oota-llvm.git
14 years agoDragonegg will be released along side llvm-2.7.
Duncan Sands [Wed, 21 Apr 2010 13:51:48 +0000 (13:51 +0000)]
Dragonegg will be released along side llvm-2.7.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@102001 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agofinal hacking for tonight, still more to go.
Chris Lattner [Wed, 21 Apr 2010 06:42:24 +0000 (06:42 +0000)]
final hacking for tonight, still more to go.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101995 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agocontinue the process of detangling this.
Chris Lattner [Wed, 21 Apr 2010 06:23:40 +0000 (06:23 +0000)]
continue the process of detangling this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101992 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agorough pass moving stuff into relevant sections, still much
Chris Lattner [Wed, 21 Apr 2010 05:17:40 +0000 (05:17 +0000)]
rough pass moving stuff into relevant sections, still much
editing to do.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101987 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoremove ldc, rubinious, macruby, icedtea, llvm-lua, which
Chris Lattner [Wed, 21 Apr 2010 04:28:21 +0000 (04:28 +0000)]
remove ldc, rubinious, macruby, icedtea, llvm-lua, which
don't have updates for 2.7.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101985 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoImplement -disable-non-leaf-fp-elim which disable frame pointer elimination
Evan Cheng [Wed, 21 Apr 2010 03:18:23 +0000 (03:18 +0000)]
Implement -disable-non-leaf-fp-elim which disable frame pointer elimination
optimization for non-leaf functions. This will be hooked up to gcc's
-momit-leaf-frame-pointer option. rdar://7886181

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101984 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoisel (i32 anyext i16) as insert_subreg when 16-bit ops are being promoted.
Evan Cheng [Wed, 21 Apr 2010 01:47:12 +0000 (01:47 +0000)]
isel (i32 anyext i16) as insert_subreg when 16-bit ops are being promoted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101979 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoTrim include.
Evan Cheng [Wed, 21 Apr 2010 01:39:06 +0000 (01:39 +0000)]
Trim include.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101978 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd more const qualifiers on TargetMachine and friends.
Dan Gohman [Wed, 21 Apr 2010 01:34:56 +0000 (01:34 +0000)]
Add more const qualifiers on TargetMachine and friends.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101977 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoUpdate CMakeLists.txt.
Dan Gohman [Wed, 21 Apr 2010 01:32:29 +0000 (01:32 +0000)]
Update CMakeLists.txt.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101976 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMove several SelectionDAG-independent utility functions out of the
Dan Gohman [Wed, 21 Apr 2010 01:22:34 +0000 (01:22 +0000)]
Move several SelectionDAG-independent utility functions out of the
SelectionDAG directory and into a new Analysis.cpp file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101975 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoThumb instructions which have reglist operands at the end and predicate operands
Johnny Chen [Wed, 21 Apr 2010 01:01:19 +0000 (01:01 +0000)]
Thumb instructions which have reglist operands at the end and predicate operands
before reglist were not properly handled with respect to IT Block.  Fix that by
creating a new method ARMBasicMCBuilder::DoPredicateOperands() used by those
instructions for disassembly.  Add a test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101974 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoImplement (but don't enable) PR6724 and rdar://6295824. In short,
Chris Lattner [Wed, 21 Apr 2010 00:47:40 +0000 (00:47 +0000)]
Implement (but don't enable) PR6724 and rdar://6295824.  In short,
we have RefreshCallGraph detect when a function pass devirtualizes
a call, and have CGSCCPassMgr iterate (up to a count) when this
happens.  This allows (in the example) GVN to devirtualize the
call in foo, then the inliner to inline it away.

This is not currently enabled because I haven't done any analysis
on the (potentially substantial) code size or performance impact of
doing this, and guess what, it exposes callgraph updating bugs in
various passes.  This is progress though, and you can play with it
by passing -max-cg-scc-iterations=5 to opt.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101973 91177308-0d34-0410-b5e6-96231b3b80d8

