OSDN Git Service

android-x86/external-llvm.git
8 years agoRevert r268832 "Refactor stripDebugInfo(Function) to handle intrinsic"
Vitaly Buka [Sat, 7 May 2016 02:10:59 +0000 (02:10 +0000)]
Revert r268832 "Refactor stripDebugInfo(Function) to handle intrinsic"

It breaks many bots

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268837 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[ValueTracking] Hoist some computation out of a loop; NFC
Sanjoy Das [Sat, 7 May 2016 02:08:24 +0000 (02:08 +0000)]
[ValueTracking] Hoist some computation out of a loop; NFC

There is no need to match the comparison instruction repeatedly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268836 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoClean up comment; NFC
Sanjoy Das [Sat, 7 May 2016 02:08:22 +0000 (02:08 +0000)]
Clean up comment; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268835 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoDelete trailing whitespace; NFC
Sanjoy Das [Sat, 7 May 2016 02:08:15 +0000 (02:08 +0000)]
Delete trailing whitespace; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268834 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRevert r268810 becase it brakes msan bot.
Vitaly Buka [Sat, 7 May 2016 01:54:00 +0000 (01:54 +0000)]
Revert r268810 becase it brakes msan bot.

16802==WARNING: MemorySanitizer: use-of-uninitialized-value
    lib/Target/ARM/ARMFrameLowering.cpp:1632

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268833 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRefactor stripDebugInfo(Function) to handle intrinsic
Mehdi Amini [Sat, 7 May 2016 01:42:36 +0000 (01:42 +0000)]
Refactor stripDebugInfo(Function) to handle intrinsic

This moves the code that handles stripping debug info intrinsic from
 StripDebugInfo(Module) to StripDebugInfo(Function). The latter is
already walking every instructions so it makes sense to do it at the
same time.
This makes also stripDebugInfo(Function) as an API more useful: it
is really dropping every debug info in the Function.
Finally the existing code is trigerring an assertion when the Module
is not fully materialized.

From: Mehdi Amini <mehdi.amini@apple.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268832 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Teach X86FixupBWInsts to promote MOV8rr/MOV16rr to MOV32rr.
Ahmed Bougacha [Sat, 7 May 2016 01:11:17 +0000 (01:11 +0000)]
[X86] Teach X86FixupBWInsts to promote MOV8rr/MOV16rr to MOV32rr.

This re-applies r268760, reverted in r268794.
Fixes http://llvm.org/PR27670

The original imp-defs assertion was way overzealous: forward all
implicit operands, except imp-defs of the new super-reg def (r268787
for GR64, but also possible for GR16->GR32), or imp-uses of the new
super-reg use.
While there, mark the source use as Undef, and add an imp-use of the
old source reg: that should cover any case of dead super-regs.

At the stage the pass runs, flags are unlikely to matter anyway;
still, let's be as correct as possible.

Also add MIR tests for the various interesting cases.

Original commit message:
Codesize is less (16) or equal (8), and we avoid partial
dependencies.

Differential Revision: http://reviews.llvm.org/D19999

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268831 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Register and initialize the FixupBW pass.
Ahmed Bougacha [Sat, 7 May 2016 01:11:10 +0000 (01:11 +0000)]
[X86] Register and initialize the FixupBW pass.

That lets us use it in MIR tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268830 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[libFuzzer] modify the docs for startup/init
Kostya Serebryany [Fri, 6 May 2016 23:51:28 +0000 (23:51 +0000)]
[libFuzzer] modify the docs for startup/init

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268824 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[libFuzzer] tweak the documentation about rss_limit
Kostya Serebryany [Fri, 6 May 2016 23:41:11 +0000 (23:41 +0000)]
[libFuzzer] tweak the documentation about rss_limit

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268822 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[libFuzzer] enhance -rss_limit_mb and enable by default. Now it will print the OOM...
Kostya Serebryany [Fri, 6 May 2016 23:38:07 +0000 (23:38 +0000)]
[libFuzzer] enhance -rss_limit_mb and enable by default. Now it will print the OOM reproducer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268821 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[PGO] Use rsplit to parse value-data line in text profile file.
Rong Xu [Fri, 6 May 2016 23:20:58 +0000 (23:20 +0000)]
[PGO] Use rsplit to parse value-data line in text profile file.
The value-data line is <PGOFuncName>:<Count_Value>. PGOFuncName might contain
':' for the internal linkage functions. We therefore need to use rsplit,
rather split, to extract the data from the line. This fixes the error when
merging a text profile file to an indexed profile file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268818 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoImplement a safer bitcode upgrade for DISubprogram.
Adrian Prantl [Fri, 6 May 2016 22:53:06 +0000 (22:53 +0000)]
Implement a safer bitcode upgrade for DISubprogram.

The bitcode upgrade I added for DISubprogram in r266446 was based on the
assumption that the CU node for the subprogram was already materialized by the
time the DISubprogram is visited. This assumption may not hold true as future
versions of LLVM may decide to write out bitcode in a different order. This
patch corrects this by introducing a versioning bit next to the distinct flag to
unambiguously differentiate the new from the old record layouts.

