oota-llvm.git
13 years agoFollowup to 132458, omit unnecessary stack copy when x87 input is a
Stuart Hastings [Mon, 6 Jun 2011 23:15:58 +0000 (23:15 +0000)]
Followup to 132458, omit unnecessary stack copy when x87 input is a
load.  rdar://problem/6373334

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132696 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd -O1/2/3 to bugpoint, so when you conclude opt -O2 reproduces an issue, you can...
Eli Friedman [Mon, 6 Jun 2011 22:45:46 +0000 (22:45 +0000)]
Add -O1/2/3 to bugpoint, so when you conclude opt -O2 reproduces an issue, you can just run bugpoint -O2. :)  My implementation isn't precisely equivalent to what opt does, but as far as I can tell, it's close enough.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132695 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoWe only do always-inlining at -O1; make opt reflect that.
Eli Friedman [Mon, 6 Jun 2011 22:13:27 +0000 (22:13 +0000)]
We only do always-inlining at -O1; make opt reflect that.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132693 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoGet allocation orders from RegisterClassInfo when possible.
Jakob Stoklund Olesen [Mon, 6 Jun 2011 21:02:04 +0000 (21:02 +0000)]
Get allocation orders from RegisterClassInfo when possible.

Only target-dependent hints require callbacks. The RCI allocation order
has CSR aliases last according to their order of appearance in the
getCalleeSavedRegs list. This can depend on the calling convention.

This way, AllocationOrder::next doesn't have to check for reserved
registers, and CSRs are always allocated last, even with weird calling
conventions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132690 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd methods to support the integer-promotion of vector types. Methods to
Nadav Rotem [Mon, 6 Jun 2011 20:55:56 +0000 (20:55 +0000)]
Add methods to support the integer-promotion of vector types. Methods to
legalize SDNodes such as BUILD_VECTOR, EXTRACT_VECTOR_ELT, etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132689 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoThis directory was missing the dg.exp to cause the tests to run. Some time since
Nick Lewycky [Mon, 6 Jun 2011 20:23:00 +0000 (20:23 +0000)]
This directory was missing the dg.exp to cause the tests to run. Some time since
it was added, the test has regressed, so XFAIL it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132686 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTest case for PR10085.
Stuart Hastings [Mon, 6 Jun 2011 20:03:22 +0000 (20:03 +0000)]
Test case for PR10085.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132682 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAvoid FGETSIGN of 80-bit types. Fixes PR10085.
Stuart Hastings [Mon, 6 Jun 2011 16:44:31 +0000 (16:44 +0000)]
Avoid FGETSIGN of 80-bit types.  Fixes PR10085.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132681 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoDon't try to be clever, just preserve the target's allocation order.
Jakob Stoklund Olesen [Mon, 6 Jun 2011 16:36:30 +0000 (16:36 +0000)]
Don't try to be clever, just preserve the target's allocation order.

The order of registers returned by getCalleeSavedRegs is used to lay out
the fixed stack slots for CSRs. Some targets like their CSRs used from
one end, and some targets want them used from the other end.

When computing an allocation order, simply preserve the relative
ordering of CSRs that the target specifies in its allocation order.
Reordering CSRs would break some targets, ARM in particular.

We still place volatiles before the CSRs, providing slightly better
results with different calling conventions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132680 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoPR10077: fix fast-isel of extractvalue of aggregate constants.
Eli Friedman [Mon, 6 Jun 2011 05:46:34 +0000 (05:46 +0000)]
PR10077: fix fast-isel of extractvalue of aggregate constants.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132676 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoHarden tests for windows path separators.
Benjamin Kramer [Sun, 5 Jun 2011 18:20:05 +0000 (18:20 +0000)]
Harden tests for windows path separators.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132671 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoUse path API for path concatenation.
Benjamin Kramer [Sun, 5 Jun 2011 14:36:47 +0000 (14:36 +0000)]
Use path API for path concatenation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132668 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoBasic support for macros with explicit arguments.
Rafael Espindola [Sun, 5 Jun 2011 02:43:45 +0000 (02:43 +0000)]
Basic support for macros with explicit arguments.

