oota-llvm.git
13 years agoMC/Mach-O/ARM: Stub out RecordARMRelocation, which is mostly a copy of
Daniel Dunbar [Wed, 22 Dec 2010 13:50:05 +0000 (13:50 +0000)]
MC/Mach-O/ARM: Stub out RecordARMRelocation, which is mostly a copy of
RecordRelocation with lots of FIXMEs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122402 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoSimplify.
Daniel Dunbar [Wed, 22 Dec 2010 13:49:56 +0000 (13:49 +0000)]
Simplify.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122401 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoMC/Mach-O: Split out RecordARMRelocation for now, it is weird enough it isn't
Daniel Dunbar [Wed, 22 Dec 2010 13:49:43 +0000 (13:49 +0000)]
MC/Mach-O: Split out RecordARMRelocation for now, it is weird enough it isn't
clear how to keep in the generic path (yet).
 - Will revisit when it actually works.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122400 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd a generic expansion transform: A op (B op' C) -> (A op B) op' (A op C)
Duncan Sands [Wed, 22 Dec 2010 13:36:08 +0000 (13:36 +0000)]
Add a generic expansion transform: A op (B op' C) -> (A op B) op' (A op C)
if both A op B and A op C simplify.  This fires fairly often but doesn't
make that much difference.  On gcc-as-one-file it removes two "and"s and
turns one branch into a select.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122399 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoptx: add ld instruction and test
Che-Liang Chiou [Wed, 22 Dec 2010 10:38:51 +0000 (10:38 +0000)]
ptx: add ld instruction and test

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122398 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd some statistics, good for understanding how much more powerful
Duncan Sands [Wed, 22 Dec 2010 09:40:51 +0000 (09:40 +0000)]
Add some statistics, good for understanding how much more powerful
instcombine is compared to instsimplify.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122397 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFixes file extension for loadable modules on OS X.
Oscar Fuentes [Wed, 22 Dec 2010 08:30:17 +0000 (08:30 +0000)]
Fixes file extension for loadable modules on OS X.

Patch by Wesley Peck!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122395 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFix a bug in ReduceLoadWidth that wasn't handling extending
Chris Lattner [Wed, 22 Dec 2010 08:02:57 +0000 (08:02 +0000)]
Fix a bug in ReduceLoadWidth that wasn't handling extending
loads properly.  We miscompiled the testcase into:

_test:                                  ## @test
movl $128, (%rdi)
movzbl 1(%rdi), %eax
ret

Now we get a proper:

_test:                                  ## @test
movl $128, (%rdi)
movsbl (%rdi), %eax
movzbl %ah, %eax
ret

This fixes PR8757.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122392 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agomore cleanups, move a check for "roundedness" earlier to reject
Chris Lattner [Wed, 22 Dec 2010 08:01:44 +0000 (08:01 +0000)]
more cleanups, move a check for "roundedness" earlier to reject
unhanded cases faster and simplify code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122391 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoreduce indentation and improve comments, no functionality change.
Chris Lattner [Wed, 22 Dec 2010 07:36:50 +0000 (07:36 +0000)]
reduce indentation and improve comments, no functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122389 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoDon't generate carry bit when loading immediate values on the Microblaze.
Wesley Peck [Wed, 22 Dec 2010 01:29:32 +0000 (01:29 +0000)]
Don't generate carry bit when loading immediate values on the Microblaze.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122385 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd support for some of the LLVM atomic operations to the MBlaze backend.
Wesley Peck [Wed, 22 Dec 2010 01:15:01 +0000 (01:15 +0000)]
Add support for some of the LLVM atomic operations to the MBlaze backend.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122384 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoModeling the carry bit in the MSR register of the MicroBlaze.
Wesley Peck [Wed, 22 Dec 2010 00:53:07 +0000 (00:53 +0000)]
Modeling the carry bit in the MSR register of the MicroBlaze.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122381 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFix a regression introduced into the MBlaze delay slot filler.
Wesley Peck [Wed, 22 Dec 2010 00:22:59 +0000 (00:22 +0000)]
Fix a regression introduced into the MBlaze delay slot filler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122379 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoGive GVN back the ability to perform simple conditional propagation on conditional...
Owen Anderson [Tue, 21 Dec 2010 23:54:34 +0000 (23:54 +0000)]
Give GVN back the ability to perform simple conditional propagation on conditional branch values.
I still think that LVI should be handling this, but that capability is some ways off in the future,
and this matters for some significant benchmarks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122378 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoGCC objects to the two sides of a conditional expression having different enum
Matt Beaumont-Gay [Tue, 21 Dec 2010 23:43:23 +0000 (23:43 +0000)]
GCC objects to the two sides of a conditional expression having different enum
types, but they're just getting converted to unsigned anyway, so cast first
(and ask questions later).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122377 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoRemove dead code.
Owen Anderson [Tue, 21 Dec 2010 22:31:24 +0000 (22:31 +0000)]
Remove dead code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122371 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoIn DelayForLiveRegsBottomUp, handle instructions that read and write
Andrew Trick [Tue, 21 Dec 2010 22:27:44 +0000 (22:27 +0000)]
In DelayForLiveRegsBottomUp, handle instructions that read and write
the same physical register. Simplifies the fix from the previous
checkin r122211.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122370 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agowhitespace
Andrew Trick [Tue, 21 Dec 2010 22:25:04 +0000 (22:25 +0000)]
whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122368 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoReapply 122353-122355 with fixes. 122354 was wrong;
Dale Johannesen [Tue, 21 Dec 2010 21:55:50 +0000 (21:55 +0000)]
Reapply 122353-122355 with fixes.  122354 was wrong;
the shift type was needed one place, the shift count
type another.  The transform in 123555 had the same
problem.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122366 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd some x86 specific dagcombines for conditional increments.
Benjamin Kramer [Tue, 21 Dec 2010 21:41:44 +0000 (21:41 +0000)]
Add some x86 specific dagcombines for conditional increments.