Note for people stabilizing LLVM out-of-tree: This patch introduces a bitcode
incompatibility with llvm trunk revisions from r266446 — this commit. (But
D19987 will ensure that it degrades gracefully).

http://reviews.llvm.org/D20004
rdar://problem/26074194

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268816 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoDetectDeadLanes: Increase precision when detecting undef inputs
Matthias Braun [Fri, 6 May 2016 22:43:50 +0000 (22:43 +0000)]
DetectDeadLanes: Increase precision when detecting undef inputs

In case of COPY-like instruction we may be able to deduce that a certain
input is unused, based on the used lanes of the register defined by the
instruction.
This even works accross otherwise incompatible copies (no need to have
compatible lanemasks, completely unused operands are still completely
unused). It even makes sense to redo the analysis in this case since we
gained information for a case we previously stopped at because of the
incompatible masks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268815 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoDetectDeadLanes: Cleanup, assert on some impossible cases.
Matthias Braun [Fri, 6 May 2016 22:43:46 +0000 (22:43 +0000)]
DetectDeadLanes: Cleanup, assert on some impossible cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268814 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoDrop error when trying to fallback from PDB to DWARF.
Zachary Turner [Fri, 6 May 2016 22:29:34 +0000 (22:29 +0000)]
Drop error when trying to fallback from PDB to DWARF.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268813 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agocmake: Avoid continue, apparently that's new
Justin Bogner [Fri, 6 May 2016 22:22:25 +0000 (22:22 +0000)]
cmake: Avoid continue, apparently that's new

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268812 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[SLPVectorizer][X86] Regenerated SEXT/ZEXT cast vectorization tests
Simon Pilgrim [Fri, 6 May 2016 22:22:18 +0000 (22:22 +0000)]
[SLPVectorizer][X86] Regenerated SEXT/ZEXT cast vectorization tests

Added 256-bit vector test as well

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268811 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[ARM] Fix Scavenger assert due to underestimated stack size
Weiming Zhao [Fri, 6 May 2016 22:20:13 +0000 (22:20 +0000)]
[ARM] Fix Scavenger assert due to underestimated stack size

(this is resubmit of r268529 with minor refactoring. r268529 was reverted
 at r268536 due a memory sanitizer failure.  I have not been able to
 reproduce that failure and I checked all the variable used in my change
 but I could not spot an issue. I did some refactoring and see if it will
 give a clearer hint)

Summary:
Currently, when checking if a stack is "BigStack" or not, it doesn't count into spills and arguments. Therefore, LLVM won't reserve spill slot for this actually "BigStack". This may cause scavenger failure.

Reviewers: rengolin

Subscribers: vitalybuka, aemerson, rengolin, tberghammer, danalbert, srhines, llvm-commits

Differential Revision: http://reviews.llvm.org/D19896

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268810 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoReapply 267210 with fix for PR27490
Philip Reames [Fri, 6 May 2016 22:17:01 +0000 (22:17 +0000)]
Reapply 267210 with fix for PR27490

Original Commit Message
Extend load/store type canonicalization to handle unordered operations

Extend the type canonicalization logic to work for unordered atomic loads and stores.  Note that while this change itself is fairly simple and low risk, there's a reasonable chance this will expose problems in the backends by suddenly generating IR they wouldn't have seen before.  Anything of this nature will be an existing bug in the backend (you could write an atomic float load), but this will definitely change the frequency with which such cases are encountered.  If you see problems, feel free to revert this change, but please make sure you collect a test case.

Note that the concern about lowering is now much less likely.  PR27490 proved that we already *were* mucking with the types of ordered atomics and volatiles.  As a result, this change doesn't introduce as much new behavior as originally thought.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268809 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoMake llvm-pdbdump print CV type records
Zachary Turner [Fri, 6 May 2016 22:15:42 +0000 (22:15 +0000)]
Make llvm-pdbdump print CV type records

This reuses the CVTypeDumper from libcodeview to dump full
information about type records within a PDB file.

Differential Revision: http://reviews.llvm.org/D20022
Reviewed By: rnk

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268808 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[libFuzzer] add exeprimental -rss_limit_mb flag to fight against OOMs
Kostya Serebryany [Fri, 6 May 2016 21:58:35 +0000 (21:58 +0000)]
[libFuzzer] add exeprimental -rss_limit_mb flag to fight against OOMs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268807 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoCMake: generate check targets for lit suites without their own lit.cfgs
Justin Bogner [Fri, 6 May 2016 21:57:30 +0000 (21:57 +0000)]
CMake: generate check targets for lit suites without their own lit.cfgs

Currently our cmake generates targets like check-llvm-unit and
check-llvm-transforms-loopunroll-x86, but not check-llvm-transforms or
check-llvm-transforms-adce. This is because the search for test suites
only lists the ones with a custom lit.cfg or lit.local.cfg.

