oota-llvm.git
9 years agoSparc: Remove implicit ilist iterator conversions, NFC
Duncan P. N. Exon Smith [Tue, 20 Oct 2015 00:59:43 +0000 (00:59 +0000)]
Sparc: Remove implicit ilist iterator conversions, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250781 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoNVPTX: Remove implicit ilist iterator conversions, NFC
Duncan P. N. Exon Smith [Tue, 20 Oct 2015 00:54:09 +0000 (00:54 +0000)]
NVPTX: Remove implicit ilist iterator conversions, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250779 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoHexagon: Remove implicit ilist iterator conversions, NFC
Duncan P. N. Exon Smith [Tue, 20 Oct 2015 00:46:39 +0000 (00:46 +0000)]
Hexagon: Remove implicit ilist iterator conversions, NFC

There are two things out of the ordinary in this commit.  First, I made
a loop obviously "infinite" in HexagonInstrInfo.cpp.  After checking if
an instruction was at the beginning of a basic block (in which case,
`break`), the loop decremented and checked the iterator for `nullptr` as
the loop condition.  This has never been possible (the prev pointers are
always been circular, so even with the weird ilist/iplist
implementation, this isn't been possible), so I removed the condition.

Second, in HexagonAsmPrinter.cpp there was another case of comparing a
`MachineBasicBlock::instr_iterator` against `MachineBasicBlock::end()`
(which returns `MachineBasicBlock::iterator`).  While not incorrect,
it's fragile.  I switched this to `::instr_end()`.

All that said, no functionality change intended here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250778 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoWebAssembly: fix syntax for br_if.
JF Bastien [Tue, 20 Oct 2015 00:37:42 +0000 (00:37 +0000)]
WebAssembly: fix syntax for br_if.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250777 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAsmPrinter: Remove implicit ilist iterator conversion, NFC
Duncan P. N. Exon Smith [Tue, 20 Oct 2015 00:36:08 +0000 (00:36 +0000)]
AsmPrinter: Remove implicit ilist iterator conversion, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250776 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMips: Remove implicit ilist iterator conversions, NFC
Duncan P. N. Exon Smith [Tue, 20 Oct 2015 00:15:20 +0000 (00:15 +0000)]
Mips: Remove implicit ilist iterator conversions, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250769 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCppBackend: Remove implicit ilist iterator conversions, NFC
Duncan P. N. Exon Smith [Tue, 20 Oct 2015 00:06:41 +0000 (00:06 +0000)]
CppBackend: Remove implicit ilist iterator conversions, NFC

Mostly just converted to range-based for loops.  May have converted a
couple of extra loops as a drive-by (not sure).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250766 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoBPF: Remove implicit ilist iterator conversion, NFC
Duncan P. N. Exon Smith [Tue, 20 Oct 2015 00:02:50 +0000 (00:02 +0000)]
BPF: Remove implicit ilist iterator conversion, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250765 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoARM: Remove implicit ilist iterator conversions, NFC
Duncan P. N. Exon Smith [Mon, 19 Oct 2015 23:25:57 +0000 (23:25 +0000)]
ARM: Remove implicit ilist iterator conversions, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250759 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Orc] Fix MSVC bugs introduced in r250749.
Lang Hames [Mon, 19 Oct 2015 23:23:17 +0000 (23:23 +0000)]
[Orc] Fix MSVC bugs introduced in r250749.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250758 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoObjCARC: Remove implicit ilist iterator conversions, NFC
Duncan P. N. Exon Smith [Mon, 19 Oct 2015 23:20:14 +0000 (23:20 +0000)]
ObjCARC: Remove implicit ilist iterator conversions, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250756 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoEnhance loop rotation with existence of profile data in MachineBlockPlacement pass.
Cong Hou [Mon, 19 Oct 2015 23:16:40 +0000 (23:16 +0000)]
Enhance loop rotation with existence of profile data in MachineBlockPlacement pass.

Currently, in MachineBlockPlacement pass the loop is rotated to let the best exit to be the last BB in the loop chain, to maximize the fall-through from the loop to outside. With profile data, we can determine the cost in terms of missed fall through opportunities when rotating a loop chain and select the best rotation. Basically, there are three kinds of cost to consider for each rotation:

1. The possibly missed fall through edge (if it exists) from BB out of the loop to the loop header.
2. The possibly missed fall through edges (if they exist) from the loop exits to BB out of the loop.
3. The missed fall through edge (if it exists) from the last BB to the first BB in the loop chain.

Therefore, the cost for a given rotation is the sum of costs listed above. We select the best rotation with the smallest cost. This is only for PGO mode when we have more precise edge frequencies.

Differential revision: http://reviews.llvm.org/D10717

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250754 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Orc] Use '= default' for move constructor/assignment as per dblaikie's review.
Lang Hames [Mon, 19 Oct 2015 22:49:18 +0000 (22:49 +0000)]
[Orc] Use '= default' for move constructor/assignment as per dblaikie's review.

