oota-llvm.git
11 years agoInstCombine: Fold more shuffles of shuffles.
Jim Grosbach [Tue, 30 Apr 2013 20:43:52 +0000 (20:43 +0000)]
InstCombine: Fold more shuffles of shuffles.

Always fold a shuffle-of-shuffle into a single shuffle when there's only one
input vector in the first place. Continue to be more conservative when there's
multiple inputs.

rdar://13402653
PR15866

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180802 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[mips] Clear isCommutable bit of instructions which are not commutable.
Akira Hatanaka [Tue, 30 Apr 2013 20:40:39 +0000 (20:40 +0000)]
[mips] Clear isCommutable bit of instructions which are not commutable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180801 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLocalStackSlotAllocation improvements
Hal Finkel [Tue, 30 Apr 2013 20:04:37 +0000 (20:04 +0000)]
LocalStackSlotAllocation improvements

First, taking advantage of the fact that the virtual base registers are allocated in order of the local frame offsets, remove the quadratic register-searching behavior. Because of the ordering, we only need to check the last virtual base register created.

Second, store the frame index in the FrameRef structure, and get the frame index and the local offset from this structure at the top of the loop iteration. This allows us to de-nest the loops in insertFrameReferenceRegisters (and I think makes the code cleaner). I also moved the needsFrameBaseReg check into the first loop over instructions so that we don't bother pushing FrameRefs for instructions that don't want a virtual base register anyway.

Lastly, and this is the only functionality change, avoid the creation of single-use virtual base registers. These are currently not useful because, in general, they end up replacing what would be one r+r instruction with an add and a r+i instruction. Committing this removes the XFAIL in CodeGen/PowerPC/2007-09-07-LoadStoreIdxForms.ll

Jim has okayed this off-list.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180799 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoText files should not be marked executable.
Rafael Espindola [Tue, 30 Apr 2013 19:06:15 +0000 (19:06 +0000)]
Text files should not be marked executable.

Patch by Oliver Pinter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180797 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTBAA: remove !tbaa from testing cases if not used.
Manman Ren [Tue, 30 Apr 2013 17:52:57 +0000 (17:52 +0000)]
TBAA: remove !tbaa from testing cases if not used.

This will make it easier to turn on struct-path aware TBAA since the metadata
format will change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180796 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSpelling. Thanks, Eric.
Adrian Prantl [Tue, 30 Apr 2013 17:33:32 +0000 (17:33 +0000)]
Spelling. Thanks, Eric.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180794 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSet debug locations for branch instructions created during inlining, even
Adrian Prantl [Tue, 30 Apr 2013 17:08:16 +0000 (17:08 +0000)]
Set debug locations for branch instructions created during inlining, even
the inlined function has multiple returns.

rdar://problem/12415623

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180793 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoChange getSlotIndex to return unsigned.
Rafael Espindola [Tue, 30 Apr 2013 16:53:38 +0000 (16:53 +0000)]
Change getSlotIndex to return unsigned.

The actual storage was already using unsigned, but the interface was using
uint64_t. This is wasteful on 32 bits and looks to be the root causes of
a miscompilation on Windows where a value was being sign extended to 64bits
to compare with the result of getSlotIndex.

Patch by Pasi Parviainen!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180791 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix Addend computation for non external relocations on Macho.
Rafael Espindola [Tue, 30 Apr 2013 15:40:54 +0000 (15:40 +0000)]
Fix Addend computation for non external relocations on Macho.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180790 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: fix loop-address.ll test
Vincent Lejeune [Tue, 30 Apr 2013 12:47:56 +0000 (12:47 +0000)]
R600: fix loop-address.ll test

Texture cache is now used when shader type is not specified

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180785 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix a bug in foldSelectICmpAndOr.
David Majnemer [Tue, 30 Apr 2013 10:36:33 +0000 (10:36 +0000)]
Fix a bug in foldSelectICmpAndOr.

Differences in bitwidth between X and Y could exist even if C1 and C2 have
the same Log2 representation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180779 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agos tightens up the encoding description for ARM post-indexed ldr instructions. All...
Mihai Popa [Tue, 30 Apr 2013 09:00:12 +0000 (09:00 +0000)]
s tightens up the encoding description for ARM post-indexed ldr instructions. All instructions in this class have bit 4 cleared. It turns out that there is a test case for this, but it was marked XFAIL.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180778 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix "Combine bit test + conditional or into simple math"
David Majnemer [Tue, 30 Apr 2013 08:57:58 +0000 (08:57 +0000)]
Fix "Combine bit test + conditional or into simple math"

This fixes the optimization introduced in r179748 and reverted in r179750.