Instead, we can do something a little smarter - any directory under
test that isn't called Inputs or inside a directory called Inputs is a
test suite.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268806 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoLiveIntervalAnalysis: Fix handleMove() extending liverange for undef inputs
Matthias Braun [Fri, 6 May 2016 21:47:41 +0000 (21:47 +0000)]
LiveIntervalAnalysis: Fix handleMove() extending liverange for undef inputs

Fix handleMove() incorrectly extending liveranges when an undef input of
a vreg was moved past the (current) end of the liverange.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268805 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[GVN] PRE of unordered loads
Philip Reames [Fri, 6 May 2016 21:43:51 +0000 (21:43 +0000)]
[GVN] PRE of unordered loads

Again, fairly simple.  Only change is ensuring that we actually copy the property of the load correctly.  The aliasing legality constraints were already handled by the FRE patches.  There's nothing special about unorder atomics from the perspective of the PRE algorithm itself.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268804 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[SLPVectorizer][X86] Added BSWAP/BITREVERSE vectorization tests
Simon Pilgrim [Fri, 6 May 2016 21:41:55 +0000 (21:41 +0000)]
[SLPVectorizer][X86] Added BSWAP/BITREVERSE vectorization tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268803 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[SLPVectorizer][X86] Added CTPOP/CTLZ/CTTZ vectorization tests
Simon Pilgrim [Fri, 6 May 2016 21:33:01 +0000 (21:33 +0000)]
[SLPVectorizer][X86] Added CTPOP/CTLZ/CTTZ vectorization tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268800 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRevert "[X86] Add a new LOW32_ADDR_ACCESS_RBP register class."
Quentin Colombet [Fri, 6 May 2016 21:21:50 +0000 (21:21 +0000)]
Revert "[X86] Add a new LOW32_ADDR_ACCESS_RBP register class."

This reverts commit r268796.
I believe it breaks test/CodeGen/X86/asm-mismatched-types.ll with:
Cannot emit physreg copy instruction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268799 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix failing test due to merge conflict.
Zachary Turner [Fri, 6 May 2016 21:19:29 +0000 (21:19 +0000)]
Fix failing test due to merge conflict.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268798 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Add a new LOW32_ADDR_ACCESS_RBP register class.
Quentin Colombet [Fri, 6 May 2016 21:10:53 +0000 (21:10 +0000)]
[X86] Add a new LOW32_ADDR_ACCESS_RBP register class.

ABIs like NaCl uses 32-bit addresses but have 64-bit frame.
The new register class reflects those constraints when choosing a
register class for a address access.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268796 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Rename the X32_ADDR_ACCESS register class into LOW32_ADDR_ACCESS.
Quentin Colombet [Fri, 6 May 2016 21:10:43 +0000 (21:10 +0000)]
[X86] Rename the X32_ADDR_ACCESS register class into LOW32_ADDR_ACCESS.

This register class may be used by any ABIs that uses x86_64 ISA while
using 32-bit addresses, not just in X32 cases. Make sure the name
reflects that.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268795 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRevert r268760, it caused PR27670.
Nico Weber [Fri, 6 May 2016 21:07:02 +0000 (21:07 +0000)]
Revert r268760, it caused PR27670.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268794 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAdd missing include.
Zachary Turner [Fri, 6 May 2016 20:59:35 +0000 (20:59 +0000)]
Add missing include.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268792 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoPort DebugInfoPDB over to using llvm::Error.
Zachary Turner [Fri, 6 May 2016 20:51:57 +0000 (20:51 +0000)]
Port DebugInfoPDB over to using llvm::Error.

Differential Revision: http://reviews.llvm.org/D19940
Reviewed By: rnk

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268791 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[RS4GC] Fix typo in comment
Sanjoy Das [Fri, 6 May 2016 20:39:33 +0000 (20:39 +0000)]
[RS4GC] Fix typo in comment

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268790 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoChange GenericBinaryError to no longer include a FileName, which is then not
Kevin Enderby [Fri, 6 May 2016 20:16:28 +0000 (20:16 +0000)]
Change GenericBinaryError to no longer include a FileName, which is then not
part of the error message.

As the caller is the one that needs to add the name of where the "object file"
comes from to the error message as the object file could be in an archive, or
coming from a slice of a Mach-O universal file or a buffer created by a JIT.

In the cases of a Mach-O universal file the architecture name may or may not
also need to be printed which is up to the tool code.  For example if the tool
code is only selecting the host architecture slice then that architecture name
is never printed.

This patch is the change to the libObject code and there will be follow on
commits for changes to the code for each tool.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268789 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[libFuzzer] more trophies
Kostya Serebryany [Fri, 6 May 2016 20:14:48 +0000 (20:14 +0000)]
[libFuzzer] more trophies

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268788 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Accept imp-defs of GR64 super-registers in FixupBW MOVrr.
Ahmed Bougacha [Fri, 6 May 2016 20:03:03 +0000 (20:03 +0000)]
[X86] Accept imp-defs of GR64 super-registers in FixupBW MOVrr.

Testcase will follow shortly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268787 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[MSan] [X86] Fix vararg helper for fixed arguments in overflow area.
Marcin Koscielnicki [Fri, 6 May 2016 19:36:56 +0000 (19:36 +0000)]
[MSan] [X86] Fix vararg helper for fixed arguments in overflow area.

This fixes http://llvm.org/PR27646 on x86_64.

Differential Revision: http://reviews.llvm.org/D19997

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268783 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[AMDGPU][llvm-mc] Some refactoring of .td files
Artem Tamazov [Fri, 6 May 2016 19:32:38 +0000 (19:32 +0000)]
[AMDGPU][llvm-mc] Some refactoring of .td files

Some custom Operands and AsmOperandClasses moved to proper place.
No functional changes.

Differential Revision: http://reviews.llvm.org/D20012

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268780 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRefactor the Verifier so it can diagnose IR validation errors and debug
Adrian Prantl [Fri, 6 May 2016 19:26:47 +0000 (19:26 +0000)]
Refactor the Verifier so it can diagnose IR validation errors and debug
info metadata errors separately. (NFC)

This patch refactors the Verifier so it can diagnose IR validation errors
and debug info metadata errors separately.
The motivation behind this change is that broken (or outdated) debug info
can be "recovered" from by stripping the debug info.

