oota-llvm.git
14 years agoconvert cycle checker to smallptrset, add comments and make it
Chris Lattner [Wed, 24 Feb 2010 21:34:04 +0000 (21:34 +0000)]
convert cycle checker to smallptrset, add comments and make it
more elegant.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97059 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agorevert david's patch which does not even build.
Chris Lattner [Wed, 24 Feb 2010 21:25:08 +0000 (21:25 +0000)]
revert david's patch which does not even build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97057 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoUse a SmallPtrSet as suggested by Chris.
David Greene [Wed, 24 Feb 2010 20:59:49 +0000 (20:59 +0000)]
Use a SmallPtrSet as suggested by Chris.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97056 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdding function "lookupGCCName" to MBlazeIntrinsicInfo
Wesley Peck [Wed, 24 Feb 2010 20:16:27 +0000 (20:16 +0000)]
Adding function "lookupGCCName" to MBlazeIntrinsicInfo

Adding the function "lookupGCCName" to the MBlazeIntrinsicInfo
class to support the Clang MicroBlaze target.

Additionally, minor fixes which remove some unused PIC code
(PIC is not supported yet in the MicroBlaze backend) and
removed some unused variables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97054 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agocontract movechild+checktype into a new checkchild node, shrinking the
Chris Lattner [Wed, 24 Feb 2010 20:15:25 +0000 (20:15 +0000)]
contract movechild+checktype into a new checkchild node, shrinking the
x86 table by 1200 bytes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97053 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdded Vector Swap (VSWPd and VSWPq) instructions for disassembly only.
Johnny Chen [Wed, 24 Feb 2010 20:06:07 +0000 (20:06 +0000)]
Added Vector Swap (VSWPd and VSWPq) instructions for disassembly only.

A8.6.405

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97052 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agosplit the movechild/record/moveparent -> recordchild optzn into a
Chris Lattner [Wed, 24 Feb 2010 19:52:48 +0000 (19:52 +0000)]
split the movechild/record/moveparent -> recordchild optzn into a
movechild/record -> recordchild/movechild and
movechild/moveparent -> noop xforms.  This slightly shrinks the tables
(x86 to 117454) and enables adding future improvements.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97051 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoConstantFoldInstOperands can theoretically return null if it
Dan Gohman [Wed, 24 Feb 2010 19:31:47 +0000 (19:31 +0000)]
ConstantFoldInstOperands can theoretically return null if it
didn't fold anything.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97049 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSimplify this code; these casts aren't necessary.
Dan Gohman [Wed, 24 Feb 2010 19:31:06 +0000 (19:31 +0000)]
Simplify this code; these casts aren't necessary.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97048 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoemit a histogram of the opcodes in comments.
Chris Lattner [Wed, 24 Feb 2010 19:17:12 +0000 (19:17 +0000)]
emit a histogram of the opcodes in comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97047 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoStay away from str <undef> in ARMLoadStoreOpt. This pass does not understand
Jakob Stoklund Olesen [Wed, 24 Feb 2010 18:57:08 +0000 (18:57 +0000)]
Stay away from str <undef> in ARMLoadStoreOpt. This pass does not understand
<undef> operands, and can cause scavenger failures when it translates
<kill,undef> to <kill>.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97046 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFixed typo of opcodestr, should be "vst1", not "vld1".
Johnny Chen [Wed, 24 Feb 2010 18:00:40 +0000 (18:00 +0000)]
Fixed typo of opcodestr, should be "vst1", not "vld1".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97044 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoConvert a few more backedge-taken count functions to use BackedgeTakenInfo.
Dan Gohman [Wed, 24 Feb 2010 17:31:30 +0000 (17:31 +0000)]
Convert a few more backedge-taken count functions to use BackedgeTakenInfo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97042 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSpeculatively revert r97011, "Re-apply 96540 and 96556 with fixes.", again in
Daniel Dunbar [Wed, 24 Feb 2010 17:05:47 +0000 (17:05 +0000)]
Speculatively revert r97011, "Re-apply 96540 and 96556 with fixes.", again in
the hopes of fixing PPC bootstrap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97040 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoReapply r97010, the speculative revert failed.
Daniel Dunbar [Wed, 24 Feb 2010 08:48:04 +0000 (08:48 +0000)]
Reapply r97010, the speculative revert failed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97036 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agolets not break the old isel.
Chris Lattner [Wed, 24 Feb 2010 07:41:11 +0000 (07:41 +0000)]
lets not break the old isel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97034 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSince the new instruction selector now works, I don't need to keep
Chris Lattner [Wed, 24 Feb 2010 07:35:09 +0000 (07:35 +0000)]
Since the new instruction selector now works, I don't need to keep
the old one around for comparative purposes: have the
ENABLE_NEW_ISEL #define (which is not enabled on mainline) stop
emitting the old isel at all, yay for build time win.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97033 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoimplement a simple proof-of-concept optimization for
Chris Lattner [Wed, 24 Feb 2010 07:31:45 +0000 (07:31 +0000)]
implement a simple proof-of-concept optimization for
the new isel: fold movechild+record+moveparent into a
single recordchild N node.  This shrinks the X86 table
from 125443 to 117502 bytes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97031 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoThe new isel passes all tests, time to start making it go fast.
Chris Lattner [Wed, 24 Feb 2010 07:06:50 +0000 (07:06 +0000)]
The new isel passes all tests, time to start making it go fast.
Also add an easy macro at the top of DAGISelEmitter.cpp to enable
it.  Lets see if I can avoid accidentally turning it on :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97029 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd svn:ignore properties.
Dan Gohman [Wed, 24 Feb 2010 07:06:20 +0000 (07:06 +0000)]
Add svn:ignore properties.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97028 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSpeculatively revert r97010, "Add an argument to PHITranslateValue to specify
Daniel Dunbar [Wed, 24 Feb 2010 06:55:22 +0000 (06:55 +0000)]
Speculatively revert r97010, "Add an argument to PHITranslateValue to specify
the DominatorTree. ...", in hopes of restoring poor old PPC bootstrap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97027 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoWhen forming SSE min and max nodes for UGE and ULE comparisons, it's
Dan Gohman [Wed, 24 Feb 2010 06:52:40 +0000 (06:52 +0000)]
When forming SSE min and max nodes for UGE and ULE comparisons, it's
necessary to swap the operands to handle NaN and negative zero properly.