We still don't handle

* default values
* :req
* :vararg
* \()

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132656 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoProduce an undefined reference to _GLOBAL_OFFSET_TABLE_ if we have a
Rafael Espindola [Sun, 5 Jun 2011 01:20:06 +0000 (01:20 +0000)]
Produce an undefined reference to _GLOBAL_OFFSET_TABLE_ if we have a
VK_GOTOFF reloc. This matches as' behavior, but it is not clear why the linker
might need this, so I added a FIXME.

I could test this by duplicating test/MC/ELF/got.s, but it doesn't look
worthwhile.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132655 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFix a test that keeps breaking when allocation orders change.
Jakob Stoklund Olesen [Sat, 4 Jun 2011 23:34:40 +0000 (23:34 +0000)]
Fix a test that keeps breaking when allocation orders change.

Who said FileCheck couldn't handle arbitrarily complex conditions?

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132654 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTypeLegalizer: Add support for passing of vector-promoted types in registers (copyFro...
Nadav Rotem [Sat, 4 Jun 2011 20:58:08 +0000 (20:58 +0000)]
TypeLegalizer: Add support for passing of vector-promoted types in registers (copyFromParts/copyToParts).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132649 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTypeLegalizer: Fix a bug in the promotion of elements of integer vectors.
Nadav Rotem [Sat, 4 Jun 2011 20:32:01 +0000 (20:32 +0000)]
TypeLegalizer: Fix a bug in the promotion of elements of integer vectors.
(only happens when using the -promote-elements option).

The correct legalization order is to first try to promote element. Next, we try
to widen vectors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132648 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoRefactor parsing of variable names (ie., %foo and @foo) since they have the same
Nick Lewycky [Sat, 4 Jun 2011 18:16:26 +0000 (18:16 +0000)]
Refactor parsing of variable names (ie., %foo and @foo) since they have the same
rules. Also refactor "read string until quote" into its own function. No
functionality change!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132645 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd support for @GOTPTOFF in i386 mode.
Nick Lewycky [Sat, 4 Jun 2011 17:38:07 +0000 (17:38 +0000)]
Add support for @GOTPTOFF in i386 mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132643 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoCast unused parameter to void.
Hans Wennborg [Sat, 4 Jun 2011 16:00:19 +0000 (16:00 +0000)]
Cast unused parameter to void.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132641 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoIf the block that we're threading through is jumped to by an indirect branch,
Bill Wendling [Sat, 4 Jun 2011 09:42:04 +0000 (09:42 +0000)]
If the block that we're threading through is jumped to by an indirect branch,
then we don't want to set the destination in the indirect branch to the
destination. This is because the indirect branch needs its destinations to have
had their block addresses taken. This isn't so of the new critical edge that's
split during this process. If it turns out that the destination block has only
one predecessor, and that being a BB with an indirect branch, then it won't be
marked as 'used' and may be removed.
PR10072

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132638 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoDrop a RecordKeeper reference that wasn't necessary.
Jakob Stoklund Olesen [Sat, 4 Jun 2011 07:49:55 +0000 (07:49 +0000)]
Drop a RecordKeeper reference that wasn't necessary.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132636 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd a testcase to demonstrate the problem where phi translation is
Dan Gohman [Sat, 4 Jun 2011 07:05:05 +0000 (07:05 +0000)]
Add a testcase to demonstrate the problem where phi translation is
ignored for clobbering partial-alias loads.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132633 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoReapply r131781, now that the GVN bug with partially-aliasing loads
Dan Gohman [Sat, 4 Jun 2011 06:50:18 +0000 (06:50 +0000)]
Reapply r131781, now that the GVN bug with partially-aliasing loads
is disabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132632 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoDisable the main feature of 130180, the elimination of loads that are
Dan Gohman [Sat, 4 Jun 2011 06:48:50 +0000 (06:48 +0000)]
Disable the main feature of 130180, the elimination of loads that are
redundant with partially-aliasing loads.