While the optimization was sound, it did not properly respect differences in
bit-width.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180777 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRewrite X86 codegen regression test with FileCheck
Michael Liao [Tue, 30 Apr 2013 07:51:08 +0000 (07:51 +0000)]
Rewrite X86 codegen regression test with FileCheck

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180776 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRefactoring patch.
Stepan Dyatkovskiy [Tue, 30 Apr 2013 07:19:58 +0000 (07:19 +0000)]
Refactoring patch.
1. VarArgStyleRegisters: functionality that emits "store" instructions for byval regs moved out into separated method "StoreByValRegs". Before this patch VarArgStyleRegisters had confused use-cases. It was used for both variadic functions and for regular functions with byval parameters. In last case it created new stack-frame and registered it as VarArg frame, that is wrong.

This patch replaces VarArgsStyleRegisters usage for byval parameters with StoreByValRegs method.

2. In ARMMachineFunctionInfo, "get/setVarArgsRegSaveSize" was renamed to "get/setArgRegsSaveSize". By the same reason. Sometimes it was used for variadic functions, and sometimes for byval parameters in regular functions. Actually, this property means the size of registers, that keeps arguments, and thats why it was renamed.

3. In ARMISelLowering.cpp, ARMTargetLowering class, in methods computeRegArea and StoreByValRegs, VARegXXXXXX was renamed to ArgRegsXXXXXX still by the same reasons.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180774 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTry to fix ProgramTest on FreeBSD
Reid Kleckner [Tue, 30 Apr 2013 04:30:41 +0000 (04:30 +0000)]
Try to fix ProgramTest on FreeBSD

This seemed like the cleanest way to find the test executable.  Also fix
the file mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180770 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCollect the Addend for external relocs.
Rafael Espindola [Tue, 30 Apr 2013 01:29:57 +0000 (01:29 +0000)]
Collect the Addend for external relocs.

This fixes 2013-04-04-RelocAddend.ll. We don't have a testcase for non external
relocs with an Addend. I will try to write one.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180767 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: Always use texture cache for compute shaders
Vincent Lejeune [Tue, 30 Apr 2013 00:14:44 +0000 (00:14 +0000)]
R600: Always use texture cache for compute shaders

This will improve the performance of memory reads.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180762 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: use native for alu
Vincent Lejeune [Tue, 30 Apr 2013 00:14:38 +0000 (00:14 +0000)]
R600: use native for alu

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180761 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: Packetize instructions
Vincent Lejeune [Tue, 30 Apr 2013 00:14:27 +0000 (00:14 +0000)]
R600: Packetize instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180760 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: Rework Scheduling to handle difference between VLIW4 and VLIW5 chips
Vincent Lejeune [Tue, 30 Apr 2013 00:14:17 +0000 (00:14 +0000)]
R600: Rework Scheduling to handle difference between VLIW4 and VLIW5 chips

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180759 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: Add a Bank Swizzle operand
Vincent Lejeune [Tue, 30 Apr 2013 00:14:08 +0000 (00:14 +0000)]
R600: Add a Bank Swizzle operand

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180758 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: Take inner dependency into tex/vtx clauses
Vincent Lejeune [Tue, 30 Apr 2013 00:14:00 +0000 (00:14 +0000)]
R600: Take inner dependency into tex/vtx clauses

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180757 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: Turn TEX/VTX into native instructions
Vincent Lejeune [Tue, 30 Apr 2013 00:13:53 +0000 (00:13 +0000)]
R600: Turn TEX/VTX into native instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180756 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: Add FetchInst bit to instruction defs to denote vertex/tex instructions
Vincent Lejeune [Tue, 30 Apr 2013 00:13:39 +0000 (00:13 +0000)]
R600: Add FetchInst bit to instruction defs to denote vertex/tex instructions

v2[Vincent Lejeune]: Split FetchInst into usesTextureCache/usesVertexCache

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180755 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRewrite test in FileCheck instead of grep in X86 codegen
Michael Liao [Tue, 30 Apr 2013 00:13:38 +0000 (00:13 +0000)]
Rewrite test in FileCheck instead of grep in X86 codegen

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180754 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: Add some new processor variants
Vincent Lejeune [Tue, 30 Apr 2013 00:13:27 +0000 (00:13 +0000)]
R600: Add some new processor variants

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180753 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: Clean up instruction class definitions
Vincent Lejeune [Tue, 30 Apr 2013 00:13:20 +0000 (00:13 +0000)]
R600: Clean up instruction class definitions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180752 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: config section now reports use of killgt
Vincent Lejeune [Tue, 30 Apr 2013 00:13:13 +0000 (00:13 +0000)]
R600: config section now reports use of killgt

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180751 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRevert the command line option patch. However, keep the part that makes this pass...
Bill Wendling [Mon, 29 Apr 2013 23:48:06 +0000 (23:48 +0000)]
Revert the command line option patch. However, keep the part that makes this pass on Windows. I.e., we don't emit the target dependent attributes in a comment before the function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180750 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTBAA: remove !tbaa from testing cases if not used.
Manman Ren [Mon, 29 Apr 2013 22:58:55 +0000 (22:58 +0000)]
TBAA: remove !tbaa from testing cases if not used.

