oota-llvm.git
9 years agovpblend intrinsics combines as shifts intrinsics due to absence return stmt between...
Robert Khasanov [Tue, 24 Jun 2014 18:08:04 +0000 (18:08 +0000)]
vpblend intrinsics combines as shifts intrinsics due to absence return stmt between them
Fix PR20088

Differential Revision: http://reviews.llvm.org/D4277

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211617 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600: Remove DIV_INF
Matt Arsenault [Tue, 24 Jun 2014 17:42:16 +0000 (17:42 +0000)]
R600: Remove DIV_INF

This corresponded to an amdil instruction which there is
a 2 instruction equivalent for.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211616 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600/SI: Move pattern to instruction definition
Matt Arsenault [Tue, 24 Jun 2014 17:17:06 +0000 (17:17 +0000)]
R600/SI: Move pattern to instruction definition

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211614 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix test case in r211605/r211533
Weiming Zhao [Tue, 24 Jun 2014 17:05:43 +0000 (17:05 +0000)]
Fix test case in r211605/r211533

The test case in
"Fix PR20056: Implement pseudo LDR <reg>, =<literal/label> for AArch64" should
only work with Linux.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211613 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd new debug kind LocTrackingOnly.
Diego Novillo [Tue, 24 Jun 2014 17:02:03 +0000 (17:02 +0000)]
Add new debug kind LocTrackingOnly.

Summary:
This new debug emission kind supports emitting line location
information in all instructions, but stops code generation
from emitting debug info to the final output.

This mode is useful when the backend wants to track source
locations during code generation, but it does not want to
produce debug info. This is currently used by optimization
remarks (-pass-remarks, -pass-remarks-missed and
-pass-remarks-analysis).

To prevent debug info emission, DIBuilder never inserts the
annotation 'llvm.dbg.cu' when LocTrackingOnly is enabled.

Reviewers: echristo, dblaikie

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D4234

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211609 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoResubmit commit r211533
Weiming Zhao [Tue, 24 Jun 2014 16:21:38 +0000 (16:21 +0000)]
Resubmit commit r211533

"Fix PR20056: Implement pseudo LDR <reg>, =<literal/label> for AArch64"
Missed files are added in this commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211605 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: Avoid multiple strlen calls
David Majnemer [Tue, 24 Jun 2014 16:01:53 +0000 (16:01 +0000)]
CodeGen: Avoid multiple strlen calls

Use a StringRef to hold our section prefix.  This avoids multiple calls
to strlen.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211602 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoARM: Fix TPsoft for Thumb mode
Christian Pirker [Tue, 24 Jun 2014 15:45:59 +0000 (15:45 +0000)]
ARM: Fix TPsoft for Thumb mode

Reviewed at http://reviews.llvm.org/D4230

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211601 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove an empty directory left after r211508.
Alexander Kornienko [Tue, 24 Jun 2014 15:38:38 +0000 (15:38 +0000)]
Remove an empty directory left after r211508.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211600 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoReplace two release calls with std::move. I missed this on the previous commit.
Rafael Espindola [Tue, 24 Jun 2014 14:25:17 +0000 (14:25 +0000)]
Replace two release calls with std::move. I missed this on the previous commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211597 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoPass a unique_ptr<MemoryBuffer> to the constructors in the Binary hierarchy.
Rafael Espindola [Tue, 24 Jun 2014 13:56:32 +0000 (13:56 +0000)]
Pass a unique_ptr<MemoryBuffer> to the constructors in the  Binary hierarchy.

Once the objects are constructed, they own the buffer. Passing a unique_ptr
makes that clear.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211595 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert: r211588 - [mips] Use __clear_cache builtin instead of cacheflush() in Unix...
Daniel Sanders [Tue, 24 Jun 2014 13:53:56 +0000 (13:53 +0000)]
Revert: r211588 - [mips] Use __clear_cache builtin instead of cacheflush() in Unix Memory::InvalidateInstructionCache()

Buildbot reports a test failure on the llvm-mips-linux builder and blames r211588.
Although it doesn't appear in the blamelist, it seems it could also be r211587
(because it's committed to compiler-rt?) since they were tested together.

Reverting the most likely suspect (r211588) to confirm one way or the other.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211594 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert r211287, "Remove support for LLVM runtime multi-threading."
NAKAMURA Takumi [Tue, 24 Jun 2014 13:36:31 +0000 (13:36 +0000)]
Revert r211287, "Remove support for LLVM runtime multi-threading."

libclang still requires it on cygming, lack of incomplete <mutex>.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211592 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips] Added support for assembling sdbbp.
Daniel Sanders [Tue, 24 Jun 2014 13:00:32 +0000 (13:00 +0000)]
[mips] Added support for assembling sdbbp.