The problem I'm trying to solve with this sequence of patches is that
historically we've done a really bad job at verifying debug info.
We want to be able to make the verifier stricter without having to worry
about breaking bitcode compatibility with existing producers. For example,
we don't necessarily want IR produced by an older version of clang to be
rejected by an LTO link just because of malformed debug info, and rather
provide an option to strip it. Note that merely outdated (but well-formed)
debug info would continue to be auto-upgraded in this scenario.

http://reviews.llvm.org/D19986
rdar://problem/25818489

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268778 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[Hexagon] Be careful about anti-dependencies with a call in packetizer
Krzysztof Parzyszek [Fri, 6 May 2016 19:13:38 +0000 (19:13 +0000)]
[Hexagon] Be careful about anti-dependencies with a call in packetizer

In a case like
  J2_callr <ga:@foo>, %R0<imp-use>, ...
  R0<def> = ...
the anti-dependency on R0 cannot be ignored and the two instructions
cannot be packetized together, since if they were, the assignment to
R0 would take place before the call.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268776 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[GVN] Handle unordered atomics in cross block FRE
Philip Reames [Fri, 6 May 2016 18:46:45 +0000 (18:46 +0000)]
[GVN] Handle unordered atomics in cross block FRE

You'll note there are essentially no code changes here.  Cross block FRE heavily reuses code from the block local FRE.  All of the tricky parts were done as part of the previous patch and the refactoring that removed the original code duplication.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268775 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoSDAG: Don't leave dangling dead nodes after SelectCodeCommon
Justin Bogner [Fri, 6 May 2016 18:42:16 +0000 (18:42 +0000)]
SDAG: Don't leave dangling dead nodes after SelectCodeCommon

Relying on the caller to clean up after we've replaced all uses of a
node won't work when we've migrated to the `void Select(...)` API.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268774 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoThe associated PR for this test was PR27135 not PR27132.
Eric Christopher [Fri, 6 May 2016 18:23:14 +0000 (18:23 +0000)]
The associated PR for this test was PR27135 not PR27132.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268772 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Get rid of X32_NOREX_ADDR_ACCESS register class.
Quentin Colombet [Fri, 6 May 2016 18:22:48 +0000 (18:22 +0000)]
[X86] Get rid of X32_NOREX_ADDR_ACCESS register class.

According to H.J. Lu <hjl.tools@gmail.com>, this register class is never
used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268771 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[GVN] Do local FRE for unordered atomic loads
Philip Reames [Fri, 6 May 2016 18:17:13 +0000 (18:17 +0000)]
[GVN] Do local FRE for unordered atomic loads

This patch is the first in a small series teaching GVN to optimize unordered loads aggressively. This change just handles block local FRE because that's the simplest thing which lets me test MDA, and the AvailableValue pieces. Somewhat suprisingly, MDA appears fine and only a couple of small changes are needed in GVN.

Once this is in, I'll tackle non-local FRE and PRE. The former looks like a natural extension of this, the later will require a couple of minor changes.

Differential Revision: http://reviews.llvm.org/D19440

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268770 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoTweak the ThinLTO pass pipeline
Mehdi Amini [Fri, 6 May 2016 18:17:03 +0000 (18:17 +0000)]
Tweak the ThinLTO pass pipeline

Summary:
The original ThinLTO pipeline was derived from some
work I did tuning FullLTO on the test suite and SPEC. This
patch reduces the amount of work done in the "linker phase" of
the build, and extend the function simplifications passes
performed during the "compile phase". This helps the build time
by reducing the IR as much as possible during the compile phase
and limiting the work to be performed during the "link phase",
while keeping the performance "on par" with the existing pipeline.

Reviewers: tejohnson

Subscribers: llvm-commits, joker.eph

Differential Revision: http://reviews.llvm.org/D19773

From: Mehdi Amini <mehdi.amini@apple.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268769 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[SimplifyCFG] propagate branch metadata when creating select (retry r268550 / r268751...
Sanjay Patel [Fri, 6 May 2016 18:07:46 +0000 (18:07 +0000)]
[SimplifyCFG] propagate branch metadata when creating select (retry r268550 / r268751 with possible fix)

Retrying r268550/r268751 which were reverted at r268577/r268765 due a memory sanitizer failure.
I have not been able to reproduce that failure, but I've taken another guess at fixing
the problem in this version of the patch and will watch for another failure.

Original commit message:
Unlike earlier similar fixes, we need to recalculate the branch weights
in this case.

Differential Revision: http://reviews.llvm.org/D19674

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268767 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agorevert r268751 - caused same failures on msan bot
Sanjay Patel [Fri, 6 May 2016 17:51:37 +0000 (17:51 +0000)]
revert r268751 - caused same failures on msan bot

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268765 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[CostModel][X86] Tweak 'SSE2-only' test CPU as it was only disabling SSE41 not SSE3...
Simon Pilgrim [Fri, 6 May 2016 17:50:07 +0000 (17:50 +0000)]
[CostModel][X86] Tweak 'SSE2-only' test CPU as it was only disabling SSE41 not SSE3/SSSE3 etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268763 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[AMDGPU][llvm-mc] Add support for sendmsg(...) syntax.
Artem Tamazov [Fri, 6 May 2016 17:48:48 +0000 (17:48 +0000)]
[AMDGPU][llvm-mc] Add support for sendmsg(...) syntax.