Thanks Dave!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250749 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoLinker: Remove implicit ilist iterator conversion, NFC
Duncan P. N. Exon Smith [Mon, 19 Oct 2015 22:23:36 +0000 (22:23 +0000)]
Linker: Remove implicit ilist iterator conversion, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250748 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix -Wdeprecated regarding ORC copying ValueMaterializers
David Blaikie [Mon, 19 Oct 2015 22:15:55 +0000 (22:15 +0000)]
Fix -Wdeprecated regarding ORC copying ValueMaterializers

As usual, this is a polymorphic hierarchy without polymorphic ownership,
so simply make the dtor protected non-virtual, protected default copy
ctor/assign, and make derived classes final. The derived classes will
pick up correct default public copy ops (and dtor) implicitly.

(wish I could add -Wdeprecated to the build, but last time I tried it
triggered on some system headers I still need to look into/figure out)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250747 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[InstCombine] Optimize icmp of inc/dec at RHS
Michael Liao [Mon, 19 Oct 2015 22:08:14 +0000 (22:08 +0000)]
[InstCombine] Optimize icmp of inc/dec at RHS

Allow LLVM to optimize the sequence like the following:

  %inc = add nsw i32 %i, 1
  %cmp = icmp slt %n, %inc

into:

  %cmp = icmp sle i32 %n, %i

The case is not handled previously due to the complexity of compuation of %n.
Hence, LLVM cannot swap operands of icmp accordingly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250746 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoVectorize: Remove implicit ilist iterator conversions, NFC
Duncan P. N. Exon Smith [Mon, 19 Oct 2015 22:06:09 +0000 (22:06 +0000)]
Vectorize: Remove implicit ilist iterator conversions, NFC

Besides the usual, I finally added an overload to
`BasicBlock::splitBasicBlock()` that accepts an `Instruction*` instead
of `BasicBlock::iterator`.  Someone can go back and remove this overload
later (after updating the callers I'm going to skip going forward), but
the most common call seems to be
`BB->splitBasicBlock(BB->getTerminator(), ...)` and I'm not sure it's
better to add `->getIterator()` to every one than have the overload.
It's pretty hard to get the usage wrong.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250745 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[CGP] transform select instructions into branches and sink expensive operands
Sanjay Patel [Mon, 19 Oct 2015 21:59:12 +0000 (21:59 +0000)]
[CGP] transform select instructions into branches and sink expensive operands

This was originally checked in at r250527, but reverted at r250570 because of PR25222.
There were at least 2 problems:
1. The cost check was checking for an instruction with an exact cost of TCC_Expensive;
that should have been >=.
2. The cause of the clang stage 1 failures was illegally sinking 'call' instructions;
we can't sink instructions that may have side effects / are not safe to execute speculatively.

Fixed those conditions in sinkSelectOperand() and added test cases.

Original commit message:
This is a follow-up to the discussion in D12882.

Ideally, we would like SimplifyCFG to be able to form select instructions even when the operands
are expensive (as defined by the TTI cost model) because that may expose further optimizations.
However, we would then like a later pass like CodeGenPrepare to undo that transformation if the
target would likely benefit from not speculatively executing an expensive op (this patch).

Once we have this safety mechanism in place, we can adjust SimplifyCFG to restore its
select-formation behavior that changed with r248439.

Differential Revision: http://reviews.llvm.org/D13297

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250743 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoX86: Remove implicit ilist iterator conversions, NFC
Duncan P. N. Exon Smith [Mon, 19 Oct 2015 21:48:29 +0000 (21:48 +0000)]
X86: Remove implicit ilist iterator conversions, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250741 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[RuntimeDyld][COFF] Fix some endianness issues, re-enable the regression test.
Lang Hames [Mon, 19 Oct 2015 20:37:52 +0000 (20:37 +0000)]
[RuntimeDyld][COFF] Fix some endianness issues, re-enable the regression test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250733 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRestore the original behavior of SelectionDAG::getTargetIndex().
Owen Anderson [Mon, 19 Oct 2015 19:27:40 +0000 (19:27 +0000)]
Restore the original behavior of SelectionDAG::getTargetIndex().

It looks like an extra negation snuck in as apart of restoring it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250726 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Remove unnecessary argument sign extends
Krzysztof Parzyszek [Mon, 19 Oct 2015 19:10:48 +0000 (19:10 +0000)]
[Hexagon] Remove unnecessary argument sign extends

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250724 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoPass FunctionInfoIndex by reference to WriteFunctionSummaryToFile (NFC)
Teresa Johnson [Mon, 19 Oct 2015 19:06:06 +0000 (19:06 +0000)]
Pass FunctionInfoIndex by reference to WriteFunctionSummaryToFile (NFC)

