oota-llvm.git
16 years agoconstified operator~().
Ted Kremenek [Mon, 4 Feb 2008 16:57:26 +0000 (16:57 +0000)]
constified operator~().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46707 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoCrashed the new type legalizer. Not likely to catch
Duncan Sands [Mon, 4 Feb 2008 09:40:27 +0000 (09:40 +0000)]
Crashed the new type legalizer.  Not likely to catch
any bugs in the future since to get the crash you also
need hacked in fake libcall support (which creates odd
but legal trees), but since adding it doesn't hurt...
Thanks to Chris for this ultimately reduced version.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46706 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoI don't see how NodeUpdated can be called with a
Duncan Sands [Mon, 4 Feb 2008 09:29:17 +0000 (09:29 +0000)]
I don't see how NodeUpdated can be called with a
ReadyToProcess node - add an assertion to check
this.  Add an assertion to NodeDeleted that checks
that processed/ready nodes are indeed not deleted.
It is because they are never deleted that none of
the maps can have a deleted node as the source of
a mapping.  It does however seem to be possible in
theory to have a deleted value as the target of a
mapping, however this has not yet been spotted in
the wild.  Still mulling on what to do about this.
[The theoretical situation is this: a node A is
expanded/promoted/whatever to a newly created node
B.  Thus A->B is added to a map.  When the subtree
rooted at B is legalized it is conceivable that B
is deleted due to RAUW on a node somewhere above
it].

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46705 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoThe rest of the SSE4.1 intrinsic patterns that are obvious to me. Getting
Nate Begeman [Mon, 4 Feb 2008 06:00:24 +0000 (06:00 +0000)]
The rest of the SSE4.1 intrinsic patterns that are obvious to me.  Getting
Evan's help with the rest.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46697 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoSome more SSE 4.1 intrinsic patterns.
Nate Begeman [Mon, 4 Feb 2008 05:34:34 +0000 (05:34 +0000)]
Some more SSE 4.1 intrinsic patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46696 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoMake this test more aggressive, to cover recent improvements.
Owen Anderson [Mon, 4 Feb 2008 04:55:24 +0000 (04:55 +0000)]
Make this test more aggressive, to cover recent improvements.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46695 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoBe more precise when eliminating pointers bue to memcpy's. This allows more
Owen Anderson [Mon, 4 Feb 2008 04:53:00 +0000 (04:53 +0000)]
Be more precise when eliminating pointers bue to memcpy's.  This allows more
stores to be deleted in some cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46694 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAllow GVN to hack on memcpy's, making them open to further optimization.
Owen Anderson [Mon, 4 Feb 2008 02:59:58 +0000 (02:59 +0000)]
Allow GVN to hack on memcpy's, making them open to further optimization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46693 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoTag this test with the PR reference.
Nick Lewycky [Sun, 3 Feb 2008 16:35:19 +0000 (16:35 +0000)]
Tag this test with the PR reference.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46688 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoThere are some cases where icmp(add) can be folded into a new icmp. Handle them.
Nick Lewycky [Sun, 3 Feb 2008 16:33:09 +0000 (16:33 +0000)]
There are some cases where icmp(add) can be folded into a new icmp. Handle them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46687 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoHack on vectors too.
Nick Lewycky [Sun, 3 Feb 2008 08:19:11 +0000 (08:19 +0000)]
Hack on vectors too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46684 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFold away one multiply in instcombine. This would normally be caught in
Nick Lewycky [Sun, 3 Feb 2008 07:42:09 +0000 (07:42 +0000)]
Fold away one multiply in instcombine. This would normally be caught in
reassociate anyways, but they could be generated during instcombine's run.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46683 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix typo
Chris Lattner [Sun, 3 Feb 2008 07:30:27 +0000 (07:30 +0000)]
Fix typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46682 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoSSE 4.1 Intrinsics and detection
Nate Begeman [Sun, 3 Feb 2008 07:18:54 +0000 (07:18 +0000)]
SSE 4.1 Intrinsics and detection

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46681 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agohandle the case where a node can become ready to process
Chris Lattner [Sun, 3 Feb 2008 07:13:32 +0000 (07:13 +0000)]
handle the case where a node can become ready to process
multiple times due to a RAUW.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46680 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoUse the new infrastructure for listening to node updates to
Chris Lattner [Sun, 3 Feb 2008 07:08:51 +0000 (07:08 +0000)]
Use the new infrastructure for listening to node updates to
keep the LegalizeTypes node flags up to date when doing a RAUW.