Also, reintroduce logic for checking for NaN conditions when forming
SSE min and max instructions, fixed to take into consideration NaNs and
negative zeros. This allows forming min and max instructions in more
cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97025 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix indentation.
Dan Gohman [Wed, 24 Feb 2010 06:46:09 +0000 (06:46 +0000)]
Fix indentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97024 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoChange the scheduler from adding nodes in allnodes order
Chris Lattner [Wed, 24 Feb 2010 06:11:37 +0000 (06:11 +0000)]
Change the scheduler from adding nodes in allnodes order
to adding them in a determinstic order (bottom up from
the root) based on the structure of the graph itself.

This updates tests for some random changes, interesting
bits: CodeGen/Blackfin/promote-logic.ll no longer crashes.
I have no idea why, but that's good right?

CodeGen/X86/2009-07-16-LoadFoldingBug.ll also fails, but
now compiles to have one fewer constant pool entry, making
the expected load that was being folded disappear.  Since it
is an unreduced mass of gnast, I just removed it.

This fixes PR6370

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97023 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemove an unused variable. Was this intentional?
Chandler Carruth [Wed, 24 Feb 2010 06:09:03 +0000 (06:09 +0000)]
Remove an unused variable. Was this intentional?

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97022 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoThe new isel was not properly handling patterns that covered
Chris Lattner [Wed, 24 Feb 2010 05:33:42 +0000 (05:33 +0000)]
The new isel was not properly handling patterns that covered
internal nodes with flag results.  Record these with a new
OPC_MarkFlagResults opcode and use this to update the interior
nodes' flag results properly.  This fixes CodeGen/X86/i256-add.ll
with the new isel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97021 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoadd node #'s to debug dumps.
Chris Lattner [Wed, 24 Feb 2010 04:24:44 +0000 (04:24 +0000)]
add node #'s to debug dumps.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97019 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdded for disassembly VST1 (multiple single elements) which stores elements to
Johnny Chen [Wed, 24 Feb 2010 02:57:20 +0000 (02:57 +0000)]
Added for disassembly VST1 (multiple single elements) which stores elements to
memory from three or four registers and VST2 (multiple two-element structures)
which stores to memory from two double-spaced registers.