14 years ago- Clean up some crappy code which deals with coalescing of copies which look at
Evan Cheng [Wed, 21 Apr 2010 00:44:22 +0000 (00:44 +0000)]
- Clean up some crappy code which deals with coalescing of copies which look at
  extract_subreg / insert_subreg, etc.
- Add support for more aggressive insert_subreg coalescing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101971 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRevert r101471. For tight recursive functions which have multiple
Dan Gohman [Wed, 21 Apr 2010 00:43:30 +0000 (00:43 +0000)]
Revert r101471. For tight recursive functions which have multiple
recursive callsites, inlining can reduce the number of calls by
exponential factors, as it does in
MultiSource/Benchmarks/Olden/treeadd. More involved heuristics
will be needed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101969 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoHandle a displacement location in 64-bit as an RIP-relative displacement. It
Bill Wendling [Wed, 21 Apr 2010 00:34:04 +0000 (00:34 +0000)]
Handle a displacement location in 64-bit as an RIP-relative displacement. It
fixes a bug (<rdar://problem/7880900>) in the JIT. This code wouldn't work:

target triple = "x86_64-apple-darwin"

define double @func(double %a) {
  %tmp1 = fmul double %a, 5.000000e-01            ; <double> [#uses=1]
  ret double %tmp1
}

define i32 @main() nounwind {
  %1 = call double @func(double 4.770000e-04) ; <i64> [#uses=0]
  ret i32 0
}

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101965 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRewrite machine cse to avoid recursion.
Evan Cheng [Wed, 21 Apr 2010 00:21:07 +0000 (00:21 +0000)]
Rewrite machine cse to avoid recursion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101964 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd another variant of this test which found a place where
Dan Gohman [Wed, 21 Apr 2010 00:19:28 +0000 (00:19 +0000)]
Add another variant of this test which found a place where
CodeGen's ComputeMaskedBits was being over-conservative when computing
bits for an ADD.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101963 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoteach the x86 address matching stuff to handle
Chris Lattner [Tue, 20 Apr 2010 23:18:40 +0000 (23:18 +0000)]
teach the x86 address matching stuff to handle
(shl (or x,c), 3) the same as (shl (add x, c), 3)
when x doesn't have any bits from c set.

This finishes off PR1135.  Before we compiled the block to:
to:

LBB0_3:                                 ## %bb
cmpb $4, %dl
sete %dl
addb %dl, %cl
movb %cl, %dl
shlb $2, %dl
addb %r8b, %dl
shlb $2, %dl
movzbl %dl, %edx
movl %esi, (%rdi,%rdx,4)
leaq 2(%rdx), %r9
movl %esi, (%rdi,%r9,4)
leaq 1(%rdx), %r9
movl %esi, (%rdi,%r9,4)
addq $3, %rdx
movl %esi, (%rdi,%rdx,4)
incb %r8b
decb %al
movb %r8b, %dl
jne LBB0_1

Now we produce:

LBB0_3:                                 ## %bb
cmpb $4, %dl
sete %dl
addb %dl, %cl
movb %cl, %dl
shlb $2, %dl
addb %r8b, %dl
shlb $2, %dl
movzbl %dl, %edx
movl %esi, (%rdi,%rdx,4)
movl %esi, 8(%rdi,%rdx,4)
movl %esi, 4(%rdi,%rdx,4)
movl %esi, 12(%rdi,%rdx,4)
incb %r8b
decb %al
movb %r8b, %dl
jne LBB0_1

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101958 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoBecause of the EMMS problem, right now we have to support
Dale Johannesen [Tue, 20 Apr 2010 22:34:09 +0000 (22:34 +0000)]
Because of the EMMS problem, right now we have to support
user-defined operations that use MMX register types, but
the compiler shouldn't generate them on its own.  This adds
a Synthesizable abstraction to represent this, and changes
the vector widening computation so it won't produce MMX types.
(The motivation is to remove noise from the ABI compatibility
part of the gcc test suite, which has some breakage right now.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101951 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRename ValueMapTy as ValueToValueMapTy to clearly indicate that this has no replation...
Devang Patel [Tue, 20 Apr 2010 22:24:18 +0000 (22:24 +0000)]
Rename ValueMapTy as ValueToValueMapTy to clearly indicate that this has no replationship with ADT/ValueMap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101950 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoThere is no need to install ValueMapper.h header.
Devang Patel [Tue, 20 Apr 2010 22:18:31 +0000 (22:18 +0000)]
There is no need to install ValueMapper.h header.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101949 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoBetter error-handling of getBitFieldInvMask() where msb < lsb (encoding error),
Johnny Chen [Tue, 20 Apr 2010 21:29:28 +0000 (21:29 +0000)]
Better error-handling of getBitFieldInvMask() where msb < lsb (encoding error),
instead of just asserting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101942 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoI plan to release a version of dragonegg based on llvm-2.7 shortly
Duncan Sands [Tue, 20 Apr 2010 19:40:58 +0000 (19:40 +0000)]
I plan to release a version of dragonegg based on llvm-2.7 shortly
after the llvm-2.7 release.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101934 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoWhen MachineLICM is hoisting a physical register after regalloc, make sure the
Jakob Stoklund Olesen [Tue, 20 Apr 2010 18:45:47 +0000 (18:45 +0000)]
When MachineLICM is hoisting a physical register after regalloc, make sure the
register is not killed in the loop.

