oota-llvm.git
10 years agoFix assembly operands for the SSE2 cvtsd2ss instruction.
Cameron McInally [Tue, 19 Nov 2013 14:36:00 +0000 (14:36 +0000)]
Fix assembly operands for the SSE2 cvtsd2ss instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195129 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Mips] Adjust float ABI settings in case of MIPS16 mode.
Simon Atanasyan [Tue, 19 Nov 2013 12:20:17 +0000 (12:20 +0000)]
[Mips] Adjust float ABI settings in case of MIPS16 mode.

Hard float for mips16 means essentially to compile as soft float but to
use a runtime library for soft float that is written with native mips32
floating point instructions (those runtime routines run in mips32 hard
float mode).

The patch reviewed by Reed Kotler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195123 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFormatting and 80-col.
Eric Christopher [Tue, 19 Nov 2013 09:28:34 +0000 (09:28 +0000)]
Formatting and 80-col.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195122 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix comment.
Eric Christopher [Tue, 19 Nov 2013 09:11:26 +0000 (09:11 +0000)]
Fix comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195121 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRefactor the section emission code to remove duplicates now that
Eric Christopher [Tue, 19 Nov 2013 09:04:50 +0000 (09:04 +0000)]
Refactor the section emission code to remove duplicates now that
we can emit various sections in any order.
No functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195120 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoReformat file.
Eric Christopher [Tue, 19 Nov 2013 09:04:36 +0000 (09:04 +0000)]
Reformat file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195119 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix an issue where SROA computed different results based on the relative
Chandler Carruth [Tue, 19 Nov 2013 09:03:18 +0000 (09:03 +0000)]
Fix an issue where SROA computed different results based on the relative
order of slices of the alloca which have exactly the same size and other
properties. This was found by a perniciously unstable sort
implementation used to flush out buggy uses of the algorithm.

The fundamental idea is that findCommonType should return the best
common type it can find across all of the slices in the range. There
were two bugs here previously:

1) We would accept an integer type smaller than a byte-width multiple,
   and if there were different bit-width integer types, we would accept
   the first one. This caused an actual failure in the testcase updated
   here when the sort order changed.
2) If we found a bad combination of types or a non-load, non-store use
   before an integer typed load or store we would bail, but if we found
   the integere typed load or store, we would use it. The correct
   behavior is to always use an integer typed operation which covers the
   partition if one exists.

While a clever debugging sort algorithm found problem #1 in our existing
test cases, I have no useful test case ideas for #2. I spotted in by
inspection when looking at this code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195118 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd support for software expansion of 64-bit integer division instructions.
Michael Ilseman [Tue, 19 Nov 2013 06:54:19 +0000 (06:54 +0000)]
Add support for software expansion of 64-bit integer division instructions.

Patch by Dmitri Shtilman!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195116 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix patchpoint comments.
Andrew Trick [Tue, 19 Nov 2013 05:05:43 +0000 (05:05 +0000)]
Fix patchpoint comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195103 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd lld to projects to tag.
Bill Wendling [Tue, 19 Nov 2013 04:58:46 +0000 (04:58 +0000)]
Add lld to projects to tag.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195102 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd .clang-format without column limit to subdirectory tests/.
Daniel Jasper [Tue, 19 Nov 2013 04:26:05 +0000 (04:26 +0000)]
Add .clang-format without column limit to subdirectory tests/.

A column limit in the test folder can lead to trouble as the RUN, CHECK,
etc. comments can potentially be broken over multiple lines changing
their meaning. Without column limit, clang-format will simply keep the
test author's line breaks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195100 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse symbolic operands in the patchpoint folding routine and fix a spilling bug.
Andrew Trick [Tue, 19 Nov 2013 03:29:59 +0000 (03:29 +0000)]
Use symbolic operands in the patchpoint folding routine and fix a spilling bug.

Fixes <rdar://15487687> [JS] AnyRegCC argument ends up being spilled

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195094 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd an abstraction to handle patchpoint operands.
Andrew Trick [Tue, 19 Nov 2013 03:29:56 +0000 (03:29 +0000)]
Add an abstraction to handle patchpoint operands.

Hard-coded operand indices were scattered throughout lowering stages
and layers. It was super bug prone.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195093 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[weak vtables] Place class definitions into anonymous namespaces to prevent weak...
Juergen Ributzka [Tue, 19 Nov 2013 03:08:35 +0000 (03:08 +0000)]
[weak vtables] Place class definitions into anonymous namespaces to prevent weak vtables.

