oota-llvm.git
15 years agoIn the list-burr's pseudo two-addr dependency heuristics, don't
Dan Gohman [Thu, 16 Apr 2009 20:59:02 +0000 (20:59 +0000)]
In the list-burr's pseudo two-addr dependency heuristics, don't
add dependencies on nodes with exactly one successor which is a
COPY_TO_REGCLASS node. In the case that the copy is coalesced
away, the dependence should be on the user of the copy, rather
than the copy itself.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69309 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoHandle SUBREG_TO_REG instructions with the same heuristics
Dan Gohman [Thu, 16 Apr 2009 20:57:10 +0000 (20:57 +0000)]
Handle SUBREG_TO_REG instructions with the same heuristics
as INSERT_SUBREG instructions in the list-burr scheduler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69308 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix a bug with inttoptr/ptrtoint casts where the pointer has a different
Dan Gohman [Thu, 16 Apr 2009 19:25:55 +0000 (19:25 +0000)]
Fix a bug with inttoptr/ptrtoint casts where the pointer has a different
size from the integer, requiring zero extension or truncation. Don't
create ZExtInsts with pointer types. This fixes a regression in
consumer-jpeg.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69307 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoDo not treat beginning of inlined scope as beginning of normal function scope if...
Devang Patel [Thu, 16 Apr 2009 17:55:30 +0000 (17:55 +0000)]
Do not treat beginning of inlined scope as beginning of normal function scope if the location info is missing.

Insetad of doing ...
if (inlined_subroutine && known_location)
  DW_TAG_inline_subroutine
else
  DW_TAG_subprogram

do

if (inlined_subroutine) {
 if (known_location)
   DW_TAG_inline_subroutine
} else {
 DW_TAG_subprogram
}

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69300 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoMinor code simplifications. Don't attempt LSR on theoretical
Dan Gohman [Thu, 16 Apr 2009 16:49:48 +0000 (16:49 +0000)]
Minor code simplifications. Don't attempt LSR on theoretical
targets with pointers larger than 64 bits, due to the code not
yet being APInt clean.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69296 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoLSR is no longer a GEP optimizer. It is now an IV expression
Dan Gohman [Thu, 16 Apr 2009 16:46:01 +0000 (16:46 +0000)]
LSR is no longer a GEP optimizer. It is now an IV expression
optimizer, which just happen to frequently involve optimizing GEPs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69295 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix SCEVExpander::visitSMaxExpr and SCEVExpander::visitUMaxExpr to
Dan Gohman [Thu, 16 Apr 2009 16:15:25 +0000 (16:15 +0000)]
Fix SCEVExpander::visitSMaxExpr and SCEVExpander::visitUMaxExpr to
not create ICmpInsts with operands of different types. This fixes
a regression in Applications/d/make_dparser.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69294 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoTeach SCEVExpander::InsertCastOfTo to avoid creating inttoptr-of-ptrtoint
Dan Gohman [Thu, 16 Apr 2009 15:52:57 +0000 (15:52 +0000)]
Teach SCEVExpander::InsertCastOfTo to avoid creating inttoptr-of-ptrtoint
and ptrtoint-of-inttoptr expressions. This fixes a regression in 300.twolf.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69293 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoUse ConstantExpr::getIntToPtr instead of SCEVExpander::InsertCastOfTo,
Dan Gohman [Thu, 16 Apr 2009 15:48:38 +0000 (15:48 +0000)]
Use ConstantExpr::getIntToPtr instead of SCEVExpander::InsertCastOfTo,
since the operand is always a constant.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69291 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoUse a SCEV expression cast instead of immediately inserting a
Dan Gohman [Thu, 16 Apr 2009 15:47:35 +0000 (15:47 +0000)]
Use a SCEV expression cast instead of immediately inserting a
new instruction with SCEVExpander::InsertCastOfTo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69290 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoRecord line number at the beginning of a func.start.
Devang Patel [Thu, 16 Apr 2009 15:07:09 +0000 (15:07 +0000)]
Record line number at the beginning of a func.start.