This fixes 188.ammp on ARM where the post-ra scheduler would grab a register
that looked available but wasn't.

A testcase would be huge and fragile, sorry.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101930 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoWhen doing Thumb disassembly, there's no need to consider t2ADDrSPi12/t2SUBrSPi12,
Johnny Chen [Tue, 20 Apr 2010 18:45:24 +0000 (18:45 +0000)]
When doing Thumb disassembly, there's no need to consider t2ADDrSPi12/t2SUBrSPi12,
as their generic counterparts t2ADDri12/t2SUBri12 should suffice.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101929 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMove CodeGen/X86/2010-04-19-DAGCombineCrash.ll into CodeGen/X86/crash.ll. Also
Bill Wendling [Tue, 20 Apr 2010 18:14:47 +0000 (18:14 +0000)]
Move CodeGen/X86/2010-04-19-DAGCombineCrash.ll into CodeGen/X86/crash.ll. Also
reduce.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101925 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFor t2LDRT, t2LDRBT, t2LDRHT, t2LDRSBT, and t2LDRSHT, if Rn(Inst{19-16})=='1111',
Johnny Chen [Tue, 20 Apr 2010 17:28:50 +0000 (17:28 +0000)]
For t2LDRT, t2LDRBT, t2LDRHT, t2LDRSBT, and t2LDRSHT, if Rn(Inst{19-16})=='1111',
transform the Opcode to the corresponding t2LDR*pci counterpart.

Ref: A8.6.86 LDRT, A8.6.65 LDRBT, A8.6.77 LDRHT, A8.6.81 LDRSBT, A8.6.85 LDRSHT

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101915 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoTypo.
Evan Cheng [Tue, 20 Apr 2010 17:27:38 +0000 (17:27 +0000)]
Typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101914 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd RUN:
Devang Patel [Tue, 20 Apr 2010 17:20:10 +0000 (17:20 +0000)]
Add RUN:

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101913 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSink the CopyToExportRegsIfNeeded calls out of SelectionDAGISel
Dan Gohman [Tue, 20 Apr 2010 15:03:56 +0000 (15:03 +0000)]
Sink the CopyToExportRegsIfNeeded calls out of SelectionDAGISel
into SelectionDAGBuilder. This avoids a separate pass over the
instructions, and has the side effect of providing debug location
information to the copy.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101906 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoDon't send PHI nodes down to SelectionDAGBuilder of FastISel, since
Dan Gohman [Tue, 20 Apr 2010 15:00:41 +0000 (15:00 +0000)]
Don't send PHI nodes down to SelectionDAGBuilder of FastISel, since
they end up doing nothing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101904 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoDocument that TargetRegisterInfo::contains does not cover virtual registers.
Dan Gohman [Tue, 20 Apr 2010 14:51:20 +0000 (14:51 +0000)]
Document that TargetRegisterInfo::contains does not cover virtual registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101903 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSink this use_empty() check into isUsedOutsideOfDefiningBlock.
Dan Gohman [Tue, 20 Apr 2010 14:50:13 +0000 (14:50 +0000)]
Sink this use_empty() check into isUsedOutsideOfDefiningBlock.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101902 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoIf a PHI node somehow has debug info, propogate it to the MachineInstr PHI.
Dan Gohman [Tue, 20 Apr 2010 14:48:02 +0000 (14:48 +0000)]
If a PHI node somehow has debug info, propogate it to the MachineInstr PHI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101901 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoDon't iterate through the whole block just to find the PHI nodes.
Dan Gohman [Tue, 20 Apr 2010 14:46:25 +0000 (14:46 +0000)]
Don't iterate through the whole block just to find the PHI nodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101900 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agouse abstract accessors to CallInst
Gabor Greif [Tue, 20 Apr 2010 13:13:04 +0000 (13:13 +0000)]
use abstract accessors to CallInst