(add Y, (sete  X, 0)) -> cmp X, 1; adc  0, Y
(add Y, (setne X, 0)) -> cmp X, 1; sbb -1, Y
(sub (sete  X, 0), Y) -> cmp X, 1; sbb  0, Y
(sub (setne X, 0), Y) -> cmp X, 1; adc -1, Y

for
  unsigned foo(unsigned a, unsigned b) {
    if (a == 0) b++;
    return b;
  }
we now get:
  foo:
    cmpl  $1, %edi
    movl  %esi, %eax
    adcl  $0, %eax
    ret
instead of:
  foo:
    testl %edi, %edi
    sete  %al
    movzbl  %al, %eax
    addl  %esi, %eax
    ret

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122364 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoGVN's Expression is not POD-like (it contains a SmallVector). Simplify code while...
Benjamin Kramer [Tue, 21 Dec 2010 21:30:19 +0000 (21:30 +0000)]
GVN's Expression is not POD-like (it contains a SmallVector). Simplify code while at it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122362 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoRevert 122353-122355 for the moment, they broke stuff.
Dale Johannesen [Tue, 21 Dec 2010 21:22:27 +0000 (21:22 +0000)]
Revert 122353-122355 for the moment, they broke stuff.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122360 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoSimplify EvaluateAsAbsolute now that EvaluateAsRelocatableImpl does all
Rafael Espindola [Tue, 21 Dec 2010 20:51:42 +0000 (20:51 +0000)]
Simplify EvaluateAsAbsolute now that EvaluateAsRelocatableImpl does all
the folding it can.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122359 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoDon't relax org or align. They change size as the relaxation happens, but they
Rafael Espindola [Tue, 21 Dec 2010 20:35:18 +0000 (20:35 +0000)]
Don't relax org or align. They change size as the relaxation happens, but they
are not actually relaxed. For example, a section with only alignments will never
needs relaxation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122356 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd a new transform to DAGCombiner.
Dale Johannesen [Tue, 21 Dec 2010 20:10:51 +0000 (20:10 +0000)]
Add a new transform to DAGCombiner.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122355 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoGet the type of a shift from the shift, not from its shift
Dale Johannesen [Tue, 21 Dec 2010 20:06:19 +0000 (20:06 +0000)]
Get the type of a shift from the shift, not from its shift
count operand.  These should be the same but apparently are
not always, and this is cleaner anyway.  This improves the
code in an existing test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122354 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoShift by the word size is invalid IR; don't create it.
Dale Johannesen [Tue, 21 Dec 2010 20:00:06 +0000 (20:00 +0000)]
Shift by the word size is invalid IR; don't create it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122353 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agofix some typos
Chris Lattner [Tue, 21 Dec 2010 18:05:22 +0000 (18:05 +0000)]
fix some typos