Implemented suggestion by dblakie in review for r250704.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250723 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Orc] Add explicit move constructor and assignment operator to make MSVC happy.
Lang Hames [Mon, 19 Oct 2015 18:59:22 +0000 (18:59 +0000)]
[Orc] Add explicit move constructor and assignment operator to make MSVC happy.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250722 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd missing override noticed by Clang's -Winconsistent-missing-override.
Benjamin Kramer [Mon, 19 Oct 2015 18:41:23 +0000 (18:41 +0000)]
Add missing override noticed by Clang's -Winconsistent-missing-override.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250720 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AArch64]Merge halfword loads into a 32-bit load
Jun Bum Lim [Mon, 19 Oct 2015 18:34:53 +0000 (18:34 +0000)]
[AArch64]Merge halfword loads into a 32-bit load

Convert two halfword loads into a single 32-bit word load with bitfield extract
instructions. For example :
  ldrh w0, [x2]
  ldrh w1, [x2, #2]
becomes
  ldr w0, [x2]
  ubfx w1, w0, #16, #16
  and  w0, w0, #ffff

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250719 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Fix debug information for local objects
Krzysztof Parzyszek [Mon, 19 Oct 2015 18:30:27 +0000 (18:30 +0000)]
[Hexagon] Fix debug information for local objects

- Isolate the check for the existence of a stack frame into hasFP.
- Implement getFrameIndexReference for DWARF address computation.
- Use getFrameIndexReference for offset computation in eliminateFrameIndex.
- Preserve debug information for dynamically allocated stack objects.
- Prefer FP to access local objects at -O0.
- Add experimental code to skip allocframe when not strictly necessary
  (disabled by default).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250718 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoPut back SelectionDAG::getTargetIndex.
Benjamin Kramer [Mon, 19 Oct 2015 18:26:16 +0000 (18:26 +0000)]
Put back SelectionDAG::getTargetIndex.

While technically this is untested dead code, it has out-of-tree users.
This reverts a part of r250434.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250717 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Orc] Lambda needs to capture 'this'.
Lang Hames [Mon, 19 Oct 2015 17:53:43 +0000 (17:53 +0000)]
[Orc] Lambda needs to capture 'this'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250716 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Orc] Remove extraneous semicolon that found its way into r250712.
Lang Hames [Mon, 19 Oct 2015 17:49:37 +0000 (17:49 +0000)]
[Orc] Remove extraneous semicolon that found its way into r250712.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250715 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Delay emission of CFI instructions
Krzysztof Parzyszek [Mon, 19 Oct 2015 17:46:01 +0000 (17:46 +0000)]
[Hexagon] Delay emission of CFI instructions

Emit the CFI instructions after all code transformation have been done.
This will avoid any interference between CFI instructions and packetization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250714 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert "RegisterPressure: allocatable physreg uses are always kills"
Matthias Braun [Mon, 19 Oct 2015 17:44:22 +0000 (17:44 +0000)]
Revert "RegisterPressure: allocatable physreg uses are always kills"

This reverts commit r250596.

Reverted for now as the commit triggers assert in the AMDGPU target
pending investigation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250713 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Orc] Add support for emitting indirect stubs directly into the JIT target's
Lang Hames [Mon, 19 Oct 2015 17:43:51 +0000 (17:43 +0000)]
[Orc] Add support for emitting indirect stubs directly into the JIT  target's
memory, rather than representing the stubs in IR. Update the CompileOnDemand
layer to use this functionality.

Directly emitting stubs is much cheaper than building them in IR and codegen'ing
them (see below). It also plays well with remote JITing - stubs can be emitted
directly in the target process, rather than having to send them over the wire.

The downsides are:

(1) Care must be taken when resolving symbols, as stub symbols are held in a
    separate symbol table. This is only a problem for layer writers and other
    people using this API directly. The CompileOnDemand layer hides this detail.

(2) Aliases of function stubs can't be symbolic any more (since there's no
    symbol definition in IR), but must be converted into a constant pointer
    expression. This means that modules containing aliases of stubs cannot be
    cached. In practice this is unlikely to be a problem: There's no benefit to
    caching such a module anyway.

On balance I think the extra performance is more than worth the trade-offs: In a
simple stress test with 10000 dummy functions requiring stubs and a single
executed "hello world" main function, directly emitting stubs reduced user time
for JITing / executing by over 90% (1.5s for IR stubs vs 0.1s for direct
emission).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250712 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoConvert gold-plugin unnecessary unique_ptr into local (NFC)
Teresa Johnson [Mon, 19 Oct 2015 15:23:03 +0000 (15:23 +0000)]
Convert gold-plugin unnecessary unique_ptr into local (NFC)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250704 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix required library for r250699 to BitWriter instead of BitReader.
Teresa Johnson [Mon, 19 Oct 2015 15:21:46 +0000 (15:21 +0000)]
Fix required library for r250699 to BitWriter instead of BitReader.