This line was accidently lost yesterday.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69286 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agofix PR3995. A scale must be 1, 2, 4 or 8.
Rafael Espindola [Thu, 16 Apr 2009 12:34:53 +0000 (12:34 +0000)]
fix PR3995. A scale must be 1, 2, 4 or 8.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69284 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoprove diagnostic -> group mapping information.
Chris Lattner [Thu, 16 Apr 2009 05:52:18 +0000 (05:52 +0000)]
prove diagnostic -> group mapping information.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69270 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoExpand GEPs in ScalarEvolution expressions. SCEV expressions can now
Dan Gohman [Thu, 16 Apr 2009 03:18:22 +0000 (03:18 +0000)]
Expand GEPs in ScalarEvolution expressions. SCEV expressions can now
have pointer types, though in contrast to C pointer types, SCEV
addition is never implicitly scaled. This not only eliminates the
need for special code like IndVars' EliminatePointerRecurrence
and LSR's own GEP expansion code, it also does a better job because
it lets the normal optimizations handle pointer expressions just
like integer expressions.

Also, since LLVM IR GEPs can't directly index into multi-dimensional
VLAs, moving the GEP analysis out of client code and into the SCEV
framework makes it easier for clients to handle multi-dimensional
VLAs the same way as other arrays.

Some existing regression tests show improved optimization.
test/CodeGen/ARM/2007-03-13-InstrSched.ll in particular improved to
the point where if-conversion started kicking in; I turned it off
for this test to preserve the intent of the test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69258 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoencode subgroups into the clang .inc file. -Wall now works!
Chris Lattner [Thu, 16 Apr 2009 03:16:12 +0000 (03:16 +0000)]
encode subgroups into the clang .inc file.  -Wall now works!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69257 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoIn -fast mode do what FastISel does.
Devang Patel [Thu, 16 Apr 2009 02:33:41 +0000 (02:33 +0000)]
In -fast mode do what FastISel does.
This code could use some refactoring help!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69254 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoIf FastISel is run and it has known DebugLoc then use it.
Devang Patel [Thu, 16 Apr 2009 01:33:10 +0000 (01:33 +0000)]
If FastISel is run and it has known DebugLoc then use it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69253 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoIf location where the function was inlined is not know then do not emit debug info...
Devang Patel [Thu, 16 Apr 2009 01:31:54 +0000 (01:31 +0000)]
If location where the function was inlined is not know then do not emit debug info describing inlinied region.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69252 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agostart producing subgroup info.
Chris Lattner [Thu, 16 Apr 2009 00:53:25 +0000 (00:53 +0000)]
start producing subgroup info.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69249 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAnother testcase for IV shortening.
Dale Johannesen [Thu, 16 Apr 2009 00:45:21 +0000 (00:45 +0000)]
Another testcase for IV shortening.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69247 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoEliminate zext over (iv | const) or (signed iv),
Dale Johannesen [Wed, 15 Apr 2009 23:31:51 +0000 (23:31 +0000)]
Eliminate zext over (iv | const) or (signed iv),
and sext over (iv | const), if a longer iv is
available.  Allow expressions to have more than
one zext/sext parent.  All from OpenSSL.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69241 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agomake sure that empty diag groups get known by clang.
Chris Lattner [Wed, 15 Apr 2009 22:33:02 +0000 (22:33 +0000)]
make sure that empty diag groups get known by clang.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69235 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoimplement support for writing out diagnostic group tables.
Chris Lattner [Wed, 15 Apr 2009 20:55:08 +0000 (20:55 +0000)]
implement support for writing out diagnostic group tables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69219 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agos/RootDbgScope/FunctionDbgScope/g
Devang Patel [Wed, 15 Apr 2009 20:41:31 +0000 (20:41 +0000)]
s/RootDbgScope/FunctionDbgScope/g