This will make it easier to turn on struct-path aware TBAA since the metadata
format will change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180745 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDuplicate a testcase.
Bill Wendling [Mon, 29 Apr 2013 22:42:47 +0000 (22:42 +0000)]
Duplicate a testcase.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180744 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTBAA: remove !tbaa from testing cases if not used.
Manman Ren [Mon, 29 Apr 2013 22:42:01 +0000 (22:42 +0000)]
TBAA: remove !tbaa from testing cases if not used.

This will make it easier to turn on struct-path aware TBAA since the metadata
format will change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180743 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRewrite some tests with FileCHeck in X86 codegen
Michael Liao [Mon, 29 Apr 2013 22:41:29 +0000 (22:41 +0000)]
Rewrite some tests with FileCHeck in X86 codegen

- Revise previous patches of the same purpose by fixing
  *) grep <PA> | not grep <PB> semantically is not the same as
     CHECK: <PA>{{^<PB>.*$}} as the former will check all occurrences of <PA>
     while the later only check the first match. As the result, CHECK needs
     putting in all place where <PA> occurs.
  *) grep <PA> | count <N> needs a final CHECK-NOT of the same pattern.
     (As 'CHECK-<N>' is proposed for discussion, converting 'grep | count <N>'
      where N > 1 is postponed.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180742 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoImprove documentation.
Adrian Prantl [Mon, 29 Apr 2013 22:25:52 +0000 (22:25 +0000)]
Improve documentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180738 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoEmit the TLS initialization function pointers into the correct section.
Bill Wendling [Mon, 29 Apr 2013 22:25:40 +0000 (22:25 +0000)]
Emit the TLS initialization function pointers into the correct section.

The `llvm.tls_init_funcs' (created by the front-end) holds pointers to the TLS
initialization functions. These need to be placed into the correct section so
that they are run before `main()'.

<rdar://problem/13733006>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180737 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd getSymbolAlignment to the ObjectFile interface.
Rafael Espindola [Mon, 29 Apr 2013 22:24:22 +0000 (22:24 +0000)]
Add getSymbolAlignment to the ObjectFile interface.

For regular object files this is only meaningful for common symbols. An object
file format with direct support for atoms should be able to provide alignment
information for all symbols.

This replaces getCommonSymbolAlignment and fixes
test-common-symbols-alignment.ll on darwin. This also includes a fix to
MachOObjectFile::getSymbolFlags. It was marking undefined symbols as common
(already tested by existing mcjit tests now that it is used).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180736 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: Use correct CF_END instruction on Northern Island GPUs
Tom Stellard [Mon, 29 Apr 2013 22:23:58 +0000 (22:23 +0000)]
R600: Use correct CF_END instruction on Northern Island GPUs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180735 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: Fix encoding of CF_END_{EG, R600} instructions
Tom Stellard [Mon, 29 Apr 2013 22:23:54 +0000 (22:23 +0000)]
R600: Fix encoding of CF_END_{EG, R600} instructions

The EOP bit was not being encoded.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180734 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRationalize what is public in RuntimeDyldMachO and RuntimeDyldELF.
Rafael Espindola [Mon, 29 Apr 2013 22:06:33 +0000 (22:06 +0000)]
Rationalize what is public in RuntimeDyldMachO and RuntimeDyldELF.

The implemented RuntimeDyldImpl interface is public. Everything else is private.
Since these classes are not inherited from (yet), there is no need to have
protected members.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180733 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSimplifyCFG: If convert single conditional stores
Arnold Schwaighofer [Mon, 29 Apr 2013 21:28:24 +0000 (21:28 +0000)]
SimplifyCFG: If convert single conditional stores

This resurrects r179957, but adds code that makes sure we don't touch
atomic/volatile stores:

This transformation will transform a conditional store with a preceeding
uncondtional store to the same location:

 a[i] =
 may-alias with a[i] load
 if (cond)
   a[i] = Y

into an unconditional store.

 a[i] = X
 may-alias with a[i] load
 tmp = cond ? Y : X;
 a[i] = tmp

We assume that on average the cost of a mispredicted branch is going to be
higher than the cost of a second store to the same location, and that the
secondary benefits of creating a bigger basic block for other optimizations to
work on outway the potential case where the branch would be correctly predicted
and the cost of the executing the second store would be noticably reflected in
performance.

hmmer's execution time improves by 30% on an imac12,2 on ref data sets. With
this change we are on par with gcc's performance (gcc also performs this
transformation). There was a 1.2 % performance improvement on a ARM swift chip.
Other tests in the test-suite+external seem to be mostly uninfluenced in my
experiments:
This optimization was triggered on 41 tests such that the executable was
different before/after the patch. Only 1 out of the 40 tests (dealII) was
reproducable below 100% (by about .4%). Given that hmmer benefits so much I
believe this to be a fair trade off.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180731 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDisable the MCJIT tests on 32 bit darwin.
Rafael Espindola [Mon, 29 Apr 2013 21:09:32 +0000 (21:09 +0000)]
Disable the MCJIT tests on 32 bit darwin.