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101899 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoPR6880: Don't dereference CallsExternalNode if it's NULL.
Benjamin Kramer [Tue, 20 Apr 2010 12:16:50 +0000 (12:16 +0000)]
PR6880: Don't dereference CallsExternalNode if it's NULL.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101897 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix -Wcast-qual warning.
Benjamin Kramer [Tue, 20 Apr 2010 11:50:37 +0000 (11:50 +0000)]
Fix -Wcast-qual warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101894 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agostop computing InstImpInputs, it is dead
Chris Lattner [Tue, 20 Apr 2010 06:30:25 +0000 (06:30 +0000)]
stop computing InstImpInputs, it is dead

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101881 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoDAGInstruction::ImpOperands is dead after my recent tblgen work, zap it.
Chris Lattner [Tue, 20 Apr 2010 06:28:43 +0000 (06:28 +0000)]
DAGInstruction::ImpOperands is dead after my recent tblgen work, zap it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101880 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoBill's change in r95336 broke empty aggregates embedded
Chris Lattner [Tue, 20 Apr 2010 06:20:21 +0000 (06:20 +0000)]
Bill's change in r95336 broke empty aggregates embedded
in other types.  fix this by only bumping zero-byte globals
up to a single byte if the *entire global* is zero size,
fixing PR6340.

This also fixes empty arrays etc to be handled correctly,
and only does this on subsection-via-symbols targets (aka
darwin) which is the only place where this matters.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101879 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoteach cellspu how to return i8 and i16 from calls,
Chris Lattner [Tue, 20 Apr 2010 05:36:09 +0000 (05:36 +0000)]
teach cellspu how to return i8 and i16 from calls,
patch by Kalle Raiskila!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101875 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoremove a bunch of ad-hoc code to simplify instructions from
Chris Lattner [Tue, 20 Apr 2010 05:33:18 +0000 (05:33 +0000)]
remove a bunch of ad-hoc code to simplify instructions from
loop unswitch, and use inst simplify instead.  It is more
powerful and less duplication.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101874 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agomove some select simplifications out out instcombine into
Chris Lattner [Tue, 20 Apr 2010 05:32:14 +0000 (05:32 +0000)]
move some select simplifications out out instcombine into
inst simplify.  No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101873 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRewriteLoopBodyWithConditionConstant can end up rewriting the
Chris Lattner [Tue, 20 Apr 2010 05:09:16 +0000 (05:09 +0000)]
RewriteLoopBodyWithConditionConstant can end up rewriting the
condition we're unswitching on.  In this case, don't try to
simplify the second copy of the loop which may be dead or not,
but is probably a constant now.  This fixes PR6879

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101870 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoreapply 'reject forward references to functions whose type don't match'
Chris Lattner [Tue, 20 Apr 2010 04:49:11 +0000 (04:49 +0000)]
reapply 'reject forward references to functions whose type don't match'
now that the testsuite has been updated.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101866 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoDelete a redundant return statement.
Dan Gohman [Tue, 20 Apr 2010 01:58:20 +0000 (01:58 +0000)]
Delete a redundant return statement.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101860 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoThe visitXOR method can return the same SDNode. If so, we don't want to delete
Bill Wendling [Tue, 20 Apr 2010 01:25:01 +0000 (01:25 +0000)]
The visitXOR method can return the same SDNode. If so, we don't want to delete
it as it's not dead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101855 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agodisable optimizations in this directory for MSVC9. This avoids
Chris Lattner [Tue, 20 Apr 2010 01:11:32 +0000 (01:11 +0000)]
disable optimizations in this directory for MSVC9.  This avoids
an optimizer infinite loop on the file, PR6866.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101854 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoBetter error-handling for DisassembleThumb2DPModImm() with 2-reg operands where
Johnny Chen [Tue, 20 Apr 2010 01:01:57 +0000 (01:01 +0000)]
Better error-handling for DisassembleThumb2DPModImm() with 2-reg operands where
d==15 is considered illegal.  Return false instead of assert().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101852 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemove the palignr intrinsics now that we lower them to vector shuffles,
Eric Christopher [Tue, 20 Apr 2010 00:59:54 +0000 (00:59 +0000)]
Remove the palignr intrinsics now that we lower them to vector shuffles,
shifts and null vectors. Autoupgrade these to what we'd lower them to.