When computing what portion of a clobbering load value is needed,
it doesn't consider phi-translation which may have occurred
between the clobbing load and the redundant load.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132631 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoRemove this test, which should have been reverted along with r131781.
Dan Gohman [Sat, 4 Jun 2011 06:21:23 +0000 (06:21 +0000)]
Remove this test, which should have been reverted along with r131781.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132628 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoUse processor name in the target triple. This fixes PowerPC64 to generate
Roman Divacky [Sat, 4 Jun 2011 06:15:42 +0000 (06:15 +0000)]
Use processor name in the target triple. This fixes PowerPC64 to generate
64bit code for example.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132627 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoRevert r131781 again. Apparently there is more going on here.
Dan Gohman [Sat, 4 Jun 2011 05:11:22 +0000 (05:11 +0000)]
Revert r131781 again. Apparently there is more going on here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132625 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoSilence compiler warnings.
Jakob Stoklund Olesen [Sat, 4 Jun 2011 05:09:36 +0000 (05:09 +0000)]
Silence compiler warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132624 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTeach TableGen to evaluate DAG expressions as set operations.
Jakob Stoklund Olesen [Sat, 4 Jun 2011 04:11:37 +0000 (04:11 +0000)]
Teach TableGen to evaluate DAG expressions as set operations.

A TableGen backend can define how certain classes can be expanded into
ordered sets of defs, typically by evaluating a specific field in the
record. The SetTheory class can then evaluate DAG expressions that refer
to these named sets.

A number of standard set and list operations are predefined, and the
backend can add more specialized operators if needed. The -print-sets
backend is used by SetTheory.td to provide examples.

This is intended to simplify how register classes are defined:

  def GR32_NOSP : RegisterClass<"X86", [i32], 32, (sub GR32, ESP)>;

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132621 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFold assert-only-used variable into the assert.
Nick Lewycky [Sat, 4 Jun 2011 02:07:10 +0000 (02:07 +0000)]
Fold assert-only-used variable into the assert.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132620 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoMissing include of climits in the new BranchProbability pass.
Andrew Trick [Sat, 4 Jun 2011 01:30:52 +0000 (01:30 +0000)]
Missing include of climits in the new BranchProbability pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132616 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoNew BranchProbabilityInfo analysis. Patch by Jakub Staszak!
Andrew Trick [Sat, 4 Jun 2011 01:16:30 +0000 (01:16 +0000)]
New BranchProbabilityInfo analysis. Patch by Jakub Staszak!

BranchProbabilityInfo provides an interface for IR passes to query the
likelihood that control follows a CFG edge. This patch provides an
initial implementation of static branch predication that will populate
BranchProbabilityInfo for branches with no external profile
information using very simple heuristics. It currently isn't hooked up
to any external profile data, so static prediction does all the work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132613 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoReapply r131781 (revert r131809), now that some BasicAA shortcomings
Dan Gohman [Sat, 4 Jun 2011 00:46:31 +0000 (00:46 +0000)]
Reapply r131781 (revert r131809), now that some BasicAA shortcomings
it exposed are fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132611 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFix BasicAA's recursion detection so that it doesn't pessimize
Dan Gohman [Sat, 4 Jun 2011 00:31:50 +0000 (00:31 +0000)]
Fix BasicAA's recursion detection so that it doesn't pessimize
queries in the case of a DAG, where a query reaches a node
visited earlier, but it's not on a cycle. This avoids
MayAlias results in cases where BasicAA is expected to
return MustAlias or PartialAlias in order to protect TBAA.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132609 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoReapply 132424 with fixes. This fixes PR10068.
Stuart Hastings [Fri, 3 Jun 2011 23:53:54 +0000 (23:53 +0000)]
Reapply 132424 with fixes.  This fixes PR10068.
rdar://problem/5993888

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132606 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFix some tests that depend on register allocation.
Jakob Stoklund Olesen [Fri, 3 Jun 2011 22:45:21 +0000 (22:45 +0000)]
Fix some tests that depend on register allocation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132602 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoBlackfin always uses a reserved call frame.
Jakob Stoklund Olesen [Fri, 3 Jun 2011 22:45:18 +0000 (22:45 +0000)]
Blackfin always uses a reserved call frame.