This fixes a nasty bug that Duncan ran into and makes the
previous (nonbuggy case) more efficent.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46679 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agothe world doesn't need my debugging code.
Chris Lattner [Sun, 3 Feb 2008 07:01:05 +0000 (07:01 +0000)]
the world doesn't need my debugging code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46678 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoChange the 'global modification' APIs in SelectionDAG to take a new
Chris Lattner [Sun, 3 Feb 2008 06:49:24 +0000 (06:49 +0000)]
Change the 'global modification' APIs in SelectionDAG to take a new
DAGUpdateListener object pointer instead of just returning a vector
of deleted nodes.  This makes the interfaces more efficient (no more
allocating a vector [at least a malloc], filling it in, then walking
it) and more clean.  This also allows the client to be notified of
nodes that are *changed* but not deleted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46677 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoexplicitly include Compiler.h instead of getting it from tblgen in the middle of...
Chris Lattner [Sun, 3 Feb 2008 05:43:57 +0000 (05:43 +0000)]
explicitly include Compiler.h instead of getting it from tblgen in the middle of a class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46676 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoGeneralize the SDOperand->SDOperand form of
Chris Lattner [Sun, 3 Feb 2008 03:35:22 +0000 (03:35 +0000)]
Generalize the SDOperand->SDOperand form of
SelectionDAG::ReplaceAllUsesWith to handle replacement of
an SDOperand with *any* sdoperand, not just one for a node with
a single result.  Note that this has a horrible FIXME'd hack in it
to work around PR1975.  This should be removed when PR1975 is fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46674 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agodon't do ReplaceUses on a result that doesn't exist.
Chris Lattner [Sun, 3 Feb 2008 03:20:59 +0000 (03:20 +0000)]
don't do ReplaceUses on a result that doesn't exist.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46673 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoadd a -view-legalize-types-dags option, for viewing the dags going into legalize...
Chris Lattner [Sun, 3 Feb 2008 02:05:04 +0000 (02:05 +0000)]
add a -view-legalize-types-dags option, for viewing the dags going into legalize types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46672 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoGet rid of the annoying blank lines before labels.
Evan Cheng [Sat, 2 Feb 2008 08:39:46 +0000 (08:39 +0000)]
Get rid of the annoying blank lines before labels.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46667 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoDon't use uninitialized values. Fixes vec_align.ll on X86 Linux.
Nick Lewycky [Sat, 2 Feb 2008 08:29:58 +0000 (08:29 +0000)]
Don't use uninitialized values. Fixes vec_align.ll on X86 Linux.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46666 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoUnbreak ppc debug support.
Evan Cheng [Sat, 2 Feb 2008 05:06:29 +0000 (05:06 +0000)]
Unbreak ppc debug support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46665 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoSDIsel processes llvm.dbg.declare by recording the variable debug information descrip...
Evan Cheng [Sat, 2 Feb 2008 04:07:54 +0000 (04:07 +0000)]
SDIsel processes llvm.dbg.declare by recording the variable debug information descriptor and its corresponding stack frame index in MachineModuleInfo. This only works if the local variable is "homed" in the stack frame. It does not work for byval parameter, etc.
Added ISD::DECLARE node type to represent llvm.dbg.declare intrinsic. Now the intrinsic calls are lowered into a SDNode and lives on through out the codegen passes.
For now, since all the debugging information recording is done at isel time, when a ISD::DECLARE node is selected, it has the side effect of also recording the variable. This is a short term solution that should be fixed in time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46659 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd comment explaining what is lower level analysis pass.
Devang Patel [Sat, 2 Feb 2008 01:43:30 +0000 (01:43 +0000)]
Add comment explaining what is lower level analysis pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46658 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoDo not strip symbols when EH is ON.
Devang Patel [Sat, 2 Feb 2008 01:32:38 +0000 (01:32 +0000)]
Do not strip symbols when EH is ON.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46657 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFixing a bug creating floating point constants of type other
Gordon Henriksen [Sat, 2 Feb 2008 01:07:50 +0000 (01:07 +0000)]
Fixing a bug creating floating point constants of type other
than double through the C bindings. Thanks to Tomas Lindquist
Olsen for reporting it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46656 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFrame index can be negative.
