpdate the Ada instructions to LLVM 2.7 (from LLVM 2.5).
Duncan Sands [Tue, 31 Aug 2010 19:40:21 +0000 (19:40 +0000)]
Update the Ada instructions to LLVM 2.7 (from LLVM 2.5).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112630 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd an RAII helper to make cleanup of the RecursionSet more fool-proof.
Owen Anderson [Tue, 31 Aug 2010 19:24:27 +0000 (19:24 +0000)]
Add an RAII helper to make cleanup of the RecursionSet more fool-proof.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112628 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoOnly try to clean up the current block if we changed that block already.
Owen Anderson [Tue, 31 Aug 2010 18:55:52 +0000 (18:55 +0000)]
Only try to clean up the current block if we changed that block already.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112625 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSP relative offsets need to be adjusted by the local allocation size when
Jim Grosbach [Tue, 31 Aug 2010 18:52:31 +0000 (18:52 +0000)]
SP relative offsets need to be adjusted by the local allocation size when
determining if they're likely to be in range of the SP when resolving
frame references.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112624 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemember byval argument's frame index during argument lowering and use this info...
Devang Patel [Tue, 31 Aug 2010 18:50:09 +0000 (18:50 +0000)]
Remember byval argument's frame index during argument lowering and use this info to emit debug info.
Fixes Radar 8367011.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112623 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agothis assert should just be a condition, since this function is just asking if
Jim Grosbach [Tue, 31 Aug 2010 18:49:31 +0000 (18:49 +0000)]
this assert should just be a condition, since this function is just asking if
the offset is legally encodable, not actually trying to do the encoding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112622 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd a test for the duplicated-conditional situation illutrated by PR5652.
Owen Anderson [Tue, 31 Aug 2010 18:49:12 +0000 (18:49 +0000)]
Add a test for the duplicated-conditional situation illutrated by PR5652.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112621 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRefactor my fix for PR5652 to terminate the predecessor lookups after the first failure.
Owen Anderson [Tue, 31 Aug 2010 18:48:48 +0000 (18:48 +0000)]
Refactor my fix for PR5652 to terminate the predecessor lookups after the first failure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112620 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agomerge two tests.
Chris Lattner [Tue, 31 Aug 2010 18:44:03 +0000 (18:44 +0000)]
merge two tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112617 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoManually reduce this testcase.
Owen Anderson [Tue, 31 Aug 2010 18:16:29 +0000 (18:16 +0000)]
Manually reduce this testcase.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112615 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agomerge two tests and convert to filecheck.
Chris Lattner [Tue, 31 Aug 2010 18:05:08 +0000 (18:05 +0000)]
merge two tests and convert to filecheck.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112613 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd a micro-test for the transforms I added to JumpThreading.
Owen Anderson [Tue, 31 Aug 2010 17:59:07 +0000 (17:59 +0000)]
Add a micro-test for the transforms I added to JumpThreading.

I have not been able to find a way to test each in isolation, for a few reasons:
1) The ability to look-through non-i1 BinaryOperator's requires the ability to look through non-constant
   ICmps in order for it to ever trigger.
2) The ability to do LVI-powered PHI value determination only matters in cases that ProcessBranchOnPHI
   can't handle.  Since it already handles all the cases without other instructions in the def-use chain
   between the PHI and the branch, it requires the ability to look through ICmps and/or BinaryOperators
   as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112611 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoUpdate test for 112609
Jim Grosbach [Tue, 31 Aug 2010 17:58:47 +0000 (17:58 +0000)]
Update test for 112609

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112610 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoImprove virtual frame base register allocation heuristics.
Jim Grosbach [Tue, 31 Aug 2010 17:58:19 +0000 (17:58 +0000)]
Improve virtual frame base register allocation heuristics.