Summary:
This instruction is re-encoded in MIPS32r6/MIPS64r6 without changing the
restrictions. We hadn't implemented it for earlier ISA's so it has been added to those too.

Differential Revision: http://reviews.llvm.org/D4265

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211590 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips] Use __clear_cache builtin instead of cacheflush() in Unix Memory::InvalidateIn...
Daniel Sanders [Tue, 24 Jun 2014 12:26:49 +0000 (12:26 +0000)]
[mips] Use __clear_cache builtin instead of cacheflush() in Unix Memory::InvalidateInstructionCache()

MIPS64 Android bionic has removed cacheflush(). Use __clear_cache() instead for 32-bit and 64-bit MIPS

Patch by Duane Sand <Duane.Sand@imgtec.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211588 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoInstCombine: Disable umul.with.overflow recognition for vectors.
Benjamin Kramer [Tue, 24 Jun 2014 10:47:52 +0000 (10:47 +0000)]
InstCombine: Disable umul.with.overflow recognition for vectors.

It doesn't make a lot on most targets and the code isn't ready for it. PR20113.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211583 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoInstCombine: Don't try to reorder shuffles where the mask is a ConstantExpr.
Benjamin Kramer [Tue, 24 Jun 2014 10:38:10 +0000 (10:38 +0000)]
InstCombine: Don't try to reorder shuffles where the mask is a ConstantExpr.

We can't analyze the individual values of a vector expression. PR20114.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211581 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoGlobalOpt: Don't optimize dllimport for initializers
David Majnemer [Tue, 24 Jun 2014 06:53:45 +0000 (06:53 +0000)]
GlobalOpt: Don't optimize dllimport for initializers

Referencing a dllimport variable requires actually instructions, not
just a relocation.  This fixes PR19955.

Differential Revision: http://reviews.llvm.org/D4249

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211571 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoInclude <tuple> for std::tie
Alexey Samsonov [Tue, 24 Jun 2014 05:59:52 +0000 (05:59 +0000)]
Include <tuple> for std::tie

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211569 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AArch64] Fix a build_vector pattern match fail
Kevin Qin [Tue, 24 Jun 2014 05:37:27 +0000 (05:37 +0000)]
[AArch64] Fix a build_vector pattern match fail
caused by defect in isBuildVectorAllZeros().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211567 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Disasm][AVX512] Implement decoding of top bit for non-destructive reg fields
Adam Nemet [Tue, 24 Jun 2014 01:42:32 +0000 (01:42 +0000)]
[Disasm][AVX512] Implement decoding of top bit for non-destructive reg fields

V' bit in the P2 byte of the EVEX prefix provides the top bit of the NDD and
NDS register fields.  This was simply not used in the decoder until now.

Fixes <rdar://problem/17402661>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211565 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Emacs] Use spaces rather than tabs for indentation in tablegen-mode
Adam Nemet [Tue, 24 Jun 2014 01:42:29 +0000 (01:42 +0000)]
[Emacs] Use spaces rather than tabs for indentation in tablegen-mode

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211564 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSupport: Move class ScaledNumber
Duncan P. N. Exon Smith [Tue, 24 Jun 2014 00:38:09 +0000 (00:38 +0000)]
Support: Move class ScaledNumber

ScaledNumber has been cleaned up enough to pull out of BFI now.  Still
work to do there (tests for shifting, bloated printing code, etc.), but
it seems clean enough for its new home.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211562 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoBFI: Un-floatify more language
Duncan P. N. Exon Smith [Tue, 24 Jun 2014 00:26:13 +0000 (00:26 +0000)]
BFI: Un-floatify more language

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211561 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoBFI: Remove more float-like language from ScaledNumber
Duncan P. N. Exon Smith [Tue, 24 Jun 2014 00:26:10 +0000 (00:26 +0000)]
BFI: Remove more float-like language from ScaledNumber

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211560 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSupport: Return ScaledNumbers::MaxScale from getQuotient()
Duncan P. N. Exon Smith [Tue, 24 Jun 2014 00:26:08 +0000 (00:26 +0000)]
Support: Return ScaledNumbers::MaxScale from getQuotient()

Return MaxScale now that it's available.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211559 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSupport: Extract ScaledNumbers::MinScale and MaxScale
Duncan P. N. Exon Smith [Tue, 24 Jun 2014 00:15:19 +0000 (00:15 +0000)]
Support: Extract ScaledNumbers::MinScale and MaxScale

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211558 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoBFI: Change language from "exponent" to "scale"
Duncan P. N. Exon Smith [Mon, 23 Jun 2014 23:57:12 +0000 (23:57 +0000)]
BFI: Change language from "exponent" to "scale"