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69216 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoEliminate zext over (iv & const) or ((iv+const)&const)
Dale Johannesen [Wed, 15 Apr 2009 20:41:02 +0000 (20:41 +0000)]
Eliminate zext over (iv & const) or ((iv+const)&const)
if a longer iv is available.  These subscript forms are
not common; they're a bottleneck in OpenSSL.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69215 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agouse UppercaseString instead of EmitAllCaps
Chris Lattner [Wed, 15 Apr 2009 20:16:12 +0000 (20:16 +0000)]
use UppercaseString instead of EmitAllCaps

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69213 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agouse escape string.
Chris Lattner [Wed, 15 Apr 2009 20:13:18 +0000 (20:13 +0000)]
use escape string.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69212 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoteach EscapeString and UnescapeString to handle ".
Chris Lattner [Wed, 15 Apr 2009 20:12:52 +0000 (20:12 +0000)]
teach EscapeString and UnescapeString to handle ".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69211 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd DISubprogram is not null check.
Devang Patel [Wed, 15 Apr 2009 20:11:08 +0000 (20:11 +0000)]
Add DISubprogram is not null check.
This fixes test/CodeGen//2009-01-21-invalid-debug-info.m test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69210 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoGeneralize one of the SelectionDAG::ReplaceAllUsesWith overloads
Dan Gohman [Wed, 15 Apr 2009 20:06:30 +0000 (20:06 +0000)]
Generalize one of the SelectionDAG::ReplaceAllUsesWith overloads
to support replacing a node with another that has a superset of
the result types. Use this instead of calling
ReplaceAllUsesOfValueWith for each value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69209 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agorename -gen-clang-diags-options -> -gen-clang-diag-groups
Chris Lattner [Wed, 15 Apr 2009 20:02:32 +0000 (20:02 +0000)]
rename -gen-clang-diags-options -> -gen-clang-diag-groups

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69208 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agomove clang-specific makefile goop to clang makefile.
Chris Lattner [Wed, 15 Apr 2009 19:57:42 +0000 (19:57 +0000)]
move clang-specific makefile goop to clang makefile.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69206 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix 80-column violations.
Dan Gohman [Wed, 15 Apr 2009 19:48:57 +0000 (19:48 +0000)]
Fix 80-column violations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69204 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd a folding table entry for MOV8rr_NOREX.
Dan Gohman [Wed, 15 Apr 2009 19:48:28 +0000 (19:48 +0000)]
Add a folding table entry for MOV8rr_NOREX.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69203 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoCheck isInlinedSubroutine() before creating DW_TAG_inlined_subroutine.
Devang Patel [Wed, 15 Apr 2009 19:42:57 +0000 (19:42 +0000)]
Check isInlinedSubroutine() before creating DW_TAG_inlined_subroutine.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69202 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix MachineInstr::getNumExplicitOperands to count
Dan Gohman [Wed, 15 Apr 2009 17:59:11 +0000 (17:59 +0000)]
Fix MachineInstr::getNumExplicitOperands to count
variadic operands correctly. Patch by Jakob Stoklund Olesen!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69190 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agodon't infer diag class from parenting relations, make it an explicit field
Chris Lattner [Wed, 15 Apr 2009 16:55:46 +0000 (16:55 +0000)]
don't infer diag class from parenting relations, make it an explicit field
in the record.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69176 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoinclude the default mapping in the clang diagnostic .inc files.
Chris Lattner [Wed, 15 Apr 2009 16:43:18 +0000 (16:43 +0000)]
include the default mapping in the clang diagnostic .inc files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69173 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agominor cleanups
Chris Lattner [Wed, 15 Apr 2009 06:26:49 +0000 (06:26 +0000)]
minor cleanups