Add a testcase to exercise this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101851 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemove this debug output; it isn't that useful, and it's incomplete
Dan Gohman [Tue, 20 Apr 2010 00:56:44 +0000 (00:56 +0000)]
Remove this debug output; it isn't that useful, and it's incomplete
in the case where a basic block is split.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101850 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSink DebugLoc handling out of SelectionDAGISel into FastISel and
Dan Gohman [Tue, 20 Apr 2010 00:48:35 +0000 (00:48 +0000)]
Sink DebugLoc handling out of SelectionDAGISel into FastISel and
SelectionDAGBuilder, where it doesn't have to be as complicated.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101848 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agomake CallGraphNode dtor abort if a node is deleted when there are still
Chris Lattner [Tue, 20 Apr 2010 00:47:34 +0000 (00:47 +0000)]
make CallGraphNode dtor abort if a node is deleted when there are still
references to it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101847 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agomake the inliner do less work for leaf functions.
Chris Lattner [Tue, 20 Apr 2010 00:47:08 +0000 (00:47 +0000)]
make the inliner do less work for leaf functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101846 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix rdar://7879828 - crash in CallGraph, a self host issue.
Chris Lattner [Tue, 20 Apr 2010 00:46:50 +0000 (00:46 +0000)]
Fix rdar://7879828 - crash in CallGraph, a self host issue.
Arg promotion was deleting call graph nodes that still had references
from the 'indirect' CGN.  Like the inliner, it should only delete the
function if all references are gone.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101845 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemove MachineFunction's DefaultDebugLoc member, and make DwarfDebug.cpp
Dan Gohman [Tue, 20 Apr 2010 00:37:27 +0000 (00:37 +0000)]
Remove MachineFunction's DefaultDebugLoc member, and make DwarfDebug.cpp
responsible for figuring out what that's supposed to be on its own.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101844 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoReapply the removal of SelectionDAGISel's BB, with a fix for the case
Dan Gohman [Tue, 20 Apr 2010 00:29:35 +0000 (00:29 +0000)]
Reapply the removal of SelectionDAGISel's BB, with a fix for the case
where multiple blocks are emitted; functions which do this need to return
the new BB so that their callers can stay current.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101843 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix tests for Neon load/store intrinsics to match the i8* types expected by
Bob Wilson [Tue, 20 Apr 2010 00:17:16 +0000 (00:17 +0000)]
Fix tests for Neon load/store intrinsics to match the i8* types expected by
the intrinsics.  The reason for those i8* types is that the intrinsics are
overloaded on the vector type and we don't have a way to declare an intrinsic
where one argument is an overloaded vector type and another argument is a
pointer to the vector element type.  The bitcasts added here will match what
the frontend will typically generate when these intrinsics are used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101840 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMore IT instruction error-handling improvements from fuzzing.
Johnny Chen [Tue, 20 Apr 2010 00:15:41 +0000 (00:15 +0000)]
More IT instruction error-handling improvements from fuzzing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101839 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRevert 101825, which is causing trouble.
Dan Gohman [Mon, 19 Apr 2010 23:34:15 +0000 (23:34 +0000)]
Revert 101825, which is causing trouble.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101832 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMake BreakAntiDependencies' SUnits argument const, and make the Begin
Dan Gohman [Mon, 19 Apr 2010 23:11:58 +0000 (23:11 +0000)]
Make BreakAntiDependencies' SUnits argument const, and make the Begin
and End arguments by-value rather than by-reference.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101830 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoBetter error handling of invalid IT mask '0000', instead of just asserting.
Johnny Chen [Mon, 19 Apr 2010 23:02:58 +0000 (23:02 +0000)]
Better error handling of invalid IT mask '0000', instead of just asserting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101827 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoEliminate SelectionDAGISel's "current block" member. Just pass it as
Dan Gohman [Mon, 19 Apr 2010 22:51:14 +0000 (22:51 +0000)]
Eliminate SelectionDAGISel's "current block" member. Just pass it as
an argument to things that need it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101825 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoDelete an unnecessary reference to SelectionDAGISel::BB.
Dan Gohman [Mon, 19 Apr 2010 22:48:45 +0000 (22:48 +0000)]
Delete an unnecessary reference to SelectionDAGISel::BB.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101824 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoEliminate the CurMBB member from SelectionDAGBuilder. For places that
Dan Gohman [Mon, 19 Apr 2010 22:41:47 +0000 (22:41 +0000)]
Eliminate the CurMBB member from SelectionDAGBuilder. For places that
need it, just pass around the parent block of the current instruction
explicitly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101822 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemove the Expr member from IVUsers. Instead of remembering the expression,
Dan Gohman [Mon, 19 Apr 2010 21:48:58 +0000 (21:48 +0000)]
Remove the Expr member from IVUsers. Instead of remembering the expression,
just ask ScalarEvolution for it on demand. This helps IVUsers be more robust
in the case of expressions changing underneath it. This fixes PR6862.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101819 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAccording to A8.6.16 B (Encoding T3) and A8.3 Conditional execution -- A8.3.1
Johnny Chen [Mon, 19 Apr 2010 21:19:52 +0000 (21:19 +0000)]
According to A8.6.16 B (Encoding T3) and A8.3 Conditional execution -- A8.3.1
Pseudocode details of conditional, Condition bits '111x' indicate the
instruction is always executed.  That is, '1111' is a leagl condition field
value, which is now mapped to ARMCC::AL.