This should fix the mingw3 bot failure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250703 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix windows bot failures from r250699 by removing "/" from expected path
Teresa Johnson [Mon, 19 Oct 2015 15:19:02 +0000 (15:19 +0000)]
Fix windows bot failures from r250699 by removing "/" from expected path
in test output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250701 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agollvm-lto support for generating combined function indexes
Teresa Johnson [Mon, 19 Oct 2015 14:30:44 +0000 (14:30 +0000)]
llvm-lto support for generating combined function indexes

Summary:
This patch adds support to llvm-lto that mirrors the support added by
r249270 to the gold plugin. This enables better testing of combined
index generation for ThinLTO.

Added a new test, and this support will be used in the test in D13515.

Reviewers: joker.eph

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D13847

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250699 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove CRLF newlines. NFC.
Benjamin Kramer [Mon, 19 Oct 2015 13:05:25 +0000 (13:05 +0000)]
Remove CRLF newlines. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250698 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix mapping of @llvm.arm.ssat/usat intrinsics to ssat/usat instructions
Asiri Rathnayake [Mon, 19 Oct 2015 11:44:24 +0000 (11:44 +0000)]
Fix mapping of @llvm.arm.ssat/usat intrinsics to ssat/usat instructions

The mapping of these two intrinsics in ARMInstrInfo.td had a small
omission which lead to their operands not being validated/transformed
before being lowered into usat and ssat instructions. This can cause
incorrect instructions to be emitted.

I've also added tests for the remaining two saturating arithmatic
intrinsics @llvm.arm.qadd and @llvm.arm.qsub as they are missing
codegen tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250697 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[GlobalsAA] Fix a really horrible iterator invalidation bug
James Molloy [Mon, 19 Oct 2015 08:54:59 +0000 (08:54 +0000)]
[GlobalsAA] Fix a really horrible iterator invalidation bug

We were keeping a reference to an object in a DenseMap then mutating it. At the end of the function we were attempting to clone that reference into other keys in the DenseMap, but DenseMap may well decide to resize its hashtable which would invalidate the reference!

It took an extremely complex testcase to catch this - many thanks to Zhendong Su for catching it in PR25225.

This fixes PR25225.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250692 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemoved parameter "Consecutive" from isLegalMaskedLoad() / isLegalMaskedStore().
Elena Demikhovsky [Mon, 19 Oct 2015 07:43:38 +0000 (07:43 +0000)]
Removed parameter "Consecutive" from isLegalMaskedLoad() / isLegalMaskedStore().
Originally I planned to use the same interface for masked gather/scatter and set isConsecutive to "false" in this case.

Now I'm implementing masked gather/scatter and see that the interface is inconvenient. I want to add interfaces isLegalMaskedGather() / isLegalMaskedScatter() instead of using the "Consecutive" parameter in the existing interfaces.

Differential Revision: http://reviews.llvm.org/D13850

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250686 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips][microMIPS] Implement ADDQ.PH, ADDQ_S.W, ADDQH.PH, ADDQH.W, ADDSC, ADDU.PH...
Zlatko Buljan [Mon, 19 Oct 2015 07:16:26 +0000 (07:16 +0000)]
[mips][microMIPS] Implement ADDQ.PH, ADDQ_S.W, ADDQH.PH, ADDQH.W, ADDSC, ADDU.PH, ADDU_S.QB, ADDWC and ADDUH.QB instructions
Differential Revision: http://reviews.llvm.org/D13130

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250685 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips][microMIPS] Implement ABSQ.QB, ABSQ_S.PH, ABSQ_S.W, ABSQ_S.QB, INSV, MADD,...
Zlatko Buljan [Mon, 19 Oct 2015 06:34:44 +0000 (06:34 +0000)]
[mips][microMIPS] Implement ABSQ.QB, ABSQ_S.PH, ABSQ_S.W, ABSQ_S.QB, INSV, MADD, MADDU, MSUB, MSUBU, MULT and MULTU instructions
Differential Revision: http://reviews.llvm.org/D13721

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250683 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[PGO] Eliminate prof data register calls on FreeBSD platform
Xinliang David Li [Mon, 19 Oct 2015 04:17:10 +0000 (04:17 +0000)]
[PGO] Eliminate prof data register calls on FreeBSD platform

This is a follow up patch of r250199 after verifying the start/stop
section symbols work as spected on FreeBSD.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250679 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoPreserve CFG in MergedLoadStoreMotion. This fixes PR24426.
Jakub Staszak [Sun, 18 Oct 2015 19:34:10 +0000 (19:34 +0000)]
Preserve CFG in MergedLoadStoreMotion. This fixes PR24426.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250660 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd hashing and DenseMapInfo for ArrayRef
Rafael Espindola [Sun, 18 Oct 2015 14:04:56 +0000 (14:04 +0000)]
Add hashing and DenseMapInfo for ArrayRef

Sometimes it is more natural to use a ArrayRef<uint8_t> than a StringRef to
represent a range of bytes that is not, semantically, a string.