Evan Cheng [Sat, 2 Feb 2008 00:17:00 +0000 (00:17 +0000)]
Frame index can be negative.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46655 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoCBackend: Implement unaligned load/store.
Lauro Ramos Venancio [Fri, 1 Feb 2008 21:25:59 +0000 (21:25 +0000)]
CBackend: Implement unaligned load/store.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46646 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoDon't drop function/call return attributes like 'nounwind'.
Duncan Sands [Fri, 1 Feb 2008 20:37:16 +0000 (20:37 +0000)]
Don't drop function/call return attributes like 'nounwind'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46645 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoUpdate llvm.dbg.func.start.
Evan Cheng [Fri, 1 Feb 2008 09:12:11 +0000 (09:12 +0000)]
Update llvm.dbg.func.start.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46636 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRemove the nasty LABEL hack with a much less evil one. Now llvm.dbg.func.start implie...
Evan Cheng [Fri, 1 Feb 2008 09:10:45 +0000 (09:10 +0000)]
Remove the nasty LABEL hack with a much less evil one. Now llvm.dbg.func.start implies a stoppoint is set. SelectionDAGISel records a new source line but does not create a ISD::LABEL node for this special stoppoint. Asm printer will magically print this label. This ensures nothing is emitted before.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46635 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRename RecordLabel to RecordSourceLine because that's what it is doing.
Evan Cheng [Fri, 1 Feb 2008 02:05:57 +0000 (02:05 +0000)]
Rename RecordLabel to RecordSourceLine because that's what it is doing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46628 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd convenient std::string helpers to StringMap. Patch by Mikhail Glushenkov!
Anton Korobeynikov [Thu, 31 Jan 2008 23:02:33 +0000 (23:02 +0000)]
Add convenient std::string helpers to StringMap. Patch by Mikhail Glushenkov!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46625 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRevert 46556 and 46585. Dan please fix the PseudoSourceValue problem and re-commit.
Evan Cheng [Thu, 31 Jan 2008 21:00:00 +0000 (21:00 +0000)]
Revert 46556 and 46585. Dan please fix the PseudoSourceValue problem and re-commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46623 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoChris' change to print an approximation to long doubles
Dale Johannesen [Thu, 31 Jan 2008 18:34:01 +0000 (18:34 +0000)]
Chris' change to print an approximation to long doubles
exposed a bug in APFloat's long double->double conversion of
NaNs.  Broke several things in the ieee part of gcc testsuite.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46617 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoReapply the patch with fix.
Anton Korobeynikov [Thu, 31 Jan 2008 17:43:25 +0000 (17:43 +0000)]
Reapply the patch with fix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46615 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agorevert anton's recent stringmap patch, which breaks clang.
Chris Lattner [Thu, 31 Jan 2008 17:24:51 +0000 (17:24 +0000)]
revert anton's recent stringmap patch, which breaks clang.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46614 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoMake StringMap to be more STL'ish. Patch by Mikhail Glushenkov!
Anton Korobeynikov [Thu, 31 Jan 2008 12:30:40 +0000 (12:30 +0000)]
Make StringMap to be more STL'ish. Patch by Mikhail Glushenkov!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46612 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoWhitespace cleanup
Anton Korobeynikov [Thu, 31 Jan 2008 12:10:41 +0000 (12:10 +0000)]
Whitespace cleanup

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46611 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd a comment for a nasty short term hack.
Evan Cheng [Thu, 31 Jan 2008 10:05:13 +0000 (10:05 +0000)]
Add a comment for a nasty short term hack.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46610 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd an extra operand to LABEL nodes which distinguishes between debug, EH, or misc...
Evan Cheng [Thu, 31 Jan 2008 09:59:15 +0000 (09:59 +0000)]
Add an extra operand to LABEL nodes which distinguishes between debug, EH, or misc labels. This fixes the EH breakage. However I am not convinced this is *the* solution.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46609 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAllow ComplexExpressions in InstrInfo.td files to be slightly more... complex! Comple...
Christopher Lamb [Thu, 31 Jan 2008 07:27:46 +0000 (07:27 +0000)]
Allow ComplexExpressions in InstrInfo.td files to be slightly more... complex! ComplexExpressions can now have attributes which affect how TableGen interprets
the pattern when generating matchin code.