  1. Allocate them in the entry block of the function to enable function-wide
     re-use. The instructions to create them should be re-materializable, so
     there shouldn't be additional cost compared to creating them local
     to the basic blocks where they are used.
  2. Collect all of the frame index references for the function and sort them
     by the local offset referenced. Iterate over the sorted list to
     allocate the virtual base registers. This enables creation of base
     registers optimized for positive-offset access of frame references.
     (Note: This may be appropriate to later be a target hook to do the
     sorting in a target appropriate manner. For now it's done here for
     simplicity.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112609 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSpeculatively revert r112433.
Dan Gohman [Tue, 31 Aug 2010 17:56:47 +0000 (17:56 +0000)]
Speculatively revert r112433.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112608 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAllow creation of SHT_NULL sections, from Roman Divacky.
Benjamin Kramer [Tue, 31 Aug 2010 17:03:33 +0000 (17:03 +0000)]
Allow creation of SHT_NULL sections, from Roman Divacky.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112605 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoStop using the dom frontier in DwarfEHPrepare by not promoting alloca's
Duncan Sands [Tue, 31 Aug 2010 09:05:06 +0000 (09:05 +0000)]
Stop using the dom frontier in DwarfEHPrepare by not promoting alloca's
any more.  I plan to reimplement alloca promotion using SSAUpdater later.
It looks like Bill's URoR logic really always needs domtree, so the pass
now always asks for domtree info.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112597 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix an infinite loop; merging two functions will create a new function (if the
Nick Lewycky [Tue, 31 Aug 2010 08:29:37 +0000 (08:29 +0000)]
Fix an infinite loop; merging two functions will create a new function (if the
two are weak, we make them thunks to a new strong function) so don't iterate
through the function list as we're modifying it.

Also add back the outermost loop which got removed during the cleanups.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112595 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoDon't perform an extra traversal of the function just to do cleanup. We can safely...
Owen Anderson [Tue, 31 Aug 2010 07:55:56 +0000 (07:55 +0000)]
Don't perform an extra traversal of the function just to do cleanup.  We can safely simplify instructions after each block has been processed without worrying about iterator invalidation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112594 91177308-0d34-0410-b5e6-96231b3b80d8

14 years ago- Cleanup some whitespaces.
Bill Wendling [Tue, 31 Aug 2010 07:50:46 +0000 (07:50 +0000)]
- Cleanup some whitespaces.
- Convert {0,1} and friends into 0b01, which is identical and more consistent.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112593 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRename test directory to reflect new pass name.
Owen Anderson [Tue, 31 Aug 2010 07:50:31 +0000 (07:50 +0000)]
Rename test directory to reflect new pass name.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112592 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRename ValuePropagation to a more descriptive CorrelatedValuePropagation.
Owen Anderson [Tue, 31 Aug 2010 07:48:34 +0000 (07:48 +0000)]
Rename ValuePropagation to a more descriptive CorrelatedValuePropagation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112591 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRename file to something more descriptive.
Owen Anderson [Tue, 31 Aug 2010 07:41:39 +0000 (07:41 +0000)]
Rename file to something more descriptive.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112590 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMore Chris-inspired JumpThreading fixes: use ConstantExpr to correctly constant-fold...
Owen Anderson [Tue, 31 Aug 2010 07:36:34 +0000 (07:36 +0000)]
More Chris-inspired JumpThreading fixes: use ConstantExpr to correctly constant-fold undef, and be more careful with its return value.
This actually exposed an infinite recursion bug in ComputeValueKnownInPredecessors which theoretically already existed (in JumpThreading's
handling of and/or of i1's), but never manifested before.  This patch adds a tracking set to prevent this case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112589 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoCleanup Whitespace.
Michael J. Spencer [Tue, 31 Aug 2010 06:36:46 +0000 (06:36 +0000)]
Cleanup Whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112587 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSystem: Fix getMagicNumber on windows.
Michael J. Spencer [Tue, 31 Aug 2010 06:36:33 +0000 (06:36 +0000)]
System: Fix getMagicNumber on windows.