Added support for sendmsg(MSG[, OP[, STREAM_ID]]) syntax
in s_sendmsg and s_sendmsghalt instructions.
The syntax matches the SP3 assembler/disassembler rules.
That is why implicit inputs (like M0 and EXEC) are not printed
to disassembly output anymore.

sendmsg(...) allows only known message types and attributes,
even if literals are used instead of symbolic names.
However, raw literal (without "sendmsg") still can be used,
and that allows for any 16-bit value.

Tests updated/added.

Differential Revision: http://reviews.llvm.org/D19596

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268762 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[CostModel][X86] Added ctlz/cttz undef-zero costmodel tests
Simon Pilgrim [Fri, 6 May 2016 17:48:35 +0000 (17:48 +0000)]
[CostModel][X86] Added ctlz/cttz undef-zero costmodel tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268761 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Teach X86FixupBWInsts to promote MOV8rr/MOV16rr to MOV32rr.
Ahmed Bougacha [Fri, 6 May 2016 17:42:57 +0000 (17:42 +0000)]
[X86] Teach X86FixupBWInsts to promote MOV8rr/MOV16rr to MOV32rr.

Codesize is less (16) or equal (8), and we avoid partial dependencies.

Differential Revision: http://reviews.llvm.org/D19999

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268760 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Remove \brief in FixupBW. NFC.
Ahmed Bougacha [Fri, 6 May 2016 17:28:47 +0000 (17:28 +0000)]
[X86] Remove \brief in FixupBW. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268754 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Simplify FixupBW sub_8bit_hi-related logic. NFC.
Ahmed Bougacha [Fri, 6 May 2016 17:28:42 +0000 (17:28 +0000)]
[X86] Simplify FixupBW sub_8bit_hi-related logic. NFC.

Instead of passing around sizes and asking for subregs, we can check
the subreg indices we care about: sub_8bit_hi and sub_8bit.

Differential Revision: http://reviews.llvm.org/D20006

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268753 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[AArch64] Fix test to specify triple and disable post-RA scheduling.
Geoff Berry [Fri, 6 May 2016 17:12:38 +0000 (17:12 +0000)]
[AArch64] Fix test to specify triple and disable post-RA scheduling.

This should fix bot breakage caused by r268746:
[AArch64] Combine callee-save and local stack SP adjustment instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268752 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[SimplifyCFG] propagate branch metadata when creating select (retry r268550 with...
Sanjay Patel [Fri, 6 May 2016 17:07:47 +0000 (17:07 +0000)]
[SimplifyCFG] propagate branch metadata when creating select (retry r268550 with possible fix)

Retrying r268550 which was reverted at r268577 due a memory sanitizer failure.
I have not been able to reproduce that failure, but I've taken a guess at fixing
the problem in this version of the patch and will watch for another failure.

Original commit message:
Unlike earlier similar fixes, we need to recalculate the branch weights
in this case.

Differential Revision: http://reviews.llvm.org/D19674

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268751 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[AArch64] Combine callee-save and local stack SP adjustment instructions.
Geoff Berry [Fri, 6 May 2016 16:34:59 +0000 (16:34 +0000)]
[AArch64] Combine callee-save and local stack SP adjustment instructions.

Summary:
If a function needs to allocate both callee-save stack memory and local
stack memory, we currently decrement/increment the SP in two steps:
first for the callee-save area, and then for the local stack area.  This
changes the code to allocate them both at once at the very beginning/end
of the function.  This has two benefits:

1) there is one fewer sub/add micro-op in the prologue/epilogue

2) the stack adjustment instructions act as a scheduling barrier, so
moving them to the very beginning/end of the function increases post-RA
scheduler's ability to move instructions (that only depend on argument
registers) before any of the callee-save stores

This change can cause an increase in instructions if the original local
stack SP decrement could be folded into the first store to the stack.
This occurs when the first local stack store is to stack offset 0.  In
this case we are trading off one more sub instruction for one fewer sub
micro-op (along with benefits (2) and (3) above).

Reviewers: t.p.northover

Subscribers: aemerson, rengolin, mcrosier, llvm-commits

Differential Revision: http://reviews.llvm.org/D18619

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268746 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[AArch64] Decouple zero store promotion from narrow ld merge. NFC.
Jun Bum Lim [Fri, 6 May 2016 15:08:57 +0000 (15:08 +0000)]
[AArch64] Decouple zero store promotion from narrow ld merge. NFC.

Summary: This change refactors to decouple the zero store promotion from the narrow ld merge and add a flag (enable-narrow-ld-merge=true) to control the narrow ld merge optimization.

Reviewers: jmolloy, t.p.northover, mcrosier

Subscribers: aemerson, rengolin, mcrosier, llvm-commits

Differential Revision: http://reviews.llvm.org/D19885

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268744 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRevert "AMDGPU/SI: Add amdgpu_kernel calling convention. Part 2."
Nikolay Haustov [Fri, 6 May 2016 14:59:04 +0000 (14:59 +0000)]
Revert "AMDGPU/SI: Add amdgpu_kernel calling convention. Part 2."

This reverts commit 47486d52454d60cdf6becc0b2efe533c73794380.