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211557 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoBFI: Update ScaledNumber header docs
Duncan P. N. Exon Smith [Mon, 23 Jun 2014 23:49:31 +0000 (23:49 +0000)]
BFI: Update ScaledNumber header docs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211556 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoBFI: Rename UnsignedFloat => ScaledNumber
Duncan P. N. Exon Smith [Mon, 23 Jun 2014 23:36:17 +0000 (23:36 +0000)]
BFI: Rename UnsignedFloat => ScaledNumber

A lot of the docs and API are out of date, but I'll leave that for a
separate commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211555 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix comments from r211553
Duncan P. N. Exon Smith [Mon, 23 Jun 2014 23:17:36 +0000 (23:17 +0000)]
Fix comments from r211553

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211554 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSupport: Extract ScaledNumbers::getSum() and getDifference()
Duncan P. N. Exon Smith [Mon, 23 Jun 2014 23:15:25 +0000 (23:15 +0000)]
Support: Extract ScaledNumbers::getSum() and getDifference()

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211553 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSupport: Return scale from ScaledNumbers::matchScales()
Duncan P. N. Exon Smith [Mon, 23 Jun 2014 23:14:51 +0000 (23:14 +0000)]
Support: Return scale from ScaledNumbers::matchScales()

This will be convenient when extracting `ScaledNumbers::getSum()`.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211552 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agonot: Only consider exit code 3 to be a crash with --crash
Reid Kleckner [Mon, 23 Jun 2014 22:54:33 +0000 (22:54 +0000)]
not: Only consider exit code 3 to be a crash with --crash

This fixes Clang's test/Index/comment-xml-schema.c with Cygwin's
xmllint.exe, which uses exit(3) for XML validation failure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211550 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoPass a std::unique_ptr& to the create??? methods is lib/Object.
Rafael Espindola [Mon, 23 Jun 2014 22:00:37 +0000 (22:00 +0000)]
Pass a std::unique_ptr& to the create??? methods is lib/Object.

This makes the buffer ownership on error conditions very natural. The buffer
is only moved out of the argument if an object is constructed that now
owns the buffer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211546 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[FastISel][X86] Lower unsupported selects to control-flow.
Juergen Ributzka [Mon, 23 Jun 2014 21:55:44 +0000 (21:55 +0000)]
[FastISel][X86] Lower unsupported selects to control-flow.

The extends the select lowering coverage by emiting pseudo cmov
instructions. These insturction will be later on lowered to control-flow to
simulate the select.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211545 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[FastISel][X86] Add support for floating-point select.
Juergen Ributzka [Mon, 23 Jun 2014 21:55:40 +0000 (21:55 +0000)]
[FastISel][X86] Add support for floating-point select.

This extends the select lowering to support floating-point selects. The
lowering depends on SSE instructions and that the conditon comes from a
floating-point compare. Under this conditions it is possible to emit an
optimized instruction sequence that doesn't require any branches to
simulate the select.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211544 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[FastISel][X86] Optimize selects when the condition comes from a compare.
Juergen Ributzka [Mon, 23 Jun 2014 21:55:36 +0000 (21:55 +0000)]
[FastISel][X86] Optimize selects when the condition comes from a compare.

Optimize the select instructions sequence to use the EFLAGS directly from a
compare when possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211543 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMake ObjectFile and BitcodeReader always own the MemoryBuffer.
Rafael Espindola [Mon, 23 Jun 2014 21:53:12 +0000 (21:53 +0000)]
Make ObjectFile and BitcodeReader always own the MemoryBuffer.

This allows us to just use a std::unique_ptr to store the pointer to the buffer.
The flip side is that they have to support releasing the buffer back to the
caller.

Overall this looks like a more efficient and less brittle api.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211542 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoThis reverts commit r211533 and r211539.
Rafael Espindola [Mon, 23 Jun 2014 21:20:58 +0000 (21:20 +0000)]
This reverts commit r211533 and r211539.

 Revert "Fix PR20056: Implement pseudo LDR <reg>, =<literal/label> for AArch64"
 Revert "Fix cmake build."

It was missing a file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211540 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix cmake build.
Juergen Ributzka [Mon, 23 Jun 2014 21:15:55 +0000 (21:15 +0000)]
Fix cmake build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211539 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSimplify memory management with std::unique_ptr.
Rafael Espindola [Mon, 23 Jun 2014 21:15:27 +0000 (21:15 +0000)]
Simplify memory management with std::unique_ptr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211538 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agonm-trivial-object.test requires shell since Lit internal runner isn't capable of...
NAKAMURA Takumi [Mon, 23 Jun 2014 21:07:04 +0000 (21:07 +0000)]
nm-trivial-object.test requires shell since Lit internal runner isn't capable of chdir.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211537 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix PR20056: Implement pseudo LDR <reg>, =<literal/label> for AArch64
Weiming Zhao [Mon, 23 Jun 2014 20:44:16 +0000 (20:44 +0000)]
Fix PR20056: Implement pseudo LDR <reg>, =<literal/label> for AArch64

This patch is based on the changes from ARM target [1,2]

Based on ARM doc [3], if the literal value can be loaded with a valid MOV,
it can emit that instruction. This is implemented in this patch.