getMagicNumber was treating the _binary_ data it read in as a
null terminated string. This resulted in the std::string
calculating the length, and causing an assert in other code that
assumed that the length it passed was the same as the length of
the string it would get back.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112586 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix spelling/typo.
Michael J. Spencer [Tue, 31 Aug 2010 06:36:22 +0000 (06:36 +0000)]
Fix spelling/typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112585 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoOffset is not always unsigned number.
Devang Patel [Tue, 31 Aug 2010 06:12:08 +0000 (06:12 +0000)]
Offset is not always unsigned number.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112584 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSimplify.
Devang Patel [Tue, 31 Aug 2010 06:11:28 +0000 (06:11 +0000)]
Simplify.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112583 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSwitch to DenseSet, simplifying much more code. We now have a single iteration
Nick Lewycky [Tue, 31 Aug 2010 05:53:05 +0000 (05:53 +0000)]
Switch to DenseSet, simplifying much more code. We now have a single iteration
where we hash, compare and fold, instead of one iteration where we build up
the hash buckets and a second one to fold.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112582 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemove r111665, which implemented store-narrowing in InstCombine. Chris discovered...
Owen Anderson [Tue, 31 Aug 2010 04:41:06 +0000 (04:41 +0000)]
Remove r111665, which implemented store-narrowing in InstCombine.  Chris discovered a miscompilation in it, and it's not easily
fixable at the optimizer level. I'll investigate reimplementing it in DAGCombine.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112575 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agozap unused method. x86 is the only user and already has a more powerfull version
Bruno Cardoso Lopes [Tue, 31 Aug 2010 02:36:20 +0000 (02:36 +0000)]
zap unused method. x86 is the only user and already has a more powerfull version

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112571 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoUse X86ISD::MOVSS and MOVSD to represent the movl mask pattern, also fix the handling...
Bruno Cardoso Lopes [Tue, 31 Aug 2010 02:26:40 +0000 (02:26 +0000)]
Use X86ISD::MOVSS and MOVSD to represent the movl mask pattern, also fix the handling of those nodes when seeking for scalars inside vector shuffles

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112570 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRewrite slightly so we can expand for floating point types easier.
Eric Christopher [Tue, 31 Aug 2010 01:28:42 +0000 (01:28 +0000)]
Rewrite slightly so we can expand for floating point types easier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112568 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd experimental -disable-physical-join command line option.
Jakob Stoklund Olesen [Tue, 31 Aug 2010 01:27:49 +0000 (01:27 +0000)]
Add experimental -disable-physical-join command line option.

Eventually, we want to disable physreg coalescing completely, and let the
register allocator do its job using hints.

This option makes it possible to measure the impact of disabling physreg
coalescing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112567 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix a typo.
Owen Anderson [Mon, 30 Aug 2010 23:59:30 +0000 (23:59 +0000)]
Fix a typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112560 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoIf we have an unhandled type then assert, we shouldn't get here for
Eric Christopher [Mon, 30 Aug 2010 23:48:26 +0000 (23:48 +0000)]
If we have an unhandled type then assert, we shouldn't get here for
things we can't handle.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112559 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoUpdate the descriptions of NoModRef and ModRef to be consistent
Dan Gohman [Mon, 30 Aug 2010 23:47:24 +0000 (23:47 +0000)]
Update the descriptions of NoModRef and ModRef to be consistent
with the descriptions of Mod and Ref.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112557 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix borken test
Anton Korobeynikov [Mon, 30 Aug 2010 23:41:49 +0000 (23:41 +0000)]
Fix borken test