A8.6.391 & A8.6.393

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97018 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoChanged the table generator so that the X86
Sean Callanan [Wed, 24 Feb 2010 02:56:25 +0000 (02:56 +0000)]
Changed the table generator so that the X86
disassembler never recognizes InitReg instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97017 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agohandle very large call frames when require SPAdj != 0 for Thumb1
Jim Grosbach [Wed, 24 Feb 2010 02:15:43 +0000 (02:15 +0000)]
handle very large call frames when require SPAdj != 0 for Thumb1

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97013 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoLowerCall() should always do getCopyFromReg() to reference the stack pointer.
Jim Grosbach [Wed, 24 Feb 2010 01:43:03 +0000 (01:43 +0000)]
LowerCall() should always do getCopyFromReg() to reference the stack pointer.
Machine instruction selection is much happier when operands are in virtual
registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97012 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRe-apply 96540 and 96556 with fixes.
Evan Cheng [Wed, 24 Feb 2010 01:42:31 +0000 (01:42 +0000)]
Re-apply 96540 and 96556 with fixes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97011 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd an argument to PHITranslateValue to specify the DominatorTree. If this
Bob Wilson [Wed, 24 Feb 2010 01:39:00 +0000 (01:39 +0000)]
Add an argument to PHITranslateValue to specify the DominatorTree.  If this
argument is non-null, pass it along to PHITranslateSubExpr so that it can
prefer using existing values that dominate the PredBB, instead of just
blindly picking the first equivalent value that it finds on a uselist.
Also when the DominatorTree is specified, have PHITranslateValue filter
out any result that does not dominate the PredBB.  This is basically just
refactoring the check that used to be in GetAvailablePHITranslatedSubExpr
and also in GVN.

Despite my initial expectations, this change does not affect the results
of GVN for any testcases that I could find, but it should help compile time.
Before this change, if PHITranslateSubExpr picked a value that does not
dominate, PHITranslateWithInsertion would then insert a new value, which GVN
would later determine to be redundant and would replace.  By picking a good
value to begin with, we save GVN the extra work of inserting and then
replacing a new value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97010 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoDIV8r must define %AX since X86DAGToDAGISel::Select() sometimes uses it
Jakob Stoklund Olesen [Wed, 24 Feb 2010 00:39:35 +0000 (00:39 +0000)]
DIV8r must define %AX since X86DAGToDAGISel::Select() sometimes uses it
instead of %AL/%AH.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97006 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agofix X86/uint_to_fp-2.ll, only fold loads when they have a
Chris Lattner [Tue, 23 Feb 2010 23:47:34 +0000 (23:47 +0000)]
fix X86/uint_to_fp-2.ll, only fold loads when they have a
single use.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97003 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agomake selectnodeto set the nodeid to -1. This makes it more akin to
Chris Lattner [Tue, 23 Feb 2010 23:01:35 +0000 (23:01 +0000)]
make selectnodeto set the nodeid to -1.  This makes it more akin to
creating a new node then replacing uses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97000 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemember to handle sub-registers when moving imp-defs to a rematted instruction.
Jakob Stoklund Olesen [Tue, 23 Feb 2010 22:44:02 +0000 (22:44 +0000)]
Remember to handle sub-registers when moving imp-defs to a rematted instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96995 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoKeep track of phi join registers explicitly in LiveVariables.
Jakob Stoklund Olesen [Tue, 23 Feb 2010 22:43:58 +0000 (22:43 +0000)]
Keep track of phi join registers explicitly in LiveVariables.

Previously, LiveIntervalAnalysis would infer phi joins by looking for multiply
defined registers. That doesn't work if the phi join is implicitly defined in
all but one of the predecessors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96994 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix rev 96389 by restricting the xform to mask that's either signbit or max signed...
Evan Cheng [Tue, 23 Feb 2010 21:51:54 +0000 (21:51 +0000)]
Fix rev 96389 by restricting the xform to mask that's either signbit or max signed value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96990 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoThis is the second patch to allow x86 code to be aligned with optimal nops.
Kevin Enderby [Tue, 23 Feb 2010 21:41:24 +0000 (21:41 +0000)]
This is the second patch to allow x86 code to be aligned with optimal nops.
With the compiler changed to use EmitCodeAlignment() it does change the
functionality.  But X86 assembly code assembled with llvm-mc does not change
its output.  For that we will eventually change the assembler frontend to
detect a '.align x, 0x90' when used in a section that 'hasInstructions' and use
EmitCodeAlignment, but will wait until we have better target hooks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96988 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoDon't mark call instruction as a barrier.
Richard Osborne [Tue, 23 Feb 2010 21:08:11 +0000 (21:08 +0000)]
Don't mark call instruction as a barrier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96983 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRoll back r96959 again.
Jeffrey Yasskin [Tue, 23 Feb 2010 20:53:37 +0000 (20:53 +0000)]
Roll back r96959 again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96981 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdded for disassembly VLD1 (multiple single elements) which loads memory into
Johnny Chen [Tue, 23 Feb 2010 20:51:23 +0000 (20:51 +0000)]
Added for disassembly VLD1 (multiple single elements) which loads memory into
three or four registers and VLD2 (multiple two-element structures) which loads
memory into two double-spaced registers.