I recently enabled them on 32 and 64 bit darwin, but it looks like 32 bit is
still fairly broken.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180730 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove unused method.
Rafael Espindola [Mon, 29 Apr 2013 20:00:27 +0000 (20:00 +0000)]
Remove unused method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180729 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUpdate the documentation.
Rafael Espindola [Mon, 29 Apr 2013 19:33:51 +0000 (19:33 +0000)]
Update the documentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180725 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUse a RelocationRef instead of a relocation_iterator.
Rafael Espindola [Mon, 29 Apr 2013 19:03:21 +0000 (19:03 +0000)]
Use a RelocationRef instead of a relocation_iterator.
No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180723 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRevert "revert r179735, it has no testcases, and doesn't really make sense."
Reid Kleckner [Mon, 29 Apr 2013 18:23:53 +0000 (18:23 +0000)]
Revert "revert r179735, it has no testcases, and doesn't really make sense."

This un-reverts r179735 and reverts commit r180574.

This fixes assertion failures for me locally and should fix the failures
on Windows reported widely on llvm-dev.  We should check if the bots
caught this and if so why not.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180722 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoExposing MCJIT through C API
Andrew Kaylor [Mon, 29 Apr 2013 17:49:40 +0000 (17:49 +0000)]
Exposing MCJIT through C API

Re-submitting with fix for OCaml dependency problems (removing dependency on SectionMemoryManager when it isn't used).

Patch by Fili Pizlo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180720 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoPropagate relocation info to resolveRelocation.
Rafael Espindola [Mon, 29 Apr 2013 17:24:34 +0000 (17:24 +0000)]
Propagate relocation info to resolveRelocation.

This gets most of the MCJITs tests passing with MachO.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180716 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoReplace ObjRelocationInfo with relocation_iterator.
Rafael Espindola [Mon, 29 Apr 2013 14:44:23 +0000 (14:44 +0000)]
Replace ObjRelocationInfo with relocation_iterator.

For MachO we need information that is not represented in ObjRelocationInfo.
Instead of copying the bits we think are needed from a relocation_iterator,
just pass the relocation_iterator down to the format specific functions.

No functionality change yet as we still drop the information once
processRelocationRef returns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180711 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd in some conditional compilation in order to silence an unused variable warning.
Michael Gottesman [Mon, 29 Apr 2013 07:29:08 +0000 (07:29 +0000)]
Add in some conditional compilation in order to silence an unused variable warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180700 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[objc-arc] Apply the RV optimization to retains next to calls in ObjCARCContract...
Michael Gottesman [Mon, 29 Apr 2013 06:53:53 +0000 (06:53 +0000)]
[objc-arc] Apply the RV optimization to retains next to calls in ObjCARCContract instead of ObjCARCOpts.

Turning retains into retainRV calls disrupts the data flow analysis in
ObjCARCOpts. Thus we move it as late as we can by moving it into
ObjCARCContract.

We leave in the conversion from retainRV -> retain in ObjCARCOpt since
it enables the dataflow analysis.

rdar://10813093

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180698 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdded statistics to count the number of retains/releases before/after optimization.
Michael Gottesman [Mon, 29 Apr 2013 06:16:57 +0000 (06:16 +0000)]
Added statistics to count the number of retains/releases before/after optimization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180697 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemoved trailing whitespace.
Michael Gottesman [Mon, 29 Apr 2013 06:16:55 +0000 (06:16 +0000)]
Removed trailing whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180696 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix for r180693. = /.
Michael Gottesman [Mon, 29 Apr 2013 05:25:39 +0000 (05:25 +0000)]
Fix for r180693. = /.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180694 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[objc-arc-annotations] Moved the disabling of call movement to ConnectTDBUTraversals...
Michael Gottesman [Mon, 29 Apr 2013 05:13:13 +0000 (05:13 +0000)]
[objc-arc-annotations] Moved the disabling of call movement to ConnectTDBUTraversals so that I can prevent Changed = true from being set. This prevents an infinite loop.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180693 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoInline variable into the #ifdef block where it's used.
Benjamin Kramer [Sun, 28 Apr 2013 07:47:04 +0000 (07:47 +0000)]
Inline variable into the #ifdef block where it's used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180688 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAArch64 InstrFormats:
Jia Liu [Sun, 28 Apr 2013 01:45:11 +0000 (01:45 +0000)]
AArch64 InstrFormats:
delete blank.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180687 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix typo. Stupid me.
Joerg Sonnenberger [Sat, 27 Apr 2013 22:32:54 +0000 (22:32 +0000)]
Fix typo. Stupid me.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180686 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRegen.
Eric Christopher [Sat, 27 Apr 2013 22:23:16 +0000 (22:23 +0000)]
Regen.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180685 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoOnly use cxxabi.h's demangler, if it is actually available.
Joerg Sonnenberger [Sat, 27 Apr 2013 22:12:32 +0000 (22:12 +0000)]
Only use cxxabi.h's demangler, if it is actually available.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180684 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoHandle tied sub-operands in AsmMatcherEmitter
Ulrich Weigand [Sat, 27 Apr 2013 18:48:23 +0000 (18:48 +0000)]
Handle tied sub-operands in AsmMatcherEmitter