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112555 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoCombine these two tests, and make sure there's a newline at the end of the file.
Owen Anderson [Mon, 30 Aug 2010 23:37:41 +0000 (23:37 +0000)]
Combine these two tests, and make sure there's a newline at the end of the file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112554 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoCleanups suggested by Chris.
Owen Anderson [Mon, 30 Aug 2010 23:34:17 +0000 (23:34 +0000)]
Cleanups suggested by Chris.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112553 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRe-apply r112539, being more careful to respect the return values of the constant...
Owen Anderson [Mon, 30 Aug 2010 23:22:36 +0000 (23:22 +0000)]
Re-apply r112539, being more careful to respect the return values of the constant folding methods.  Additionally,
use the ConstantExpr::get*() methods to simplify some constant folding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112550 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoExpand MOVi32imm in ARM mode after regalloc. This provides
Anton Korobeynikov [Mon, 30 Aug 2010 22:50:36 +0000 (22:50 +0000)]
Expand MOVi32imm in ARM mode after regalloc. This provides
scheduling opportunities (extra instruction can go in between
MOVT / MOVW pair removing the stall).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112546 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd statistics to evaluate this pass.
Owen Anderson [Mon, 30 Aug 2010 22:45:55 +0000 (22:45 +0000)]
Add statistics to evaluate this pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112545 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRevert r112539. It accidentally introduced a miscompilation.
Owen Anderson [Mon, 30 Aug 2010 22:33:41 +0000 (22:33 +0000)]
Revert r112539.  It accidentally introduced a miscompilation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112543 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFixes and cleanups pointed out by Chris. In general, be careful to handle 0 results...
Owen Anderson [Mon, 30 Aug 2010 22:07:52 +0000 (22:07 +0000)]
Fixes and cleanups pointed out by Chris.  In general, be careful to handle 0 results from ComputeValueKnownInPredecessors
(indicating undef), and re-use existing constant folding APIs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112539 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoUse the existing T2I_bin_s_irs pattern instead of creating T2I_bin_sw_irs, which
Bill Wendling [Mon, 30 Aug 2010 22:05:23 +0000 (22:05 +0000)]
Use the existing T2I_bin_s_irs pattern instead of creating T2I_bin_sw_irs, which
is meant to do exactly the same thing. Thanks to Jim Grosbach for pointing this
out! :-)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112538 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix a comment.
NAKAMURA Takumi [Mon, 30 Aug 2010 21:54:03 +0000 (21:54 +0000)]
Fix a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112535 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemember to clear the shadow kill flag at the same time as clearing the real
Jakob Stoklund Olesen [Mon, 30 Aug 2010 21:52:40 +0000 (21:52 +0000)]
Remember to clear the shadow kill flag at the same time as clearing the real
kill flag.

This could cause duplicate kill flags when the same register was used twice in a
continuous sequence of STRs.

There is no small test case. <rdar://problem/8218046>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112534 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix llc to run the verifier once, not twice.
Dan Gohman [Mon, 30 Aug 2010 21:41:20 +0000 (21:41 +0000)]
Fix llc to run the verifier once, not twice.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112532 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemove this from the main tree. I'll host it out of tree.
Owen Anderson [Mon, 30 Aug 2010 21:34:26 +0000 (21:34 +0000)]
Remove this from the main tree.  I'll host it out of tree.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112529 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd comments explaining why it's not necessary to include the
Dan Gohman [Mon, 30 Aug 2010 21:18:41 +0000 (21:18 +0000)]
Add comments explaining why it's not necessary to include the
is-function-local flag in metadata uniquing bits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112528 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemove NEON vmovn intrinsic, replacing it with vector truncate operations.
Bob Wilson [Mon, 30 Aug 2010 20:02:30 +0000 (20:02 +0000)]
Remove NEON vmovn intrinsic, replacing it with vector truncate operations.
Auto-upgrade the old intrinsic and update tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112507 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMake ARM add rN, sp, #imm instructions rematerializable. That's how the address of...
Jim Grosbach [Mon, 30 Aug 2010 19:49:58 +0000 (19:49 +0000)]
Make ARM add rN, sp, #imm instructions rematerializable. That's how the address of locals is calculated, so this should
help relieve register pressure a bit. Recalculating the local address is
almost always going to be better than spilling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112503 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix LLVM target initialization to deal with sociopathic outside projects
Eric Christopher [Mon, 30 Aug 2010 18:34:48 +0000 (18:34 +0000)]
Fix LLVM target initialization to deal with sociopathic outside projects
that like to randomly define things like "X86", regenerate autoconf bits
and update cmake.

Fixes PR7852.