A8.6.307 & A8.6.310

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96980 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoUse the module's context instead of the global context.
Nicolas Geoffray [Tue, 23 Feb 2010 19:42:44 +0000 (19:42 +0000)]
Use the module's context instead of the global context.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96977 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agonew test case for r96974.
Devang Patel [Tue, 23 Feb 2010 19:37:40 +0000 (19:37 +0000)]
new test case for r96974.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96975 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agono need to override IsLegalToFold, the base implementation
Chris Lattner [Tue, 23 Feb 2010 19:33:11 +0000 (19:33 +0000)]
no need to override IsLegalToFold, the base implementation
disables load folding at -O0.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96973 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agofix a bug in findNonImmUse (used by IsLegalToFold) where nodes with
Chris Lattner [Tue, 23 Feb 2010 19:32:27 +0000 (19:32 +0000)]
fix a bug in findNonImmUse (used by IsLegalToFold) where nodes with
no id's would cause early exit allowing IsLegalToFold to return true
instead of false, producing a cyclic dag.

This was striking the new isel because it isn't using SelectNodeTo yet,
which theoretically is just an optimization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96972 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoPrint node ID's in dumps and views if set.
Chris Lattner [Tue, 23 Feb 2010 19:31:18 +0000 (19:31 +0000)]
Print node ID's in dumps and views if set.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96971 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdding the MicroBlaze backend.
Wesley Peck [Tue, 23 Feb 2010 19:15:24 +0000 (19:15 +0000)]
Adding the MicroBlaze backend.

The MicroBlaze is a highly configurable 32-bit soft-microprocessor for
use on Xilinx FPGAs. For more information see:
http://www.xilinx.com/tools/microblaze.htm
http://en.wikipedia.org/wiki/MicroBlaze

The current LLVM MicroBlaze backend generates assembly which can be
compiled using the an appropriate binutils assembler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96969 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agouse OutStreamer.EmitCodeAlignment for alignment in the text
Chris Lattner [Tue, 23 Feb 2010 18:46:22 +0000 (18:46 +0000)]
use OutStreamer.EmitCodeAlignment for alignment in the text
segment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96967 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agogenericize MCAsmStreamer::EmitCodeAlignment to support other targets
Chris Lattner [Tue, 23 Feb 2010 18:44:31 +0000 (18:44 +0000)]
genericize MCAsmStreamer::EmitCodeAlignment to support other targets
so that it doesn't break them when the code generator starts using it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96966 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoOops. Pass -lgcc _only_ on ARM, not on everything except ARM.
Jeffrey Yasskin [Tue, 23 Feb 2010 18:40:48 +0000 (18:40 +0000)]
Oops.  Pass -lgcc _only_ on ARM, not on everything except ARM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96965 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoECALLF, ECALLT shouldn't be marked as barriers.
Richard Osborne [Tue, 23 Feb 2010 18:29:49 +0000 (18:29 +0000)]
ECALLF, ECALLT shouldn't be marked as barriers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96964 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoThis is the first patch to put the needed bits in place to eventually allow code
Kevin Enderby [Tue, 23 Feb 2010 18:26:34 +0000 (18:26 +0000)]
This is the first patch to put the needed bits in place to eventually allow code
to be aligned with optimal nops.  This patch does not change any functionality
and when the compiler is changed to use EmitCodeAlignment() it should also not
change the resulting output.  Once the compiler change is made and everything
looks good the next patch with the table of optimal X86 nops will be added to
WriteNopData() changing the output.  There are many FIXMEs in this patch which
will be removed when we have better target hooks (coming soon I hear).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96963 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoMark unconditional branches as barriers. Found using -verify-machineinstrs
Richard Osborne [Tue, 23 Feb 2010 18:13:38 +0000 (18:13 +0000)]
Mark unconditional branches as barriers. Found using -verify-machineinstrs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96960 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRoll r96559 forward again, adding libLLVM-2.7svn.so to LLVM. This links 3 of
Jeffrey Yasskin [Tue, 23 Feb 2010 18:10:07 +0000 (18:10 +0000)]
Roll r96559 forward again, adding libLLVM-2.7svn.so to LLVM.  This links 3 of
the examples shared to make sure the shared library keeps working.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96959 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSpeed up cycle checking significantly by caching results.
David Greene [Tue, 23 Feb 2010 17:37:50 +0000 (17:37 +0000)]
Speed up cycle checking significantly by caching results.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96956 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoDon't do (X != Y) ? X : Y -> X for floating-point values; it doesn't
Dan Gohman [Tue, 23 Feb 2010 17:17:57 +0000 (17:17 +0000)]
Don't do (X != Y) ? X : Y  -> X for floating-point values; it doesn't
handle NaN properly.