This patch places class definitions in implementation files into anonymous
namespaces to prevent weak vtables. This eliminates the need of providing an
out-of-line definition to pin the vtable explicitly to the file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195092 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoImplement AArch64 neon instructions class SIMD lsone and SIMD lone-post.
Hao Liu [Tue, 19 Nov 2013 02:17:05 +0000 (02:17 +0000)]
Implement AArch64 neon instructions class SIMD lsone and SIMD lone-post.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195078 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove unused special member functions and reformat.
Eric Christopher [Tue, 19 Nov 2013 02:01:07 +0000 (02:01 +0000)]
Remove unused special member functions and reformat.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195077 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix previous commit and fully remove variable.
Eric Christopher [Tue, 19 Nov 2013 01:52:38 +0000 (01:52 +0000)]
Fix previous commit and fully remove variable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195076 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove unused variable.
Eric Christopher [Tue, 19 Nov 2013 01:50:29 +0000 (01:50 +0000)]
Remove unused variable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195075 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoImplement AArch64 SISD intrinsics for vget_high and vget_low.
Jiangning Liu [Tue, 19 Nov 2013 01:46:48 +0000 (01:46 +0000)]
Implement AArch64 SISD intrinsics for vget_high and vget_low.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195074 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoimplement MC layer of AArch64 neon instruction PMULL and PMULL2 with 128 bit integer.
Kevin Qin [Tue, 19 Nov 2013 01:40:25 +0000 (01:40 +0000)]
implement MC layer of AArch64 neon instruction PMULL and PMULL2 with 128 bit integer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195072 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd predicate for AArch64 crypto instructions.
Jiangning Liu [Tue, 19 Nov 2013 01:38:31 +0000 (01:38 +0000)]
Add predicate for AArch64 crypto instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195071 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Mips] Support for MicroMips STO refactoring.
Jack Carter [Tue, 19 Nov 2013 01:25:18 +0000 (01:25 +0000)]
[Mips] Support for MicroMips STO refactoring.

No true functional changes.

Change the "hack" name of emitMipsHackSTOCG to emitSymSTO.

Remove demonstration code in AsmParser for emitMipsHackSTOCG and
emitMipsHackELFFlags. The STO field is in an ELF symbol and is not
an explicit directive. That said, we are missing the compliment call
in AsmParser and that will need to be addressed soon.

XFAIL dummy tests for emitMipsHackELFFlags and emitMipsHackELFFlags.
These will built out with following patches.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195067 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[weak vtables] Remove a bunch of weak vtables
Juergen Ributzka [Tue, 19 Nov 2013 00:57:56 +0000 (00:57 +0000)]
[weak vtables] Remove a bunch of weak vtables

This patch removes most of the trivial cases of weak vtables by pinning them to
a single object file. The memory leaks in this version have been fixed. Thanks
Alexey for pointing them out.

Differential Revision: http://llvm-reviews.chandlerc.com/D2068

Reviewed by Andy

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195064 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agollvm-dwarfdump: support for emitting only the debug_types section using -debug-dump
David Blaikie [Tue, 19 Nov 2013 00:29:42 +0000 (00:29 +0000)]
llvm-dwarfdump: support for emitting only the debug_types section using -debug-dump

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195063 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDwarfDebug: Move trailing else to the same line as prior closing brace
David Blaikie [Mon, 18 Nov 2013 23:59:04 +0000 (23:59 +0000)]
DwarfDebug: Move trailing else to the same line as prior closing brace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195060 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDwarfDebug: Remove some more redundant explicit constructions.
David Blaikie [Mon, 18 Nov 2013 23:57:26 +0000 (23:57 +0000)]
DwarfDebug: Remove some more redundant explicit constructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195059 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[Mips] MipsTargetStreamer refactoring.
Jack Carter [Mon, 18 Nov 2013 23:55:27 +0000 (23:55 +0000)]
[Mips] MipsTargetStreamer refactoring.