The first (and currently, only) attribute causes the immediate parent node of the ComplexPattern operand to be passed into the matching code rather than the node at the root of the entire DAG containing the pattern.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46606 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd more thorough error checking for NULL register classes.
Christopher Lamb [Thu, 31 Jan 2008 07:09:08 +0000 (07:09 +0000)]
Add more thorough error checking for NULL register classes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46605 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd x86 specific getFrameIndexOffset(). This fixes local variable debugging info.
Evan Cheng [Thu, 31 Jan 2008 04:06:00 +0000 (04:06 +0000)]
Add x86 specific getFrameIndexOffset(). This fixes local variable debugging info.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46598 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoMRegisterInfo::getLocation() is a really bad idea. Its function is to calculate the...
Evan Cheng [Thu, 31 Jan 2008 03:37:28 +0000 (03:37 +0000)]
MRegisterInfo::getLocation() is a really bad idea. Its function is to calculate the offset from frame pointer to a stack slot and then storing the delta in a MachineLocation object. The name is bad (it implies a getter), and MRegisterInfo doesn't need to know about MachineLocation.
Replace getLocation() with getFrameIndexOffset() which returns the delta from frame pointer to stack slot. Dwarf writer can then use the information for whatever it wants.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46597 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoMakes the same change in ppc backend: avoid inserting prologue before debug labels.
Evan Cheng [Thu, 31 Jan 2008 03:33:38 +0000 (03:33 +0000)]
Makes the same change in ppc backend: avoid inserting prologue before debug labels.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46596 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAvoid unnecessarily casting away const, fixing a FIXME.
Dan Gohman [Thu, 31 Jan 2008 01:05:10 +0000 (01:05 +0000)]
Avoid unnecessarily casting away const, fixing a FIXME.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46591 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAvoid unnecessarily casting away const.
Dan Gohman [Thu, 31 Jan 2008 01:01:48 +0000 (01:01 +0000)]
Avoid unnecessarily casting away const.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46590 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix a typo in a comment.
Dan Gohman [Thu, 31 Jan 2008 00:44:33 +0000 (00:44 +0000)]
Fix a typo in a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46588 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRename ISD::FLT_ROUNDS to ISD::FLT_ROUNDS_ to avoid conflicting
Dan Gohman [Thu, 31 Jan 2008 00:41:03 +0000 (00:41 +0000)]
Rename ISD::FLT_ROUNDS to ISD::FLT_ROUNDS_ to avoid conflicting
with the real FLT_ROUNDS (defined in <float.h>).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46587 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAlso avoid adding callee save code before debug labels.
Evan Cheng [Thu, 31 Jan 2008 00:27:49 +0000 (00:27 +0000)]
Also avoid adding callee save code before debug labels.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46586 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoCreate a new class, MemOperand, for describing memory references
Dan Gohman [Thu, 31 Jan 2008 00:25:39 +0000 (00:25 +0000)]
Create a new class, MemOperand, for describing memory references
in the backend. Introduce a new SDNode type, MemOperandSDNode, for
holding a MemOperand in the SelectionDAG IR, and add a MemOperand
list to MachineInstr, and code to manage them. Remove the offset
field from SrcValueSDNode; uses of SrcValueSDNode that were using
it are all all using MemOperandSDNode now.

Also, begin updating some getLoad and getStore calls to use the
PseudoSourceValue objects.

Most of this was written by Florian Brander, some
reorganization and updating to TOT by me.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46585 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFixed a bug in MergeValueInAsValue() pointed out by David Greene. Replace val# with...
Evan Cheng [Wed, 30 Jan 2008 22:44:55 +0000 (22:44 +0000)]
Fixed a bug in MergeValueInAsValue() pointed out by David Greene. Replace val# with previous liverange's.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46579 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoTreat the label for the first @llvm.dbg.stoppoint the same way as the dbg_func_start...
Evan Cheng [Wed, 30 Jan 2008 20:08:35 +0000 (20:08 +0000)]
Treat the label for the first @llvm.dbg.stoppoint the same way as the dbg_func_start label. Make sure nothing else is inserted before them.