Materializing the stack pointer update before a call requires a scratch
register that may not be available.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132601 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAnother possible bug. Stopgap until we can autogenerate tables and
Eric Christopher [Fri, 3 Jun 2011 22:09:12 +0000 (22:09 +0000)]
Another possible bug. Stopgap until we can autogenerate tables and
constraint lengths.

Part of rdar://9037836 and rdar://9119939

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132598 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTestcase for llvm-gcc commit r132591.
Eric Christopher [Fri, 3 Jun 2011 20:56:31 +0000 (20:56 +0000)]
Testcase for llvm-gcc commit r132591.

Part of rdar://9037836 and rdar://9119939

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132592 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFix an off by one error.
Eric Christopher [Fri, 3 Jun 2011 20:44:52 +0000 (20:44 +0000)]
Fix an off by one error.

Part of rdar://9037836 and rdar://9119939

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132590 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoRemove testcase. It's problematic for the buildbots.
Bill Wendling [Fri, 3 Jun 2011 20:43:58 +0000 (20:43 +0000)]
Remove testcase. It's problematic for the buildbots.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132589 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoSwitch AllocationOrder to using RegisterClassInfo instead of a BitVector
Jakob Stoklund Olesen [Fri, 3 Jun 2011 20:34:53 +0000 (20:34 +0000)]
Switch AllocationOrder to using RegisterClassInfo instead of a BitVector
of reserved registers.

Use RegisterClassInfo in RABasic as well. This slightly changes som
allocation orders because RegisterClassInfo puts CSR aliases last.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132581 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoPreserve the original ordering when a CSR has multiple aliases.
Jakob Stoklund Olesen [Fri, 3 Jun 2011 20:34:50 +0000 (20:34 +0000)]
Preserve the original ordering when a CSR has multiple aliases.

Previously, these aliases would be ordered alphabetically. (BH, BL)

Print out the computed allocation orders.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132580 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoWhen merging MustAlias and PartialAlias, chose PartialAlias instead
Dan Gohman [Fri, 3 Jun 2011 20:17:36 +0000 (20:17 +0000)]
When merging MustAlias and PartialAlias, chose PartialAlias instead
of conservatively choosing MayAlias.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132579 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoUse IRBuilder, preserve line numbers.
Devang Patel [Fri, 3 Jun 2011 19:46:19 +0000 (19:46 +0000)]
Use IRBuilder, preserve line numbers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132578 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAccomodate front-ends which use private instead of internal here.
Bill Wendling [Fri, 3 Jun 2011 19:21:05 +0000 (19:21 +0000)]
Accomodate front-ends which use private instead of internal here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132575 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoMake the Uv constraint a memory operand. This doesn't solve the
Eric Christopher [Fri, 3 Jun 2011 17:24:37 +0000 (17:24 +0000)]
Make the Uv constraint a memory operand. This doesn't solve the
addressing mode problem mentioned in r132559.

Backend part of rdar://9037836 and part of rdar://9119939

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132561 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd a TODO about memory operands.
Eric Christopher [Fri, 3 Jun 2011 17:21:23 +0000 (17:21 +0000)]
Add a TODO about memory operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132559 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTest commit.
Hans Wennborg [Fri, 3 Jun 2011 17:15:37 +0000 (17:15 +0000)]
Test commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132558 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoA typedef's context is not the same as type's context. It is the context of typedef...
Devang Patel [Fri, 3 Jun 2011 17:04:51 +0000 (17:04 +0000)]
A typedef's context is not the same as type's context. It is the context of typedef decl itself. Use extra parameter to communicate this to DIBuilder.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132556 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoRevert name change from r132533. Lower case naming was intended per style guidelines.
Chad Rosier [Fri, 3 Jun 2011 17:02:19 +0000 (17:02 +0000)]
Revert name change from r132533.  Lower case naming was intended per style guidelines.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132555 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFix wrong usages of CTR/MCTR where CTR8/MCTR8 was meant.
Roman Divacky [Fri, 3 Jun 2011 15:47:49 +0000 (15:47 +0000)]
Fix wrong usages of CTR/MCTR where CTR8/MCTR8 was meant.