[1] Fix PR18345: ldr= pseudo instruction produces incorrect code when using in inline assembly
Author: David Peixotto <dpeixott@codeaurora.org>
commit b92cca222898d87bbc764fa22e805adb04ef7f13 (r200777)
[2] Implement the ldr-pseudo opcode for ARM assembly
Author: David Peixotto <dpeixott@codeaurora.org>
commit 0fa193b08627927ccaa0804a34d80480894614b8 (r197708)
[3] http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0802a/CJAHAIBC.html

Differential Revision: http://reviews.llvm.org/D4163

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211533 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoConvert a few methods to use ErrorOr.
Rafael Espindola [Mon, 23 Jun 2014 20:41:02 +0000 (20:41 +0000)]
Convert a few methods to use ErrorOr.

It used to be inconvenient to mix ErrorOr and UniquePtr, but with c++11
they work OK together.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211532 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSupport: Extract ScaledNumbers::matchScale()
Duncan P. N. Exon Smith [Mon, 23 Jun 2014 20:40:45 +0000 (20:40 +0000)]
Support: Extract ScaledNumbers::matchScale()

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211531 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoChange the default input for llvm-nm to be a.out instead of standard input
Kevin Enderby [Mon, 23 Jun 2014 20:27:53 +0000 (20:27 +0000)]
Change the default input for llvm-nm to be a.out instead of standard input
to match llvm-size and other UNIX systems for their nm(1).

Tweak test cases that used llvm-nm with standard input to add a "-" to
indicate that and add a test case to check the default of a.out for llvm-nm.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211529 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Mips] Add a target streamer when creating a null streamer.
Rafael Espindola [Mon, 23 Jun 2014 19:43:40 +0000 (19:43 +0000)]
[Mips] Add a target streamer when creating a null streamer.

Should fix DebugInfo/global.ll on the mips bot.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211527 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRecommit 211309 (StringMap::insert), reverted in 211328 due to issues with private...
David Blaikie [Mon, 23 Jun 2014 18:28:53 +0000 (18:28 +0000)]
Recommit 211309 (StringMap::insert), reverted in 211328 due to issues with private, but non-deleted, move members.

Certain versions of GCC (~4.7) couldn't handle the SFINAE on access
control, but with "= delete" (hidden behind a macro for portability)
this issue is worked around/addressed.

Patch by Agustín Bergé

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211525 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600/SI: Verify restrictions on div_scale operands.
Matt Arsenault [Mon, 23 Jun 2014 18:28:31 +0000 (18:28 +0000)]
R600/SI: Verify restrictions on div_scale operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211524 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600/SI: Fix div_scale intrinsic.
Matt Arsenault [Mon, 23 Jun 2014 18:28:28 +0000 (18:28 +0000)]
R600/SI: Fix div_scale intrinsic.

The operand that must match one of the others does matter,
and implement selecting for it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211523 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600: Fix formatting of intrinsic definitions.
Matt Arsenault [Mon, 23 Jun 2014 18:28:23 +0000 (18:28 +0000)]
R600: Fix formatting of intrinsic definitions.

It makes more sense for the GCCBuiltin name to come immediately
after the ID name.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211522 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCleanup r211507
Duncan P. N. Exon Smith [Mon, 23 Jun 2014 18:08:58 +0000 (18:08 +0000)]
Cleanup r211507

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211521 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoARMEB: Vector extend operations
Christian Pirker [Mon, 23 Jun 2014 18:05:53 +0000 (18:05 +0000)]
ARMEB: Vector extend operations

Reviewed at http://reviews.llvm.org/D4043

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211520 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600: Remove AMDILISelLowering
Matt Arsenault [Mon, 23 Jun 2014 18:00:55 +0000 (18:00 +0000)]
R600: Remove AMDILISelLowering

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211519 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600: Select is not expensive.
Matt Arsenault [Mon, 23 Jun 2014 18:00:52 +0000 (18:00 +0000)]
R600: Select is not expensive.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211518 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600: Move add/sub with overflow out of AMDILISelLowering
Matt Arsenault [Mon, 23 Jun 2014 18:00:49 +0000 (18:00 +0000)]
R600: Move add/sub with overflow out of AMDILISelLowering

Add more tests for these.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211517 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600: Move more out of AMDILISelLowering
Matt Arsenault [Mon, 23 Jun 2014 18:00:44 +0000 (18:00 +0000)]
R600: Move more out of AMDILISelLowering

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211516 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600: Don't set fp_round_inreg action.
Matt Arsenault [Mon, 23 Jun 2014 18:00:41 +0000 (18:00 +0000)]
R600: Don't set fp_round_inreg action.