No functionality changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195057 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugInfo: Simplify a few more explicit constructions, underconstrained types, and...
David Blaikie [Mon, 18 Nov 2013 23:33:32 +0000 (23:33 +0000)]
DebugInfo: Simplify a few more explicit constructions, underconstrained types, and make DIType(MDNode*) explicit like all the other DI* node ctors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195055 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert "COFF: Emit all MCSymbols rather than filtering out some of them"
Reid Kleckner [Mon, 18 Nov 2013 23:08:12 +0000 (23:08 +0000)]
Revert "COFF: Emit all MCSymbols rather than filtering out some of them"

This reverts commit r190888, to fix PR17967.  The original change wasn't
the right way to get @feat.00 into the object file.  The right fix is to
make @feat.00 be a global symbol.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195053 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebug info: Let LowerDbgDeclare perfom the dbg.declare -> dbg.value
Adrian Prantl [Mon, 18 Nov 2013 23:04:38 +0000 (23:04 +0000)]
Debug info: Let LowerDbgDeclare perfom the dbg.declare -> dbg.value
lowering only for load/stores to scalar allocas. The resulting values
confuse the backend and don't add anything because we can describe
array-allocas with a dbg.declare intrinsic just fine.

rdar://problem/15464571

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195052 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDocs: tweak code-block spacing.
Ahmed Bougacha [Mon, 18 Nov 2013 22:26:59 +0000 (22:26 +0000)]
Docs: tweak code-block spacing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195049 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoThe 'optnone' attribute means don't inline anything into this function
Paul Robinson [Mon, 18 Nov 2013 21:44:03 +0000 (21:44 +0000)]
The 'optnone' attribute means don't inline anything into this function
(except functions marked always_inline).
Functions with 'optnone' must also have 'noinline' so they don't get
inlined into any other function.

Based on work by Andrea Di Biagio.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195046 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Fix moveToVALU when the first operand is VSrc.
Matt Arsenault [Mon, 18 Nov 2013 20:09:55 +0000 (20:09 +0000)]
R600/SI: Fix moveToVALU when the first operand is VSrc.

Moving into a VSrc doesn't always work, since it could be
replaced with an SGPR later.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195042 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Fix multiple SGPR reads when using VCC.
Matt Arsenault [Mon, 18 Nov 2013 20:09:50 +0000 (20:09 +0000)]
R600/SI: Fix multiple SGPR reads when using VCC.

No other SGPR operands are allowed, so if VCC is
used, move the other to a VGPR.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195041 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Implement add i64, but do not yet enable.
Matt Arsenault [Mon, 18 Nov 2013 20:09:47 +0000 (20:09 +0000)]
R600/SI: Implement add i64, but do not yet enable.

Test doesn't actually check the output. I need
to fix add i64 being matched for the addressing
calculations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195040 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Specify SSrc operands
Matt Arsenault [Mon, 18 Nov 2013 20:09:43 +0000 (20:09 +0000)]
R600/SI: Specify SSrc operands

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195039 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: addc / adde i32 are legal
Matt Arsenault [Mon, 18 Nov 2013 20:09:40 +0000 (20:09 +0000)]
R600/SI: addc / adde i32 are legal

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195038 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Match addc to S_ADD_U32.
Matt Arsenault [Mon, 18 Nov 2013 20:09:37 +0000 (20:09 +0000)]
R600/SI: Match addc to S_ADD_U32.

The carry always goes to SCC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195037 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Match adde/sube to S_ADDC_U32/S_SUBB_U32
Matt Arsenault [Mon, 18 Nov 2013 20:09:34 +0000 (20:09 +0000)]
R600/SI: Match adde/sube to S_ADDC_U32/S_SUBB_U32

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195036 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Specify S_ADD/S_SUB set SCC and add is commutable
Matt Arsenault [Mon, 18 Nov 2013 20:09:32 +0000 (20:09 +0000)]
R600/SI: Specify S_ADD/S_SUB set SCC and add is commutable

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195035 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Move patterns to match add / sub to scalar instructions
Matt Arsenault [Mon, 18 Nov 2013 20:09:29 +0000 (20:09 +0000)]
R600/SI: Move patterns to match add / sub to scalar instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195034 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Fix extra defs of VCC / SCC.
Matt Arsenault [Mon, 18 Nov 2013 20:09:21 +0000 (20:09 +0000)]
R600/SI: Fix extra defs of VCC / SCC.