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122349 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoRevert 122341. It breaks some darwin tests.
David Greene [Tue, 21 Dec 2010 17:25:43 +0000 (17:25 +0000)]
Revert 122341.  It breaks some darwin tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122346 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFix indentation, add comment.
Stuart Hastings [Tue, 21 Dec 2010 17:16:58 +0000 (17:16 +0000)]
Fix indentation, add comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122345 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoVisit instructions deterministically. Use a FIFO so as to approximately
Duncan Sands [Tue, 21 Dec 2010 17:08:55 +0000 (17:08 +0000)]
Visit instructions deterministically.  Use a FIFO so as to approximately
visit instructions before their uses, since InstructionSimplify does a
better job in that case.  All this prompted by Frits van Bommel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122343 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoMissing logic for nested CALLSEQ_START/END.
Stuart Hastings [Tue, 21 Dec 2010 17:07:24 +0000 (17:07 +0000)]
Missing logic for nested CALLSEQ_START/END.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122342 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFix PR 8199. This patch prepends the build tool dir to LLVM programs
David Greene [Tue, 21 Dec 2010 16:55:53 +0000 (16:55 +0000)]
Fix PR 8199.  This patch prepends the build tool dir to LLVM programs
being tested.  This ensures that we test the tools just built and not
some random tools that might happen to be in the user's PATH.  This
makes LLVM testing much more stable and predictable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122341 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTidy up a bit. Trailing whitespace, hard tabs and 80-columns.
Jim Grosbach [Tue, 21 Dec 2010 16:16:00 +0000 (16:16 +0000)]
Tidy up a bit. Trailing whitespace, hard tabs and 80-columns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122337 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoIf an instruction simplifies, try again to simplify any uses of it. This is
Duncan Sands [Tue, 21 Dec 2010 16:12:03 +0000 (16:12 +0000)]
If an instruction simplifies, try again to simplify any uses of it.  This is
not very important since the pass is only used for testing, but it does make
it more realistic.  Suggested by Frits van Bommel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122336 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoMC/Mach-O: Shuffle enums a bit to make it harder to inadvertently use the wrong
Daniel Dunbar [Tue, 21 Dec 2010 15:26:45 +0000 (15:26 +0000)]
MC/Mach-O: Shuffle enums a bit to make it harder to inadvertently use the wrong
type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122334 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd an additional InstructionSimplify factorization test.
Duncan Sands [Tue, 21 Dec 2010 15:12:22 +0000 (15:12 +0000)]
Add an additional InstructionSimplify factorization test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122333 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoWhile I don't think any later transforms can fire, it seems cleaner to
Duncan Sands [Tue, 21 Dec 2010 15:03:43 +0000 (15:03 +0000)]
While I don't think any later transforms can fire, it seems cleaner to
not assume this (for example in case more transforms get added below
it).  Suggested by Frits van Bommel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122332 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFix inverted condition noticed by Frits van Bommel.
Duncan Sands [Tue, 21 Dec 2010 14:48:48 +0000 (14:48 +0000)]
Fix inverted condition noticed by Frits van Bommel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122331 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoPull a few more simplifications out of instcombine (there are still
Duncan Sands [Tue, 21 Dec 2010 14:00:22 +0000 (14:00 +0000)]
Pull a few more simplifications out of instcombine (there are still
plenty left though!), in particular for multiplication.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122330 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFix typo in comment, spotted by Deewiant.
Duncan Sands [Tue, 21 Dec 2010 13:39:20 +0000 (13:39 +0000)]
Fix typo in comment, spotted by Deewiant.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122329 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTeach InstructionSimplify about distributive laws. These transforms fire
Duncan Sands [Tue, 21 Dec 2010 13:32:22 +0000 (13:32 +0000)]
Teach InstructionSimplify about distributive laws.  These transforms fire
quite often, but don't make much difference in practice presumably because
instcombine also knows them and more.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122328 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoMove checking of the recursion limit into the various Thread methods.
Duncan Sands [Tue, 21 Dec 2010 09:09:15 +0000 (09:09 +0000)]
Move checking of the recursion limit into the various Thread methods.
No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122327 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd generic simplification of associative operations, generalizing
Duncan Sands [Tue, 21 Dec 2010 08:49:00 +0000 (08:49 +0000)]
Add generic simplification of associative operations, generalizing
a couple of existing transforms.  This fires surprisingly often, for
example when compiling gcc "(X+(-1))+1->X" fires quite a lot as well
as various "and" simplifications (usually with a phi node operand).
Most of the time this doesn't make a real difference since the same
thing would have been done elsewhere anyway, eg: by instcombine, but
there are a few places where this results in simplifications that we
were not doing before.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122326 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoIncremental progress towards a new implementation of StrongPHIElimination. Most
Cameron Zwarich [Tue, 21 Dec 2010 06:54:43 +0000 (06:54 +0000)]
Incremental progress towards a new implementation of StrongPHIElimination. Most
of the problems with my last attempt were in the updating of LiveIntervals
rather than the coalescing itself. Therefore, I decided to get that right first
by essentially reimplementing the existing PHIElimination using LiveIntervals.