The problem this patch addresses is the handling of register tie
constraints in AsmMatcherEmitter, where one operand is tied to a
sub-operand of another operand.  The typical scenario for this to
happen is the tie between the "write-back" register of a pre-inc
instruction, and the base register sub-operand of the memory address
operand of that instruction.

The current AsmMatcherEmitter code attempts to handle tied
operands by emitting the operand as usual first, and emitting
a CVT_Tied node when handling the second (tied) operand.  However,
this really only works correctly if the tied operand does not
have sub-operands (and isn't a sub-operand itself).  Under those
circumstances, a wrong MC operand list is generated.

In discussions with Jim Grosbach, it turned out that the MC operand
list really ought not to contain tied operands in the first place;
instead, it ought to consist of exactly those operands that are
named in the AsmString.  However, getting there requires significant
rework of (some) targets.

This patch fixes the immediate problem, and at the same time makes
one (small) step in the direction of the long-term solution, by
implementing two changes:

1. Restricts the AsmMatcherEmitter handling of tied operands to
   apply solely to simple operands (not complex operands or
   sub-operand of such).

This means that at least we don't get silently corrupt MC operand
lists as output.  However, if we do have tied sub-operands, they
would now no longer be handled at all, except for:

2. If we have an operand that does not occur in the AsmString,
   and also isn't handled as tied operand, simply emit a dummy
   MC operand (constant 0).

This works as long as target code never attempts to access
MC operands that do no not occur in the AsmString (and are
not tied simple operands), which happens to be the case for
all targets where this situation can occur (ARM and PowerPC).

[ Note that this change means that many of the ARM custom
  converters are now superfluous, since the implement the
  same "hack" now performed already by common code. ]

Longer term, we ought to fix targets to never access *any*
MC operand that does not occur in the AsmString (including
tied simple operands), and then finally completely remove
all such operands from the MC operand list.

Patch approved by Jim Grosbach.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180677 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix a XOR reassociation bug.
Shuxin Yang [Sat, 27 Apr 2013 18:02:12 +0000 (18:02 +0000)]
Fix a XOR reassociation bug.

When Reassociator optimize "(x | C1)" ^ "(X & C2)", it may swap the two
subexpressions, however, it forgot to swap cached constants (of C1 and C2)
accordingly.

rdar://13739160

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180676 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDocumentation: end option description with a period
Dmitri Gribenko [Sat, 27 Apr 2013 16:34:24 +0000 (16:34 +0000)]
Documentation: end option description with a period

Patch by Dimitry Andric.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180675 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAArch64: convert MC-layer test to .s file
Tim Northover [Sat, 27 Apr 2013 11:56:14 +0000 (11:56 +0000)]
AArch64: convert MC-layer test to .s file

The CodeGen aspects of this test are already covered by cfi-frame.ll;
making it an assembly file reduces the risk of incidental changes
affecting the test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180671 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[objc-arc] Test cleanups.
Michael Gottesman [Sat, 27 Apr 2013 05:25:54 +0000 (05:25 +0000)]
[objc-arc] Test cleanups.

Mainly adding paranoid checks for the closing brace of a function to
help with FileCheck error readability. Also some other minor changes.

No actual CHECK changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180668 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoGeneralize the MachineTraceMetrics public API.
Andrew Trick [Sat, 27 Apr 2013 03:54:20 +0000 (03:54 +0000)]
Generalize the MachineTraceMetrics public API.

Naturally, we should be able to pass in extra instructions, not just
extra blocks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180667 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd target flags to MachineMemOperands.
Andrew Trick [Sat, 27 Apr 2013 03:54:17 +0000 (03:54 +0000)]
Add target flags to MachineMemOperands.

This seems to me an obvious place to allow target passes to annotate
memory operations. There are plenty of bits, and I'm not aware of
another good way for early target passes to propagate hints along to
later passes. Target independent transforms can simply preserve them,
the way they preserve the other flags. Like MachineMemOperands in
general, if the target flags are lost we must still generate correct
code.

This has lots of uses, but I want this flexibility now to make it
easier to work with the new MachineTraceMetrics
analysis. MachineTraceMetrics can gather a lot of information about
instructions based on the surrounding code. This information can be
used to influence postRA machine passes that don't work on SSA form.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180666 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agowhitespace
Andrew Trick [Sat, 27 Apr 2013 03:54:14 +0000 (03:54 +0000)]
whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180665 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUse the target triple from the target machine rather than the module
Eric Christopher [Sat, 27 Apr 2013 01:07:52 +0000 (01:07 +0000)]
Use the target triple from the target machine rather than the module
to determine whether or not we're on a darwin platform for debug code
emitting.

