OSDN Git Service

android-x86/external-llvm.git
7 years ago[Hexagon] Mark PS_jumpret as pseudo-instruction, expand it into J2_jumpr
Krzysztof Parzyszek [Fri, 19 Aug 2016 14:04:45 +0000 (14:04 +0000)]
[Hexagon] Mark PS_jumpret as pseudo-instruction, expand it into J2_jumpr

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279241 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Hexagon] Improvements to handling and generation of FP instructions
Krzysztof Parzyszek [Fri, 19 Aug 2016 13:34:31 +0000 (13:34 +0000)]
[Hexagon] Improvements to handling and generation of FP instructions

Improved handling of fma, floating point min/max, additional load/store
instructions for floating point types.

Patch by Jyotsna Verma.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279239 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LoopVectorize] Don't copy std::vector in for-range loop.
Benjamin Kramer [Fri, 19 Aug 2016 12:44:24 +0000 (12:44 +0000)]
[LoopVectorize] Don't copy std::vector in for-range loop.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279233 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM] Revert r279227 and r279228 until I can find someone to help me
Chandler Carruth [Fri, 19 Aug 2016 10:51:55 +0000 (10:51 +0000)]
[PM] Revert r279227 and r279228 until I can find someone to help me
solve completely opaque MSVC build errors. It complains about lots of
stuff with this change without givin nearly enough information to even
try to fix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279231 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][SSE] Add support for matching commuted insertps patterns
Simon Pilgrim [Fri, 19 Aug 2016 10:31:53 +0000 (10:31 +0000)]
[X86][SSE] Add support for matching commuted insertps patterns

INSERTPS doesn't fit well with our shuffle mask canonicalization, so we need to attempt both the original mask and the commuted mask to more likely get a match

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279230 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SimplifyCFG] Rewrite SinkThenElseCodeToEnd
James Molloy [Fri, 19 Aug 2016 10:10:27 +0000 (10:10 +0000)]
[SimplifyCFG] Rewrite SinkThenElseCodeToEnd

The new version has several advantages:
  1) IMSHO it's more readable and neater
  2) It handles loads and stores properly
  3) It can handle any number of incoming blocks rather than just two. I'll be taking advantage of this in a followup patch.

With this change we can now finally sink load-modify-store idioms such as:

    if (a)
      return *b += 3;
    else
      return *b += 4;

    =>

    %z = load i32, i32* %y
    %.sink = select i1 %a, i32 5, i32 7
    %b = add i32 %z, %.sink
    store i32 %b, i32* %y
    ret i32 %b

When this works for switches it'll be even more powerful.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279229 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM] Fix a compile error with GCC. NFC.
Chandler Carruth [Fri, 19 Aug 2016 09:53:10 +0000 (09:53 +0000)]
[PM] Fix a compile error with GCC. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279228 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM] Make the the new pass manager support fully generic extra arguments
Chandler Carruth [Fri, 19 Aug 2016 09:45:16 +0000 (09:45 +0000)]
[PM] Make the the new pass manager support fully generic extra arguments
to run methods, both for transform passes and analysis passes.

This also allows the analysis manager to use a different set of extra
arguments from the pass manager where useful. Consider passes over
analysis produced units of IR like SCCs of the call graph or loops.
Passes of this nature will often want to refer to the analysis result
that was used to compute their IR units (the call graph or LoopInfo).
And for transformations, they may want to communicate special update
information to the outer pass manager. With this change, it becomes
possible to have a run method for a loop pass that looks more like:

  PreservedAnalyses run(Loop &L, AnalysisManager<Loop, LoopInfo> &AM,
                        LoopInfo &LI, LoopUpdateRecord &UR);

And to query the analysis manager like:

    AM.getResult<MyLoopAnalysis>(L, LI);

This makes accessing the known-available analyses convenient and clear,
and it makes passing customized data structures around easy.

My initial use case is going to be in updating the pass manager layers
when the analysis units of IR change. But there are more use cases here
such as having a layer that lets inner passes signal whether certain
additional passes should be run because of particular simplifications
made. Two desires for this have come up in the past: triggering
additional optimization after successfully unrolling loops, and
triggering additional inlining after collapsing indirect calls to direct
calls.

Despite adding this layer of generic extensibility, the *only* change to
existing, simple usage are for places where we forward declare the
AnalysisManager template. We really shouldn't be doing this because of
the fragility exposed here, but currently it makes coping with the
legacy PM code easier.

Differential Revision: http://reviews.llvm.org/D21462

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279227 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM] Try to work-around what appears to be an MSVC SFINAE issue with
Chandler Carruth [Fri, 19 Aug 2016 09:26:00 +0000 (09:26 +0000)]
[PM] Try to work-around what appears to be an MSVC SFINAE issue with
r279217 where it fails to select the path that other compilers select.

The workaround won't be as careful to produce an error when an analysis
result is incorrect, but we can rely on non-MSVC builds to catch such
errors it seems and MSVC doesn't seem to support the alternative
techniques.

Hoping this brings the windows bots back to life. If not, will have to
revert all of this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279225 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CodeGen] Fix a trivial type conversion bug dating back to pre-2008
James Molloy [Fri, 19 Aug 2016 08:38:50 +0000 (08:38 +0000)]
[CodeGen] Fix a trivial type conversion bug dating back to pre-2008

The heuristic above this code is incredibly suspect, but disregarding that it mutates the cast opcode so we need to check the *mutated* opcode later to see if we need to emit an AssertSext or AssertZext node.