It works correctly, with only a few tests failing (which may not be legitimate
failures) and no new verifier failures (at least as far as I can tell, I didn't
count the number per file).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122321 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd ARM-specific DAG combining to cast i64 vector element load/stores to f64.
Bob Wilson [Tue, 21 Dec 2010 06:43:19 +0000 (06:43 +0000)]
Add ARM-specific DAG combining to cast i64 vector element load/stores to f64.
Type legalization splits up i64 values into pairs of i32 values, which leads
to poor quality code when inserting or extracting i64 vector elements.
If the vector element is loaded or stored, it can be treated as an f64 value
and loaded or stored directly from a VPR register.  Use the pre-legalization
DAG combiner to cast those vector elements to f64 types so that the type
legalizer won't mess them up.  Radar 8755338.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122319 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoLayout one section until no relaxations are done and then move to the next
Rafael Espindola [Tue, 21 Dec 2010 04:22:09 +0000 (04:22 +0000)]
Layout one section until no relaxations are done and then move to the next
section.

This helps because in practice sections form a dag with debug sections pointing
to text sections. Finishing up the text sections first makes the debug section
relaxation trivial.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122314 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoArm and thumb call instructions are also in different orders.
Eric Christopher [Tue, 21 Dec 2010 03:50:43 +0000 (03:50 +0000)]
Arm and thumb call instructions are also in different orders.

Fixes rdar://8782223

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122313 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoGrammar corrections and some formatting changes.
Bill Wendling [Tue, 21 Dec 2010 03:31:05 +0000 (03:31 +0000)]
Grammar corrections and some formatting changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122312 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agorename MVT::Flag to MVT::Glue. "Flag" is a terrible name for
Chris Lattner [Tue, 21 Dec 2010 02:38:05 +0000 (02:38 +0000)]
rename MVT::Flag to MVT::Glue.  "Flag" is a terrible name for
something that just glues two nodes together, even if it is
sometimes used for flags.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122310 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoIf we're not using reg+reg offset we're using reg+imm, set the opcode
Eric Christopher [Tue, 21 Dec 2010 02:12:07 +0000 (02:12 +0000)]
If we're not using reg+reg offset we're using reg+imm, set the opcode
to be the one we want to use. bugpoint reduced testcase is a little large,
I'll see if I can simplify it down more.

Fixes part of rdar://8782207

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122307 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoimprove "cannot yet select" errors a trivial amount: now
Chris Lattner [Tue, 21 Dec 2010 02:07:03 +0000 (02:07 +0000)]
improve "cannot yet select" errors a trivial amount: now
they are just as useless, but at least a bit more gramatical

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122305 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFix a copy-pasto. When the tBR_JTr instruction was converted to using the
Bill Wendling [Tue, 21 Dec 2010 01:57:15 +0000 (01:57 +0000)]
Fix a copy-pasto. When the tBR_JTr instruction was converted to using the
tPseudoInst class, its size was changed from "special" to "2 bytes". This is
incorrect because the jump table will no longer be taken into account when
calculating branch offsets.
<rdar://problem/8782216>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122303 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoComment cleanups.
Bill Wendling [Tue, 21 Dec 2010 01:54:40 +0000 (01:54 +0000)]
Comment cleanups.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122302 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd EdgeBundles to SplitKit.
Jakob Stoklund Olesen [Tue, 21 Dec 2010 01:50:21 +0000 (01:50 +0000)]
Add EdgeBundles to SplitKit.