Solves the problem of a module with no triple on the command line
and no triple in the module using non-gdb ok features on darwin. Fix
up the member-pointers test to check the correct things for cross
platform (DW_FORM_flag is a good prefix).

Unfortunately no testcase because I have no ideas how to test something
without a triple and without a triple in the module yet check
precisely on two platforms. Ideas welcome.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180660 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMove the XFAIL out of the middle of a comment.
Eric Christopher [Sat, 27 Apr 2013 01:07:22 +0000 (01:07 +0000)]
Move the XFAIL out of the middle of a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180659 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMake all darwin ppc stubs local.
Rafael Espindola [Sat, 27 Apr 2013 00:43:16 +0000 (00:43 +0000)]
Make all darwin ppc stubs local.

This fixes pr15763.
Patch by David Fang.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180657 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoStruct-path aware TBAA: change the format of TBAAStructType node.
Manman Ren [Sat, 27 Apr 2013 00:26:11 +0000 (00:26 +0000)]
Struct-path aware TBAA: change the format of TBAAStructType node.

We switch the order of offset and field type to make TBAAStructType node
(name, parent node, offset) similar to scalar TBAA node (name, parent node).
TypeIsImmutable is added to TBAAStructTag node.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180654 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCleanup and document MachineLocation.
Adrian Prantl [Fri, 26 Apr 2013 21:57:17 +0000 (21:57 +0000)]
Cleanup and document MachineLocation.
Clarify documentation and API to make the difference between register and
register-indirect addressed locations more explicit. Put in a comment
to point out that with the current implementation we cannot specify
a register-indirect location with offset 0 (a breg 0 in DWARF).
No functionality change intended.

rdar://problem/13658587

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180641 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMicro-optimization
Bill Wendling [Fri, 26 Apr 2013 21:15:08 +0000 (21:15 +0000)]
Micro-optimization

TLVs probably won't be as common as the other types of variables. Check for them
last before defaulting to "DATA".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180631 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMake CHECK lines a bit less strict so they also match code generated for win64.
Benjamin Kramer [Fri, 26 Apr 2013 21:04:21 +0000 (21:04 +0000)]
Make CHECK lines a bit less strict so they also match code generated for win64.

Hopefully brings the windows buildbots back to life.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180630 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTeach the interpreter to handle vector compares and additional vector arithmetic...
Nadav Rotem [Fri, 26 Apr 2013 20:19:41 +0000 (20:19 +0000)]
Teach the interpreter to handle vector compares and additional vector arithmetic operations.

Patch by Yuri Veselov.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180626 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUse llvm/Object/MachO.h in macho-dumper. Drop the old macho parser.
Rafael Espindola [Fri, 26 Apr 2013 20:07:33 +0000 (20:07 +0000)]
Use llvm/Object/MachO.h in macho-dumper. Drop the old macho parser.

For Mach-O there were 2 implementations for parsing object files. A
standalone llvm/Object/MachOObject.h and llvm/Object/MachO.h which
implements the generic interface in llvm/Object/ObjectFile.h.

This patch adds the missing features to MachO.h, moves macho-dump to
use MachO.h and removes ObjectFile.h.

In addition to making sure that check-all is clean, I checked that the
new version produces exactly the same output in all Mach-O files in a
llvm+clang build directory (including executables and shared
libraries).

To test the performance, I ran macho-dump over all the files in a
llvm+clang build directory again, but this time redirecting the output
to /dev/null. Both the old and new versions take about 4.6 seconds
(2.5 user) to finish.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180624 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoThe exception demo needs its symbols exported.
Rafael Espindola [Fri, 26 Apr 2013 19:07:40 +0000 (19:07 +0000)]
The exception demo needs its symbols exported.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180622 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: Initialize AMDGPUMachineFunction::ShaderType to ShaderType::COMPUTE
Tom Stellard [Fri, 26 Apr 2013 18:32:24 +0000 (18:32 +0000)]
R600: Initialize AMDGPUMachineFunction::ShaderType to ShaderType::COMPUTE

We need to intialize this to something and since clang does not set
the shader type attribute and clang is used only for compute shaders,
initializing it to COMPUTE seems like the best choice.