When replacing scalar operations with vector,
the wrong implicit output register was used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195033 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600: Enable the IR structurizer by default
Tom Stellard [Mon, 18 Nov 2013 19:43:44 +0000 (19:43 +0000)]
R600: Enable the IR structurizer by default

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195031 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600: Fix a crash in the AMDILCFGStrucurizer
Tom Stellard [Mon, 18 Nov 2013 19:43:38 +0000 (19:43 +0000)]
R600: Fix a crash in the AMDILCFGStrucurizer

The ifPatternMatch() function was not correctly reporting the number
of matches in some cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195030 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600: Add a SubtargetFeatture for disabling the ifcvt pass.
Tom Stellard [Mon, 18 Nov 2013 19:43:33 +0000 (19:43 +0000)]
R600: Add a SubtargetFeatture for disabling the ifcvt pass.

This is useful when writing test cases for the AMDIL structurizer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195029 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600: Use lower-case for EnableIRStructurizer feature
Tom Stellard [Mon, 18 Nov 2013 19:43:29 +0000 (19:43 +0000)]
R600: Use lower-case for EnableIRStructurizer feature

llc converts all values passed to -mattr= to lowercase, so this
enables us to toggle this feature when using llc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195028 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Fix illegal VGPR->SGPR copy inside of loop
Tom Stellard [Mon, 18 Nov 2013 18:50:20 +0000 (18:50 +0000)]
R600/SI: Fix illegal VGPR->SGPR copy inside of loop

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195026 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600/SI: Fix another case of illegal VGPR->SGPR copy
Tom Stellard [Mon, 18 Nov 2013 18:50:15 +0000 (18:50 +0000)]
R600/SI: Fix another case of illegal VGPR->SGPR copy

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195025 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoChecking for a return value with FormatMessage; if the call fails, there's no guarant...
Aaron Ballman [Mon, 18 Nov 2013 17:43:22 +0000 (17:43 +0000)]
Checking for a return value with FormatMessage; if the call fails, there's no guarantee that the buffer will be non-null.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195019 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFixing a possible memory leak from a failing realloc() call.
Aaron Ballman [Mon, 18 Nov 2013 17:33:32 +0000 (17:33 +0000)]
Fixing a possible memory leak from a failing realloc() call.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195018 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRecover gracefully when deserializing invalid YAML input.
Alexander Kornienko [Mon, 18 Nov 2013 15:50:04 +0000 (15:50 +0000)]
Recover gracefully when deserializing invalid YAML input.
Fixes http://llvm.org/PR16221, http://llvm.org/PR15927
Phabricator: http://llvm-reviews.chandlerc.com/D1236

Patch by Andrew Tulloch!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195016 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[ASan] Fix PR17867 - make sure ASan doesn't crash if use-after-scope and use-after...
Alexey Samsonov [Mon, 18 Nov 2013 14:53:55 +0000 (14:53 +0000)]
[ASan] Fix PR17867 - make sure ASan doesn't crash if use-after-scope and use-after-return are combined.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195014 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[PR17978] Mark two ARM/fast-isel tests as XFAIL:vg_leak due to GV.
NAKAMURA Takumi [Mon, 18 Nov 2013 13:50:19 +0000 (13:50 +0000)]
[PR17978] Mark two ARM/fast-isel tests as XFAIL:vg_leak due to GV.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195010 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoLoopVectorizer: Extend the induction variable to a larger type
Arnold Schwaighofer [Mon, 18 Nov 2013 13:14:32 +0000 (13:14 +0000)]
LoopVectorizer: Extend the induction variable to a larger type

In some case the loop exit count computation can overflow. Extend the type to
prevent most of those cases.

The problem is loops like:
int main ()
{
  int a = 1;
  char b = 0;
  lbl:
    a &= 4;
    b--;
    if (b) goto lbl;
  return a;
}

The backedge count is 255. The induction variable type is i8. If we add one to
255 to get the exit count we overflow to zero.

To work around this issue we extend the type of the induction variable to i32 in
the case of i8 and i16.

PR17532

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195008 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[mips] Fix 'ran out of registers' in MIPS32 with FP64 when generating code for (Const...
Daniel Sanders [Mon, 18 Nov 2013 13:12:43 +0000 (13:12 +0000)]
[mips] Fix 'ran out of registers' in MIPS32 with FP64 when generating code for (ConstantFP 0.0)

Fixed an inappropriate use of BuildPairF64 when compiling for MIPS32 with FP64
which resulted in an impossible constraint on the register allocation. It now
uses BuildPairF64_64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195007 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[mips][msa] Update encoding of bnz.v (typo).
Matheus Almeida [Mon, 18 Nov 2013 13:09:54 +0000 (13:09 +0000)]
[mips][msa] Update encoding of bnz.v (typo).