- Check for MTCTR8 in addition to MTCTR when looking up a hazard.

- When lowering an indirect call use CTR8 when targeting 64bit.

- Introduce BCTR8 that uses CTR8 and use it on 64bit when expanding ISD::BRIND.

The last change fixes PR8487. With those changes, we are able to compile a
running "ls" and "sh" on FreeBSD/PowerPC64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132552 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agosinged int causes signed extension, which contradicts the intention to pick up
Zhongxing Xu [Fri, 3 Jun 2011 08:29:51 +0000 (08:29 +0000)]
singed int causes signed extension, which contradicts the intention to pick up
integers with high 32 bits being zero.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132538 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoRework the logic to not rely on undefined behaviour (1LL << 64). Also simplify.
Nick Lewycky [Fri, 3 Jun 2011 08:25:39 +0000 (08:25 +0000)]
Rework the logic to not rely on undefined behaviour (1LL << 64). Also simplify.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132537 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoBail on unswitching a switch statement for a case with a critical edge. We name
Nick Lewycky [Fri, 3 Jun 2011 06:27:15 +0000 (06:27 +0000)]
Bail on unswitching a switch statement for a case with a critical edge. We name
which edge to split by pred/succ pair, which means that we can end up splitting
the wrong edge (by case value) in the switch statement entirely. Fixes PR10031!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132535 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoWhitespace and other cleanup. Functionallity unchanged.
Chad Rosier [Fri, 3 Jun 2011 05:09:12 +0000 (05:09 +0000)]
Whitespace and other cleanup.  Functionallity unchanged.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132533 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTreat Win64 EH as a DWARF EH scheme. For GCC-style exceptions, the layout of
Charles Davis [Fri, 3 Jun 2011 05:09:12 +0000 (05:09 +0000)]
Treat Win64 EH as a DWARF EH scheme. For GCC-style exceptions, the layout of
the handler's data area is similar to a DWARF-format LSDA. (It is, in fact,
a 32-bit pointer to the personality routine followed by the DWARF LSDA.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132532 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoCorrections and additional information for "Creating and LLVM Project"
Andrew Trick [Fri, 3 Jun 2011 02:20:48 +0000 (02:20 +0000)]
Corrections and additional information for "Creating and LLVM Project"
documentation. This should now reflect the current state of LLVM Makefiles.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132527 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agowhitespace
Andrew Trick [Fri, 3 Jun 2011 02:16:53 +0000 (02:16 +0000)]
whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132526 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd ARM fast-isel support for materializing the address of a global in cases where...
Eli Friedman [Fri, 3 Jun 2011 01:13:19 +0000 (01:13 +0000)]
Add ARM fast-isel support for materializing the address of a global in cases where the global uses an indirect symbol.

rdar://9431157

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132522 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoBasic PassManager diagnostics.
Andrew Trick [Fri, 3 Jun 2011 00:48:58 +0000 (00:48 +0000)]
Basic PassManager diagnostics.

Added asserts whenever attempting to use a potentially
uninitialized pass. This helps people trying to develop a new pass and
people trying to understand the bug reports filed by the former people.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132520 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agowhitespace
Andrew Trick [Fri, 3 Jun 2011 00:44:32 +0000 (00:44 +0000)]
whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132519 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTest case pasto (failed when run with IR verifier).
Andrew Trick [Thu, 2 Jun 2011 23:57:27 +0000 (23:57 +0000)]
Test case pasto (failed when run with IR verifier).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132516 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAvoid calling TRI->getAllocatableSet in RAFast.
Jakob Stoklund Olesen [Thu, 2 Jun 2011 23:41:40 +0000 (23:41 +0000)]
Avoid calling TRI->getAllocatableSet in RAFast.