Reviewed-by: Christian König <christian.koenig@amd.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180620 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agocleanup testcase some more
Adrian Prantl [Fri, 26 Apr 2013 18:10:54 +0000 (18:10 +0000)]
cleanup testcase some more
rdar://problem/13056109

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180619 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agofix a typo that due to cu&paste quadrupled itself
Adrian Prantl [Fri, 26 Apr 2013 18:10:50 +0000 (18:10 +0000)]
fix a typo that due to cu&paste quadrupled itself
rdar://problem/13056109

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180618 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoARM: Fix encoding of hint instruction for Thumb.
Quentin Colombet [Fri, 26 Apr 2013 17:54:54 +0000 (17:54 +0000)]
ARM: Fix encoding of hint instruction for Thumb.
"hint" space for Thumb actually overlaps the encoding space of the CPS
instruction. In actuality, hints can be defined as CPS instructions where imod
and M bits are all nil.

Handle decoding of permitted nop-compatible hints (i.e. nop, yield, wfi, wfe,
sev) in DecodeT2CPSInstruction.

This commit adds a proper diagnostic message for Imm0_4 and updates all tests.

Patch by Mihail Popa <Mihail.Popa@arm.com>.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180617 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd missing ':'.
Rafael Espindola [Fri, 26 Apr 2013 17:54:46 +0000 (17:54 +0000)]
Add missing ':'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180616 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoBugfix for the debug intrinsic handling in InstCombiner:
Adrian Prantl [Fri, 26 Apr 2013 17:48:33 +0000 (17:48 +0000)]
Bugfix for the debug intrinsic handling in InstCombiner:
Since we can't guarantee that the original dbg.declare instrinsic
is removed by LowerDbgDeclare(), we need to make sure that we are
not inserting the same dbg.value intrinsic over and over.
This removes tons of redundant DIEs when compiling optimized code.

rdar://problem/13056109

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180615 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoPowerPC: Use RegisterOperand instead of RegisterClass operands
Ulrich Weigand [Fri, 26 Apr 2013 16:53:15 +0000 (16:53 +0000)]
PowerPC: Use RegisterOperand instead of RegisterClass operands

In the default PowerPC assembler syntax, registers are specified simply
by number, so they cannot be distinguished from immediate values (without
looking at the opcode).  This means that the default operand matching logic
for the asm parser does not work, and we need to specify custom matchers.
Since those can only be specified with RegisterOperand classes and not
directly on the RegisterClass, all instructions patterns used by the asm
parser need to use a RegisterOperand (instead of a RegisterClass) for
all their register operands.

This patch adds one RegisterOperand for each RegisterClass, using the
same name as the class, just in lower case, and updates all instruction
patterns to use RegisterOperand instead of RegisterClass operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180611 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRe-write the address propagation code for pre-indexed loads/stores to take into accou...
Silviu Baranga [Fri, 26 Apr 2013 15:52:24 +0000 (15:52 +0000)]
Re-write the address propagation code for pre-indexed loads/stores to take into account some previously misssed cases (PRE_DEC addressing mode, the offset and base address are swapped, etc). This should fix PR15581.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180609 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoPowerPC: Fix encoding of vsubcuw and vsum4sbs instructions
Ulrich Weigand [Fri, 26 Apr 2013 15:39:57 +0000 (15:39 +0000)]
PowerPC: Fix encoding of vsubcuw and vsum4sbs instructions

When testing the asm parser, I noticed wrong encodings for the
above instructions (wrong sub-opcodes).

Tests will be added together with the asm parser.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180608 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoPowerPC: Fix encoding of stfsu and stfdu instructions
Ulrich Weigand [Fri, 26 Apr 2013 15:39:40 +0000 (15:39 +0000)]
PowerPC: Fix encoding of stfsu and stfdu instructions

When testing the asm parser, I noticed wrong encodings for the
above instructions (wrong sub-opcodes).  Note that apparently
the compiler currently never generates pre-inc instructions
for floating point types for some reason ...

Tests will be added together with the asm parser.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180607 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoPowerPC: Fix encoding of rldimi and rldcl instructions
Ulrich Weigand [Fri, 26 Apr 2013 15:39:12 +0000 (15:39 +0000)]
PowerPC: Fix encoding of rldimi and rldcl instructions

When testing the asm parser, I noticed wrong encodings for the
above instructions (wrong operand name in rldimi, wrong form
and sub-opcode for rldcl).

Tests will be added together with the asm parser.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180606 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoPowerPC: Support PC-relative fixup_ppc_brcond14.
Ulrich Weigand [Fri, 26 Apr 2013 15:38:30 +0000 (15:38 +0000)]
PowerPC: Support PC-relative fixup_ppc_brcond14.

When testing the asm parser, I ran into an error when using a conditional
branch to an external symbol (this doesn't occur in compiler-generated
code) due to missing support in PPCELFObjectWriter::getRelocTypeInner.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180605 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoARM/NEON: Pattern match vector integer abs to vabs.
Benjamin Kramer [Fri, 26 Apr 2013 15:00:57 +0000 (15:00 +0000)]
ARM/NEON: Pattern match vector integer abs to vabs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180604 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoX86: Now that we have a canonical form for vector integer abs, match it into pabs.
Benjamin Kramer [Fri, 26 Apr 2013 12:05:21 +0000 (12:05 +0000)]
X86: Now that we have a canonical form for vector integer abs, match it into pabs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180600 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDAGCombiner: Canonicalize vector integer abs in the same way we do it for scalars.
Benjamin Kramer [Fri, 26 Apr 2013 09:19:19 +0000 (09:19 +0000)]
DAGCombiner: Canonicalize vector integer abs in the same way we do it for scalars.