Note that there's no hardware yet that relies on that encoding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195006 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[mips][msa] Fix immediate value of LSA instruction as it was being wrongly encoded.
Matheus Almeida [Mon, 18 Nov 2013 12:32:49 +0000 (12:32 +0000)]
[mips][msa] Fix immediate value of LSA instruction as it was being wrongly encoded.

The immediate field should be encoded as "imm - 1" as the CPU always adds one to that field.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195004 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix forgotten member initialization detected by MSan bootstrap bot
Alexey Samsonov [Mon, 18 Nov 2013 11:06:01 +0000 (11:06 +0000)]
Fix forgotten member initialization detected by MSan bootstrap bot

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195003 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[mips][msa] Add MSA to the release notes.
Daniel Sanders [Mon, 18 Nov 2013 10:38:47 +0000 (10:38 +0000)]
[mips][msa] Add MSA to the release notes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195001 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agotest-release.sh: tweak RPATH for the binary packages.
Arnaud A. de Grandmaison [Mon, 18 Nov 2013 10:34:59 +0000 (10:34 +0000)]
test-release.sh: tweak RPATH for the binary packages.

libtool sets RPATH to "$ORIGIN/../lib:/the/directory/where/it/was/built/lib" so that a developper can use the built or the installed version seamlessly. Our binary packages should not have this developer friendly tweak, as the users of the binaries will not have the build tree.

Beside, in case the development tree is a possibly on an automounted share, this can create very bad user experience : they will incur an automount timeout penalty and will get a very bad feeling of llvm/clang's speed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194999 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUnbreak the build after r194997
Alexey Samsonov [Mon, 18 Nov 2013 09:44:36 +0000 (09:44 +0000)]
Unbreak the build after r194997

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194998 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert r194865 and r194874.
Alexey Samsonov [Mon, 18 Nov 2013 09:31:53 +0000 (09:31 +0000)]
Revert r194865 and r194874.

This change is incorrect. If you delete virtual destructor of both a base class
and a subclass, then the following code:
  Base *foo = new Child();
  delete foo;
will not cause the destructor for members of Child class. As a result, I observe
plently of memory leaks. Notable examples I investigated are:
ObjectBuffer and ObjectBufferStream, AttributeImpl and StringSAttributeImpl.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194997 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[AArch64 NEON]Add mov alias for simd copy instructions.
Kevin Qin [Mon, 18 Nov 2013 09:20:32 +0000 (09:20 +0000)]
[AArch64 NEON]Add mov alias for simd copy instructions.
Set some unspecified bits of INS/DUP to zero as ARMARM requested.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194996 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoImplement the newly added ACLE functions for ld1/st1 with 2/3/4 vectors.
Hao Liu [Mon, 18 Nov 2013 06:31:53 +0000 (06:31 +0000)]
Implement the newly added ACLE functions for ld1/st1 with 2/3/4 vectors.
The functions are like: vst1_s8_x2 ...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194990 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebug Info Verifier: disable it by default.
Manman Ren [Mon, 18 Nov 2013 03:19:31 +0000 (03:19 +0000)]
Debug Info Verifier: disable it by default.

Debug info verifier is part of the verifier which is a Function Pass.
Tot currently tries to pull all reachable debug info MDNodes in each function,
which is too time-consuming. The correct fix seems to be separating debug info
verification to its own module pass.

I will disable the debug info verifier until a correct fix is found.

For Bill's testing case, enabling debug info verifier increase compile
time from 11s to 11m.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194986 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix spacing, forward declare order.
Matt Arsenault [Mon, 18 Nov 2013 02:51:33 +0000 (02:51 +0000)]
Fix spacing, forward declare order.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194985 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove unnecessary temporary construction.
David Blaikie [Sun, 17 Nov 2013 21:59:31 +0000 (21:59 +0000)]
Remove unnecessary temporary construction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194981 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRemove redundant explicit default initialization.
David Blaikie [Sun, 17 Nov 2013 21:57:33 +0000 (21:57 +0000)]
Remove redundant explicit default initialization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194980 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDwarfCompileUnit: Add type safety to createGlobalVariableDIE
David Blaikie [Sun, 17 Nov 2013 21:55:13 +0000 (21:55 +0000)]
DwarfCompileUnit: Add type safety to createGlobalVariableDIE