Fixes PR29041.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279223 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[asan] Fix size of shadow incorrectly calculated in r279178
Vitaly Buka [Fri, 19 Aug 2016 08:33:53 +0000 (08:33 +0000)]
[asan] Fix size of shadow incorrectly calculated in r279178

Summary: r279178 generates 8 times more stores than necessary.

Reviewers: eugenis

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D23708

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279222 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM] NFC refactoring: remove the AnalysisManagerBase class, folding it
Chandler Carruth [Fri, 19 Aug 2016 08:31:47 +0000 (08:31 +0000)]
[PM] NFC refactoring: remove the AnalysisManagerBase class, folding it
into the AnalysisManager class template.

Back when I first added this base class there were separate analysis
managers and some plausible reason why it would be a useful factoring of
common code between them. However, after a lot of refactoring cleaning,
we now have *entirely* shared code. The base class was just an arbitrary
division between code in one class template and a separate class
template. It didn't add anything and forced lots of indirection through
"derived_this" for no real gain.

We can always factor a base CRTP class out with common code if there is
ever some *other* analysis manager that wants to share a subset of
logic. But for now, folding things into the primary template is
a non-trivial simplification with no down sides I see. It shortens the
code considerably, removes an unhelpful abstraction, and will make
subsequent patches *dramatically* less complex which enhance the
analysis manager infrastructure to effectively cope with invalidation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279221 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[modules] Add missing include.
Vassil Vassilev [Fri, 19 Aug 2016 08:30:42 +0000 (08:30 +0000)]
[modules] Add missing include.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279219 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM] Redesign how the new PM detects whether an analysis result provides
Chandler Carruth [Fri, 19 Aug 2016 07:49:23 +0000 (07:49 +0000)]
[PM] Redesign how the new PM detects whether an analysis result provides
its own invalidate method.

Previously, the technique would assume that if a result didn't have an
invalidate method that didn't exactly match the expected signature it
didn't have one at all. This is in fact not the case. And we had
analyses with incorrect signatures for the invalidate method in the
tree that would be erroneously invalidated in certain cases! Yikes.

Moreover a result might legitimately want to have multiple overloads for
the invalidate method, and if one changes or a new one is needed we
again really want a compiler error. For example in the tree we had not
added the overload for a *function* IR unit to the invalidate routine
for TLI. Doh.

So a new techique for the SFINAE detection here: if the result has *any*
member spelled "invalidate" we turn off the synthesis of a default
version. We don't care if it is a member function or a member variable
or how many overloads there are. Once a result has something by that
name it must provide suitable overloads for the contexts in which it is
used. This seems much more resilient and durable.

Huge props to Richard Smith who helped me figure out how on earth we
could even do this in C++. It took quite some doing. The technique is
remarkably clean however, and merely requires that the analysis results
are not *final* classes. I think that's a requirement we can live with
even if it is a bit odd.

I've fixed the two bad in-tree analysis results. And this will make my
next change which changes the API for invalidate much easier to
validate as correct.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279217 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM] Rework the new PM support for building the ModuleSummaryIndex to
Chandler Carruth [Fri, 19 Aug 2016 07:49:19 +0000 (07:49 +0000)]
[PM] Rework the new PM support for building the ModuleSummaryIndex to
directly produce the index as the value type result.

This requires making the index movable which is straightforward. It
greatly simplifies things by allowing us to completely avoid the builder
API and the layers of abstraction inherent there. Instead both pass
managers can directly construct these when run by value. They still
won't be constructed truly eagerly thanks to the optional in the legacy
PM. The code that directly builds the index can also just share a direct
function.

A notable change here is that the result type of the analysis for the
new PM is no longer a reference type. This was really problematic when
making changes to how we handle result types to make our interface
requirements *much* more strict and precise. But I think this is an
overall improvement.

Differential Revision: https://reviews.llvm.org/D23701

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279216 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix tests in llvm/test/tools/gold/X86 to satisfy r279014.
NAKAMURA Takumi [Fri, 19 Aug 2016 06:44:44 +0000 (06:44 +0000)]
Fix tests in llvm/test/tools/gold/X86 to satisfy r279014.

They would unexpectedly pass if test/tools/gold/X86/Output had outputs of previous tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279214 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Profile] Fix edge count read bug
Xinliang David Li [Fri, 19 Aug 2016 06:31:45 +0000 (06:31 +0000)]
[Profile] Fix edge count read bug

 Use uint64_t to avoid value truncation before scaling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279213 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LTO] Move callback member from base class to the derived where it is used (NFC)
Mehdi Amini [Fri, 19 Aug 2016 06:10:03 +0000 (06:10 +0000)]
[LTO] Move callback member from base class to the derived where it is used (NFC)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279212 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoConstify some path in the bitcode writer (NFC)
Mehdi Amini [Fri, 19 Aug 2016 06:06:18 +0000 (06:06 +0000)]
Constify some path in the bitcode writer (NFC)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279211 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LTO] Add a move to inialize member in ctor initialization list (NFC)
Mehdi Amini [Fri, 19 Aug 2016 05:56:37 +0000 (05:56 +0000)]
[LTO] Add a move to inialize member in ctor initialization list (NFC)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279210 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Profile] Simple code refactoring for reuse /NFC
Xinliang David Li [Fri, 19 Aug 2016 05:31:33 +0000 (05:31 +0000)]
[Profile] Simple code refactoring for reuse /NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279209 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[XRay] Synthesize a reference to the xray_instr_map
Dean Michael Berris [Fri, 19 Aug 2016 04:44:30 +0000 (04:44 +0000)]
[XRay] Synthesize a reference to the xray_instr_map

Without the synthesized reference to a symbol in the xray_instr_map,
linker section garbage collection will helpfully remove the whole
xray_instr_map section from the final executable (or archive). This will
cause the runtime to not be able to identify the sleds and hot-patch the
calls/jumps into the runtime trampolines.