Note this solution might be somewhat fragile since ISD::LABEL may be used for other
purposes. If that ends up to be an issue, we may need to introduce a different node
for debug labels.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46571 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdjust loop per review feedback.
Dale Johannesen [Wed, 30 Jan 2008 19:44:39 +0000 (19:44 +0000)]
Adjust loop per review feedback.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46569 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoA semi-gross fix for a debug info issue. When inserting the "function start" label...
Evan Cheng [Wed, 30 Jan 2008 19:35:32 +0000 (19:35 +0000)]
A semi-gross fix for a debug info issue. When inserting the "function start" label (i.e. first label in the entry block) take care to insert it at the beginning of the block.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46568 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoSupply appropriate arch info to darwin assembler.
Devang Patel [Wed, 30 Jan 2008 19:19:31 +0000 (19:19 +0000)]
Supply appropriate arch info to darwin assembler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46567 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAccept getelementptr starting at GV with all 0 indices as a
Dale Johannesen [Wed, 30 Jan 2008 19:00:21 +0000 (19:00 +0000)]
Accept getelementptr starting at GV with all 0 indices as a
legitimate way of representing global variable GV in debug info.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46565 91177308-0d34-0410-b5e6-96231b3b80d8

16 years ago Always put version numbers at the end because they are optional.
Devang Patel [Wed, 30 Jan 2008 18:30:11 +0000 (18:30 +0000)]
 Always put version numbers at the end because they are optional.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46563 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoEven though InsertAtEndOfBasicBlock is an ugly hack it still deserves a proper name...
Evan Cheng [Wed, 30 Jan 2008 18:18:23 +0000 (18:18 +0000)]
Even though InsertAtEndOfBasicBlock is an ugly hack it still deserves a proper name. Rename it to EmitInstrWithCustomInserter since it does not necessarily insert
instruction at the end.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46562 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoEnable exception handling if it is supported.
Devang Patel [Wed, 30 Jan 2008 17:43:03 +0000 (17:43 +0000)]
Enable exception handling if it is supported.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46560 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd a new PseudoSourceValue class, which will be used to help track
Dan Gohman [Wed, 30 Jan 2008 16:35:31 +0000 (16:35 +0000)]
Add a new PseudoSourceValue class, which will be used to help track
memory reference information in the backend. Most of this was written by
Florian Brander, cleanup and updating to TOT by me.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46556 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRemove a couple more cases of "getNumUses() == 0". No need to walk the linked
Nick Lewycky [Wed, 30 Jan 2008 08:01:28 +0000 (08:01 +0000)]
Remove a couple more cases of "getNumUses() == 0". No need to walk the linked
list just to see if whether the list is empty.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46555 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoUse empty() instead of comparing size() with zero.
Nick Lewycky [Wed, 30 Jan 2008 07:54:16 +0000 (07:54 +0000)]
Use empty() instead of comparing size() with zero.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46554 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoDon't DCE FreeInst's. We were using those! Patch from Owen Anderson.
Nick Lewycky [Wed, 30 Jan 2008 07:45:37 +0000 (07:45 +0000)]
Don't DCE FreeInst's. We were using those! Patch from Owen Anderson.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46553 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoPointers change size depending upon the target. Remove them to make the test more...
Tanya Lattner [Wed, 30 Jan 2008 05:15:15 +0000 (05:15 +0000)]
Pointers change size depending upon the target. Remove them to make the test more stable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46548 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoSkip over the label which marks the beginning of the function before inserting prolog...
Evan Cheng [Wed, 30 Jan 2008 03:57:33 +0000 (03:57 +0000)]
Skip over the label which marks the beginning of the function before inserting prologue code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46546 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoBack out mods to llvmAsmParser, incorrectly committed
Scott Michel [Wed, 30 Jan 2008 03:10:00 +0000 (03:10 +0000)]
Back out mods to llvmAsmParser, incorrectly committed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46545 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoMore cleanups for CellSPU:
Scott Michel [Wed, 30 Jan 2008 02:55:46 +0000 (02:55 +0000)]
More cleanups for CellSPU:

- Expand tabs... (poss 80-col violations, will get them later...)
- Consolidate logic for SelectDFormAddr and SelectDForm2Addr into a single
  function, simplifying maintenance. Also reduced custom instruction
  generation for SPUvecinsert/INSERT_MASK.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46544 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoMake DSE much more aggressive by performing DCE earlier. Update a testcase to reflec...
