1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-25 12:12:47 +01:00
Commit Graph

10583 Commits

Author SHA1 Message Date
Jakob Stoklund Olesen
cb8f334578 Refix MSVC9 and upper_bound. It actually needs a fully symmetric comparator.
llvm-svn: 114469
2010-09-21 20:16:12 +00:00
Chris Lattner
32ec32b690 finish pushing MachinePointerInfo through selectiondags. At this point,
I think I've audited all uses, so it should be dependable for address spaces,
and the pointer+offset info should also be accurate when there.

llvm-svn: 114464
2010-09-21 18:58:22 +00:00
Chris Lattner
3dde58c15a convert a couple more places to use the new getStore()
llvm-svn: 114463
2010-09-21 18:51:21 +00:00
Chris Lattner
8cdc5e75f7 update a bunch of code to use the MachinePointerInfo version of getStore.
llvm-svn: 114461
2010-09-21 18:41:36 +00:00
Jakob Stoklund Olesen
d09d4a9693 Don't pollute the global namespace.
llvm-svn: 114459
2010-09-21 18:34:17 +00:00
Jakob Stoklund Olesen
375867aa60 MSVC9 does not support upper_bound with an asymmetric comparator.
llvm-svn: 114455
2010-09-21 18:24:30 +00:00
Bob Wilson
c4345abcc0 Define the TargetLowering::getTgtMemIntrinsic hook for ARM so that NEON load
and store intrinsics are represented with MemIntrinsicSDNodes.

llvm-svn: 114454
2010-09-21 17:56:22 +00:00
Chris Lattner
86b3f287ce eliminate an old SelectionDAG::getTruncStore method, propagating
MachinePointerInfo around more.

llvm-svn: 114452
2010-09-21 17:42:31 +00:00
Chris Lattner
bf98f86fed eliminate last SelectionDAG::getLoad old entrypoint, on to stores.
llvm-svn: 114450
2010-09-21 17:28:52 +00:00
Chris Lattner
8af4fb7aed fix the code that infers SV info to be correct when dealing
with an indexed load/store that has an offset in the index.

llvm-svn: 114449
2010-09-21 17:24:05 +00:00
Jakob Stoklund Olesen
03451a0e51 Add LiveInterval::find and use it for most LiveRange searching operations
instead of calling lower_bound or upper_bound directly.

This cleans up the search logic a bit because {lower,upper}_bound compare
LR->start by default, and it is usually simpler to search LR->end.

Funnelling all searches through one function also makes it possible to replace
the search algorithm with something faster than binary search.

llvm-svn: 114448
2010-09-21 17:12:18 +00:00
Jakob Stoklund Olesen
73d2940daa Remove dead method.
llvm-svn: 114447
2010-09-21 17:12:15 +00:00
Chris Lattner
cdfd993df0 propagate MachinePointerInfo through various uses of the old
SelectionDAG::getExtLoad overload, and eliminate it.

llvm-svn: 114446
2010-09-21 17:04:51 +00:00
Chris Lattner
0d430648ae continue MachinePointerInfo'izing, eliminating use of one of the old
getLoad overloads.

llvm-svn: 114443
2010-09-21 16:36:31 +00:00
Chris Lattner
5c0bdf4543 convert dagcombine off the old form of getLoad. This fixes several bugs
with SVOffset computation.

llvm-svn: 114442
2010-09-21 16:08:50 +00:00
Chris Lattner
637762e6d1 simplify DAGCombiner::SimplifySelectOps step #2/2.
llvm-svn: 114437
2010-09-21 15:58:55 +00:00
Chris Lattner
e3616071c5 substantially reduce indentation and simplify DAGCombiner::SimplifySelectOps.
no functionality change (step #1)

llvm-svn: 114436
2010-09-21 15:46:59 +00:00
Lang Hames
f2e621dcb8 Fixed ambiguous call.
llvm-svn: 114431
2010-09-21 13:47:10 +00:00
Lang Hames
eae68e1117 Added an additional PBQP problem builder which adds coalescing costs (both between pairs of virtuals, and between virtuals and physicals).
llvm-svn: 114429
2010-09-21 13:19:36 +00:00
Gabor Greif
99c07b1d95 Move the search for the appropriate AND instruction
into OptimizeCompareInstr.
This necessitates the passing of CmpValue around,
so widen the virtual functions to accomodate.