This change adds a reference from the text section at the end of the
function to keep around the associated xray_instr_map section as well.

We also make sure that we catch this reference in the test.

Reviewers: chandlerc, echristo, majnemer, mehdi_amini

Subscribers: mehdi_amini, llvm-commits, dberris

Differential Revision: https://reviews.llvm.org/D23398

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279204 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[RuntimeDyld][MCJIT] Un-XFAIL some tests that were fixed by r279182.
Lang Hames [Fri, 19 Aug 2016 03:12:16 +0000 (03:12 +0000)]
[RuntimeDyld][MCJIT] Un-XFAIL some tests that were fixed by r279182.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279201 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "RegScavenging: Add scavengeRegisterBackwards()"
Matthias Braun [Fri, 19 Aug 2016 03:03:24 +0000 (03:03 +0000)]
Revert "RegScavenging: Add scavengeRegisterBackwards()"

The ppc64 multistage bot fails on this.

This reverts commit r279124.

Also Revert "CodeGen: Add/Factor out LiveRegUnits class; NFCI" because it depends on the previous change
This reverts commit r279171.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279199 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ADT] Add the worlds simplest STL extra. Or at least close to it.
Chandler Carruth [Fri, 19 Aug 2016 02:07:51 +0000 (02:07 +0000)]
[ADT] Add the worlds simplest STL extra. Or at least close to it.

This is a little class template that just builds an inheritance chain of
empty classes. Despite how simple this is, it can be used to really
nicely create ranked overload sets. I've added a unittest as much to
document this as test it. You can pass an object of this type as an
argument to a function overload set an it will call the first viable and
enabled candidate at or below the rank of the object.

I'm planning to use this in a subsequent commit to more clearly rank
overload candidates used for SFINAE. All credit for this technique and
both lines of code here to Richard Smith who was helping me rewrite the
SFINAE check in question to much more effectively capture the intended
set of checks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279197 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[RuntimeDyld] Add support for ELF R_ARM_REL32 and R_ARM_GOT_PREL.
Lang Hames [Fri, 19 Aug 2016 01:15:39 +0000 (01:15 +0000)]
[RuntimeDyld] Add support for ELF R_ARM_REL32 and R_ARM_GOT_PREL.

Patch by William Dillon. Thanks William!

This patch adds support for the R_ARM_REL32 and R_ARM_GOT_PREL ELF ARM
relocations to RuntimeDyld, which should allow JITing of code that
produces these relocations.

No test case: Unfortunately RuntimeDyldELF's GOT building mechanism (which
uses a separate section for GOT entries) isn't compatible with
RuntimeDyldChecker. The correct fix for this is to fix RuntimeDyldELF's GOT
support (it's fundamentally broken at the moment: separate sections aren't
guaranteed to be in range of a GOT entry load), but that's a non-trivial job.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279182 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[asan] Optimize store size in FunctionStackPoisoner::poisonRedZones
Vitaly Buka [Thu, 18 Aug 2016 23:51:15 +0000 (23:51 +0000)]
[asan] Optimize store size in FunctionStackPoisoner::poisonRedZones

Summary: Reduce store size to avoid leading and trailing zeros.

Reviewers: kcc, eugenis

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D23648

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279178 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoInclude X86CallFrameOptimization in the opt-bisect process.
Andrew Kaylor [Thu, 18 Aug 2016 22:49:51 +0000 (22:49 +0000)]
Include X86CallFrameOptimization in the opt-bisect process.

Differential Revision: https://reviews.llvm.org/D23683

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279175 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAArch64: remove extraneous padding
Saleem Abdulrasool [Thu, 18 Aug 2016 22:35:06 +0000 (22:35 +0000)]
AArch64: remove extraneous padding

The structs BarrierOp, PrefetchOp, PSBHintOp are in AArch64AsmParser.cpp
(inside anonymous namespace).  This diff changes the order of fields and
removes the excessive padding (8 bytes).

Patch by Alexander Shaposhnikov!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279173 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CMake] Add variables for tracking which runtimes are included
Chris Bieneman [Thu, 18 Aug 2016 22:18:11 +0000 (22:18 +0000)]
[CMake] Add variables for tracking which runtimes are included

This allows sub-projects to have conditionals based on the presence of other projects.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279172 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoCodeGen: Add/Factor out LiveRegUnits class; NFCI
Matthias Braun [Thu, 18 Aug 2016 22:11:28 +0000 (22:11 +0000)]
CodeGen: Add/Factor out LiveRegUnits class; NFCI

This is a set of register units intended to track register liveness, it
is similar in spirit to LivePhysRegs.
You can also think of this as the liveness tracking parts of the
RegisterScavenger factored out into an own class.