This will be used in lld in a sec.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250658 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86][SSE] Add vector bit rotation tests.
Simon Pilgrim [Sun, 18 Oct 2015 12:54:37 +0000 (12:54 +0000)]
[X86][SSE] Add vector bit rotation tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250656 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse SDValue bool check. NFCI.
Simon Pilgrim [Sun, 18 Oct 2015 12:33:54 +0000 (12:33 +0000)]
Use SDValue bool check. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250653 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMove one-use variable inside test. NFC.
Simon Pilgrim [Sun, 18 Oct 2015 11:47:23 +0000 (11:47 +0000)]
Move one-use variable inside test. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250651 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86][AVX512DQ] add scalar fpclass
Asaf Badouh [Sun, 18 Oct 2015 11:04:38 +0000 (11:04 +0000)]
[X86][AVX512DQ] add scalar fpclass

Differential Revision: http://reviews.llvm.org/D13769

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250650 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAVX512: Lowering i8/i16 vector CTLZ using the dword LZCNT vector instruction
Igor Breger [Sun, 18 Oct 2015 09:56:39 +0000 (09:56 +0000)]
AVX512: Lowering i8/i16 vector CTLZ using the dword LZCNT vector instruction

Differential Revision: http://reviews.llvm.org/D13632

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250649 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Sparc] Use MCPhysReg instead of unsigned to size static arrays of registers. Should...
Craig Topper [Sun, 18 Oct 2015 05:29:05 +0000 (05:29 +0000)]
[Sparc] Use MCPhysReg instead of unsigned to size static arrays of registers. Should reduce the table size.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250644 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse array_lengthof. NFC
Craig Topper [Sun, 18 Oct 2015 05:15:38 +0000 (05:15 +0000)]
Use array_lengthof. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250643 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMake a bunch of static arrays const.
Craig Topper [Sun, 18 Oct 2015 05:15:34 +0000 (05:15 +0000)]
Make a bunch of static arrays const.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250642 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[RuntimeDyld] Add support for absolute symbols.
Lang Hames [Sun, 18 Oct 2015 01:41:37 +0000 (01:41 +0000)]
[RuntimeDyld] Add support for absolute symbols.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250639 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMinor Instr PGO code restructuring
Xinliang David Li [Sun, 18 Oct 2015 01:02:29 +0000 (01:02 +0000)]
Minor Instr PGO code restructuring

1. Key constant values (version, magic) and data structures related to raw and
   indexed profile format are moved into one centralized file: InstrProf.h.
2. Utility function such as MD5Hash computation is also moved to the common
   header to allow sharing with other components in the future.
3. A header data structure is introduced for Indexed format so that the reader
   and writer can always be in sync.
4. Added some comments to document different places where multiple definition
   of the data structure must be kept in sync (reader/writer, runtime, lowering
   etc).  No functional change is intended.

Differential Revision:  http://reviews.llvm.org/D13758

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250638 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SCEV] Fix whitespace issues and remove extra braces; NFC
Sanjoy Das [Sun, 18 Oct 2015 00:29:27 +0000 (00:29 +0000)]
[SCEV] Fix whitespace issues and remove extra braces; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250636 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SCEV] Use std::all_of and std::any_of; NFC
Sanjoy Das [Sun, 18 Oct 2015 00:29:23 +0000 (00:29 +0000)]
[SCEV] Use std::all_of and std::any_of; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250635 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SCEV] Use auto where it helps remove line breaks; NFC
Sanjoy Das [Sun, 18 Oct 2015 00:29:20 +0000 (00:29 +0000)]
[SCEV] Use auto where it helps remove line breaks; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250634 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SCEV] Use range for loops; NFC
Sanjoy Das [Sun, 18 Oct 2015 00:29:16 +0000 (00:29 +0000)]
[SCEV] Use range for loops; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250633 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse std::find instead of manual loop.
Craig Topper [Sat, 17 Oct 2015 21:32:28 +0000 (21:32 +0000)]
Use std::find instead of manual loop.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250624 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse std::is_sorted to replace a custom version. Also replace a comparison predicate...
Craig Topper [Sat, 17 Oct 2015 21:32:26 +0000 (21:32 +0000)]
Use std::is_sorted to replace a custom version. Also replace a comparison predicate struct with a lambda.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250623 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86][XOP] Add VPROT instruction opcodes
Simon Pilgrim [Sat, 17 Oct 2015 19:04:24 +0000 (19:04 +0000)]
[X86][XOP] Add VPROT instruction opcodes

Added X86ISD opcodes for VPROT vector rotate by variable and by immediate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250620 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove unnecessary 'const' pointed out by David Blaikie.
Craig Topper [Sat, 17 Oct 2015 18:22:46 +0000 (18:22 +0000)]
Remove unnecessary 'const' pointed out by David Blaikie.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250619 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86][XOP] Add VPROT rotate by immediate intrinsics tests
Simon Pilgrim [Sat, 17 Oct 2015 18:21:53 +0000 (18:21 +0000)]
[X86][XOP] Add VPROT rotate by immediate intrinsics tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250618 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[DAG] Ensure vector constant folding uses correct scalar undef types
Simon Pilgrim [Sat, 17 Oct 2015 16:49:43 +0000 (16:49 +0000)]
[DAG] Ensure vector constant folding uses correct scalar undef types