Owen Anderson [Wed, 30 Jan 2008 01:24:47 +0000 (01:24 +0000)]
Make DSE much more aggressive by performing DCE earlier.  Update a testcase to reflect this increased aggressiveness.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46542 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix an issue where, under very specific circumstances, memdep could end up dereferenc...
Owen Anderson [Wed, 30 Jan 2008 01:24:05 +0000 (01:24 +0000)]
Fix an issue where, under very specific circumstances, memdep could end up dereferencing the end
of one of its internal maps.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46541 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix a bug where scalarrepl would discard offset if type would match.
Chris Lattner [Wed, 30 Jan 2008 00:39:15 +0000 (00:39 +0000)]
Fix a bug where scalarrepl would discard offset if type would match.
In practice this can only happen on code with already undefined behavior,
but this is still a good thing to handle correctly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46539 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFactor the addressing mode and the load/store VT out of LoadSDNode
Dan Gohman [Wed, 30 Jan 2008 00:15:11 +0000 (00:15 +0000)]
Factor the addressing mode and the load/store VT out of LoadSDNode
and StoreSDNode into their common base class LSBaseSDNode. Member
functions getLoadedVT and getStoredVT are replaced with the common
getMemoryVT to simplify code that will handle both loads and stores.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46538 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoWork in progress. This patch *fixes* x86-64 calls which are modelled as StructRet...
Evan Cheng [Tue, 29 Jan 2008 19:34:22 +0000 (19:34 +0000)]
Work in progress. This patch *fixes* x86-64 calls which are modelled as StructRet but really should be return in registers, e.g. _Complex long double, some 128-bit aggregates. This is a short term solution that is necessary only because llvm, for now, cannot model i128 nor call's with multiple results.
Status: This only works for direct calls, and only the caller side is done. Disabled for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46527 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoUpdate this test case.
Evan Cheng [Tue, 29 Jan 2008 19:30:05 +0000 (19:30 +0000)]
Update this test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46526 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoWhen expanding an operand, it's not the result value
Duncan Sands [Tue, 29 Jan 2008 19:29:08 +0000 (19:29 +0000)]
When expanding an operand, it's not the result value
type that matters but the operand type.  This fixes
2008-01-08-IllegalCMP.ll which crashed with the new
legalize infrastructure because SETCC with result
type i8 and operand type i64 was being custom expanded
by the X86 backend.  With this fix, the gcc build gets
as far as the first libcall.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46525 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoDon't let globalopt hack on volatile loads or stores.
Chris Lattner [Tue, 29 Jan 2008 19:01:37 +0000 (19:01 +0000)]
Don't let globalopt hack on volatile loads or stores.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46523 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAltered ctor for APSInt to accept an optional "isUnsigned" flag. Default
Ted Kremenek [Tue, 29 Jan 2008 18:55:14 +0000 (18:55 +0000)]
Altered ctor for APSInt to accept an optional "isUnsigned" flag.  Default
behavior is unchanged.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46521 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoUse empty() instead of comparing size() with zero.
Dan Gohman [Tue, 29 Jan 2008 13:02:09 +0000 (13:02 +0000)]
Use empty() instead of comparing size() with zero.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46514 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix a typo in a comment.
Dan Gohman [Tue, 29 Jan 2008 12:43:50 +0000 (12:43 +0000)]
Fix a typo in a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46513 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoUpdate this file for 2.0 syntax.
Dan Gohman [Tue, 29 Jan 2008 12:35:50 +0000 (12:35 +0000)]
Update this file for 2.0 syntax.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46512 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd comments describing what the loop depth values mean. Also, make a
Dan Gohman [Tue, 29 Jan 2008 12:12:40 +0000 (12:12 +0000)]
Add comments describing what the loop depth values mean. Also, make a
minor simplification to the loop depth calculation code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46511 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix 80-col violations.
Dan Gohman [Tue, 29 Jan 2008 12:09:55 +0000 (12:09 +0000)]
Fix 80-col violations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46510 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRemove top-level const qualifiers from casts, avoiding associated
Dan Gohman [Tue, 29 Jan 2008 12:08:20 +0000 (12:08 +0000)]
Remove top-level const qualifiers from casts, avoiding associated
compiler warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46509 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix a typo in a comment.