Do (X une Y) ? X : Y  -> X if one of X and Y is not zero.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96955 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSpelling. s/suppor /support /
Jim Grosbach [Tue, 23 Feb 2010 17:16:27 +0000 (17:16 +0000)]
Spelling. s/suppor /support /

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96954 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemove the code which constant-folded ptrtoint(inttoptr(x)+c) to
Dan Gohman [Tue, 23 Feb 2010 16:35:41 +0000 (16:35 +0000)]
Remove the code which constant-folded ptrtoint(inttoptr(x)+c) to
getelementptr. Despite only doing so in the case where x is a known
array object and c can be converted to an index within range, this
could still be invalid if c is actually the address of an object
allocated outside of LLVM. Also, SCEVExpander, the original motivation
for this code, has since been improved to avoid inttoptr+ptroint in
more cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96950 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoModified examples Makefile to only build the ExceptionDemo example for
Garrison Venn [Tue, 23 Feb 2010 16:27:59 +0000 (16:27 +0000)]
Modified examples Makefile to only build the ExceptionDemo example for
x86 and x86_64 on UNIX systems. Only OS X 10.6.2 (x86_64) and 32bit CentOS 5.2
with gcc 4.1.2 were tested. ARM UNIX build triggered failure motivating this
modification, as it seems that the ARM ABI does not support _Unwind_GetIP(...),
_Unwind_SetGR(...), and _Unwind_SetIP(...). From doing a quick browse of:
http://infocenter.arm.com/help/topic/com.arm.doc.ihi0038a/IHI0038A_ehabi.pdf,
it seems as if all other exception related apis are supported. Looks like
the port can be done to ARM. Thanks to Xerxes RĂ„nby <xerxes@zafena.se> for
pointing out this error.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96949 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoTesting subversion commit access
Wesley Peck [Tue, 23 Feb 2010 15:11:17 +0000 (15:11 +0000)]
Testing subversion commit access

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96948 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix -mtune forwarding.
Mikhail Glushenkov [Tue, 23 Feb 2010 14:29:42 +0000 (14:29 +0000)]
Fix -mtune forwarding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96944 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemove unused lowering function LowerJumpTable
Richard Osborne [Tue, 23 Feb 2010 14:17:20 +0000 (14:17 +0000)]
Remove unused lowering function LowerJumpTable

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96943 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoLower BR_JT on the XCore to a jump into a series of jump instructions.
Richard Osborne [Tue, 23 Feb 2010 13:25:07 +0000 (13:25 +0000)]
Lower BR_JT on the XCore to a jump into a series of jump instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96942 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agotests: Don't make a missing llvm-gcc dir a fatal error.
Daniel Dunbar [Tue, 23 Feb 2010 11:34:12 +0000 (11:34 +0000)]
tests: Don't make a missing llvm-gcc dir a fatal error.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96938 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSwitch .bc/.ll Makefile rules to use LLVM{CC,CXX} instead of LLVMG{CC,XX}
Daniel Dunbar [Tue, 23 Feb 2010 10:28:06 +0000 (10:28 +0000)]
Switch .bc/.ll Makefile rules to use LLVM{CC,CXX} instead of LLVMG{CC,XX}

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96936 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd LLVM{CC,CXX} make variables, which specify the configured path the LLVM
Daniel Dunbar [Tue, 23 Feb 2010 10:00:53 +0000 (10:00 +0000)]
Add LLVM{CC,CXX} make variables, which specify the configured path the LLVM
capable compilers (which could be llvm-gcc or clang).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96935 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoInitial configure support for using Clang as the LLVM capable compiler.
Daniel Dunbar [Tue, 23 Feb 2010 10:00:49 +0000 (10:00 +0000)]
Initial configure support for using Clang as the LLVM capable compiler.