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69152 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoLimit the number of times we're willing to chase pointers. Removes an O(n^2)
Nick Lewycky [Wed, 15 Apr 2009 06:23:41 +0000 (06:23 +0000)]
Limit the number of times we're willing to chase pointers. Removes an O(n^2)
problem from instcombine.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69151 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAllow jumping to the end of a bitstream while reading
Douglas Gregor [Wed, 15 Apr 2009 04:53:47 +0000 (04:53 +0000)]
Allow jumping to the end of a bitstream while reading

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69145 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoCheck for alignment.
Bill Wendling [Wed, 15 Apr 2009 04:51:05 +0000 (04:51 +0000)]
Check for alignment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69140 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoMore obsessive reformatting. Fixed some validation errors.
Bill Wendling [Wed, 15 Apr 2009 02:12:37 +0000 (02:12 +0000)]
More obsessive reformatting. Fixed some validation errors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69130 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoDon't use "protected:" in classes that aren't intended to be
Dan Gohman [Wed, 15 Apr 2009 01:47:03 +0000 (01:47 +0000)]
Don't use "protected:" in classes that aren't intended to be
subclassed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69129 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix doxygen comment syntax.
Dan Gohman [Wed, 15 Apr 2009 01:44:07 +0000 (01:44 +0000)]
Fix doxygen comment syntax.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69128 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix X86MachineFunctionInfo's doxygen comment.
Dan Gohman [Wed, 15 Apr 2009 01:20:18 +0000 (01:20 +0000)]
Fix X86MachineFunctionInfo's doxygen comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69127 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoMove MachineRegisterInfo::setRegClass out of line.
Dan Gohman [Wed, 15 Apr 2009 01:19:35 +0000 (01:19 +0000)]
Move MachineRegisterInfo::setRegClass out of line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69126 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoMove MachineJumpTableInfo::ReplaceMBBInJumpTables out of line.
Dan Gohman [Wed, 15 Apr 2009 01:18:49 +0000 (01:18 +0000)]
Move MachineJumpTableInfo::ReplaceMBBInJumpTables out of line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69125 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoGive RemoveRegOperandFromRegInfo a comment and move the
Dan Gohman [Wed, 15 Apr 2009 01:17:37 +0000 (01:17 +0000)]
Give RemoveRegOperandFromRegInfo a comment and move the
code out of line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69124 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoEnhance induction variable code to remove the
Dale Johannesen [Wed, 15 Apr 2009 01:10:12 +0000 (01:10 +0000)]
Enhance induction variable code to remove the
sext around sext(shorter IV + constant), using a
longer IV instead, when it can figure out the
add can't overflow.  This comes up a lot in
subscripting; mainly affects 64 bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69123 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAvoid making the transformation enabled by my last patch if the new destinations...
Evan Cheng [Wed, 15 Apr 2009 00:43:54 +0000 (00:43 +0000)]
Avoid making the transformation enabled by my last patch if the new destinations have phi nodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69121 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoWhile inlining, clone llvm.dbg.func.start intrinsic and adjust
Devang Patel [Wed, 15 Apr 2009 00:17:06 +0000 (00:17 +0000)]
While inlining, clone llvm.dbg.func.start intrinsic and adjust
llvm.dbg.region.end instrinsic. This nested llvm.dbg.func.start/llvm.dbg.region.end pair now enables DW_TAG_inlined_subroutine support in code generator.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69118 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agosilence a warning.
Chris Lattner [Wed, 15 Apr 2009 00:16:05 +0000 (00:16 +0000)]
silence a warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69117 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoConstruct and emit DW_TAG_inlined_subroutine DIEs for inlined subroutine scopes ...
Devang Patel [Wed, 15 Apr 2009 00:10:26 +0000 (00:10 +0000)]
Construct and emit DW_TAG_inlined_subroutine DIEs for inlined subroutine scopes (only in FastISel mode).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69116 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoDo for GR16_NOREX what r69049 did for GR8_NOREX, to avoid trouble with
Dan Gohman [Wed, 15 Apr 2009 00:10:16 +0000 (00:10 +0000)]
Do for GR16_NOREX what r69049 did for GR8_NOREX, to avoid trouble with
the local register allocator.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69115 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd a method to check that the subprogram holds debug info for the given Function...
Devang Patel [Wed, 15 Apr 2009 00:06:07 +0000 (00:06 +0000)]
Add a method to check that the subprogram holds debug info for the given Function or not.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69113 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd a new MOV8rr_NOREX, and make X86's copyRegToReg use it when
Dan Gohman [Wed, 15 Apr 2009 00:04:23 +0000 (00:04 +0000)]
Add a new MOV8rr_NOREX, and make X86's copyRegToReg use it when
either the source or destination is a physical h register.