Dan Gohman [Tue, 29 Jan 2008 12:07:11 +0000 (12:07 +0000)]
Fix a typo in a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46508 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix a typo in a comment.
Dan Gohman [Tue, 29 Jan 2008 12:06:33 +0000 (12:06 +0000)]
Fix a typo in a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46507 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd explicit keywords.
Dan Gohman [Tue, 29 Jan 2008 11:36:12 +0000 (11:36 +0000)]
Add explicit keywords.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46506 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agollvm.sqrt(-0.0) is defined.
Chris Lattner [Tue, 29 Jan 2008 07:00:44 +0000 (07:00 +0000)]
llvm.sqrt(-0.0) is defined.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46500 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoeliminate additions of 0.0 when they are obviously dead. This has to be careful to
Chris Lattner [Tue, 29 Jan 2008 06:52:45 +0000 (06:52 +0000)]
eliminate additions of 0.0 when they are obviously dead.  This has to be careful to
avoid turning -0.0 + 0.0 -> -0.0 which is incorrect.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46499 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd a testcase for eliminating memcpy's at the end of functions. Forgot to commit...
Owen Anderson [Tue, 29 Jan 2008 06:40:32 +0000 (06:40 +0000)]
Add a testcase for eliminating memcpy's at the end of functions.  Forgot to commit this with my last commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46497 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd target triples to these so they don't fail on linux.
Chris Lattner [Tue, 29 Jan 2008 06:26:07 +0000 (06:26 +0000)]
Add target triples to these so they don't fail on linux.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46496 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoUse getPreferredAlignmentLog or getPreferredAlignment
Duncan Sands [Tue, 29 Jan 2008 06:23:44 +0000 (06:23 +0000)]
Use getPreferredAlignmentLog or getPreferredAlignment
to get the alignment of global variables, rather than
using hand-made versions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46495 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd support for eliminating memcpy's at the end of functions. Also fix some errors...
Owen Anderson [Tue, 29 Jan 2008 06:18:36 +0000 (06:18 +0000)]
Add support for eliminating memcpy's at the end of functions.  Also fix some errors I noticed in
the handling of eliminating stores to byval arguments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46494 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoThis would be better done as an executable test.
Duncan Sands [Tue, 29 Jan 2008 06:04:54 +0000 (06:04 +0000)]
This would be better done as an executable test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46493 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAfter recent changes we fail to optimize this test
Duncan Sands [Tue, 29 Jan 2008 05:57:23 +0000 (05:57 +0000)]
After recent changes we fail to optimize this test
sufficiently to have it pass.  I'm removing it from
the testsuite and adding it to PR452 instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46492 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRegAllocBigBlock doesn't need LiveVariables either.
Owen Anderson [Tue, 29 Jan 2008 02:32:13 +0000 (02:32 +0000)]
RegAllocBigBlock doesn't need LiveVariables either.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46488 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix to bug 1951: tblgen gratuitously renames variables when no temporary was
Scott Michel [Tue, 29 Jan 2008 02:29:31 +0000 (02:29 +0000)]
Fix to bug 1951: tblgen gratuitously renames variables when no temporary was
generated. This feature would only show up in fairly complex patterns, such
as this one in CellSPU:

  def : Pat<(add (SPUhi tconstpool:$in, 0), (SPUlo tconstpool:$in, 0)),
    (IOHLlo (ILHUhi tconstpool:$in), tconstpool:$in)>;

which generated the following emit code:

SDNode *Emit_0(const SDOperand &N, unsigned Opc0, unsigned Opc1, MVT::ValueType VT0, MVT::ValueType VT1) DISABLE_INLINE {
  SDOperand N0 = N.getOperand(0);
  SDOperand N00 = N0.getOperand(0);
  SDOperand N01 = N0.getOperand(1);
  SDOperand N1 = N.getOperand(1);
  SDOperand N10 = N1.getOperand(0);
  SDOperand N11 = N1.getOperand(1);
  SDOperand Tmp3(CurDAG->getTargetNode(Opc0, VT0, N00), 0);
  return CurDAG->SelectNodeTo(N.Val, Opc1, VT1, Tmp3, Tmp2); /* Tmp2 s/b N00 */
}

Tested against the test suites without incident.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46487 91177308-0d34-0410-b5e6-96231b3b80d8