Comes in two parts:
 1. Use --with-clang=path/to/clang/compiler to select an installed clang, or
    --with-built-clang to have the makefiles use the clang which will be built
    as the LLVM capable compiler. If neither is given, --with-built-clang will
    be used if the Clang sources are checked out into the standard location
    (tools/clang).

 2. Use --with-llvmcc={llvm-gcc,clang,none} to specify which LLVM capable
    compiler to use. If not given, then llvm-gcc will be used if available,
    otherwise Clang.

Makefile support still to come.

Eric, Doug, Chris, seem reasonable?

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96934 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoUpdate mcc16 and the ancient Clang plugin for the 'cmd_line' -> 'command' change.
Mikhail Glushenkov [Tue, 23 Feb 2010 09:59:30 +0000 (09:59 +0000)]
Update mcc16 and the ancient Clang plugin for the 'cmd_line' -> 'command' change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96933 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoEliminate CFERuntimeLibDir make variable, this shouldn't be needed.
Daniel Dunbar [Tue, 23 Feb 2010 09:28:50 +0000 (09:28 +0000)]
Eliminate CFERuntimeLibDir make variable, this shouldn't be needed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96932 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoFix a thinko in the lit.cfg.
Daniel Dunbar [Tue, 23 Feb 2010 09:28:48 +0000 (09:28 +0000)]
Fix a thinko in the lit.cfg.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96931 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoInput files with empty suffixes must be passed to linker.
Mikhail Glushenkov [Tue, 23 Feb 2010 09:05:21 +0000 (09:05 +0000)]
Input files with empty suffixes must be passed to linker.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96927 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSupport -Xlinker et al.
Mikhail Glushenkov [Tue, 23 Feb 2010 09:05:15 +0000 (09:05 +0000)]
Support -Xlinker et al.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96926 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoTemporary disable response files.
Mikhail Glushenkov [Tue, 23 Feb 2010 09:05:10 +0000 (09:05 +0000)]
Temporary disable response files.

They are giving us problems on Mac.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96925 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoTypo.
Mikhail Glushenkov [Tue, 23 Feb 2010 09:05:06 +0000 (09:05 +0000)]
Typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96924 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoTrailing whitespace.
Mikhail Glushenkov [Tue, 23 Feb 2010 09:05:01 +0000 (09:05 +0000)]
Trailing whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96923 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoImplement order-preserving option forwarding.
Mikhail Glushenkov [Tue, 23 Feb 2010 09:04:57 +0000 (09:04 +0000)]
Implement order-preserving option forwarding.

Needed to correctly handle things like 'llvmc -framework Foo foo.o -framework
Bar bar.o' - before this commit all '-framework' options would've been grouped
together in the beginning.

Due to our dependence on CommandLine this turned out to be a giant hack; we will
migrate away from CommandLine eventually.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96922 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoUpdate the test suite.
Mikhail Glushenkov [Tue, 23 Feb 2010 09:04:51 +0000 (09:04 +0000)]
Update the test suite.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96921 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoCorrect option forwarding: initial implementation.
Mikhail Glushenkov [Tue, 23 Feb 2010 09:04:44 +0000 (09:04 +0000)]
Correct option forwarding: initial implementation.

Does not work, but the infrastructure changes are in place.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96920 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoPrecompiled headers: initial support.
Mikhail Glushenkov [Tue, 23 Feb 2010 09:04:33 +0000 (09:04 +0000)]
Precompiled headers: initial support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96919 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoNew experimental/undocumented feature: 'works_on_empty'.
Mikhail Glushenkov [Tue, 23 Feb 2010 09:04:28 +0000 (09:04 +0000)]
New experimental/undocumented feature: 'works_on_empty'.