This was proposed in http://llvm.org/PR27609

Differential Revision: http://reviews.llvm.org/D21916

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279171 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix link quotes on AArch64's CompilerWriterInfo section.
Jacques Pienaar [Thu, 18 Aug 2016 22:10:06 +0000 (22:10 +0000)]
Fix link quotes on AArch64's CompilerWriterInfo section.

Reviewers: t.p.northover

Subscribers: t.p.northover, aemerson, rengolin

Differential Revision: https://reviews.llvm.org/D23697

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279169 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoCodeGen: If Convert blocks that would form a diamond when tail-merged.
Kyle Butt [Thu, 18 Aug 2016 22:09:27 +0000 (22:09 +0000)]
CodeGen: If Convert blocks that would form a diamond when tail-merged.

The following function currently relies on tail-merging for if
conversion to succeed. The common tail of cond_true and cond_false is
extracted, and this then forms a diamond pattern that can be
successfully if converted.

If this block does not get extracted, either because tail-merging is
disabled or the threshold is higher, we should still recognize this
pattern and if-convert it.

Fixed a regression in the original commit. Need to un-reverse branches after
reversing them, or other conversions go awry.

Regression on self-hosting bots with no obvious explanation. Tidied up range
handling to be more obviously correct, but there was no smoking gun.

define i32 @t2(i32 %a, i32 %b) nounwind {
entry:
        %tmp1434 = icmp eq i32 %a, %b           ; <i1> [#uses=1]
        br i1 %tmp1434, label %bb17, label %bb.outer

bb.outer:               ; preds = %cond_false, %entry
        %b_addr.021.0.ph = phi i32 [ %b, %entry ], [ %tmp10, %cond_false ]
        %a_addr.026.0.ph = phi i32 [ %a, %entry ], [ %a_addr.026.0, %cond_false ]
        br label %bb

bb:             ; preds = %cond_true, %bb.outer
        %indvar = phi i32 [ 0, %bb.outer ], [ %indvar.next, %cond_true ]
        %tmp. = sub i32 0, %b_addr.021.0.ph
        %tmp.40 = mul i32 %indvar, %tmp.
        %a_addr.026.0 = add i32 %tmp.40, %a_addr.026.0.ph
        %tmp3 = icmp sgt i32 %a_addr.026.0, %b_addr.021.0.ph
        br i1 %tmp3, label %cond_true, label %cond_false

cond_true:              ; preds = %bb
        %tmp7 = sub i32 %a_addr.026.0, %b_addr.021.0.ph
        %tmp1437 = icmp eq i32 %tmp7, %b_addr.021.0.ph
        %indvar.next = add i32 %indvar, 1
        br i1 %tmp1437, label %bb17, label %bb

cond_false:             ; preds = %bb
        %tmp10 = sub i32 %b_addr.021.0.ph, %a_addr.026.0
        %tmp14 = icmp eq i32 %a_addr.026.0, %tmp10
        br i1 %tmp14, label %bb17, label %bb.outer

bb17:           ; preds = %cond_false, %cond_true, %entry
        %a_addr.026.1 = phi i32 [ %a, %entry ], [ %tmp7, %cond_true ], [ %a_addr.026.0, %cond_false ]
        ret i32 %a_addr.026.1
}

Without tail-merging or diamond-tail if conversion:
LBB1_1:                                 @ %bb
                                        @ =>This Inner Loop Header: Depth=1
        cmp     r0, r1
        ble     LBB1_3
@ BB#2:                                 @ %cond_true
                                        @   in Loop: Header=BB1_1 Depth=1
        subs    r0, r0, r1
        cmp     r1, r0
        it      ne
        cmpne   r0, r1
        bgt     LBB1_4
LBB1_3:                                 @ %cond_false
                                        @   in Loop: Header=BB1_1 Depth=1
        subs    r1, r1, r0
        cmp     r1, r0
        bne     LBB1_1
LBB1_4:                                 @ %bb17
        bx      lr

With diamond-tail if conversion, but without tail-merging:
@ BB#0:                                 @ %entry
        cmp     r0, r1
        it      eq
        bxeq    lr
LBB1_1:                                 @ %bb
                                        @ =>This Inner Loop Header: Depth=1
        cmp     r0, r1
        ite     le
        suble   r1, r1, r0
        subgt   r0, r0, r1
        cmp     r1, r0
        bne     LBB1_1
@ BB#2:                                 @ %bb17
        bx      lr

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279168 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoIfConversion: Rescan diamonds.
Kyle Butt [Thu, 18 Aug 2016 22:09:25 +0000 (22:09 +0000)]
IfConversion: Rescan diamonds.

The cost of predicating a diamond is only the instructions that are not shared
between the two branches. Additionally If a predicate clobbering instruction
occurs in the shared portion of the branches (e.g. a cond move), it may still
be possible to if convert the sub-cfg. This change handles these two facts by
rescanning the non-shared portion of a diamond sub-cfg to recalculate both the
predication cost and whether both blocks are pred-clobbering.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279167 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoIfConversion: Handle inclusive ranges more carefully.
Kyle Butt [Thu, 18 Aug 2016 22:09:23 +0000 (22:09 +0000)]
IfConversion: Handle inclusive ranges more carefully.

This may affect calculations for thresholds, but is not a significant change
in behavior.