This fixes sqlite3 with the post-RA scheduler enabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69111 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoTestcase for r69104.
Bill Wendling [Wed, 15 Apr 2009 00:04:11 +0000 (00:04 +0000)]
Testcase for r69104.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69110 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoGR8_NOREX can contain the H registers, since they don't require
Dan Gohman [Wed, 15 Apr 2009 00:00:48 +0000 (00:00 +0000)]
GR8_NOREX can contain the H registers, since they don't require
REX prefixes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69108 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoOptimize conditional branch on i1 phis with non-constant inputs.
Evan Cheng [Tue, 14 Apr 2009 23:40:03 +0000 (23:40 +0000)]
Optimize conditional branch on i1 phis with non-constant inputs.

This turns:

eq:
        %3 = icmp eq i32 %1, %2
        br label %join

ne:
        %4 = icmp ne i32 %1, %2
        br label %join

join:
        %5 = phi i1 [%3, %eq], [%4, %ne]
        br i1 %5, label %yes, label %no

=>

eq:
        %3 = icmp eq i32 %1, %2
        br i1 %3, label %yes, label %no

ne:
        %4 = icmp ne i32 %1, %2
        br i1 %4, label %yes, label %no

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69102 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix the RUN lines so that this test actually tests.
Dan Gohman [Tue, 14 Apr 2009 22:50:17 +0000 (22:50 +0000)]
Fix the RUN lines so that this test actually tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69096 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFor the h-register addressing-mode trick, use the correct value for
Dan Gohman [Tue, 14 Apr 2009 22:45:05 +0000 (22:45 +0000)]
For the h-register addressing-mode trick, use the correct value for
any non-address uses of the address value. This fixes 186.crafty.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69094 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoMac OS X 10.6 and above do not use key manager to register EH frames.
Evan Cheng [Tue, 14 Apr 2009 22:31:59 +0000 (22:31 +0000)]
Mac OS X 10.6 and above do not use key manager to register EH frames.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69090 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoWhen the result of an EXTRACT_SUBREG, INSERT_SUBREG, or SUBREG_TO_REG
Dan Gohman [Tue, 14 Apr 2009 22:17:14 +0000 (22:17 +0000)]
When the result of an EXTRACT_SUBREG, INSERT_SUBREG, or SUBREG_TO_REG
operator is used by a CopyToReg to export the value to a different
block, don't reuse the CopyToReg's register for the subreg operation
result if the register isn't precisely the right class for the
subreg operation.

Also, rename the h-registers.ll test, now that there are more
than one.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69087 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoCall CreateProcess with bInheritHandles = TRUE.
Mikhail Glushenkov [Tue, 14 Apr 2009 21:31:36 +0000 (21:31 +0000)]
Call CreateProcess with bInheritHandles = TRUE.