No functionality changes.

llvm-svn: 114428
2010-09-21 12:01:15 +00:00
Chris Lattner
b68bd70665 a few more trivial updates. This fixes PerformInsertVectorEltInMemory to not
pass a completely incorrect SrcValue, which would result in a miscompile with
combiner-aa.

llvm-svn: 114411
2010-09-21 07:32:19 +00:00
Chris Lattner
1cad885bf7 add some accessors
llvm-svn: 114409
2010-09-21 06:43:24 +00:00
Chris Lattner
112cf9bc89 it's more elegant to put the "getConstantPool" and
"getFixedStack" on the MachinePointerInfo class.  While
this isn't the problem I'm setting out to solve, it is the
right way to eliminate PseudoSourceValue, so lets go with it.

llvm-svn: 114406
2010-09-21 06:22:23 +00:00
Chris Lattner
f94de5bf46 reimplement memcpy/memmove/memset lowering to use MachinePointerInfo
instead of srcvalue/offset pairs.  This corrects SV info for mem 
operations whose size is > 32-bits.

llvm-svn: 114401
2010-09-21 05:40:29 +00:00
Chris Lattner
b6d15db75c add some helpful accessors.
llvm-svn: 114400
2010-09-21 05:39:30 +00:00
Chris Lattner
dbe51ad1b8 add overloads for SelectionDAG::getLoad, getStore, getTruncStore that take a
MachinePointerInfo.  Among other virtues, this doesn't silently  truncate the
svoffset to 32-bits.

llvm-svn: 114399
2010-09-21 05:10:45 +00:00
Chris Lattner
e1fc671030 simplify interface to SelectionDAG::getMemIntrinsicNode, making it take a MachinePointerInfo
llvm-svn: 114397
2010-09-21 04:57:15 +00:00
Chris Lattner
e4db4cad3b chagne interface to SelectionDAG::getAtomic to take a MachinePointerInfo,
eliminating some weird "infer a frame address" logic which was dead.

llvm-svn: 114396
2010-09-21 04:53:42 +00:00
Chris Lattner
5de5fada20 don't implicitly drop the offset of a machinememoperand when legalizing atomics.
llvm-svn: 114395
2010-09-21 04:51:11 +00:00
Chris Lattner
af01f8d142 force clients of MachineFunction::getMachineMemOperand to provide a
MachinePointerInfo, propagating the type out a level of API.  Remove
the old MachineFunction::getMachineMemOperand impl.

llvm-svn: 114393
2010-09-21 04:46:39 +00:00
Chris Lattner
940c35a3c3 start pushing MachinePointerInfo out through the MachineMemOperand interface
to the MachineFunction construction methods.

llvm-svn: 114390
2010-09-21 04:32:08 +00:00
Chris Lattner
7fdf193383 refactor the Value*/offset pair from MachineMemOperand out to a new
MachinePointerInfo struct, no functionality change.

This also adds an assert to MachineMemOperand::MachineMemOperand
that verifies that the Value* is either null or is an IR pointer type.

llvm-svn: 114389
2010-09-21 04:23:39 +00:00
Evan Cheng
1ce02d180e Enable machine sinking critical edge splitting. e.g.
define double @foo(double %x, double %y, i1 %c) nounwind {
  %a = fdiv double %x, 3.2
  %z = select i1 %c, double %a, double %y
  ret double %z
}

Was:
_foo:
        divsd   LCPI0_0(%rip), %xmm0
        testb   $1, %dil
        jne     LBB0_2
        movaps  %xmm1, %xmm0
LBB0_2:
        ret

Now:
_foo:
        testb   $1, %dil
        je      LBB0_2
        divsd   LCPI0_0(%rip), %xmm0
        ret
LBB0_2:
        movaps  %xmm1, %xmm0
        ret

This avoids the divsd when early exit is taken.
rdar://8454886

llvm-svn: 114372
2010-09-20 22:52:00 +00:00
Owen Anderson
fc94b337eb When TCO is turned on, it is possible to end up with aliasing FrameIndex's. Therefore,
CombinerAA cannot assume that different FrameIndex's never alias, but can instead use
MachineFrameInfo to get the actual offsets of these slots and check for actual aliasing.