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194979 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTableGen: Generate an enum for all named Operand types in tblgen'd InstrInfo.
Ahmed Bougacha [Sun, 17 Nov 2013 21:24:41 +0000 (21:24 +0000)]
TableGen: Generate an enum for all named Operand types in tblgen'd InstrInfo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194978 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDocs: Clearly separate Operand-related paragraphs.
Ahmed Bougacha [Sun, 17 Nov 2013 21:24:37 +0000 (21:24 +0000)]
Docs: Clearly separate Operand-related paragraphs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194977 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDocs: Fix typo: NoIntinerary -> NoItinerary.
Ahmed Bougacha [Sun, 17 Nov 2013 21:24:34 +0000 (21:24 +0000)]
Docs: Fix typo: NoIntinerary -> NoItinerary.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194976 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebug Info: fix typo in function name.
Manman Ren [Sun, 17 Nov 2013 19:35:03 +0000 (19:35 +0000)]
Debug Info: fix typo in function name.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194975 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebug Info Verifier: fix when to find debug info nodes and when to verify them.
Manman Ren [Sun, 17 Nov 2013 18:48:57 +0000 (18:48 +0000)]
Debug Info Verifier: fix when to find debug info nodes and when to verify them.

We used to collect debug info MDNodes in doInitialization and verify them in
doFinalization. That is incorrect since MDNodes can be modified by passes run
between doInitialization and doFinalization.

To fix the problem, we handle debug info MDNodes that can be reached from a
function in runOnFunction (i.e we collect those nodes by calling processDeclare,
processValue and processLocation, and then verify them in runOnFunction).

We handle debug info MDNodes that can be reached from named metadata in
doFinalization. This is in line with how Verifier handles module-level data
(they are verified in doFinalization).

rdar://15472296

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194974 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebug Info Verifier: enable public functions of Finder to update the type map.
Manman Ren [Sun, 17 Nov 2013 18:42:37 +0000 (18:42 +0000)]
Debug Info Verifier: enable public functions of Finder to update the type map.

We used to depend on running processModule before the other public functions
such as processDeclare, processValue and processLocation. We are now relaxing
the constraint by adding a module argument to the three functions and
letting the three functions to initialize the type map. This will be used in
a follow-on patch that collects nodes reachable from a Function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194973 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUtils/LoopUnroll.cpp: Tweak (StringRef)OldName to be valid until it is used, since...
NAKAMURA Takumi [Sun, 17 Nov 2013 18:05:34 +0000 (18:05 +0000)]
Utils/LoopUnroll.cpp: Tweak (StringRef)OldName to be valid until it is used, since r194601.

eraseFromParent() invalidates OldName.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194970 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd a loop rerolling flag to the PassManagerBuilder
Hal Finkel [Sun, 17 Nov 2013 16:02:50 +0000 (16:02 +0000)]
Add a loop rerolling flag to the PassManagerBuilder

This adds a boolean member variable to the PassManagerBuilder to control loop
rerolling (just like we have for unrolling and the various vectorization
options). This is necessary for control by the frontend. Loop rerolling remains
disabled by default at all optimization levels.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194966 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agopython: Fix check for disasm creation failure
Anders Waldenborg [Sun, 17 Nov 2013 15:40:57 +0000 (15:40 +0000)]
python: Fix check for disasm creation failure

Check should be for pointer being NULL, not what it points to.

Also adds a test for this case.

Reviewed By: indygreg

Differential Revision: http://llvm-reviews.chandlerc.com/D1878

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194965 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agopython: Properly initialize before trying to create disasm
Anders Waldenborg [Sun, 17 Nov 2013 15:17:08 +0000 (15:17 +0000)]
python: Properly initialize before trying to create disasm

As the "LLVMInitializeAll*" functions are not available as symbols in
the shared library they can't be used, and as a workaround a list of
the targets is kept and the individual symbols tried. As soon as the
"All"-functions are changed to proper symbols (as opposed to static
inlines in the headers) this hack will be replace with simple calls
to the corresponding "LLVMInitializeAll*" functions.