Patch by Xerxes RĂ„nby!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112499 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoKill a couple of unused variables.
Eric Christopher [Mon, 30 Aug 2010 18:31:44 +0000 (18:31 +0000)]
Kill a couple of unused variables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112498 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agonuke dead ivar which was supposed to be committed with r112496
Chris Lattner [Mon, 30 Aug 2010 18:16:27 +0000 (18:16 +0000)]
nuke dead ivar which was supposed to be committed with r112496

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112497 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agotwo changes:
Chris Lattner [Mon, 30 Aug 2010 18:12:35 +0000 (18:12 +0000)]
two changes:
1) nuke ConstDataCoalSection, which is dead.
2) revise my previous patch for rdar://8018335,
  which was completely wrong.  Specifically, it doesn't
  make sense to mark __TEXT,__const_coal as PURE_INSTRUCTIONS,
  because it is for readonly data.  templates (it turns out)
  go to const_coal_nt.  The real fix for rdar://8018335 was
  to give ConstTextCoalSection a section kind of ReadOnly
  instead of Text.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112496 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoWhen expanding NEON VST pseudo instructions, if the original super-register
Bob Wilson [Mon, 30 Aug 2010 18:10:48 +0000 (18:10 +0000)]
When expanding NEON VST pseudo instructions, if the original super-register
operand is killed, add it to the expanded instruction as an implicit kill
operand instead of marking the individual subregs with kill flags.  This
should work better in general and also handles the case for VST3 where one
of the subregs was not referenced in the expanded instruction and so was
not marked killed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112494 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMCELF: The value of all common symbols is the offset from the start of the section...
Benjamin Kramer [Mon, 30 Aug 2010 17:20:17 +0000 (17:20 +0000)]
MCELF: The value of all common symbols is the offset from the start of the section.  Patch by Roman Divacky.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112492 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoIt is possible to try to merge a not-constant with a constantrage, when dealing with...
Owen Anderson [Mon, 30 Aug 2010 17:03:45 +0000 (17:03 +0000)]
It is possible to try to merge a not-constant with a constantrage, when dealing with ptrtoint ConstantExpr's.
Unfortunately, the only testcase I have for this is huge and doesn't reduce well because the error is
sensitive to iteration-order issues, since the problem only occurs when merging values in a particular order.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112489 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoPartially revert r112480. Caused test failures.
Michael J. Spencer [Mon, 30 Aug 2010 15:34:08 +0000 (15:34 +0000)]
Partially revert r112480. Caused test failures.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112486 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agocoff-dump.py: Fix PR7996. Now it is compatible to Python-2.4.
NAKAMURA Takumi [Mon, 30 Aug 2010 15:19:56 +0000 (15:19 +0000)]
coff-dump.py: Fix PR7996. Now it is compatible to Python-2.4.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112485 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix constant-over-index.ll test on windows.
Michael J. Spencer [Mon, 30 Aug 2010 15:08:02 +0000 (15:08 +0000)]
Fix constant-over-index.ll test on windows.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112483 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoTest: Fix LLVMC tests on CMake.
Michael J. Spencer [Mon, 30 Aug 2010 14:49:00 +0000 (14:49 +0000)]
Test: Fix LLVMC tests on CMake.

The CMake build didn't define TEST_COMPILE_CXX_CMD. The tests assumed gcc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112480 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoDon't print two "0x" prefixes. Use a raw_ostream overload instead of llvm::format.
Benjamin Kramer [Mon, 30 Aug 2010 14:46:53 +0000 (14:46 +0000)]
Don't print two "0x" prefixes. Use a raw_ostream overload instead of llvm::format.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112479 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoEE/JIT: Do not invoke parent's ctors/dtors from main()! (PR3897)
NAKAMURA Takumi [Mon, 30 Aug 2010 14:00:29 +0000 (14:00 +0000)]
EE/JIT: Do not invoke parent's ctors/dtors from main()! (PR3897)