Makes llvmc show error messages printed by child processes when run from the
Cygwin/MSYS shell. Since ExecuteAndWait does not return until the child program
has finished execution, this change should be harmless.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69082 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoDelete trailing whitespace.
Mikhail Glushenkov [Tue, 14 Apr 2009 21:31:14 +0000 (21:31 +0000)]
Delete trailing whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69081 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoSome of GR8_NOREX registers are only available in 64-bit mode.
Evan Cheng [Tue, 14 Apr 2009 16:57:43 +0000 (16:57 +0000)]
Some of GR8_NOREX registers are only available in 64-bit mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69049 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoHandle aggregate type arguments to direct and indirect calls.
Sanjiv Gupta [Tue, 14 Apr 2009 02:49:52 +0000 (02:49 +0000)]
Handle aggregate type arguments to direct and indirect calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69022 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoUse the output of the asm so the optimizer won't
Dale Johannesen [Tue, 14 Apr 2009 01:51:40 +0000 (01:51 +0000)]
Use the output of the asm so the optimizer won't
delete it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69018 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoLoopIndexSplit needs to inform the loop pass manager of the instructions it is
Owen Anderson [Tue, 14 Apr 2009 01:04:19 +0000 (01:04 +0000)]
LoopIndexSplit needs to inform the loop pass manager of the instructions it is
deleting, not just the basic block.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69011 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoDo not force asm's to be chained if they don't touch
Dale Johannesen [Tue, 14 Apr 2009 00:56:56 +0000 (00:56 +0000)]
Do not force asm's to be chained if they don't touch
memory and aren't volatile.  This was interfering with
good scheduling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69008 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix PR3934 part 2. findOnlyInterestingUse() was not setting IsCopy and IsDstPhys...
Evan Cheng [Tue, 14 Apr 2009 00:32:25 +0000 (00:32 +0000)]
Fix PR3934 part 2. findOnlyInterestingUse() was not setting IsCopy and IsDstPhys which are returned by value and used by callee. This happened to work on the earlier test cases because of a logic error in the caller side.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@69006 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoMake these errors more noticable in build logs.
Daniel Dunbar [Mon, 13 Apr 2009 22:26:09 +0000 (22:26 +0000)]
Make these errors more noticable in build logs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68998 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoChange SelectionDAG type legalization to allow BUILD_VECTOR operands to be
Bob Wilson [Mon, 13 Apr 2009 22:05:19 +0000 (22:05 +0000)]
Change SelectionDAG type legalization to allow BUILD_VECTOR operands to be
promoted to legal types without changing the type of the vector.  This is
following a suggestion from Duncan
(http://lists.cs.uiuc.edu/pipermail/llvmdev/2009-February/019923.html).
The transformation that used to be done during type legalization is now
postponed to DAG legalization.  This allows the BUILD_VECTORs to be optimized
and potentially handled specially by target-specific code.

It turns out that this is also consistent with an optimization done by the
DAG combiner: a BUILD_VECTOR and INSERT_VECTOR_ELT may be combined by
replacing one of the BUILD_VECTOR operands with the newly inserted element;
but INSERT_VECTOR_ELT allows its scalar operand to be larger than the
element type, with any extra high bits being implicitly truncated.  The
result is a BUILD_VECTOR where one of the operands has a type larger the
the vector element type.

Any code that operates on BUILD_VECTORs may now need to be aware of the
potential type discrepancy between the vector element type and the
BUILD_VECTOR operands.  This patch updates all of the places that I could
find to handle that case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68996 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoRename COPY_TO_SUBCLASS to COPY_TO_REGCLASS, and generalize
Dan Gohman [Mon, 13 Apr 2009 21:06:25 +0000 (21:06 +0000)]
Rename COPY_TO_SUBCLASS to COPY_TO_REGCLASS, and generalize
it accordingly. Thanks to Jakob Stoklund Olesen for pointing
out how this might be useful.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68986 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoRefactor some code in SelectionDAGLegalize::ExpandBUILD_VECTOR.
Bob Wilson [Mon, 13 Apr 2009 20:20:30 +0000 (20:20 +0000)]
Refactor some code in SelectionDAGLegalize::ExpandBUILD_VECTOR.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68981 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoPR3934: Fix a bogus two-address pass assertion.
Evan Cheng [Mon, 13 Apr 2009 20:04:24 +0000 (20:04 +0000)]
PR3934: Fix a bogus two-address pass assertion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68979 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoGet rid of some compile warnings.
Bill Wendling [Mon, 13 Apr 2009 19:45:05 +0000 (19:45 +0000)]
Get rid of some compile warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68978 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd a static APInt::getNumWords
Douglas Gregor [Mon, 13 Apr 2009 19:43:10 +0000 (19:43 +0000)]
Add a static APInt::getNumWords