The problem was that an inclusive range must have an additonal flag to showr
that it is empty, because otherwise begin == end implies that the range has one
element, and it may not be possible to move past on either side.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279166 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agollvm-objdump: Add Hexagon printer changes for -S/-l options
Hemant Kulkarni [Thu, 18 Aug 2016 21:50:13 +0000 (21:50 +0000)]
llvm-objdump: Add Hexagon printer changes for -S/-l options

Differential Revision: https://reviews.llvm.org/D23521

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279161 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CMake] Create convenience targets for runtime projects
Chris Bieneman [Thu, 18 Aug 2016 21:47:18 +0000 (21:47 +0000)]
[CMake] Create convenience targets for runtime projects

Each runtime project has a top-level target that is the name of the runtime (minus the "lib" prefix if applicable). This creates top-level targets mapping to runtime projects.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279160 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SystemZ] Use valid base/index regs for inline asm
Zhan Jun Liau [Thu, 18 Aug 2016 21:44:15 +0000 (21:44 +0000)]
[SystemZ] Use valid base/index regs for inline asm

Summary:
Inline asm memory constraints can have the base or index register be assigned
to %r0 right now. Make sure that we assign only ADDR64 registers to the base
and index.

Reviewers: uweigand

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D23367

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279157 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Analysis] Change several Analysis pieces to use NodeRef. NFC.
Tim Shen [Thu, 18 Aug 2016 21:41:24 +0000 (21:41 +0000)]
[Analysis] Change several Analysis pieces to use NodeRef. NFC.

Reviewers: dblaikie, grosser

Subscribers: mzolotukhin, llvm-commits

Differential Revision: https://reviews.llvm.org/D23625

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279156 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CMake] Make llvm-config implicit dependency for subprojects
Chris Bieneman [Thu, 18 Aug 2016 21:41:21 +0000 (21:41 +0000)]
[CMake] Make llvm-config implicit dependency for subprojects

The subproject interface being used for runtime libraries expects that llvm-config is passed into the subproject for consumption. We currently do this for every subproject, so we should expect that all LLVM ExternalProjects depend on llvm-config for the time being.

Eventually I'd like to see the sub-projects using LLVMConfig.cmake instead of the llvm-config binary, but that will take time to roll out.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279155 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CMake] Minor fix to regex in r279152
Chris Bieneman [Thu, 18 Aug 2016 21:36:36 +0000 (21:36 +0000)]
[CMake] Minor fix to regex in r279152

The third version component is optional in Xcode's version spew, so we need to make it optional in the regex.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279153 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CMake] Support for generating Xcode 8 compatible toolchains
Chris Bieneman [Thu, 18 Aug 2016 21:32:48 +0000 (21:32 +0000)]
[CMake] Support for generating Xcode 8 compatible toolchains

Xcode 8 requires toolchain compatibility version 2. This allows us to select the correct compatibility version based on the installed version of Xcode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279152 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstCombine] add helper function for folds of icmp (shl 1, Y), C; NFCI
Sanjay Patel [Thu, 18 Aug 2016 21:28:30 +0000 (21:28 +0000)]
[InstCombine] add helper function for folds of icmp (shl 1, Y), C; NFCI

Clean up the existing code by:
1. Renaming variables
2. Adding local variables
3. Making it vector-safe

This is still guarded by a ConstantInt check, so no functional change is intended.
But this should be ready to go: if we move the ConstantInt check down, all of
these folds should do the right thing for vector types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279150 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[lanai] Add ISA document to CompilerWritersInfo
Jacques Pienaar [Thu, 18 Aug 2016 21:25:17 +0000 (21:25 +0000)]
[lanai] Add ISA document to CompilerWritersInfo

Summary: Add Lanai ISA document to CompilerWritersInfo.

Reviewers: eliben

Subscribers: aemerson, llvm-commits

Differential Revision: https://reviews.llvm.org/D23693

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279149 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU/SI: Fix a test in wqm.ll to always use s_cbranch_vcc*
Tom Stellard [Thu, 18 Aug 2016 21:21:53 +0000 (21:21 +0000)]
AMDGPU/SI: Fix a test in wqm.ll to always use s_cbranch_vcc*

Summary:
We need to use floating-point compares to ensure that s_cbranch_vcc*
instructions are always generated.  With integer compares, future
optimizations could cause s_cbranch_scc* to be generated instead.

Reviewers: arsenm, nhaehnle

Subscribers: llvm-commits, kzhuravl

Differential Revision: https://reviews.llvm.org/D23401

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279148 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[libFuzzer] add more __attribute__((visibility("default")))
Kostya Serebryany [Thu, 18 Aug 2016 20:52:52 +0000 (20:52 +0000)]
[libFuzzer] add more __attribute__((visibility("default")))

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279143 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoMake cltz and cttz zero undef when the operand cannot be zero in InstCombine
Amaury Sechet [Thu, 18 Aug 2016 20:43:50 +0000 (20:43 +0000)]
Make cltz and cttz zero undef when the operand cannot be zero in InstCombine

Summary: Also add popcount(n) == bitsize(n)  -> n == -1 transformation.