When compiling a program with lots of small functions like
483.xalancbmk, this makes RAFast 11% faster.

Add some comments to clarify the difference between unallocatable and
reserved registers. It's quite subtle.

The fast register allocator depends on EFLAGS' not being allocatable on
x86. That way it can completely avoid tracking liveness, and it won't
mind when there are multiple uses of a single def.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132514 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoHave LowerOperandForConstraint handle multiple character constraints.
Eric Christopher [Thu, 2 Jun 2011 23:16:42 +0000 (23:16 +0000)]
Have LowerOperandForConstraint handle multiple character constraints.

Part of rdar://9119939

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132510 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFlag unallocatable register classes instead of giving them empty
Jakob Stoklund Olesen [Thu, 2 Jun 2011 23:07:24 +0000 (23:07 +0000)]
Flag unallocatable register classes instead of giving them empty
allocation orders.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132509 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoMake it possible to have unallocatable register classes.
Jakob Stoklund Olesen [Thu, 2 Jun 2011 23:07:20 +0000 (23:07 +0000)]
Make it possible to have unallocatable register classes.

Some register classes are only used for instruction operand constraints.
They should never be used for virtual registers. Previously, those
register classes were given an empty allocation order, but now you can
say 'let isAllocatable=0' in the register class definition.

TableGen calculates if a register is part of any allocatable register
class, and makes that information available in TargetRegisterDesc::inAllocatableClass.

The goal here is to eliminate use cases for overriding allocation_order_*
methods.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132508 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoPreserve line number information while converting Invoke into a Call.
Devang Patel [Thu, 2 Jun 2011 22:46:58 +0000 (22:46 +0000)]
Preserve line number information while converting Invoke into a Call.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132505 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoThis should have been a C++ testcase.
Bill Wendling [Thu, 2 Jun 2011 22:26:15 +0000 (22:26 +0000)]
This should have been a C++ testcase.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132504 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoJust use a SmallVector.
Jakob Stoklund Olesen [Thu, 2 Jun 2011 22:22:43 +0000 (22:22 +0000)]
Just use a SmallVector.

I was confused whether new uint8_t[] would zero-initialize the returned
array, and it seems that so is gcc-4.0.

This should fix the test failures on darwin 9.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132500 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTestcase for r132493.
Bill Wendling [Thu, 2 Jun 2011 22:12:42 +0000 (22:12 +0000)]
Testcase for r132493.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132495 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoUpdate for r132493 change.
Bill Wendling [Thu, 2 Jun 2011 22:11:49 +0000 (22:11 +0000)]
Update for r132493 change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132494 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoRemove dead code.
Devang Patel [Thu, 2 Jun 2011 21:31:00 +0000 (21:31 +0000)]
Remove dead code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132488 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoUpdate DBG_VALUEs while breaking anti dependencies.
Devang Patel [Thu, 2 Jun 2011 21:26:52 +0000 (21:26 +0000)]
Update DBG_VALUEs while breaking anti dependencies.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132487 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoFix encoding for VEXTdf.
Tanya Lattner [Thu, 2 Jun 2011 21:25:24 +0000 (21:25 +0000)]
Fix encoding for VEXTdf.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132486 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoPR10067: Add missing safety check to call return transformation in MemCpyOpt::process...
Eli Friedman [Thu, 2 Jun 2011 21:24:42 +0000 (21:24 +0000)]
PR10067: Add missing safety check to call return transformation in MemCpyOpt::processStore.  If something accesses the dest of the "copy" between the call and the copy, the performCallSlotOptzn transformation is not valid.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132485 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoDuring post RA scheduling, do not try to chase reg defs. to preserve DBG_VALUEs....
Devang Patel [Thu, 2 Jun 2011 20:07:12 +0000 (20:07 +0000)]
During post RA scheduling, do not try to chase reg defs. to preserve DBG_VALUEs. This approach has several downsides, for example, it does not work when dbg value is a constant integer, it does not work if reg is defined more than once, it places end of debug value range markers in the wrong place. It even causes misleading incorrect debug info when duplicate DBG_VALUE instructions point to same reg def.