Also add a test case for condition field '1111'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101817 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix typo. add a test case.
Devang Patel [Mon, 19 Apr 2010 20:31:39 +0000 (20:31 +0000)]
Fix typo. add a test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101812 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMore progress on promoting i16 operations to i32 for x86. Work in progress.
Evan Cheng [Mon, 19 Apr 2010 19:29:22 +0000 (19:29 +0000)]
More progress on promoting i16 operations to i32 for x86. Work in progress.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101808 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoGive SelectionDAG a TargetMachine too, rather than having it
Dan Gohman [Mon, 19 Apr 2010 19:22:07 +0000 (19:22 +0000)]
Give SelectionDAG a TargetMachine too, rather than having it
fetch one from the MachineFunction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101807 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMore 80 col violation.
Evan Cheng [Mon, 19 Apr 2010 19:17:44 +0000 (19:17 +0000)]
More 80 col violation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101806 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd DW_AT_APPLE_omit_frame_ptr to encode -fomit-frame-pointer flag.
Devang Patel [Mon, 19 Apr 2010 19:14:02 +0000 (19:14 +0000)]
Add DW_AT_APPLE_omit_frame_ptr to encode -fomit-frame-pointer flag.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101805 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoCode that needs a TargetMachine should have access to one directly, rather
Dan Gohman [Mon, 19 Apr 2010 19:05:59 +0000 (19:05 +0000)]
Code that needs a TargetMachine should have access to one directly, rather
than just getting one through a TargetLowering.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101802 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMove isInTailCallPosition out of SelectionDAGBuilder, as it isn't
Dan Gohman [Mon, 19 Apr 2010 18:41:46 +0000 (18:41 +0000)]
Move isInTailCallPosition out of SelectionDAGBuilder, as it isn't
SelectionDAG-specific.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101801 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoEnable -Wcast-qual for C++ files, where intentional qualifier-stripping can
Dan Gohman [Mon, 19 Apr 2010 18:33:28 +0000 (18:33 +0000)]
Enable -Wcast-qual for C++ files, where intentional qualifier-stripping can
be done with const_cast.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101798 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemove -module, which appears to be a leftover from prior libtool usage.
Dan Gohman [Mon, 19 Apr 2010 18:11:02 +0000 (18:11 +0000)]
Remove -module, which appears to be a leftover from prior libtool usage.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101793 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoThis doesn't need SmallVector.h anymore.
Dan Gohman [Mon, 19 Apr 2010 17:51:31 +0000 (17:51 +0000)]
This doesn't need SmallVector.h anymore.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101790 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSupport .a files directly (without -l).
Mikhail Glushenkov [Mon, 19 Apr 2010 17:25:38 +0000 (17:25 +0000)]
Support .a files directly (without -l).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101789 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoBetter error-handling for DisassembleThumb2DPSoReg() where the 3-reg operand
Johnny Chen [Mon, 19 Apr 2010 17:16:40 +0000 (17:16 +0000)]
Better error-handling for DisassembleThumb2DPSoReg() where the 3-reg operand
instructions should have Rd (Inst{11-8}) != 0b1111.