Minor fix to D13665 found during post-commit review.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250616 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoReplace a custom table sort check with std::is_sorted. Change a function to take...
Craig Topper [Sat, 17 Oct 2015 16:37:13 +0000 (16:37 +0000)]
Replace a custom table sort check with std::is_sorted. Change a function to take ArrayRef instead of pointer and length. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250615 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse std::begin/end and std::is_sorted to simplify some code. NFC
Craig Topper [Sat, 17 Oct 2015 16:37:11 +0000 (16:37 +0000)]
Use std::begin/end and std::is_sorted to simplify some code. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250614 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse binary search in isCPUStringValid since the array is sorted.
Craig Topper [Sat, 17 Oct 2015 16:37:09 +0000 (16:37 +0000)]
Use binary search in isCPUStringValid since the array is sorted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250613 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[CostModel] Fixed AVX integer shift costs
Simon Pilgrim [Sat, 17 Oct 2015 13:23:38 +0000 (13:23 +0000)]
[CostModel] Fixed AVX integer shift costs

Targets with AVX but without AVX2 were incorrectly reporting costs of 256-bit integer shifts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250611 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86][FastISel] Teach how to select SSE4A nontemporal stores.
Simon Pilgrim [Sat, 17 Oct 2015 13:04:42 +0000 (13:04 +0000)]
[X86][FastISel] Teach how to select SSE4A nontemporal stores.

Add FastISel support for SSE4A scalar float / double non-temporal stores

Follow up to D13698

Differential Revision: http://reviews.llvm.org/D13773

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250610 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[InstCombine] SSE4A constant folding and conversion to shuffles.
Simon Pilgrim [Sat, 17 Oct 2015 11:40:05 +0000 (11:40 +0000)]
[InstCombine] SSE4A constant folding and conversion to shuffles.

This patch improves support for combining the SSE4A EXTRQ(I) and INSERTQ(I) intrinsics:

1 - Converts INSERTQ/EXTRQ calls to INSERTQI/EXTRQI if the 'bit index' and 'length' operands are constant
2 - Converts INSERTQI/EXTRQI calls to shufflevector if the bit index/length are both byte aligned (we can already lower shuffles to INSERTQI/EXTRQI if its useful)
3 - Constant folding support
4 - Add zeroinitializer handling

Differential Revision: http://reviews.llvm.org/D13348

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250609 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[JIT/Examples] Fix Fibonacci so that it runs again.
Davide Italiano [Sat, 17 Oct 2015 06:36:46 +0000 (06:36 +0000)]
[JIT/Examples] Fix Fibonacci so that it runs again.

The old JIT is (long) gone.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250604 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[libFuzzer] add -shuffle flag
Kostya Serebryany [Sat, 17 Oct 2015 04:38:26 +0000 (04:38 +0000)]
[libFuzzer] add -shuffle flag

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250603 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Reverting test file change.
Colin LeMahieu [Sat, 17 Oct 2015 01:58:51 +0000 (01:58 +0000)]
[Hexagon] Reverting test file change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250601 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Adding skeleton of HVX extension instructions.
Colin LeMahieu [Sat, 17 Oct 2015 01:33:04 +0000 (01:33 +0000)]
[Hexagon] Adding skeleton of HVX extension instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250600 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRegisterPressure: Unify the sparse sets in LiveRegsSet; NFC
Matthias Braun [Sat, 17 Oct 2015 01:03:44 +0000 (01:03 +0000)]
RegisterPressure: Unify the sparse sets in LiveRegsSet; NFC

Also do some cleanups comment improvements.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250598 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRegisterPressure: allocatable physreg uses are always kills
Matthias Braun [Sat, 17 Oct 2015 00:46:57 +0000 (00:46 +0000)]
RegisterPressure: allocatable physreg uses are always kills

This property was already used in the code path when no liveness
intervals are present. Unfortunately the code path that uses liveness
intervals tried to query a cached live interval for an allocatable
physreg, those are usually not computed so a conservative default was
used.

This doesn't affect any of the lit testcases. This is a foreclosure to
upcoming changes which should be NFC but without this patch this tidbit
wouldn't be NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250596 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRegisterPressure: Remove 0 entries from PressureChange
Matthias Braun [Sat, 17 Oct 2015 00:35:59 +0000 (00:35 +0000)]
RegisterPressure: Remove 0 entries from PressureChange

This should not change behaviour because as far as I can see all code
reading the pressure changes has no effect if the PressureInc is 0.
Removing these entries however does avoid unnecessary computation, and
results in a more stable debug output. I want the stable debug output to
check that some upcoming changes are indeed NFC and identical even at
the debug output level.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250595 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoWebAssembly: don't omit dead vregs from locals
JF Bastien [Sat, 17 Oct 2015 00:25:38 +0000 (00:25 +0000)]
WebAssembly: don't omit dead vregs from locals

Summary:
This is a temporary hack until we get around to remapping the vreg
numbers to local numbers. Dead vregs cause bad numbering and make
consumers sad.