On Mingw and Cygwin, the symbol __main is resolved to
callee's(eg. tools/lli) one, to invoke wrong duplicated ctors
(and register wrong callee's dtors with atexit(3)).
We expect, by callee, ExecutionEngine::runStaticConstructorsDestructors()
is called before ExecutionEngine::runFunctionAsMain() is called.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112474 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoThe value is offset from the start of the section for non-common symbols, submitted...
Benjamin Kramer [Mon, 30 Aug 2010 12:00:16 +0000 (12:00 +0000)]
The value is offset from the start of the section for non-common symbols, submitted by Jordan Gordeev.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112473 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoIndex external symbols by symbol table instead of parent section, by Roman Divacky.
Benjamin Kramer [Mon, 30 Aug 2010 11:59:29 +0000 (11:59 +0000)]
Index external symbols by symbol table instead of parent section, by Roman Divacky.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112472 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMark all common symbols external. This is not exactly correct but it lets apps
Benjamin Kramer [Mon, 30 Aug 2010 11:56:55 +0000 (11:56 +0000)]
Mark all common symbols external. This is not exactly correct but it lets apps
link for now and can be adjusted later. Patch by Roman Divacky.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112471 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemove a hack that tries to understand incorrect triples from the
Duncan Sands [Mon, 30 Aug 2010 10:57:54 +0000 (10:57 +0000)]
Remove a hack that tries to understand incorrect triples from the
Triple class constructor.  Only valid triples should now be used
inside LLVM - front-ends are now responsable for rejecting or
correcting invalid target triples.  The Triple::normalize method
can be used to straighten out funky triples provided by users.
Give this a whirl through the buildbots to see if I caught all
places where triples enter LLVM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112470 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoCorrect bogus module triple specifications.
Duncan Sands [Mon, 30 Aug 2010 10:48:29 +0000 (10:48 +0000)]
Correct bogus module triple specifications.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112469 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd a new example to the LLVM distribution: a trace-based Brainfuck compiler that...
Owen Anderson [Mon, 30 Aug 2010 07:33:39 +0000 (07:33 +0000)]
Add a new example to the LLVM distribution: a trace-based Brainfuck compiler that uses LLVM as its code generator.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112465 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAttempt to remove the MSIL backend from CMake as well based on Chris's r112375.
Chandler Carruth [Mon, 30 Aug 2010 07:25:54 +0000 (07:25 +0000)]
Attempt to remove the MSIL backend from CMake as well based on Chris's r112375.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112464 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRevert r112461. It was failing on PPC...
Bill Wendling [Mon, 30 Aug 2010 04:36:50 +0000 (04:36 +0000)]
Revert r112461. It was failing on PPC...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112463 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoCreate Thumb2sI_cpsr and T2sI_cpsr. These new classes indicate that CPSR is the
Bill Wendling [Mon, 30 Aug 2010 01:47:35 +0000 (01:47 +0000)]
Create Thumb2sI_cpsr and T2sI_cpsr. These new classes indicate that CPSR is the
optional modified register (instead of reg0). Along with r112461 it will make
sure that the optional define of CPSR is marked as "def" and will thus mark the
instructions using these classes (t2ANDS*) as setting the 's' flag.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112462 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoWhen adding a register, we should mark it as "def" if it can optionally define
Bill Wendling [Mon, 30 Aug 2010 01:36:05 +0000 (01:36 +0000)]
When adding a register, we should mark it as "def" if it can optionally define
said (physical) register.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112461 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agorevert 112457, it looks like it broke selfhost.
Chris Lattner [Sun, 29 Aug 2010 22:28:18 +0000 (22:28 +0000)]
revert 112457, it looks like it broke selfhost.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112459 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agorewrite DwarfEHPrepare to use SSAUpdater to promote its allocas
Chris Lattner [Sun, 29 Aug 2010 19:54:28 +0000 (19:54 +0000)]
rewrite DwarfEHPrepare to use SSAUpdater to promote its allocas
instead of PromoteMemToReg.  This allows it to stop using DF and DT,
eliminating a computation of DT and DF from clang -O3.  Clang is now
down to 2 runs of DomFrontier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112457 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoinline function into its only caller.
Chris Lattner [Sun, 29 Aug 2010 19:28:28 +0000 (19:28 +0000)]
inline function into its only caller.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112455 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agotwo changes: 1) make AliasSet hold the list of call sites with an
Chris Lattner [Sun, 29 Aug 2010 18:42:23 +0000 (18:42 +0000)]
two changes: 1) make AliasSet hold the list of call sites with an
assertingvh so we get a violent explosion if the pointer dangles.

2) Fix AliasSetTracker::deleteValue to remove call sites with
   by-pointer comparisons instead of by-alias queries.  Using
   findAliasSetForCallSite can cause alias sets to get merged
   when they shouldn't, and can also miss alias sets when the
   call is readonly.