Edge bundles is an annotation on the CFG that turns it into a bipartite directed
graph where each basic block is connected to an outgoing and an ingoing bundle.
These bundles are useful for identifying regions of the CFG for live range
splitting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122301 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoUse IntEqClasses to compute connected components of live intervals.
Jakob Stoklund Olesen [Tue, 21 Dec 2010 00:48:17 +0000 (00:48 +0000)]
Use IntEqClasses to compute connected components of live intervals.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122296 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd ADT/IntEqClasses.h as a light-weight implementation of EquivalenceClasses.h.
Jakob Stoklund Olesen [Tue, 21 Dec 2010 00:04:46 +0000 (00:04 +0000)]
Add ADT/IntEqClasses.h as a light-weight implementation of EquivalenceClasses.h.

This implementation already exists as ConnectedVNInfoEqClasses in
LiveInterval.cpp, and it seems to be generally useful to have a light-weight way
of forming equivalence classes of small integers.

IntEqClasses doesn't allow enumeration of the elements in a class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122293 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoSpeculatively revert the use of DenseMap in LazyValueInfo, which may be causing Linux...
Owen Anderson [Mon, 20 Dec 2010 23:53:19 +0000 (23:53 +0000)]
Speculatively revert the use of DenseMap in LazyValueInfo, which may be causing Linux self-host failures.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122291 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAttempt to appease the DragonEgg buildbots.
Owen Anderson [Mon, 20 Dec 2010 23:23:18 +0000 (23:23 +0000)]
Attempt to appease the DragonEgg buildbots.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122288 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoRevert r122114 (CallbackVH observing use-list changes) because it caused severe slowd...
Owen Anderson [Mon, 20 Dec 2010 22:28:03 +0000 (22:28 +0000)]
Revert r122114 (CallbackVH observing use-list changes) because it caused severe slowdowns on the Linux self-host configuration.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122279 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoImplement feedback from Bruno on making pblendvb an x86-specific ISD node in addition...
Nate Begeman [Mon, 20 Dec 2010 22:04:24 +0000 (22:04 +0000)]
Implement feedback from Bruno on making pblendvb an x86-specific ISD node in addition to being an intrinsic, and convert
lowering to use it.  Hopefully the pattern fragment is doing the right thing with XMM0, looks correct in testing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122277 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoUpdate the target feature matrix to reflect some new features in the MBlaze backend.
Wesley Peck [Mon, 20 Dec 2010 21:54:50 +0000 (21:54 +0000)]
Update the target feature matrix to reflect some new features in the MBlaze backend.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122276 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoConvert one of LVI's primary maps to a DenseMap, now that we know are more assured...
Owen Anderson [Mon, 20 Dec 2010 21:30:54 +0000 (21:30 +0000)]
Convert one of LVI's primary maps to a DenseMap, now that we know are more assured of iterator stability.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122273 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTeach the MBlaze disassembler to disassemble special purpose registers.
Wesley Peck [Mon, 20 Dec 2010 21:18:04 +0000 (21:18 +0000)]
Teach the MBlaze disassembler to disassemble special purpose registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122269 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoSet the value of absolute symbols.
Roman Divacky [Mon, 20 Dec 2010 21:14:39 +0000 (21:14 +0000)]
Set the value of absolute symbols.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122268 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoThis is still used and issuing an annoying warning. Don't deprecate something in
Bill Wendling [Mon, 20 Dec 2010 21:14:18 +0000 (21:14 +0000)]
This is still used and issuing an annoying warning. Don't deprecate something in
the library unless *all* uses have been converted over to the new form.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122267 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoOops, forgot to add the pass itself!
Duncan Sands [Mon, 20 Dec 2010 21:07:42 +0000 (21:07 +0000)]
Oops, forgot to add the pass itself!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122265 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd a new convenience pass for testing InstructionSimplify. Previously
Duncan Sands [Mon, 20 Dec 2010 20:54:37 +0000 (20:54 +0000)]
Add a new convenience pass for testing InstructionSimplify.  Previously
it could only be tested indirectly, via instcombine, gvn or some other
pass that makes use of InstructionSimplify, which means that testcases
had to be carefully contrived to dance around any other transformations
that that pass did.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122264 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoPrint all 64bits for st_value and st_size. Adjust tests accordingly.
Roman Divacky [Mon, 20 Dec 2010 20:49:43 +0000 (20:49 +0000)]
Print all 64bits for st_value and st_size. Adjust tests accordingly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122263 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTeach the MBlaze asm parser how to parse special purpose register names.
Wesley Peck [Mon, 20 Dec 2010 20:43:24 +0000 (20:43 +0000)]
Teach the MBlaze asm parser how to parse special purpose register names.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122261 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoCosmetic changes.
Dale Johannesen [Mon, 20 Dec 2010 20:10:50 +0000 (20:10 +0000)]
Cosmetic changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122259 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd a check missing from my last commit and avoid a potential overflow situation.
Benjamin Kramer [Mon, 20 Dec 2010 20:00:31 +0000 (20:00 +0000)]
Add a check missing from my last commit and avoid a potential overflow situation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122258 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoMore LVI cleanups, including trying to simplify the process of maintaining the OverDe...
Owen Anderson [Mon, 20 Dec 2010 19:33:41 +0000 (19:33 +0000)]
More LVI cleanups, including trying to simplify the process of maintaining the OverDefinedCache.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122256 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoReuse the reference into the LVI cache throughout the solver subsystem. This is...
Owen Anderson [Mon, 20 Dec 2010 18:18:16 +0000 (18:18 +0000)]
Reuse the reference into the LVI cache throughout the solver subsystem.  This is much easier to
verify as being safe thanks its recent de-recursivization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122254 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoReduce indentation.
Benjamin Kramer [Mon, 20 Dec 2010 16:21:59 +0000 (16:21 +0000)]
Reduce indentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122249 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTeach InstCombine to merge (icmp ult (X + CA), C1) | (icmp eq X, C2) into (icmp ult...
Benjamin Kramer [Mon, 20 Dec 2010 16:18:51 +0000 (16:18 +0000)]
Teach InstCombine to merge (icmp ult (X + CA), C1) | (icmp eq X, C2) into (icmp ult (X + CA), C1 + 1) if C2 + CA == C1.