It broke calling OpenCL kernel from another kernel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268739 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[CostModel][X86] Added costmodel tests for vector ctpop/ctlz/cttz/bitreverse/bswap
Simon Pilgrim [Fri, 6 May 2016 14:38:14 +0000 (14:38 +0000)]
[CostModel][X86] Added costmodel tests for vector ctpop/ctlz/cttz/bitreverse/bswap

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268738 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips] Fix inconsistent .cprestore behaviour between direct object emission and assem...
Daniel Sanders [Fri, 6 May 2016 14:37:24 +0000 (14:37 +0000)]
[mips] Fix inconsistent .cprestore behaviour between direct object emission and assembling.

Summary:
Direct object emission has an initialization order problem where an
InitMCObjectFile is called after MipsTargetELFStreamer determines whether
PIC is enabled by default or not. There doesn't seem to be point that
initializes all cases so split the responsibility between
MipsTargetELFStreamer and MipsAsmPrinter.

Reviewers: sdardis

Subscribers: dsanders, llvm-commits, sdardis

Differential Revision: http://reviews.llvm.org/D19728

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268737 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[SimplifyCFG] Prefer a simplification based on a dominating condition.
Chad Rosier [Fri, 6 May 2016 14:25:14 +0000 (14:25 +0000)]
[SimplifyCFG] Prefer a simplification based on a dominating condition.

Rather than merge two branches with a common destination.
Differential Revision: http://reviews.llvm.org/D19743

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268735 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips] Correct the ordering of HI/LO pairs in the relocation table.
Daniel Sanders [Fri, 6 May 2016 13:49:25 +0000 (13:49 +0000)]
[mips] Correct the ordering of HI/LO pairs in the relocation table.

Summary:
There seems to have been a misunderstanding as to the meaning of 'offset' in
the rules laid down by our ABI. The previous code believed that 'offset' meant
the offset within the section that the relocation is applied to. However, it
should have meant the offset from the symbol used in the relocation expression.

This patch adds two fields to ELFRelocationEntry and uses them to correct the
order of relocations for MIPS. These fields contain:
* The original symbol before shouldRelocateWithSymbol() is considered. This
  ensures that R_MIPS_GOT16 is able to correctly distinguish between local and
  external symbols, allowing us to tell whether %got() requires a matching
  %lo() or not (local symbols require one, external symbols don't). It also
  prevents confusing cases where the fuzzy matching rules cause things like
  %hi(foo)/%lo(foo+3) and %hi(bar)/%lo(bar+1) to swap their %lo()'s.
* The original offset before shouldRelocateWithSymbol() is considered. The
  existing Addend field is always zero when the object uses in place addends
  (because it's already moved it to the encoding) but MIPS needs to use the
  original offset to ensure that the linker correctly calculates the carry-in
  bit for %hi() and %got().

IAS ensures that unmatchable %hi()/%got() relocations are placed at the end of
the table to ensure that the linker rejects the table (we're unable to report
such errors directly). The alternatives to this risk accidental matching
against inappropriate relocations which may silently compute incorrect values
due to an incorrect carry bit between the %lo() and %hi()/%got().

Reviewers: sdardis

Subscribers: dsanders, sdardis, rafael, llvm-commits

Differential Revision: http://reviews.llvm.org/D19718

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268733 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips][mips16] Use isUnconditionalBranch() in AnalyzeBranch() and constant island...
Daniel Sanders [Fri, 6 May 2016 13:23:51 +0000 (13:23 +0000)]
[mips][mips16] Use isUnconditionalBranch() in AnalyzeBranch() and constant island pass.

Summary:
This stops it misidentifying unconditional branches as conditional branches
which fixes a -verify-machineinstrs error about exiting a function via fall through.

Reviewers: sdardis

Subscribers: dsanders, sdardis, llvm-commits

Differential Revision: http://reviews.llvm.org/D19864

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268731 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips][fastisel] Conditional moves do not have implicit operands.
Daniel Sanders [Fri, 6 May 2016 12:57:26 +0000 (12:57 +0000)]
[mips][fastisel] Conditional moves do not have implicit operands.

Reviewers: sdardis

Subscribers: dsanders, sdardis, llvm-commits

Differential Revision: http://reviews.llvm.org/D19862

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268730 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[TableGen] AsmMatcher: support for default values for optional operands
Sam Kolton [Fri, 6 May 2016 11:31:17 +0000 (11:31 +0000)]
[TableGen] AsmMatcher: support for default values for optional operands

Summary:
This change allows to specify "DefaultMethod" for optional operand (IsOptional = 1) in AsmOperandClass that return default value for operand. This is used in convertToMCInst to set default values in MCInst.
Previously if you wanted to set default value for operand you had to create custom converter method. With this change it is possible to use standard converters even when optional operands presented.

Reviewers: tstellarAMD, ab, craig.topper

Subscribers: jyknight, dsanders, arsenm, nhaustov, llvm-commits

Differential Revision: http://reviews.llvm.org/D18242

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268726 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix whitespace and line wrapping. NFC.
Ryan Govostes [Fri, 6 May 2016 11:22:11 +0000 (11:22 +0000)]
Fix whitespace and line wrapping. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268725 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[asan] add option to set shadow mapping offset
Ryan Govostes [Fri, 6 May 2016 10:25:22 +0000 (10:25 +0000)]
[asan] add option to set shadow mapping offset

Allowing overriding the default ASAN shadow mapping offset with the
-asan-shadow-offset option, and allow zero to be specified for both offset and
scale.