We could also just look at debug info an use named locals instead, but
vregs have to work properly anyways so there!

Reviewers: binji, sunfish

Subscribers: jfb, llvm-commits, dschuff

Differential Revision: http://reviews.llvm.org/D13839

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250594 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoWebAssembly: fix the syntax for comparisons
JF Bastien [Sat, 17 Oct 2015 00:12:29 +0000 (00:12 +0000)]
WebAssembly: fix the syntax for comparisons

Summary: It has also slightly changed.

Reviewers: binji

Subscribers: jfb, dschuff, llvm-commits, sunfish

Differential Revision: http://reviews.llvm.org/D13837

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250591 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRegisterPressure: Hide non-const iterators of PressureDiff
Matthias Braun [Sat, 17 Oct 2015 00:08:48 +0000 (00:08 +0000)]
RegisterPressure: Hide non-const iterators of PressureDiff

It is too easy to accidentally violate the ordering requirements when
modifying the PressureDiff entries through iterators.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250590 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoStreamWriter: List basic types instead of derived ones in HexNumber
Matthias Braun [Sat, 17 Oct 2015 00:08:45 +0000 (00:08 +0000)]
StreamWriter: List basic types instead of derived ones in HexNumber

This avoids problems with different (u)intXX definition on different
platforms. Specifically this fixes a case on OS/X which had uint64_t
defined as unsigned long long.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250589 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[WinEH] Fix eh.exceptionpointer intrinsic lowering
Joseph Tremoulet [Sat, 17 Oct 2015 00:08:08 +0000 (00:08 +0000)]
[WinEH] Fix eh.exceptionpointer intrinsic lowering

Summary:
Some shared code for handling eh.exceptionpointer and eh.exceptioncode
needs to not share the part that truncates to 32 bits, which is intended
just for exception codes.

Reviewers: rnk

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D13747

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250588 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDisable a test relying on symbol demangling on non-Windows platforms
Reid Kleckner [Fri, 16 Oct 2015 23:56:14 +0000 (23:56 +0000)]
Disable a test relying on symbol demangling on non-Windows platforms

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250587 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSpeculative fix for GCC build
Reid Kleckner [Fri, 16 Oct 2015 23:53:12 +0000 (23:53 +0000)]
Speculative fix for GCC build

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250585 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[WinEH] Fix stack alignment in funclets and ParentFrameOffset calculation
Reid Kleckner [Fri, 16 Oct 2015 23:43:27 +0000 (23:43 +0000)]
[WinEH] Fix stack alignment in funclets and ParentFrameOffset calculation

Our previous value of "16 + 8 + MaxCallFrameSize" for ParentFrameOffset
is incorrect when CSRs are involved. We were supposed to have a test
case to catch this, but it wasn't very rigorous.

The main effect here is that calling _CxxThrowException inside a
catchpad doesn't immediately crash on MOVAPS when you have an odd number
of CSRs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250583 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[llvm-symbolizer] Use the export table if no symbols are present
Reid Kleckner [Fri, 16 Oct 2015 23:43:22 +0000 (23:43 +0000)]
[llvm-symbolizer] Use the export table if no symbols are present

This lets us make guesses about symbols in third party DLLs without
debug info, like MSVCR120.dll or kernel32.dll. dbghelp does the same
thing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250582 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRegisterPressure: Use range based for, cleanup
Matthias Braun [Fri, 16 Oct 2015 23:25:09 +0000 (23:25 +0000)]
RegisterPressure: Use range based for, cleanup

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250579 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[llvm-readobj] Teach ELFDumper about symbol versioning.
Davide Italiano [Fri, 16 Oct 2015 23:19:01 +0000 (23:19 +0000)]
[llvm-readobj] Teach ELFDumper about symbol versioning.

Differential Revision:  http://reviews.llvm.org/D13824

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250575 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoInstroduce a template file to define InstrPGO core data structures.
Xinliang David Li [Fri, 16 Oct 2015 23:17:34 +0000 (23:17 +0000)]
Instroduce a template file to define InstrPGO core data structures.

Changing PGO data format layout can be a pain. Many different places need
to be touched and kept in sync. Failing to do so usually results in errors
very time consuming to debug.

This file is intended to be the master file that defines the layout of the
core runtime data structures. Currently only two structure is covered: Per
function ProfData structure and the function record structure used in
coverage mapping.

No client code has been made yet, so this commit is NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250574 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[CMake] Cleaning up and generalizing the LLVMInstallSymlink script so that it can...
Chris Bieneman [Fri, 16 Oct 2015 23:17:13 +0000 (23:17 +0000)]
[CMake] Cleaning up and generalizing the LLVMInstallSymlink script so that it can be used for libraries too.