There's no point in setting this since it seems to only
by created in 1 place for ppcf128

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211515 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600/SI: Handle i64 sub.
Matt Arsenault [Mon, 23 Jun 2014 18:00:38 +0000 (18:00 +0000)]
R600/SI: Handle i64 sub.

We can handle it the same way as add

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211514 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600/SI: Move selection of i64 add to separate function.
Matt Arsenault [Mon, 23 Jun 2014 18:00:34 +0000 (18:00 +0000)]
R600/SI: Move selection of i64 add to separate function.

Also don't use a SmallVector for fixed size array.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211513 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600: Rename AMDIL file
Matt Arsenault [Mon, 23 Jun 2014 18:00:31 +0000 (18:00 +0000)]
R600: Rename AMDIL file

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211512 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix missing words in sentence
Matt Arsenault [Mon, 23 Jun 2014 18:00:26 +0000 (18:00 +0000)]
Fix missing words in sentence

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211511 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse helper function
Matt Arsenault [Mon, 23 Jun 2014 18:00:24 +0000 (18:00 +0000)]
Use helper function

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211510 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAlphabetize forward declarations
Matt Arsenault [Mon, 23 Jun 2014 18:00:20 +0000 (18:00 +0000)]
Alphabetize forward declarations

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211509 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDelete utils/FileUpdate.
Rafael Espindola [Mon, 23 Jun 2014 17:58:39 +0000 (17:58 +0000)]
Delete utils/FileUpdate.

It is unused and it looks like it was never used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211508 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSupport: Extract ScaledNumbers::compare()
Duncan P. N. Exon Smith [Mon, 23 Jun 2014 17:47:40 +0000 (17:47 +0000)]
Support: Extract ScaledNumbers::compare()

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211507 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAllow using .cfi_startproc without a leading symbol.
Rafael Espindola [Mon, 23 Jun 2014 15:34:32 +0000 (15:34 +0000)]
Allow using .cfi_startproc without a leading symbol.

This is possible now that we don't produce .eh symbols. This fixes pr19430.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211502 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoclang-format initializer to reduce noise in another patch.
Rafael Espindola [Mon, 23 Jun 2014 15:17:34 +0000 (15:17 +0000)]
clang-format initializer to reduce noise in another patch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211501 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoStop producing func.eh symbols on Darwin.
Rafael Espindola [Mon, 23 Jun 2014 15:13:23 +0000 (15:13 +0000)]
Stop producing func.eh symbols on Darwin.

According Nick Kledzik (http://llvm.org/bugs/show_bug.cgi?id=19430#c2):
"... mach-o no longer needs names in the __eh_frame section (and has not for
years)."

Iain Sandoe confirms it is also unnecessary for their old darwin support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211500 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove a temporary hack.
Rafael Espindola [Mon, 23 Jun 2014 14:22:55 +0000 (14:22 +0000)]
Remove a temporary hack.

Amusingly this survived a lot longer than the CFI transition. We don't even
support non-cfi assemblers any more.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211498 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[PowerPC] Refactor getMinCallFrameSize / getMinCallArgumentsSize
Ulrich Weigand [Mon, 23 Jun 2014 14:15:53 +0000 (14:15 +0000)]
[PowerPC] Refactor getMinCallFrameSize / getMinCallArgumentsSize

As of r211495, the only remaining users of getMinCallFrameSize are in
core ABI code (LowerFormalParameter / LowerCall).  This is actually a
good thing, since the details of the parameter save area are ABI specific.

With the new ELFv2 ABI in particular, the rules defining the size of the
save area will become significantly more complex, so it wouldn't make
sense to implement those outside ABI code that has all required
information.

In preparation, this patch eliminates the getMinCallFrameSize (and
associated getMinCallArgumentsSize) routines, and inlines them into all
callers.  Note that since nearly all call arguments are constant, this
allows simplifying the inlined copies to a single line everywhere.

No change in generate code expected.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211497 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[PowerPC] Allow stack frames without parameter save area
Ulrich Weigand [Mon, 23 Jun 2014 13:47:52 +0000 (13:47 +0000)]
[PowerPC] Allow stack frames without parameter save area

The PPCFrameLowering::determineFrameLayout routine currently ensures
that every function that allocates a stack frame provides space for the
parameter save area (via PPCFrameLowering::getMinCallFrameSize).