Patch by Aaron Carroll <aaronc@apple.com>.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268724 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[AVR] Add a majority of the backend code
Dylan McKay [Fri, 6 May 2016 10:12:31 +0000 (10:12 +0000)]
[AVR] Add a majority of the backend code

Summary: This adds the majority of the AVR backend.

Reviewers: hfinkel, dsanders, vkalintiris, arsenm

Subscribers: dylanmckay

Differential Revision: http://reviews.llvm.org/D17906

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268722 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAttempt to fix the modules builder by declaring SCEV in LoopUtils.h
Silviu Baranga [Fri, 6 May 2016 09:37:14 +0000 (09:37 +0000)]
Attempt to fix the modules builder by declaring SCEV in LoopUtils.h

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268720 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago AMDGPU/SI: Add amdgpu_kernel calling convention. Part 2.
Nikolay Haustov [Fri, 6 May 2016 09:23:13 +0000 (09:23 +0000)]
AMDGPU/SI: Add amdgpu_kernel calling convention. Part 2.

    Summary:
    Check calling convention in AMDGPUMachineFunction::isKernel

    This will be used for AMDGPU_HSA_KERNEL symbol type in output ELF.

    Also, in the future unused non-kernels may be optimized.

    Reviewers: tstellarAMD, arsenm

    Subscribers: arsenm, joker.eph, llvm-commits

    Differential Revision: http://reviews.llvm.org/D19917

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268719 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU/SI: Add amdgpu_kernel calling convention. Part 1.
Nikolay Haustov [Fri, 6 May 2016 09:07:29 +0000 (09:07 +0000)]
AMDGPU/SI: Add amdgpu_kernel calling convention. Part 1.

Summary:
This will be used for AMDGPU_HSA_KERNEL symbol type in output ELF.

Also, in the future unused non-kernels may be optimized.

For now, also accept SPIR_KERNEL for HCC frontend.

Also, add bitcode compatibility tests for missing calling conventions
except AVR_BUILTIN which doesn't have parse code.

Reviewers: tstellarAMD, arsenm

Subscribers: arsenm, joker.eph, llvm-commits

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268717 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoThinLTO: fix assertion and refactor check for hidden use from inline ASM in a helper...
Mehdi Amini [Fri, 6 May 2016 08:25:33 +0000 (08:25 +0000)]
ThinLTO: fix assertion and refactor check for hidden use from inline ASM in a helper function

This test was crashing, and currently it breaks bootstrapping clang with debuginfo

Differential Revision: http://reviews.llvm.org/D20008

From: Mehdi Amini <mehdi.amini@apple.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268715 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips][microMIPS] Add CodeGen support for MUL* and DMUL* instructions
Zlatko Buljan [Fri, 6 May 2016 08:24:14 +0000 (08:24 +0000)]
[mips][microMIPS] Add CodeGen support for MUL* and DMUL* instructions
Differential Revision: http://reviews.llvm.org/D15744

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268714 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[TableGen] Fix a memory leak when creating SwitchOpcodeMatchers.
Craig Topper [Fri, 6 May 2016 06:56:14 +0000 (06:56 +0000)]
[TableGen] Fix a memory leak when creating SwitchOpcodeMatchers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268712 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[TableGen] Remove isSafeToReorderWithPatternPredicate from DAGISelMatchers as its...
Craig Topper [Fri, 6 May 2016 06:21:27 +0000 (06:21 +0000)]
[TableGen] Remove isSafeToReorderWithPatternPredicate from DAGISelMatchers as its not used anymore.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268711 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[PM] port IR based PGO prof-gen pass to new pass manager
Xinliang David Li [Fri, 6 May 2016 05:49:19 +0000 (05:49 +0000)]
[PM] port IR based PGO prof-gen pass to new pass manager

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268710 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoBitcodeWriter: Simplify. NFC.
Peter Collingbourne [Fri, 6 May 2016 02:41:23 +0000 (02:41 +0000)]
BitcodeWriter: Simplify. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268707 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[TableGen] Remove getHash support from DAGISelMatcher. It hasn't been used for some...
Craig Topper [Fri, 6 May 2016 02:37:59 +0000 (02:37 +0000)]
[TableGen] Remove getHash support from DAGISelMatcher. It hasn't been used for some time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268706 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[TableGen] Remove SinkPatternPredicates from the DAG isel matcher optimizer.
Craig Topper [Fri, 6 May 2016 02:37:56 +0000 (02:37 +0000)]
[TableGen] Remove SinkPatternPredicates from the DAG isel matcher optimizer.

Pattern predicates already appear to be emitted as far down as they can be. The optimization was making no changes on any in-tree target.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268705 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[profile] Remove another unneeded field in raw profile reader
Xinliang David Li [Fri, 6 May 2016 02:13:00 +0000 (02:13 +0000)]
[profile] Remove another unneeded field in raw profile reader

DataValueSize is now removed. The change is consolidated
with previous raw version bump.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268703 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agodocument version change
Xinliang David Li [Fri, 6 May 2016 01:41:24 +0000 (01:41 +0000)]
document version change

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268702 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[EarlyCSE] Rename a variable for clarity [NFC]
Philip Reames [Fri, 6 May 2016 01:13:58 +0000 (01:13 +0000)]
[EarlyCSE] Rename a variable for clarity [NFC]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268701 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[CodeGen] Round [SU]INT_TO_FP result when promoting from f16.
Ahmed Bougacha [Fri, 6 May 2016 00:58:00 +0000 (00:58 +0000)]
[CodeGen] Round [SU]INT_TO_FP result when promoting from f16.