Instead, use simpler approach and let DBG_VALUE follow its predecessor instruction. After live debug value analysis pass, all DBG_VALUE instruction are placed at the right place. Thanks Jakob for the hint!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132483 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd test for PR10068.
Rafael Espindola [Thu, 2 Jun 2011 20:02:48 +0000 (20:02 +0000)]
Add test for PR10068.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132482 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoRevert 132424 to fix PR10068.
Rafael Espindola [Thu, 2 Jun 2011 19:57:47 +0000 (19:57 +0000)]
Revert 132424 to fix PR10068.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132479 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAndy pointed out a dumb omission in this test case. Thanks Andy!
Stuart Hastings [Thu, 2 Jun 2011 19:26:49 +0000 (19:26 +0000)]
Andy pointed out a dumb omission in this test case.  Thanks Andy!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132477 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd a new parse hint for multi-letter constraints in inline asm.
Eric Christopher [Thu, 2 Jun 2011 19:26:37 +0000 (19:26 +0000)]
Add a new parse hint for multi-letter constraints in inline asm.

Testcase will come when we use it.

Part of rdar://9119939

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132476 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoJakob pointed out a dumb omission in this test case. Thanks Jakob!
Stuart Hastings [Thu, 2 Jun 2011 18:44:05 +0000 (18:44 +0000)]
Jakob pointed out a dumb omission in this test case.  Thanks Jakob!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132472 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoUse RegisterClassInfo::getOrder in RAFast.
Jakob Stoklund Olesen [Thu, 2 Jun 2011 18:35:30 +0000 (18:35 +0000)]
Use RegisterClassInfo::getOrder in RAFast.

This saves two virtual function calls and an Allocatable BitVector test,
making RAFast run 2% faster.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132471 91177308-0d34-0410-b5e6-96231b3b80d8

13 years ago.cfi directive register parsing flexibility.
Jim Grosbach [Thu, 2 Jun 2011 17:14:04 +0000 (17:14 +0000)]
.cfi directive register parsing flexibility.

Parsing a register name/number for .cfi directives can't assume that a
register name starts with a '%' token. Be more flexible and check for a
register number instead. Still unlikely to be perfect, but it allows us
to parse both plain identifiers as register names and integers as register
numbers, which is what we're wanting to support at this point.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132466 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoOmit unnecessary stack copy when x87 input is a load.
Stuart Hastings [Thu, 2 Jun 2011 15:57:11 +0000 (15:57 +0000)]
Omit unnecessary stack copy when x87 input is a load.
rdar://problem/6373334

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132458 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoStart with a zeroed CSRNum map.
Benjamin Kramer [Thu, 2 Jun 2011 12:07:44 +0000 (12:07 +0000)]
Start with a zeroed CSRNum map.

Found by valgrind.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132457 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoInitialize members to fix problem found by valgrind.
Jakob Stoklund Olesen [Thu, 2 Jun 2011 05:43:49 +0000 (05:43 +0000)]
Initialize members to fix problem found by valgrind.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132456 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoUse TRI::has{Sub,Super}ClassEq() where possible.
Jakob Stoklund Olesen [Thu, 2 Jun 2011 05:43:46 +0000 (05:43 +0000)]
Use TRI::has{Sub,Super}ClassEq() where possible.

No functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132455 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTweak testcase for ARM bot. rdar://problem/5993888
Stuart Hastings [Thu, 2 Jun 2011 05:05:39 +0000 (05:05 +0000)]
Tweak testcase for ARM bot.  rdar://problem/5993888

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132454 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoDon't hardcode the %reg format in the streamer.
Rafael Espindola [Thu, 2 Jun 2011 02:34:55 +0000 (02:34 +0000)]
Don't hardcode the %reg format in the streamer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132451 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoAdd a RegisterClassInfo class that lazily caches information about
Jakob Stoklund Olesen [Thu, 2 Jun 2011 02:19:35 +0000 (02:19 +0000)]
Add a RegisterClassInfo class that lazily caches information about
register classes.