Ref: A6.3 32-bit Thumb instruction encoding
     A6.3.11 Data-processing (shifted register)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101788 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix this for std::vectors which don't have .data().
Dan Gohman [Mon, 19 Apr 2010 16:33:28 +0000 (16:33 +0000)]
Fix this for std::vectors which don't have .data().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101785 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoARM disassembler did not react to recent changes to the NEON instruction table.
Johnny Chen [Mon, 19 Apr 2010 16:20:34 +0000 (16:20 +0000)]
ARM disassembler did not react to recent changes to the NEON instruction table.
VLD1q*_UPD and VST1q*_UPD have the ${dst:dregpair} operand now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101784 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoDon't write into MemoryBuffers.
Benjamin Kramer [Mon, 19 Apr 2010 16:15:31 +0000 (16:15 +0000)]
Don't write into MemoryBuffers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101783 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix -Wcast-qual warnings.
Dan Gohman [Mon, 19 Apr 2010 15:55:10 +0000 (15:55 +0000)]
Fix -Wcast-qual warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101782 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRevert 91528 and use a std::vector instead, fixing an abuse of std::string.
Dan Gohman [Mon, 19 Apr 2010 15:54:44 +0000 (15:54 +0000)]
Revert 91528 and use a std::vector instead, fixing an abuse of std::string.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101781 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix -Wcast-qual warnings.
Dan Gohman [Mon, 19 Apr 2010 15:42:05 +0000 (15:42 +0000)]
Fix -Wcast-qual warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101779 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd a few more ELF bits.
Dan Gohman [Mon, 19 Apr 2010 15:40:15 +0000 (15:40 +0000)]
Add a few more ELF bits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101778 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd a const.
Dan Gohman [Mon, 19 Apr 2010 15:39:27 +0000 (15:39 +0000)]
Add a const.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101777 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoThese functions don't need to access this.
Dan Gohman [Mon, 19 Apr 2010 15:38:38 +0000 (15:38 +0000)]
These functions don't need to access this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101776 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd missed part of prev. commit
Anton Korobeynikov [Sun, 18 Apr 2010 20:41:42 +0000 (20:41 +0000)]
Add missed part of prev. commit

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101755 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMake processor FUs unique for given itinerary. This extends the limit of 32
Anton Korobeynikov [Sun, 18 Apr 2010 20:31:01 +0000 (20:31 +0000)]
Make processor FUs unique for given itinerary. This extends the limit of 32
FU per CPU arch to 32 per intinerary allowing precise modelling of quite
complex pipelines in the future.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101754 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoavoid temporary std::string in non posix_spawn path.
Chris Lattner [Sun, 18 Apr 2010 17:34:10 +0000 (17:34 +0000)]
avoid temporary std::string in non posix_spawn path.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101723 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoEliminate temporary string.
Benjamin Kramer [Sun, 18 Apr 2010 09:19:41 +0000 (09:19 +0000)]
Eliminate temporary string.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101711 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoProperly inherit the environment on darwin where environ is not available for shared...
Benjamin Kramer [Sun, 18 Apr 2010 09:16:04 +0000 (09:16 +0000)]
Properly inherit the environment on darwin where environ is not available for shared libraries.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101710 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoThis is horrible. Split the difference, and declare 'environ' on all non-Darwin
Nick Lewycky [Sun, 18 Apr 2010 07:07:48 +0000 (07:07 +0000)]
This is horrible. Split the difference, and declare 'environ' on all non-Darwin
platforms to unbreak the darwin and linux builds. The BSD folks should feel
free to change the #if, if this breaks them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101703 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRevert r101701, Darwin doesn't have 'environ'. Go figure.
Nick Lewycky [Sun, 18 Apr 2010 06:44:21 +0000 (06:44 +0000)]
Revert r101701, Darwin doesn't have 'environ'. Go figure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101702 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix linux build. posix_spawn doesn't inherit the environment by default.
Nick Lewycky [Sun, 18 Apr 2010 06:22:26 +0000 (06:22 +0000)]
Fix linux build. posix_spawn doesn't inherit the environment by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101701 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agomake Program::Execute use posix_spawn on systems that support it,
Chris Lattner [Sun, 18 Apr 2010 04:14:37 +0000 (04:14 +0000)]
make Program::Execute use posix_spawn on systems that support it,
as it is more efficient than fork/exec.

Thanks to Eric for adding the autoconf check.  It would be nice if
a cmake guru could add a cmake check for posix_spawn as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101693 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoreduce indentation
Chris Lattner [Sun, 18 Apr 2010 03:35:23 +0000 (03:35 +0000)]
reduce indentation

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101692 91177308-0d34-0410-b5e6-96231b3b80d8