If we don't, values that aren't precisely representable in f16 could
be used as-is in a promoted f32 operation, which would produce
incorrect results.

AArch64 had the correct behavior; add a focused test.

Fixes http://llvm.org/PR26871

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268700 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoMake StringTableBuilder to cache hash values.
Rui Ueyama [Fri, 6 May 2016 00:51:58 +0000 (00:51 +0000)]
Make StringTableBuilder to cache hash values.

This change seems to speed up LLD a bit if it has a lot of mergeable
sections. The number is below. It's not too bad for a small patch.

Time to link Clang (debug build):

w/o patch 6.3696 seconds
w/patch   6.2746 seconds (-1.5%)

Differential Revision: http://reviews.llvm.org/D19933

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268698 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoObject: Fix two -Wpessimizing-move warnings after r268694
Justin Bogner [Thu, 5 May 2016 23:59:57 +0000 (23:59 +0000)]
Object: Fix two -Wpessimizing-move warnings after r268694

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268697 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix window bots failures due to r268694 - Cleanup and refactor of malformedError...
Kevin Enderby [Thu, 5 May 2016 23:55:27 +0000 (23:55 +0000)]
Fix window bots failures due to r268694 - Cleanup and refactor of malformedError() in lib/Object/MachOObjectFile.cpp .

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268696 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[LAA] Fix confusing debug message
Adam Nemet [Thu, 5 May 2016 23:41:28 +0000 (23:41 +0000)]
[LAA] Fix confusing debug message

This message used to be correct, when all we cared about was whether the
dependence was safe (i.e. NoDep) or unsafe.  With the current more
precise characterization, this is a forward dep.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268695 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoCleanup and refactor of malformedError() in lib/Object/MachOObjectFile.cpp .
Kevin Enderby [Thu, 5 May 2016 23:41:05 +0000 (23:41 +0000)]
Cleanup and refactor of malformedError() in lib/Object/MachOObjectFile.cpp .

No functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268694 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoSDAG: Rename Select->SelectImpl and repurpose Select as returning void
Justin Bogner [Thu, 5 May 2016 23:19:08 +0000 (23:19 +0000)]
SDAG: Rename Select->SelectImpl and repurpose Select as returning void

This is a step towards removing the rampant undefined behaviour in
SelectionDAG, which is a part of llvm.org/PR26808.

We rename SelectionDAGISel::Select to SelectImpl and update targets to
match, and then change Select to return void and consolidate the
sketchy behaviour we're trying to get away from there.

Next, we'll update backends to implement `void Select(...)` instead of
SelectImpl and eventually drop the base Select implementation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268693 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoSDAG: Remove OPC_MarkGlueResults and associated logic. NFC
Justin Bogner [Thu, 5 May 2016 22:37:45 +0000 (22:37 +0000)]
SDAG: Remove OPC_MarkGlueResults and associated logic. NFC

This opcode never happens in practice, and yet the logic we have in
place to handle it would be undefined behaviour if we ever executed
it. Remove it rather than trying to refactor code that's never
reached.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268692 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[scan-build] fix warnings emitted on LLVM Hexagon code base
Krzysztof Parzyszek [Thu, 5 May 2016 22:00:44 +0000 (22:00 +0000)]
[scan-build] fix warnings emitted on LLVM Hexagon code base

Patch by Apelete Seketeli.

Differential Revision: http://reviews.llvm.org/D19968

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268691 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[Hexagon] Fix the offset ranges for vector memory instructions
Krzysztof Parzyszek [Thu, 5 May 2016 21:58:02 +0000 (21:58 +0000)]
[Hexagon] Fix the offset ranges for vector memory instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268690 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix some Clang-tidy readability-simplify-boolean-expr and Include What You Use warnings.
Eugene Zelenko [Thu, 5 May 2016 21:35:47 +0000 (21:35 +0000)]
Fix some Clang-tidy readability-simplify-boolean-expr and Include What You Use warnings.

Differential revision: reviews.llvm.org/D19946

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268689 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[PM] port Branch Frequency Analaysis pass to new PM
Xinliang David Li [Thu, 5 May 2016 21:13:27 +0000 (21:13 +0000)]
[PM] port Branch Frequency Analaysis pass to new PM

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268687 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[PM] Port Interprocedural SCCP to the new pass manager.
Davide Italiano [Thu, 5 May 2016 21:05:36 +0000 (21:05 +0000)]
[PM] Port Interprocedural SCCP to the new pass manager.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268684 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[codeview] Improve some comments
Reid Kleckner [Thu, 5 May 2016 20:58:46 +0000 (20:58 +0000)]
[codeview] Improve some comments

This FIXME was already fixed, and these LF_* enum names were
inconsistent.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268683 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[AArch64] Remove unused MBP headers/dependency. NFC.
Chad Rosier [Thu, 5 May 2016 20:58:38 +0000 (20:58 +0000)]
[AArch64] Remove unused MBP headers/dependency. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268682 91177308-0d34-0410-b5e6-96231b3b80d8