Reviewers: majnemer, spatel

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D23134

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279141 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstCombine] use m_APInt to allow icmp (trunc X, Y), C folds for splat constant...
Sanjay Patel [Thu, 18 Aug 2016 20:28:54 +0000 (20:28 +0000)]
[InstCombine] use m_APInt to allow icmp (trunc X, Y), C folds for splat constant vectors

This is a sibling of:
https://reviews.llvm.org/rL278859
https://reviews.llvm.org/rL278935
https://reviews.llvm.org/rL278945
https://reviews.llvm.org/rL279066
https://reviews.llvm.org/rL279077
https://reviews.llvm.org/rL279101

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279133 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstCombine] clean up foldICmpTruncConstant(); NFCI
Sanjay Patel [Thu, 18 Aug 2016 20:25:16 +0000 (20:25 +0000)]
[InstCombine] clean up foldICmpTruncConstant(); NFCI

1. Fix variable names
2. Add local variables to reduce code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279132 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SelectionDAG] Rename fextend -> fpextend, fround -> fpround, frnd -> fround
Michael Kuperstein [Thu, 18 Aug 2016 20:08:15 +0000 (20:08 +0000)]
[SelectionDAG] Rename fextend -> fpextend, fround -> fpround, frnd -> fround

The names of the tablegen defs now match the names of the ISD nodes.
This makes the world a slightly saner place, as previously "fround" matched
ISD::FP_ROUND and not ISD::FROUND.

Differential Revision: https://reviews.llvm.org/D23597

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279129 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU : Fix QSAD and MQSAD instructions' incorrect data type.
Wei Ding [Thu, 18 Aug 2016 19:51:14 +0000 (19:51 +0000)]
AMDGPU : Fix QSAD and MQSAD instructions' incorrect data type.

Differential Revision: http://reviews.llvm.org/D23689

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279126 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SLP] Initialize VectorizedValue when gathering
Matthew Simpson [Thu, 18 Aug 2016 19:50:32 +0000 (19:50 +0000)]
[SLP] Initialize VectorizedValue when gathering

We abort building vectorizable trees in some cases (e.g., if the maximum
recursion depth is reached, if the region size is too large, etc.). If this
happens for a reduction, we can be left with a root entry that needs to be
gathered. For these cases, we need make sure we actually set VectorizedValue to
the resulting vector.

This patch ensures we properly set VectorizedValue, and it also ensures the
insertelement sequence generated for the gathers is inserted at the correct
location.

Reference: https://llvm.org/bugs/show_bug.cgi?id=28330
Differential Revison: https://reviews.llvm.org/D23410

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279125 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRegScavenging: Add scavengeRegisterBackwards()
Matthias Braun [Thu, 18 Aug 2016 19:47:59 +0000 (19:47 +0000)]
RegScavenging: Add scavengeRegisterBackwards()

Re-apply r276044 with off-by-1 instruction fix for the reload placement.

This is a variant of scavengeRegister() that works for
enterBasicBlockEnd()/backward(). The benefit of the backward mode is
that it is not affected by incomplete kill flags.

This patch also changes
PrologEpilogInserter::doScavengeFrameVirtualRegs() to use the register
scavenger in backwards mode.

Differential Revision: http://reviews.llvm.org/D21885

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279124 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoBranch Folding: Accept explicit threshold for tail merge size.
Kyle Butt [Thu, 18 Aug 2016 18:57:29 +0000 (18:57 +0000)]
Branch Folding: Accept explicit threshold for tail merge size.

This is prep work for allowing the threshold to be different during layout,
and to enforce a single threshold between merging and duplicating during
layout. No observable change intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279117 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdd a version of Intrinsic::getName which is more efficient when there are no overloads.
Pete Cooper [Thu, 18 Aug 2016 18:30:54 +0000 (18:30 +0000)]
Add a version of Intrinsic::getName which is more efficient when there are no overloads.

When running 'opt -O2 verify-uselistorder-nodbg.lto.bc', there are 33m allocations.  8.2m
come from std::string allocations in Intrinsic::getName().  Turns out this method only
returns a std::string because it needs to handle overloads, but that is not the common case.

This adds an overload of getName which just returns a StringRef when there are no overloads
and so saves on the allocations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279113 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][SSE] Missed insertps shuffle patterns
Simon Pilgrim [Thu, 18 Aug 2016 18:19:28 +0000 (18:19 +0000)]
[X86][SSE] Missed insertps shuffle patterns

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279111 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CMake] Silence message on multi-configuration generators
Chris Bieneman [Thu, 18 Aug 2016 18:17:28 +0000 (18:17 +0000)]
[CMake] Silence message on multi-configuration generators

The Xcode and Visual Studio generators always log "-- No build type selected, default to Debug". This is because CMake doesn't initialize "CMAKE_CONFIGURATION_TYPES" until the generator's EnableLanguage call gets hit.

The first place EnableLanguage gets hit in our configuration is in the project() call. Since CMAKE_BUILD_TYPE isn't used until after we call project() it is safe to just move this check down a bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279110 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[asan] Extend test
Vitaly Buka [Thu, 18 Aug 2016 18:17:19 +0000 (18:17 +0000)]
[asan] Extend test

Summary: PR27453

Reviewers: eugenis

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D23647

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279109 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AMDGPU] add s_incperflevel/s_decperflevel intrinsics.
Valery Pykhtin [Thu, 18 Aug 2016 18:06:20 +0000 (18:06 +0000)]
[AMDGPU] add s_incperflevel/s_decperflevel intrinsics.

Differential revision: https://reviews.llvm.org/D23666

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279106 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix SystemZ compilation abort caused by negative AND mask
Elliot Colp [Thu, 18 Aug 2016 18:04:26 +0000 (18:04 +0000)]
Fix SystemZ compilation abort caused by negative AND mask

Normally, when an AND with a constant is lowered to NILL, the constant value is truncated to 16 bits. However, since r274066, ANDs whose results are used in a shift are caught by a different pattern that does not truncate. The instruction printer expects a 16-bit unsigned immediate operand for NILL, so this results in an abort.