InstCombine creates these so now we compile x == 23 || x == 24 || x == 25 to
  %x.off = add i32 %x, -23
  %1 = icmp ult i32 %x.off, 3
instead of
  %x.off = add i32 %x, -23
  %1 = icmp ult i32 %x.off, 2
  %cmp3 = icmp eq i32 %x, 25
  %ret2 = or i1 %1, %cmp3

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122248 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd header...
Daniel Dunbar [Mon, 20 Dec 2010 15:45:51 +0000 (15:45 +0000)]
Add header...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122247 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoX86/MC/Mach-O: Split out createX86MachObjectWriter().
Daniel Dunbar [Mon, 20 Dec 2010 15:07:39 +0000 (15:07 +0000)]
X86/MC/Mach-O: Split out createX86MachObjectWriter().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122246 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoHave SimplifyBinOp dispatch Xor, Add and Sub to the corresponding methods
Duncan Sands [Mon, 20 Dec 2010 14:47:04 +0000 (14:47 +0000)]
Have SimplifyBinOp dispatch Xor, Add and Sub to the corresponding methods
(they had just been forgotten before).  Adding Xor causes "main" in the
existing testcase 2010-11-01-lshr-mask.ll to be hugely more simplified.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122245 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFix spelling.
Michael J. Spencer [Mon, 20 Dec 2010 13:30:37 +0000 (13:30 +0000)]
Fix spelling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122244 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoSupport/PathV2: Add missing has_relative_path impl.
Michael J. Spencer [Mon, 20 Dec 2010 13:30:28 +0000 (13:30 +0000)]
Support/PathV2: Add missing has_relative_path impl.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122243 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoThere is no need for isAssociative to take the type as an argument anymore.
Duncan Sands [Mon, 20 Dec 2010 13:10:23 +0000 (13:10 +0000)]
There is no need for isAssociative to take the type as an argument anymore.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122242 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoDiscourage people from using isWeakForLinker when they should be using
Duncan Sands [Mon, 20 Dec 2010 10:57:01 +0000 (10:57 +0000)]
Discourage people from using isWeakForLinker when they should be using
mayBeOverridden.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122241 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoNew cmake option LLVM_APPEND_VC_REV for controlling when the VC
Oscar Fuentes [Mon, 20 Dec 2010 09:47:13 +0000 (09:47 +0000)]
New cmake option LLVM_APPEND_VC_REV for controlling when the VC
revision id is appended to the LLVM version string. Defaults to
OFF.