Reviewed By: indygreg

CC: llvm-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D1879

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194964 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUpdate CREDITS
Alp Toker [Sun, 17 Nov 2013 11:44:36 +0000 (11:44 +0000)]
Update CREDITS

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194962 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoTestcase for PR17964
Bill Wendling [Sun, 17 Nov 2013 10:53:19 +0000 (10:53 +0000)]
Testcase for PR17964

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194961 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoRevert "Micro-optimization"
Bill Wendling [Sun, 17 Nov 2013 10:53:13 +0000 (10:53 +0000)]
Revert "Micro-optimization"

This reverts commit f1d9fe9d04ce93f6d5dcebbd2cb6a07414d7a029.

This was causing PR17964. We need to use thread data before regular data.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194960 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDAGCombiner: Partially revert r192795, getNOT was fixed not to create illegal constants.
Benjamin Kramer [Sun, 17 Nov 2013 10:40:03 +0000 (10:40 +0000)]
DAGCombiner: Partially revert r192795, getNOT was fixed not to create illegal constants.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194959 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoDebugLoc defines LineCol as 32 bit in comment but unsigned in code.
Yaron Keren [Sun, 17 Nov 2013 09:47:39 +0000 (09:47 +0000)]
DebugLoc defines LineCol as 32 bit in comment but unsigned in code.
This patch modifies LineCol to be a uint32_t.

See http://llvm.org/bugs/show_bug.cgi?id=17957

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194957 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[block-freq] Add BlockFrequency::scale that returns a remainder from the division...
Michael Gottesman [Sun, 17 Nov 2013 03:25:24 +0000 (03:25 +0000)]
[block-freq] Add BlockFrequency::scale that returns a remainder from the division and make the private scale in BlockFrequency more performant.

This change is the first in a series of changes improving LLVM's Block
Frequency propogation implementation to not lose probability mass in
branchy code when propogating block frequency information from a basic
block to its successors. This patch is a simple infrastructure
improvement that does not actually modify the block frequency
algorithm. The specific changes are:

1. Changes the division algorithm used when scaling block frequencies by
branch probabilities to a short division algorithm. This gives us the
remainder for free as well as provides a nice speed boost. When I
benched the old routine and the new routine on a Sandy Bridge iMac with
disabled turbo mode performing 8192 iterations on an array of length
32768, I saw ~600% increase in speed in mean/median performance.

2. Exposes a scale method that returns a remainder. This is important so
we can ensure that when we scale a block frequency by some branch
probability BP = N/D, the remainder from the division by D can be
retrieved and propagated to other children to ensure no probability mass
is lost (more to come on this).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194950 91177308-0d34-0410-b5e6-96231b3b80d8

10 years ago[PM] Completely remove support for explicit 'require' methods on the
Chandler Carruth [Sun, 17 Nov 2013 03:18:05 +0000 (03:18 +0000)]
[PM] Completely remove support for explicit 'require' methods on the
AnalysisManager. All this method did was assert something and we have
a perfectly good way to trigger that assert from the query path.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194947 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse more getZExtOrTruncs
Matt Arsenault [Sun, 17 Nov 2013 02:31:26 +0000 (02:31 +0000)]
Use more getZExtOrTruncs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194945 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse getZExtOrTrunc instead of repeating the same logic.
Matt Arsenault [Sun, 17 Nov 2013 02:24:21 +0000 (02:24 +0000)]
Use getZExtOrTrunc instead of repeating the same logic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194944 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd the cold attribute to error-reporting call sites
Hal Finkel [Sun, 17 Nov 2013 02:06:35 +0000 (02:06 +0000)]
Add the cold attribute to error-reporting call sites

Generally speaking, control flow paths with error reporting calls are cold.
So far, error reporting calls are calls to perror and calls to fprintf,
fwrite, etc. with stderr as the stream. This can be extended in the future.

The primary motivation is to improve block placement (the cold attribute
affects the static branch prediction heuristics).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194943 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdded a size field to the stack map record to handle subregister spills.
Andrew Trick [Sun, 17 Nov 2013 01:36:23 +0000 (01:36 +0000)]
Added a size field to the stack map record to handle subregister spills.