This patch adds code to manually truncate the constant in this situation. The rest of the bits are then set, so we will detect a case for NILL "naturally" rather than using peephole optimizations.

Differential Revision: http://reviews.llvm.org/D21854

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279105 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAArch64: Don't call getIterator() on iterators
Duncan P. N. Exon Smith [Thu, 18 Aug 2016 17:58:09 +0000 (17:58 +0000)]
AArch64: Don't call getIterator() on iterators

Remove an unnecessary round-trip:

    iterator => operator->() => getIterator()

In some cases, the iterator is end(), so the dereference of operator->
is invalid (UB).

The testcase only crashes with r278974 (currently reverted to
investigate this), which adds an assertion for invalid dereferences of
ilist nodes.

Fixes PR29035.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279104 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LLVM] Fix some Clang-tidy modernize-use-using and Include What You Use warnings
Eugene Zelenko [Thu, 18 Aug 2016 17:56:27 +0000 (17:56 +0000)]
[LLVM] Fix some Clang-tidy modernize-use-using and Include What You Use warnings

Differential revision: https://reviews.llvm.org/D23675

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279102 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstCombine] use m_APInt to allow icmp (udiv X, Y), C folds for splat constant vectors
Sanjay Patel [Thu, 18 Aug 2016 17:55:59 +0000 (17:55 +0000)]
[InstCombine] use m_APInt to allow icmp (udiv X, Y), C folds for splat constant vectors

This is a sibling of:
https://reviews.llvm.org/rL278859
https://reviews.llvm.org/rL278935
https://reviews.llvm.org/rL278945
https://reviews.llvm.org/rL279066
https://reviews.llvm.org/rL279077

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279101 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[WebAssembly] Disable the store-results optimization.
Dan Gohman [Thu, 18 Aug 2016 17:51:27 +0000 (17:51 +0000)]
[WebAssembly] Disable the store-results optimization.

The WebAssemly spec removing the return value from store instructions, so
remove the associated optimization from LLVM.

This patch leaves the store instruction operands in place for now, so stores
now always write to "$drop"; these will be removed in a seperate patch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279100 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Assumptions] Make collecting ephemeral values not quadratic in the
Chandler Carruth [Thu, 18 Aug 2016 17:51:24 +0000 (17:51 +0000)]
[Assumptions] Make collecting ephemeral values not quadratic in the
number of assume intrinsics.

The classical way to have a cache-friendly vector style container when
we need queue semantics for BFS instead of stack semantics for DFS is to
use an ever-growing vector and an index. Erasing from the front requires
O(size) work, and unless we expect the worklist to grow *very* large,
its probably cheaper to just grow and race down the list.

But that makes it more bad that we're putting the assume intrinsics in
this at all. We end up looking at the (by definition empty) use list to
see if they're ephemeral (when we've already put them in that set), etc.

Instead, directly populate the worklist with the operands when we mark
the assume intrinsics as ephemeral. Also, test the visited set *before*
putting things into the worklist so we don't accumulate the same value
in the list 100s of times.

It would be nice to use a set-vector for this but I think its useful to
test the set earlier to avoid repeatedly querying whether the same
instruction is safe to speculate.

Hopefully with these changes the number of values pushed onto the
worklist is smaller, and we avoid quadratic work by letting it grow as
necessary.

Differential Revision: https://reviews.llvm.org/D23396

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279099 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix -Wpessimizing-move error, NFC
Vedant Kumar [Thu, 18 Aug 2016 17:39:53 +0000 (17:39 +0000)]
Fix -Wpessimizing-move error, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279095 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstCombine] clean up foldICmpUDivConstant; NFC
Sanjay Patel [Thu, 18 Aug 2016 17:37:26 +0000 (17:37 +0000)]
[InstCombine] clean up foldICmpUDivConstant; NFC

1. Better variable names
2. Remove unnecessary check of ConstantInt

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279094 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoReapply "ADT: Remove references in has_rbegin for reverse()"
Duncan P. N. Exon Smith [Thu, 18 Aug 2016 17:15:25 +0000 (17:15 +0000)]
Reapply "ADT: Remove references in has_rbegin for reverse()"

This reverts commit r279086, reapplying r279084.  I'm not sure what I
ran before, because the compile failure for ADTTests reproduced locally.

The problem is that TestRev is calling BidirectionalVector::rbegin()
when the BidirectionalVector is const, but rbegin() is always non-const.
I've updated BidirectionalVector::rbegin() to be callable from const.

Original commit message follows.

--

As a follow-up to r278991, add some tests that check that
decltype(reverse(R).begin()) == decltype(R.rbegin()), and get them
passing by adding std::remove_reference to has_rbegin.

I'm using static_assert instead of EXPECT_TRUE (and updated the other
has_rbegin check from r278991 in the same way) since I figure that's
more helpful.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279091 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoResubmit "Write the TPI stream from a PDB to Yaml."
Zachary Turner [Thu, 18 Aug 2016 16:49:29 +0000 (16:49 +0000)]
Resubmit "Write the TPI stream from a PDB to Yaml."