Until now the VC revision id was always appended to the revision
string whenever cmake was invoked (either explicitly or implicitly
because a cmake source file changed). This was causing massive
recompilations because config.h are reconfigured with the new contents
of PACKAGE_VERSION.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122240 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoUpdate cmake library dependencies.
Oscar Fuentes [Mon, 20 Dec 2010 09:47:08 +0000 (09:47 +0000)]
Update cmake library dependencies.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122239 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agofix PR8807 by making transformConstExprCastCall aware of byval arguments.
Chris Lattner [Mon, 20 Dec 2010 08:36:38 +0000 (08:36 +0000)]
fix PR8807 by making transformConstExprCastCall aware of byval arguments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122238 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agovarious cleanups for transformConstExprCastCall
Chris Lattner [Mon, 20 Dec 2010 08:25:06 +0000 (08:25 +0000)]
various cleanups for transformConstExprCastCall

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122237 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agowhen eliding a byval copy due to inlining a readonly function, we have
Chris Lattner [Mon, 20 Dec 2010 08:10:40 +0000 (08:10 +0000)]
when eliding a byval copy due to inlining a readonly function, we have
to make sure that the reused alloca has sufficient alignment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122236 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agopull byval processing out to its own helper function.
Chris Lattner [Mon, 20 Dec 2010 07:57:41 +0000 (07:57 +0000)]
pull byval processing out to its own helper function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122235 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agofix PR8769, a miscompilation by inliner when inlining a function with a byval
Chris Lattner [Mon, 20 Dec 2010 07:45:28 +0000 (07:45 +0000)]
fix PR8769, a miscompilation by inliner when inlining a function with a byval
argument.  The generated alloca has to have at least the alignment of the
byval, if not, the client may be making assumptions that the new alloca won't
satisfy.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122234 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agomerge two tests.
Chris Lattner [Mon, 20 Dec 2010 07:39:57 +0000 (07:39 +0000)]
merge two tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122233 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agofilecheckize
Chris Lattner [Mon, 20 Dec 2010 07:38:24 +0000 (07:38 +0000)]
filecheckize

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122232 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoMachineVerifier should count landing pad successors as basic blocks rather than
Cameron Zwarich [Mon, 20 Dec 2010 04:19:48 +0000 (04:19 +0000)]
MachineVerifier should count landing pad successors as basic blocks rather than
out-edges. Fixes PR8824.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122228 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTeach MachineVerifier that early clobber defs begin at USE slots and other defs
Cameron Zwarich [Mon, 20 Dec 2010 03:15:20 +0000 (03:15 +0000)]
Teach MachineVerifier that early clobber defs begin at USE slots and other defs
begin at DEF slots. Fixes the second half of PR8813.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122225 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd a missing check from r122218.
Cameron Zwarich [Mon, 20 Dec 2010 02:59:51 +0000 (02:59 +0000)]
Add a missing check from r122218.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122224 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agotemporarily disable this: PR8823.
Chris Lattner [Mon, 20 Dec 2010 02:11:23 +0000 (02:11 +0000)]
temporarily disable this: PR8823.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122222 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoimplement type legalization promotion support for SMULO and UMULO, giving
Chris Lattner [Mon, 20 Dec 2010 02:05:39 +0000 (02:05 +0000)]
implement type legalization promotion support for SMULO and UMULO, giving
ARM (and other 32-bit-only) targets support for i8 and i16 overflow
multiplies.  The generated code isn't great, but this at least fixes
CodeGen/Generic/overflow.ll when running on ARM hosts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122221 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agonow that addc/adde are gone, "ADDC" in the X86 backend uses EFLAGS results,
Chris Lattner [Mon, 20 Dec 2010 01:37:09 +0000 (01:37 +0000)]
now that addc/adde are gone, "ADDC" in the X86 backend uses EFLAGS results,
the same as setcc.  Optimize ADDC(0,0,FLAGS) -> SET_CARRY(FLAGS).  This is
a step towards finishing off PR5443.  In the testcase in that bug we now  get:

movq %rdi, %rax
addq %rsi, %rax
sbbq %rcx, %rcx
testb $1, %cl
setne %dl
ret

instead of:

movq %rdi, %rax
addq %rsi, %rax
movl $0, %ecx
adcq $0, %rcx
testq %rcx, %rcx
setne %dl
ret

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122219 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoDon't assume that an instruction ending a register's live range always reads
Cameron Zwarich [Mon, 20 Dec 2010 01:22:37 +0000 (01:22 +0000)]
Don't assume that an instruction ending a register's live range always reads
the register; it may be a dead def instead. Fixes PR8820.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122218 91177308-0d34-0410-b5e6-96231b3b80d8