It provides information for each register class that cannot be
determined statically, like:

- The number of allocatable registers in a class after filtering out the
  reserved and invalid registers.

- The preferred allocation order with registers that overlap callee-saved
  registers last.

- The last callee-saved register that overlaps a given physical register.

This information usually doesn't change between functions, so it is
reused for compiling multiple functions when possible.  The many
possible combinations of reserved and callee saves registers makes it
unfeasible to compute this information statically in TableGen.

Use RegisterClassInfo to count available registers in various heuristics
in SimpleRegisterCoalescing, making the pass run 4% faster.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132450 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoDetect FI|cst pattern in MipsDAGToDAGISel::SelectAddr. Patch by Sasa Stankovic.
Akira Hatanaka [Thu, 2 Jun 2011 01:03:14 +0000 (01:03 +0000)]
Detect FI|cst pattern in MipsDAGToDAGISel::SelectAddr. Patch by Sasa Stankovic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132448 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTest case for r132444.
Akira Hatanaka [Thu, 2 Jun 2011 00:25:53 +0000 (00:25 +0000)]
Test case for r132444.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132445 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoCustom-lower FRAMEADDR. Patch by Sasa Stankovic.
Akira Hatanaka [Thu, 2 Jun 2011 00:24:44 +0000 (00:24 +0000)]
Custom-lower FRAMEADDR. Patch by Sasa Stankovic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132444 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoWhen marking a block as being unanalyzable, use "Clobber" on the terminator instead...
Eli Friedman [Thu, 2 Jun 2011 00:08:52 +0000 (00:08 +0000)]
When marking a block as being unanalyzable, use "Clobber" on the terminator instead of the first instruction in the block.  This is a bit of a hack; "Clobber" isn't really the right marking in the first place.  memdep doesn't really have any way of properly expressing "unanalyzable" at the moment.  Using it on the terminator is much less ambiguous than using it on an arbitrary instruction, though.

In the given testcase, the "Clobber" was pointing to a load, and GVN was incorrectly assuming that meant that the "Clobber" load overlapped the load being analyzed (when they are actually unrelated).

The included testcase tests both this commit and r132434.

Part two of rdar://9429882.  (r132434 was mislabeled.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132442 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoTypos.
Chad Rosier [Wed, 1 Jun 2011 23:32:40 +0000 (23:32 +0000)]
Typos.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132437 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoIn MemoryDependenceAnalysis::getNonLocalPointerDepFromBB, if a given block is is...
Eli Friedman [Wed, 1 Jun 2011 23:16:53 +0000 (23:16 +0000)]
In MemoryDependenceAnalysis::getNonLocalPointerDepFromBB, if a given block is is deemed unanalyzable (and we execute one of the "goto PredTranslationFailure" statements), make sure we don't put information about the predecessors of that block into the returned data structures; this can lead to, among other things, extraneous results (which will confuse passes using memdep).  Fixes an assert in GVN compiling ruby. Part of rdar://problem/9521954 .

Testcase coming up soon.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132434 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoA DBG_VALUE that truncates a range does not start another dbg value range.
Devang Patel [Wed, 1 Jun 2011 23:00:17 +0000 (23:00 +0000)]
A DBG_VALUE that truncates a range does not start another dbg value range.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132433 91177308-0d34-0410-b5e6-96231b3b80d8

13 years agoDo not drop constant values when a variable's content is described using .debug_loc...
Devang Patel [Wed, 1 Jun 2011 22:03:25 +0000 (22:03 +0000)]
Do not drop constant values when a variable's content is described using .debug_loc entries.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132427 91177308-0d34-0410-b5e6-96231b3b80d8