This is actually not necessary.  There may be functions that never call
another routine but still allocate a frame; those do not require the
parameter save area.  In the future, with the ELFv2 ABI, even some
routines that do call other functions do not need to allocate the
parameter save area.

While it is not a bug to allocate the parameter area when it is not
needed, it is better to avoid it to save stack space.

Note that when any particular function call requires the parameter save
area, this space will already have been included by ABI code in the size
the CALLSEQ_START insn is annotated with, and therefore included in the
size returned by MFI->getMaxCallFrameSize().

This means that determineFrameLayout simply does not need to care about
the parameter save area.  (It still needs to ensure that every frame
provides the linkage area.)  This is implemented by this patch.

Note that this exposed a bug in the new fast-isel code where the parameter
area was *not* included in the CALLSEQ_START size; this is also fixed.

A couple of test cases needed to be adapted for the new (smaller) stack
frame size those tests now see.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211495 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[PowerPC] Fix IsDarwin arg in PPCFrameLowering:: calls
Ulrich Weigand [Mon, 23 Jun 2014 13:21:43 +0000 (13:21 +0000)]
[PowerPC] Fix IsDarwin arg in PPCFrameLowering:: calls

As remarked in the commit message to r211493, in several places
throughout the 64-bit SVR4 ABI code there are calls to
PPCFrameLowering::getLinkageSize and getMinCallFrameSize
using an incorrect IsDarwin argument of "true".

(Some of those were made explicit by the above refactoring patch, others
have been there all along.)

This patch fixes those places to pass "false" for IsDarwin.

No change in generated code expected.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211494 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[PowerPC] Refactor setMinReservedArea and CalculateParameterAndLinkageAreaSize
Ulrich Weigand [Mon, 23 Jun 2014 13:08:27 +0000 (13:08 +0000)]
[PowerPC] Refactor setMinReservedArea and CalculateParameterAndLinkageAreaSize

The PPCISelLowering.cpp routines PPCTargetLowering::setMinReservedArea and
CalculateParameterAndLinkageAreaSize are currently used as subroutines
from both 64-bit SVR4 and Darwin ABI code.

However, the two ABIs are already quite different w.r.t. AltiVec
conventions, and they will become more different when the ELFv2 ABI is
supported.  Also, in general it seems better to disentangle ABI support
routines for different ABIs to avoid accidentally affecting one ABI when
intending to change only the other.

(Actually, the current code strictly speaking already contains a bug:
these routines call PPCFrameLowering::getMinCallFrameSize and
PPCFrameLowering::getLinkageSize with the IsDarwin parameter set to
"true" even on 64-bit SVR4.  This bug currently has no adverse effect
since those routines always return the same for 64-bit SVR4 and 64-bit
Darwin, but it still seems wrong ...  I'll fix this in a follow-up
commit shortly.)

To remove this code sharing, I'm simply inlining both routines into all
call sites (there are just two each, one for 64-bit SVR4 and one for
Darwin), and simplifying due to constant parameters where possible.

A small piece of code that *does* make sense to share is refactored into
the new routine EnsureStackAlignment, now also called from 32-bit SVR4
ABI code.

No change in generated code is expected.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211493 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[PowerPC] Fix on-stack AltiVec arguments with 64-bit SVR4
Ulrich Weigand [Mon, 23 Jun 2014 12:36:34 +0000 (12:36 +0000)]
[PowerPC] Fix on-stack AltiVec arguments with 64-bit SVR4

Current 64-bit SVR4 code seems to have some remnants of Darwin code
in AltiVec argument handing.  This had the effect that AltiVec arguments
(or subsequent arguments) were not correctly placed in the parameter area
in some cases.

The correct behaviour with the 64-bit SVR4 ABI is:
- All AltiVec arguments take up space in the parameter area, just like
  any other arguments, whether vararg or not.
- They are always 16-byte aligned, skipping a parameter area doubleword
  (and the associated GPR, if any), if necessary.

This patch implements the correct behaviour and adds a test case.
(Verified against GCC behaviour via the ABI compat test suite.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211492 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoARM: mark UBFX as not allowing PC.
Tim Northover [Mon, 23 Jun 2014 09:20:02 +0000 (09:20 +0000)]
ARM: mark UBFX as not allowing PC.

Strictly, it's unpredictable. But we don't quite model that yet and an error is
better than ignoring the issue. This one somehow got left out before though.

rdar://problem/15997748

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211490 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMC: Cleanup parseMSInlineAsm
David Majnemer [Mon, 23 Jun 2014 02:17:16 +0000 (02:17 +0000)]
MC: Cleanup parseMSInlineAsm

Utilize range based for-loops to simplify some code.
Use insert() instead of a loop for simplicity/efficiency.