#2 fixes PR6889, which only repros with a .c file :(

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112452 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoLICM does get dead instructions input to it. Instead of sinking them
Chris Lattner [Sun, 29 Aug 2010 18:22:25 +0000 (18:22 +0000)]
LICM does get dead instructions input to it. Instead of sinking them
out of loops, just delete them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112451 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agouse moveBefore instead of remove+insert, it avoids some
Chris Lattner [Sun, 29 Aug 2010 18:18:40 +0000 (18:18 +0000)]
use moveBefore instead of remove+insert, it avoids some
symtab manipulation, so its faster (in addition to being
more elegant)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112450 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agorevert 112448 for now.
Chris Lattner [Sun, 29 Aug 2010 18:11:16 +0000 (18:11 +0000)]
revert 112448 for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112449 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agooptimize LICM::hoist to use moveBefore. Correct its updating
Chris Lattner [Sun, 29 Aug 2010 18:03:33 +0000 (18:03 +0000)]
optimize LICM::hoist to use moveBefore.  Correct its updating
of AST to remove the hoisted instruction from the AST, since it
is no longer in the loop.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112448 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agofix some bugs (found by inspection) where LICM would not update
Chris Lattner [Sun, 29 Aug 2010 18:00:00 +0000 (18:00 +0000)]
fix some bugs (found by inspection) where LICM would not update
LICM correctly.  When sinking an instruction, it should not add
entries for the sunk instruction to the AST, it should remove
the entry for the sunk instruction.  The blocks being sunk to
are not in the loop, so their instructions shouldn't be in the
AST (yet)!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112447 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agorework the ownership of subloop alias information: instead of
Chris Lattner [Sun, 29 Aug 2010 17:46:00 +0000 (17:46 +0000)]
rework the ownership of subloop alias information: instead of
keeping them around until the pass is destroyed, keep them
around a) just when useful (not for outer loops) and b) destroy
them right after we use them.  This should reduce memory use
and fixes potential bugs where a loop is deleted and another
loop gets allocated to the same address.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112446 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoapparently unswitch had the same "Feature". Stop its
Chris Lattner [Sun, 29 Aug 2010 17:23:19 +0000 (17:23 +0000)]
apparently unswitch had the same "Feature".  Stop its
claims that it preserves domfrontier if it doesn't really.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112445 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agonow that loop passes don't use DomFrontier, there is no reason
Chris Lattner [Sun, 29 Aug 2010 17:21:35 +0000 (17:21 +0000)]
now that loop passes don't use DomFrontier, there is no reason
for the unroller to pretend it supports updating it.  It still
has a horrible hack for DomTree.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112444 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMake IVUsers iterative instead of recursive.
Dan Gohman [Sun, 29 Aug 2010 16:40:03 +0000 (16:40 +0000)]
Make IVUsers iterative instead of recursive.

This has the side effect of reversing the order of most of
IVUser's results.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112442 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoOptionally rerun dedicated-register filtering after applying
Dan Gohman [Sun, 29 Aug 2010 16:39:22 +0000 (16:39 +0000)]
Optionally rerun dedicated-register filtering after applying
other filtering techniques, as those may allow it to filter
out more obviously unprofitable candidates.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112441 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix several areas in LSR to do a better job keeping the main
Dan Gohman [Sun, 29 Aug 2010 16:32:54 +0000 (16:32 +0000)]
Fix several areas in LSR to do a better job keeping the main
LSRInstance data structures up to date. This fixes some
pessimizations caused by stale data which will be exposed
in an upcoming change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112440 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRefactor the three main groups of code out of
Dan Gohman [Sun, 29 Aug 2010 16:09:42 +0000 (16:09 +0000)]
Refactor the three main groups of code out of
NarrowSearchSpaceUsingHeuristics into separate functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112439 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoDelete a bogus check.
Dan Gohman [Sun, 29 Aug 2010 15:30:29 +0000 (15:30 +0000)]
Delete a bogus check.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112438 91177308-0d34-0410-b5e6-96231b3b80d8