This already helps SSE2 x86 a lot because it lacks an efficient way to
represent a vector select. The long term goal is to enable the backend to match
a canonicalized pattern into a single instruction (e.g. vabs or pabs).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180597 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLoopVectorizer: Calculate the number of pointers to disambiguate at runtime based...
Nadav Rotem [Fri, 26 Apr 2013 05:08:59 +0000 (05:08 +0000)]
LoopVectorizer:  Calculate the number of pointers to disambiguate at runtime based on the numbers of reads and writes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180593 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUse 'git svn find-rev' in git-svnrevert instead of shell script fu.
Michael Gottesman [Fri, 26 Apr 2013 03:27:39 +0000 (03:27 +0000)]
Use 'git svn find-rev' in git-svnrevert instead of shell script fu.

Thanks Chandler!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180592 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRevert "[objc-arc] Added ImpreciseAutoreleaseSet to track autorelease calls that...
Michael Gottesman [Fri, 26 Apr 2013 01:12:18 +0000 (01:12 +0000)]
Revert "[objc-arc] Added ImpreciseAutoreleaseSet to track autorelease calls that were once autoreleaseRV instructions."

This reverts commit r180222.

I think this might tie in with a different problem which will require a
different approach potentially. I am reverting this in the case I need to go
down that second path.

My apologies for the noise. = /.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180590 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUpdated GettingStarted.rst so that it references utils/git-svn for git-svnup instead...
Michael Gottesman [Fri, 26 Apr 2013 01:04:45 +0000 (01:04 +0000)]
Updated GettingStarted.rst so that it references utils/git-svn for git-svnup instead of catting it into the documentation itself.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180589 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdded the scripts git-svnup/git-svnrevert to utils/git-svn.
Michael Gottesman [Fri, 26 Apr 2013 00:58:45 +0000 (00:58 +0000)]
Added the scripts git-svnup/git-svnrevert to utils/git-svn.

It makes more sense to have git-svnup here than catting said file in the
documentation (where we should rather point users to this directory).
I included git-svnrevert as an additional gift to the community. I will update
the documentation in a second commit later today.

git-svnrevert takes in a git hash for a commit, looks up the svn revision for
said commit and then creates the normal git revert commit message with the one
liner message, except instead of saying

  Revert "<<<INSERT ONELINER HERE>>>"

  This reverts commit <<<INSERT GITHASH HERE>>>

It says:

  Revert "<<<INSERT ONELINER HERE>>>"

  This reverts commit r<<<INSERT SVN REVISION HERE>>>

so git hashes will not escape into our svn logs (which just look unseemly).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180587 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMips assembler: .set reorder support
Jack Carter [Thu, 25 Apr 2013 23:31:35 +0000 (23:31 +0000)]
Mips assembler: .set reorder support

Mips have delayslots for certain instructions
like jumps and branches. These are instructions
that follow the branch or jump and are executed
before the jump or branch is completed.

Early Mips compilers could not cope with delayslots
and left them up to the assembler. The assembler would
fill the delayslots with the appropriate instruction,
usually just a nop to allow correct runtime behavior.

The default behavior for this is set with .set reorder.
To tell the assembler that you don't want it to mess with
the delayslot one used .set noreorder.

For backwards compatibility we need to support
.set reorder and have it be the default behavior in the
assembler.

Our support for it is to insert a NOP directly after an
instruction with a delayslot when in .set reorder mode.

Contributer: Vladimir Medic

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180584 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove SMLoc paired with CHECK-NOT patterns. Not functionality change.
Michael Liao [Thu, 25 Apr 2013 21:31:34 +0000 (21:31 +0000)]
Remove SMLoc paired with CHECK-NOT patterns. Not functionality change.

Pattern has source location by itself. After adding a trivial method to
retrieve it, it's unnecessary to pair a source location for CHECK-NOT patterns.
One thing revised after this is the diagnostic info is more accurate by
pointing to the start of the CHECK-NOT pattern instead of the end of the
CHECK-NOT pattern. E.g. diagnostic message previously looks like

    <stdin>:1:1: error: CHECK-NOT: string occurred!
    test
    ^
    test.txt:1:16: note: CHECK-NOT: pattern specified here
    CHECK-NOT: test
                   ^

is changed to

    <stdin>:1:1: error: CHECK-NOT: string occurred!
    test
    ^
    test.txt:1:12: note: CHECK-NOT: pattern specified here
    CHECK-NOT: test
               ^

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180578 91177308-0d34-0410-b5e6-96231b3b80d8