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68977 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoRight now, Debugging information to encode scopes (DW_TAG_lexical_block) relies on...
Devang Patel [Mon, 13 Apr 2009 18:13:16 +0000 (18:13 +0000)]
Right now, Debugging information to encode scopes (DW_TAG_lexical_block) relies on DBG_LABEL. Unfortunately this intefers with the quality of optimized code.
This patch updates dwarf writer to encode scoping information in DWARF only in FastISel mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68973 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoReapply 68847.
Devang Patel [Mon, 13 Apr 2009 17:02:03 +0000 (17:02 +0000)]
Reapply 68847.
Now debug_inlined section is covered by TAI->doesDwarfUsesInlineInfoSection(), which is false by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68964 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoImplement x86 h-register extract support.
Dan Gohman [Mon, 13 Apr 2009 16:09:41 +0000 (16:09 +0000)]
Implement x86 h-register extract support.
 - Add patterns for h-register extract, which avoids a shift and mask,
   and in some cases a temporary register.
 - Add address-mode matching for turning (X>>(8-n))&(255<<n), where
   n is a valid address-mode scale value, into an h-register extract
   and a scaled-offset address.
 - Replace X86's MOV32to32_ and related instructions with the new
   target-independent COPY_TO_SUBREG instruction.

On x86-64 there are complicated constraints on h registers, and
CodeGen doesn't currently provide a high-level way to express all of them,
so they are handled with a bunch of special code. This code currently only
supports extracts where the result is used by a zero-extend or a store,
though these are fairly common.

These transformations are not always beneficial; since there are only
4 h registers, they sometimes require extra move instructions, and
this sometimes increases register pressure because it can force out
values that would otherwise be in one of those registers. However,
this appears to be relatively uncommon.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68962 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd a new TargetInstrInfo MachineInstr opcode, COPY_TO_SUBCLASS.
Dan Gohman [Mon, 13 Apr 2009 15:38:05 +0000 (15:38 +0000)]
Add a new TargetInstrInfo MachineInstr opcode, COPY_TO_SUBCLASS.
This will be used to replace things like X86's MOV32to32_.

Enhance ScheduleDAGSDNodesEmit to be more flexible and robust
in the presense of subregister superclasses and subclasses. It
can now cope with the definition of a virtual register being in
a subclass of a use.