No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211486 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMC: adjust text section flags for WoA
Saleem Abdulrasool [Sun, 22 Jun 2014 22:25:01 +0000 (22:25 +0000)]
MC: adjust text section flags for WoA

Correct the section flags for code built for Windows on ARM with
`-ffunction-sections`.  Windows on ARM uses solely Thumb-2 instructions, and
indicates that the function is thumb by placing it in a text section that has
IMAGE_SCN_MEM_16BIT flag set.

When we encounter a .section directive, a new section is constructed.  This may
be a text segment.  In order to identify that we need the additional flag,
expose the target triple through the ObjectFileInfo as this information is lost
otherwise.

Since any modern ARM targeting environment on Windows would be Thumb-2 (Windows
ARM NT or Windows Embedded Compact), introducing a new flag to indicate the
section attribute seems to be a bit overkill.  Simply depend on the target
triple.  Since there is one location that this information is currently needed,
creating a target specific assembly parser and delegating the parsing of section
switches also feels a bit heavy handed.  If it turns out that this information
ends up changing additional behaviour, then it may be worth considering that
alternative.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211481 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert r211399, "Generate native unwind info on Win64"
NAKAMURA Takumi [Sun, 22 Jun 2014 22:00:56 +0000 (22:00 +0000)]
Revert r211399, "Generate native unwind info on Win64"

It broke Legacy JIT Tests on x86_64-{mingw32|msvc}, aka Windows x64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211480 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600: Use LowerSDIVREM for i64 node replace
Jan Vesely [Sun, 22 Jun 2014 21:43:01 +0000 (21:43 +0000)]
R600: Use LowerSDIVREM for i64 node replace

v2: move div/rem node replacement to R600ISelLowering
    make lowerSDIVREM protected

Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211478 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600: Implement custom SDIVREM.
Jan Vesely [Sun, 22 Jun 2014 21:43:00 +0000 (21:43 +0000)]
R600: Implement custom SDIVREM.

Instead of separate SDIV/SREM. SDIV used UDIV which in turn used UDIVREM anyway.
SREM used SDIV(UDIV->UDIVREM)+MUL+SUB, using UDIVREM directly is more efficient.

v2: Don't use all caps names

Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211477 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600: Add udivrem test
Jan Vesely [Sun, 22 Jun 2014 21:42:58 +0000 (21:42 +0000)]
R600: Add udivrem test

v2: move < %s to the end of the line
    space after ;
    add v4i32 test

Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211476 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix PR20087 by using the source index when changing the vector load
Filipe Cabecinhas [Sun, 22 Jun 2014 17:21:37 +0000 (17:21 +0000)]
Fix PR20087 by using the source index when changing the vector load

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211472 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoIntroduce a Lit feature "debug_frame" and apply it to llvm/test/MC/ELF/cfi-version.ll.
NAKAMURA Takumi [Sun, 22 Jun 2014 12:35:39 +0000 (12:35 +0000)]
Introduce a Lit feature "debug_frame" and apply it to llvm/test/MC/ELF/cfi-version.ll.

.debug_frame is not emitted for targeting Windows x64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211466 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd a description to the test from r211433 explaining why it's written that way.
Benjamin Kramer [Sun, 22 Jun 2014 12:22:04 +0000 (12:22 +0000)]
Add a description to the test from r211433 explaining why it's written that way.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211465 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoLoopVectorizer: Fix a dominance issue
Arnold Schwaighofer [Sun, 22 Jun 2014 03:38:59 +0000 (03:38 +0000)]
LoopVectorizer: Fix a dominance issue

The induction variables start value needs to be defined before we branch
(overflow check) to the scalar preheader where we used it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211460 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMergeFunctions Pass, removed DenseMap helpers.
Stepan Dyatkovskiy [Sun, 22 Jun 2014 01:53:30 +0000 (01:53 +0000)]
MergeFunctions Pass, removed DenseMap helpers.

Patch removes rest part of code related to old implementation.

This patch belongs to patch series that improves MergeFunctions
performance time from O(N*N) to O(N*log(N)).

This one was the final patch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211457 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMergeFunctions Pass, updated header comments.
Stepan Dyatkovskiy [Sun, 22 Jun 2014 00:57:09 +0000 (00:57 +0000)]
MergeFunctions Pass, updated header comments.

Added short description for new comparison algorithm, that introduces
total ordering among functions set.

This patch belongs to patch series that improves MergeFunctions
performance time from O(N*N) to O(N*log(N)).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211456 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoReport error for non-zero data in .bss
Weiming Zhao [Sun, 22 Jun 2014 00:33:44 +0000 (00:33 +0000)]
Report error for non-zero data in .bss

User may initialize a var with non-zero value and specify .bss section.
E.g. : int a __attribute__((section(".bss"))) = 2;

This patch converts an assertion to error report for better user
experience.