For now, just enough support to make -filelist work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96918 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoSupport '-install_name'.
Mikhail Glushenkov [Tue, 23 Feb 2010 09:04:18 +0000 (09:04 +0000)]
Support '-install_name'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96917 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoAdd a way to enable '-opt=foo' forwarding.
Mikhail Glushenkov [Tue, 23 Feb 2010 09:04:13 +0000 (09:04 +0000)]
Add a way to enable '-opt=foo' forwarding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96916 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoInline and eliminate LLVMG{CC,XX}WITHPATH.
Daniel Dunbar [Tue, 23 Feb 2010 07:56:41 +0000 (07:56 +0000)]
Inline and eliminate LLVMG{CC,XX}WITHPATH.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96913 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemove ancient Xcode project, replaced by CMake project.
Daniel Dunbar [Tue, 23 Feb 2010 07:56:38 +0000 (07:56 +0000)]
Remove ancient Xcode project, replaced by CMake project.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96912 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoRemove dead LUPGRADE make variable.
Daniel Dunbar [Tue, 23 Feb 2010 07:56:36 +0000 (07:56 +0000)]
Remove dead LUPGRADE make variable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96911 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoKill off unused LLVMGCCLIBEXEC make variable.
Daniel Dunbar [Tue, 23 Feb 2010 07:56:34 +0000 (07:56 +0000)]
Kill off unused LLVMGCCLIBEXEC make variable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96910 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoKill off LLVMGCCARCH and LLVMGCC_VERSION make variables.
Daniel Dunbar [Tue, 23 Feb 2010 07:56:31 +0000 (07:56 +0000)]
Kill off LLVMGCCARCH and LLVMGCC_VERSION make variables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96909 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoEliminate llvmgcc_version testing variable.
Daniel Dunbar [Tue, 23 Feb 2010 07:56:28 +0000 (07:56 +0000)]
Eliminate llvmgcc_version testing variable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96908 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoKill off LLVMGCC_MAJVERS make variable.
Daniel Dunbar [Tue, 23 Feb 2010 07:56:22 +0000 (07:56 +0000)]
Kill off LLVMGCC_MAJVERS make variable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96907 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoKill unused llvmgccmajvers testing variable.
Daniel Dunbar [Tue, 23 Feb 2010 07:56:18 +0000 (07:56 +0000)]
Kill unused llvmgccmajvers testing variable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96906 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoadd some #if 0'd out code for checking that named values in
Chris Lattner [Tue, 23 Feb 2010 07:50:58 +0000 (07:50 +0000)]
add some #if 0'd out code for checking that named values in
input/output patterns have the same type.  It turns out that
this triggers all the time because we don't infer types
between these boundaries.  Until we do, don't turn this on.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96905 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoReject patterns that use a name multiple times in the src or result
Chris Lattner [Tue, 23 Feb 2010 07:22:28 +0000 (07:22 +0000)]
Reject patterns that use a name multiple times in the src or result
of a pattern and where the uses have different types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96904 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agodisable two patterns that are using non-sensical result pattern types.
Chris Lattner [Tue, 23 Feb 2010 07:21:15 +0000 (07:21 +0000)]
disable two patterns that are using non-sensical result pattern types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96903 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoremove a confused pattern that is trying to match an address
Chris Lattner [Tue, 23 Feb 2010 07:16:12 +0000 (07:16 +0000)]
remove a confused pattern that is trying to match an address
then use it as an MMX register (!?).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96901 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoreject patterns that have dead named arguments in the input pattern
Chris Lattner [Tue, 23 Feb 2010 06:55:24 +0000 (06:55 +0000)]
reject patterns that have dead named arguments in the input pattern
this is tidier and can find bugs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96900 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoremove a bunch of dead named arguments in input patterns,
Chris Lattner [Tue, 23 Feb 2010 06:54:29 +0000 (06:54 +0000)]
remove a bunch of dead named arguments in input patterns,
though some look dubious afaict, these are all ok.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96899 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agoreject patterns that mention a name in the destination pattern
Chris Lattner [Tue, 23 Feb 2010 06:35:45 +0000 (06:35 +0000)]
reject patterns that mention a name in the destination pattern
but not in the input.  Previously, this would trigger an abort
late in the isel logic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96898 91177308-0d34-0410-b5e6-96231b3b80d8

14 years agomerge some code.
Chris Lattner [Tue, 23 Feb 2010 06:16:51 +0000 (06:16 +0000)]
merge some code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96896 91177308-0d34-0410-b5e6-96231b3b80d8