Re-introduce the code for recording register superreg classes and
subreg classes. This is needed because when subreg extracts and
inserts get coalesced away, the virtual registers are left in
the correct subclass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68961 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoRemove x86's special-case handling for ISD::TRUNCATE and
Dan Gohman [Mon, 13 Apr 2009 15:29:31 +0000 (15:29 +0000)]
Remove x86's special-case handling for ISD::TRUNCATE and
ISD::SIGN_EXTEND_INREG. Tablegen-generated code can handle
these cases, and the scheduling issues observed earlier
appear to be resolved now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68959 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix copy+pastos in comments.
Dan Gohman [Mon, 13 Apr 2009 15:28:29 +0000 (15:28 +0000)]
Fix copy+pastos in comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68958 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoGeneralize getRegisterClassForRegister to handle registers
Dan Gohman [Mon, 13 Apr 2009 15:24:11 +0000 (15:24 +0000)]
Generalize getRegisterClassForRegister to handle registers
in multiple classes in the case that the classes are all
in subset/superset relations. This function is used by the
fast-isel emitter, which always wants the super-most set.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68957 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoDon't abort on an aliasing physical register that does not have
Dan Gohman [Mon, 13 Apr 2009 15:22:29 +0000 (15:22 +0000)]
Don't abort on an aliasing physical register that does not have
a live interval. This is needed for some upcoming subreg changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68956 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoWhen assigning a physical register to a MachineOperand, set
Dan Gohman [Mon, 13 Apr 2009 15:21:32 +0000 (15:21 +0000)]
When assigning a physical register to a MachineOperand, set
the subreg field to 0, since the subreg field is only used
for virtual register subregs. This doesn't change
current functionality; it just eliminates bogus noise from
debug output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68955 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoList the l registers before h registers, for consistency.
Dan Gohman [Mon, 13 Apr 2009 15:18:42 +0000 (15:18 +0000)]
List the l registers before h registers, for consistency.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68954 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd an assertion to verify that a copy was actually emitted.
Dan Gohman [Mon, 13 Apr 2009 15:16:56 +0000 (15:16 +0000)]
Add an assertion to verify that a copy was actually emitted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68953 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoUse X86::SUBREG_8BIT instead of hard-coding the equivalent constant.
Dan Gohman [Mon, 13 Apr 2009 15:14:03 +0000 (15:14 +0000)]
Use X86::SUBREG_8BIT instead of hard-coding the equivalent constant.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68951 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd a comment about MOVSX64rr8.
Dan Gohman [Mon, 13 Apr 2009 15:13:28 +0000 (15:13 +0000)]
Add a comment about MOVSX64rr8.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68950 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd comments to INSERT_SUBREG, EXTRACT_SURBEG, SUBREG_TO_REG,
Dan Gohman [Mon, 13 Apr 2009 15:12:29 +0000 (15:12 +0000)]
Add comments to INSERT_SUBREG, EXTRACT_SURBEG, SUBREG_TO_REG,
and IMPLICIT_DEF.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68949 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix another hard-coded constant to use X86AddrNumOperands.
Dan Gohman [Mon, 13 Apr 2009 15:04:25 +0000 (15:04 +0000)]
Fix another hard-coded constant to use X86AddrNumOperands.
This unbreaks the JIT on x86-64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68948 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoX86-64 TLS support for local exec and initial exec.
Rafael Espindola [Mon, 13 Apr 2009 13:02:49 +0000 (13:02 +0000)]
X86-64 TLS support for local exec and initial exec.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68947 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoWhile passing arg of types larger than char only one byte at lower end was getting...
Sanjiv Gupta [Mon, 13 Apr 2009 09:38:38 +0000 (09:38 +0000)]
While passing arg of types larger than char only one byte at lower end was getting passed. We couldn't catch this as we did not have tests that were passing an int value larger than 256.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68946 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix ocaml bindings; add "available_externally" linkage type.
Nick Lewycky [Mon, 13 Apr 2009 07:02:32 +0000 (07:02 +0000)]
Fix ocaml bindings; add "available_externally" linkage type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68945 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix warning in .ll parser, detect and reject available_externally on function
Nick Lewycky [Mon, 13 Apr 2009 07:02:02 +0000 (07:02 +0000)]
Fix warning in .ll parser, detect and reject available_externally on function
declarations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68944 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoadd AvailableExternally linkage to C bindings.
Chris Lattner [Mon, 13 Apr 2009 06:25:37 +0000 (06:25 +0000)]
add AvailableExternally linkage to C bindings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68942 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd a new "available_externally" linkage type. This is intended
Chris Lattner [Mon, 13 Apr 2009 05:44:34 +0000 (05:44 +0000)]
Add a new "available_externally" linkage type.  This is intended
to support C99 inline, GNU extern inline, etc.  Related bugzilla's
include PR3517, PR3100, & PR2933.  Nothing uses this yet, but it
appears to work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@68940 91177308-0d34-0410-b5e6-96231b3b80d8