Implementing this on bigendian platforms could get strange. I added a
target hook, getStackSlotRange, per Jakob's recommendation to make
this as explicit as possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194942 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix ndebug-build unused variable in loop rerolling
Hal Finkel [Sun, 17 Nov 2013 01:21:54 +0000 (01:21 +0000)]
Fix ndebug-build unused variable in loop rerolling

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194941 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoUse right address space pointer size
Matt Arsenault [Sun, 17 Nov 2013 00:06:39 +0000 (00:06 +0000)]
Use right address space pointer size

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194940 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAdd a loop rerolling pass
Hal Finkel [Sat, 16 Nov 2013 23:59:05 +0000 (23:59 +0000)]
Add a loop rerolling pass

This adds a loop rerolling pass: the opposite of (partial) loop unrolling. The
transformation aims to take loops like this:

for (int i = 0; i < 3200; i += 5) {
  a[i]     += alpha * b[i];
  a[i + 1] += alpha * b[i + 1];
  a[i + 2] += alpha * b[i + 2];
  a[i + 3] += alpha * b[i + 3];
  a[i + 4] += alpha * b[i + 4];
}

and turn them into this:

for (int i = 0; i < 3200; ++i) {
  a[i] += alpha * b[i];
}

and loops like this:

for (int i = 0; i < 500; ++i) {
  x[3*i] = foo(0);
  x[3*i+1] = foo(0);
  x[3*i+2] = foo(0);
}

and turn them into this:

for (int i = 0; i < 1500; ++i) {
  x[i] = foo(0);
}

There are two motivations for this transformation:

  1. Code-size reduction (especially relevant, obviously, when compiling for
code size).

  2. Providing greater choice to the loop vectorizer (and generic unroller) to
choose the unrolling factor (and a better ability to vectorize). The loop
vectorizer can take vector lengths and register pressure into account when
choosing an unrolling factor, for example, and a pre-unrolled loop limits that
choice. This is especially problematic if the manual unrolling was optimized
for a machine different from the current target.

The current implementation is limited to single basic-block loops only. The
rerolling recognition should work regardless of how the loop iterations are
intermixed within the loop body (subject to dependency and side-effect
constraints), but the significant restriction is that the order of the
instructions in each iteration must be identical. This seems sufficient to
capture all current use cases.

This pass is not currently enabled by default at any optimization level.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194939 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoThe WebKit_JS CC preserves the same registers as the C CC.
Juergen Ributzka [Sat, 16 Nov 2013 22:08:58 +0000 (22:08 +0000)]
The WebKit_JS CC preserves the same registers as the C CC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194936 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoApply the InstCombine fptrunc sqrt optimization to llvm.sqrt
Hal Finkel [Sat, 16 Nov 2013 21:29:08 +0000 (21:29 +0000)]
Apply the InstCombine fptrunc sqrt optimization to llvm.sqrt

InstCombine, in visitFPTrunc, applies the following optimization to sqrt calls:

  (fptrunc (sqrt (fpext x))) -> (sqrtf x)

but does not apply the same optimization to llvm.sqrt. This is a problem
because, to enable vectorization, Clang generates llvm.sqrt instead of sqrt in
fast-math mode, and because this optimization is being applied to sqrt and not
applied to llvm.sqrt, sometimes the fast-math code is slower.

This change makes InstCombine apply this optimization to llvm.sqrt as well.

This fixes the specific problem in PR17758, although the same underlying issue
(optimizations applied to libcalls are not applied to intrinsics) exists for
other optimizations in SimplifyLibCalls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194935 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix assert on unaligned access to global with different address space size.
Matt Arsenault [Sat, 16 Nov 2013 20:50:54 +0000 (20:50 +0000)]
Fix assert on unaligned access to global with different address space size.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194934 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoFix codegen for null different sized pointer.
Matt Arsenault [Sat, 16 Nov 2013 20:24:41 +0000 (20:24 +0000)]
Fix codegen for null different sized pointer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194932 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoScalarEvolution: Warn if the result of setFlags/clearFlags is unused.
Benjamin Kramer [Sat, 16 Nov 2013 16:25:47 +0000 (16:25 +0000)]
ScalarEvolution: Warn if the result of setFlags/clearFlags is unused.

This was a source of bugs in the past.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194929 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoAnnotate APInt methods where it's not clear whether they are in place with warn_unuse...
Benjamin Kramer [Sat, 16 Nov 2013 16:25:41 +0000 (16:25 +0000)]
Annotate APInt methods where it's not clear whether they are in place with warn_unused_result.

Fix ScalarEvolution bugs uncovered by this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194928 91177308-0d34-0410-b5e6-96231b3b80d8

10 years agoR600: Make dot_4 instructions predicable
Vincent Lejeune [Sat, 16 Nov 2013 16:24:41 +0000 (16:24 +0000)]
R600: Make dot_4 instructions predicable

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194927 91177308-0d34-0410-b5e6-96231b3b80d8