This fixes CodeGen/X86/2010-02-19-TailCallRetAddrBug.ll and CodeGen/X86/tailcallstack64.ll
when CombinerAA is enabled, modulo a different register allocation sequence.

llvm-svn: 114348
2010-09-20 20:39:59 +00:00
Evan Cheng
1c7f72d0bc Avoid splitting critical edge twice for a set of PHI uses.
llvm-svn: 114338
2010-09-20 19:12:55 +00:00
Owen Anderson
9594834ec1 Revert r114312 while I sort out some issues.
llvm-svn: 114313
2010-09-19 21:01:26 +00:00
Owen Anderson
65f23dcf7e Tentatively enabled DAGCombiner Alias Analysis by default. As far as I know,
r114268 fixed the last of the blockers to enabling it.  I will be monitoring
for failures.

llvm-svn: 114312
2010-09-19 19:51:55 +00:00
Benjamin Kramer
89a330c4b9 Unbreak msvc build.
llvm-svn: 114284
2010-09-18 14:41:26 +00:00
Lang Hames
2d87a43549 Fixed non-const iterator error.
llvm-svn: 114273
2010-09-18 09:49:08 +00:00
Lang Hames
4a8c999803 Added a separate class (PBQPBuilder) for PBQP Problem construction. This class can be extended to support custom constraints.
For now the allocator still uses the old (internal) construction mechanism by default. This will be phased out soon assuming 
no issues with the builder system come up.

To invoke the new construction mechanism just pass '-regalloc=pbqp -pbqp-builder' to llc. To provide custom constraints a
Target just needs to extend PBQPBuilder and pass an instance of their derived builder to the RegAllocPBQP constructor.

llvm-svn: 114272
2010-09-18 09:07:10 +00:00
Evan Cheng
442cb9c620 Fix code that break critical edges for PHI uses. Watch out for multiple PHIs in different blocks.
llvm-svn: 114270
2010-09-18 06:42:17 +00:00
Owen Anderson
015641f659 Invert the logic of reachesChainWithoutSideEffects(). What we want to check is that there is
NO path to the destination containing side effects, not that SOME path contains no side effects.
In  practice, this only manifests with CombinerAA enabled, because otherwise the chain has little
to no branching, so "any" is effectively equivalent to "all".

llvm-svn: 114268
2010-09-18 04:45:14 +00:00
Evan Cheng
8c2bde65f0 Teach machine sink to
1) Do forward copy propagation. This makes it easier to estimate the cost of the
   instruction being sunk.
2) Break critical edges on demand, including cases where the value is used by
   PHI nodes.
Critical edge splitting is not yet enabled by default.

llvm-svn: 114227
2010-09-17 22:28:18 +00:00
Evan Cheng
042ab036a2 Machine CSE was forgetting to clear some data structures.
llvm-svn: 114222
2010-09-17 21:59:42 +00:00
Evan Cheng
5f2926d516 Fix a potential bug that can cause miscomparison with and without debug info.
llvm-svn: 114220
2010-09-17 21:56:26 +00:00
Devang Patel
99777d1d07 If FE forgot to provide a file name (usually it uses "stdin" as name in such situation) then make one up to ensure that debug info is not malformed.
llvm-svn: 114119
2010-09-16 20:57:49 +00:00
Jakob Stoklund Olesen
e37047a62f Use the value mapping provided by LiveIntervalMap. This simplifies the code a
great deal because we don't have to worry about maintaining SSA form.

Unconditionally copy back to dupli when the register is live out of the split
range, even if the live-out value was defined outside the range. Skipping the
back-copy only makes sense when the live range is going to spill outside the
split range, and we don't know that it will. Besides, this was a hack to avoid
SSA update issues.

Clear up some confusion about the end point of a half-open LiveRange. Methinks
LiveRanges need to be closed so both start and end are included in the range.
The low bits of a SlotIndex are symbolic, so a half-open range doesn't really
make sense. This would be a pervasive change, though.

llvm-svn: 114043
2010-09-16 00:01:36 +00:00
Devang Patel
e15f83013d Check bb to ensure that alloca is in separate basic block.
This fixes funcargs.exp regression reported by gdb testsuite.

llvm-svn: 113992
2010-09-15 18:13:55 +00:00
Devang Patel
fbc9fca6ac If dbg.declare from non-entry block is using alloca from entry block then use offset available in StaticAllocaMap to emit DBG_VALUE. Right now, this has no material impact because varible info also collected using offset table maintained in machine module info.
llvm-svn: 113967
2010-09-15 14:48:53 +00:00
Gabor Greif
bd0b1357e0 must not peephole away side effects
llvm-svn: 113848
2010-09-14 20:46:08 +00:00