In order to resolve PR25059, we're going to need to be able to generate symlinks to libraries manually, so I need this code to be reusable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250573 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[libFuzzer] print a stack trace on timeout
Kostya Serebryany [Fri, 16 Oct 2015 23:04:31 +0000 (23:04 +0000)]
[libFuzzer] print a stack trace on timeout

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250571 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert "This is a follow-up to the discussion in D12882."
Benjamin Kramer [Fri, 16 Oct 2015 23:00:29 +0000 (23:00 +0000)]
Revert "This is a follow-up to the discussion in D12882."

Breaks clang selfhost, see PR25222. This reverts commits r250527 and r250528.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250570 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[libFuzzer] reduce the size of artifacts printed on the screen
Kostya Serebryany [Fri, 16 Oct 2015 22:47:20 +0000 (22:47 +0000)]
[libFuzzer] reduce the size of artifacts printed on the screen

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250565 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[libFuzzer] When -test_single_input crashes the test it is not necessary to write...
Kostya Serebryany [Fri, 16 Oct 2015 22:41:47 +0000 (22:41 +0000)]
[libFuzzer] When -test_single_input crashes the test it is not necessary to write crash-file because input is already known to the user. Patch by Mike Aizatsky

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250564 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[x86] promote 'add nsw' to a wider type to allow more combines
Sanjay Patel [Fri, 16 Oct 2015 22:14:12 +0000 (22:14 +0000)]
[x86] promote 'add nsw' to a wider type to allow more combines

The motivation for this patch starts with PR20134:
https://llvm.org/bugs/show_bug.cgi?id=20134

void foo(int *a, int i) {
  a[i] = a[i+1] + a[i+2];
}

It seems better to produce this (14 bytes):

movslq %esi, %rsi
movl 0x4(%rdi,%rsi,4), %eax
addl 0x8(%rdi,%rsi,4), %eax
movl %eax, (%rdi,%rsi,4)

Rather than this (22 bytes):

leal 0x1(%rsi), %eax
cltq
leal 0x2(%rsi), %ecx
movslq %ecx, %rcx
movl (%rdi,%rcx,4), %ecx
addl (%rdi,%rax,4), %ecx
movslq %esi, %rax
movl %ecx, (%rdi,%rax,4)

The most basic problem (the first test case in the patch combines constants) should also be fixed in InstCombine,
but it gets more complicated after that because we need to consider architecture and micro-architecture. For
example, AArch64 may not see any benefit from the more general transform because the ISA solves the sexting in
hardware. Some x86 chips may not want to replace 2 ADD insts with 1 LEA, and there's an attribute for that:
FeatureSlowLEA. But I suspect that doesn't go far enough or maybe it's not getting used when it should; I'm
also not sure if FeatureSlowLEA should also mean "slow complex addressing mode".

I see no perf differences on test-suite with this change running on AMD Jaguar, but I see small code size
improvements when building clang and the LLVM tools with the patched compiler.

A more general solution to the sext(add nsw(x, C)) problem that works for multiple targets is available
in CodeGenPrepare, but it may take quite a bit more work to get that to fire on all of the test cases that
this patch takes care of.

Differential Revision: http://reviews.llvm.org/D13757

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250560 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMC: Don't crash after issuing a diagnostic.
Jim Grosbach [Fri, 16 Oct 2015 22:07:59 +0000 (22:07 +0000)]
MC: Don't crash after issuing a diagnostic.

Crashing is bad, m'kay? Fixing a 4 year old bug of my own creation.
Adding the testcase now which I should have added then which would have
long since caught this.

The problem is that printMessage() will display the diagnostic but not
set HadError to true, resulting in the assembler continuing on its way
and trying to create relocations for things that may not allow them or
otherwise get itself into trouble. Using the Error() helper function
here rather than calling printMessage() directly resolves this.

rdar://23133240

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250557 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[WinEH] Fix CatchRetSuccessorColorMap accounting
Joseph Tremoulet [Fri, 16 Oct 2015 21:22:54 +0000 (21:22 +0000)]
[WinEH] Fix CatchRetSuccessorColorMap accounting

Summary:
We now use the block for the catchpad itself, rather than its normal
successor, as the funclet entry.
Putting the normal successor in the map leads downstream funclet
membership computations to erroneous results.

Reviewers: majnemer, rnk

Subscribers: rnk, llvm-commits

Differential Revision: http://reviews.llvm.org/D13798

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250552 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix assertion failure with fp128 to unsigned i64 conversion
Andrew Kaylor [Fri, 16 Oct 2015 20:39:20 +0000 (20:39 +0000)]
Fix assertion failure with fp128 to unsigned i64 conversion

Patch by Mitch Bodart

Differential Revision: http://reviews.llvm.org/D13780

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250550 91177308-0d34-0410-b5e6-96231b3b80d8