Differential Revision: http://reviews.llvm.org/D4199

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211455 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMergeFunctions Pass, FnSet has been replaced with FnTree.
Stepan Dyatkovskiy [Sat, 21 Jun 2014 20:54:36 +0000 (20:54 +0000)]
MergeFunctions Pass, FnSet has been replaced with FnTree.

Patch activates new implementation.
So from now, merging process should take time O(N*log(N)).
Where N size of module (we are free to measure it in
functions or in instructions). Internally FnTree represents
binary tree. So every lookup operation takes O(log(N)) time.

It is still not the last patch in series, we also have to
clean-up pass from old code, and update pass comments.

This patch belongs to patch series that improves MergeFunctions
performance time from O(N*N) to O(N*log(N)).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211445 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMergeFunctions Pass, removed unused methods from old implementation.
Stepan Dyatkovskiy [Sat, 21 Jun 2014 20:13:24 +0000 (20:13 +0000)]
MergeFunctions Pass, removed unused methods from old implementation.

Patch removed next old FunctionComparator methods:
    * enumerate
    * isEquivalentOperation
    * isEquivalentGEP
    * isEquivalentType

This patch belongs to patch series that improves MergeFunctions
performance time from O(N*N) to O(N*log(N)).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211444 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMergeFunctions, doSanityCheck: fixed body comments.
Stepan Dyatkovskiy [Sat, 21 Jun 2014 19:07:51 +0000 (19:07 +0000)]
MergeFunctions, doSanityCheck: fixed body comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211443 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMergeFunctions Pass, introduced sanity check, that checks order relation,
Stepan Dyatkovskiy [Sat, 21 Jun 2014 18:58:11 +0000 (18:58 +0000)]
MergeFunctions Pass, introduced sanity check, that checks order relation,
introduced among functions set.

This patch belongs to patch series that improves MergeFunctions
performance time from O(N*N) to O(N*log(N)).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211442 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMergeFunctions Pass, introduced total ordering among top-level comparison
Stepan Dyatkovskiy [Sat, 21 Jun 2014 17:55:51 +0000 (17:55 +0000)]
MergeFunctions Pass, introduced total ordering among top-level comparison
methods.

Patch changes return type of FunctionComparator::compare() and
FunctionComparator::compare(const BasicBlock*, const BasicBlock*)
methods from bool (equal or not) to {-1, 0, 1} (less, equal, great).

This patch belongs to patch series that improves MergeFunctions
performance time from O(N*N) to O(N*log(N)).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211437 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoLoopUnrollRuntime: Check for overflow in the trip count calculation.
Benjamin Kramer [Sat, 21 Jun 2014 13:46:25 +0000 (13:46 +0000)]
LoopUnrollRuntime: Check for overflow in the trip count calculation.

Fixes PR19823.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211436 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoLegalizer: Add support for splitting insert_subvectors.
Benjamin Kramer [Sat, 21 Jun 2014 12:56:42 +0000 (12:56 +0000)]
Legalizer: Add support for splitting insert_subvectors.

We handle this by spilling the whole thing to the stack and doing the
insertion as a store.

PR19492. This happens in real code because the vectorizer creates v2i128 when AVX is enabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211435 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSCEVExpander: Fold constant PHIs harder. The logic below only understands proper...
Benjamin Kramer [Sat, 21 Jun 2014 11:47:18 +0000 (11:47 +0000)]
SCEVExpander: Fold constant PHIs harder. The logic below only understands proper IVs.

PR20093.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211433 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd back functionality removed in r210497.
Richard Trieu [Sat, 21 Jun 2014 02:43:02 +0000 (02:43 +0000)]
Add back functionality removed in r210497.

Instead of asserting, output a message stating that a null pointer was found.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211430 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Add ISel patterns to select SSE3/AVX ADDSUB instructions.
Andrea Di Biagio [Sat, 21 Jun 2014 01:31:15 +0000 (01:31 +0000)]
[X86] Add ISel patterns to select SSE3/AVX ADDSUB instructions.

This patch adds ISel patterns to select SSE3/AVX ADDSUB instructions
from a sequence of "vadd + vsub + blend".

Example:

///
typedef float float4 __attribute__((ext_vector_type(4)));

float4 foo(float4 A, float4 B) {
  float4 X = A - B;
  float4 Y = A + B;
  return (float4){X[0], Y[1], X[2], Y[3]};
}
///

Before this patch, (with flag -mcpu=corei7) llc produced the following
assembly sequence:
  movaps  %xmm0, %xmm2
  addps   %xmm1, %xmm2
  subps   %xmm1, %xmm0
  blendps $10, %xmm2, %xmm0

With this patch, we now get a single
  addsubps  %xmm1, %xmm0

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211427 91177308-0d34-0410-b5e6-96231b3b80d8