The original patch was breaking some buildbots due to an
incorrect ordering of function definitions which caused some
compilers to recognize a definition but others to not.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279089 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agollvm-objdump: add coff import library symbol listing support
Saleem Abdulrasool [Thu, 18 Aug 2016 16:39:19 +0000 (16:39 +0000)]
llvm-objdump: add coff import library symbol listing support

This adds behaviour similar to binutils' objdump which can show symbols in an
import library.  Differences from that stem around the fact that we do not
create section symbols nor the all import import descriptor symbol reference.
However, this does mean that the tool can serve as a possible replacement for
the existing tool.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279088 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "ADT: Remove references in has_rbegin for reverse()"
Duncan P. N. Exon Smith [Thu, 18 Aug 2016 16:27:41 +0000 (16:27 +0000)]
Revert "ADT: Remove references in has_rbegin for reverse()"

This reverts commit r279084, since it failed on a bot:
  http://bb.pgr.jp/builders/cmake-llvm-x86_64-linux/builds/41733

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279086 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoADT: Remove references in has_rbegin for reverse()
Duncan P. N. Exon Smith [Thu, 18 Aug 2016 16:22:54 +0000 (16:22 +0000)]
ADT: Remove references in has_rbegin for reverse()

As a follow-up to r278991, add some tests that check that
decltype(reverse(R).begin()) == decltype(R.rbegin()), and get them
passing by adding std::remove_reference to has_rbegin.

I'm using static_assert instead of EXPECT_TRUE (and updated the other
has_rbegin check from r278991 in the same way) since I figure that's
more helpful.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279084 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoCVP. Turn marking adds as no wrap (introduced by r278107) off by default
Artur Pilipenko [Thu, 18 Aug 2016 16:08:35 +0000 (16:08 +0000)]
CVP. Turn marking adds as no wrap (introduced by r278107) off by default

It causes a regression on our internal benchmark. Introduce cvp-dont-process flag and set it off by default while investigating the regression.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279082 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AArch64][GlobalISel] Select floating-point binary ops.
Ahmed Bougacha [Thu, 18 Aug 2016 16:05:11 +0000 (16:05 +0000)]
[AArch64][GlobalISel] Select floating-point binary ops.

There is no FREM instruction, but the others are straightforward.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279081 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[GlobalISel] Add floating-point binary ops.
Ahmed Bougacha [Thu, 18 Aug 2016 16:05:06 +0000 (16:05 +0000)]
[GlobalISel] Add floating-point binary ops.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279080 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[IRCE] Switch over to LLVM_DUMP_METHOD. NFCI.
Davide Italiano [Thu, 18 Aug 2016 15:55:49 +0000 (15:55 +0000)]
[IRCE] Switch over to LLVM_DUMP_METHOD. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279079 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ARM] Correct ARMv8*-A optional extension definitions in TargetParser
Richard Barton [Thu, 18 Aug 2016 15:50:11 +0000 (15:50 +0000)]
[ARM] Correct ARMv8*-A optional extension definitions in TargetParser

The ARMv8*-A descriptions in the ARM and AArch64 TargetParsers are incorrect
architecturally and mismatched to the backend descriptions.

RAS is an optional extension to ARMv8-A and ARMv8.1-A and mandatory in
ARMv8.2-A. Correct the ARMTargetParser descriptions which had this as enabled
by default in the earlier versions.

The FP16 and SPE extensions are optional in ARMv8.2-A and the backend defaults
them as off. They are not available as extensions to earlier ARMv8-A versions.
Correct the AArch64TargetParser which had these as enabled by default in all
ARMv8-A definitions.

These macros are only used to define preprocessor macros. There are no macros
yet as ACLE has not caught up with ARMv8.2-A so not possible to add a test.

Differential Revision: https://reviews.llvm.org/D23500

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279078 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstCombine] use m_APInt to allow icmp (mul X, Y), C folds for splat constant vectors
Sanjay Patel [Thu, 18 Aug 2016 15:44:44 +0000 (15:44 +0000)]
[InstCombine] use m_APInt to allow icmp (mul X, Y), C folds for splat constant vectors

This is a sibling of:
https://reviews.llvm.org/rL278859
https://reviews.llvm.org/rL278935
https://reviews.llvm.org/rL278945
https://reviews.llvm.org/rL279066

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279077 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[WebAssembly] Refactor WebAssemblyLowerEmscriptenException pass for setjmp/longjmp
Derek Schuff [Thu, 18 Aug 2016 15:27:25 +0000 (15:27 +0000)]
[WebAssembly] Refactor WebAssemblyLowerEmscriptenException pass for setjmp/longjmp

This patch changes the code structure of
WebAssemblyLowerEmscriptenException pass to support both exception
handling and setjmp/longjmp. It also changes the name of the pass and
the source file.

1. Change the file/pass name to WebAssemblyLowerEmscriptenExceptions ->
WebAssemblyLowerEmscriptenEHSjLj to make it clear that it supports both
EH and SjLj
2. List function / global variable names at the top so they
can be changed easily
3. Some cosmetic changes

Patch by Heejin Ahn

Differential Revision: https://reviews.llvm.org/D23588

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279075 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AArch64][GlobalISel] Select G_SDIV/G_UDIV.
Ahmed Bougacha [Thu, 18 Aug 2016 15:17:13 +0000 (15:17 +0000)]
[AArch64][GlobalISel] Select G_SDIV/G_UDIV.

There is no REM instruction; that will require an expansion.
It's not obvious that should be done in select, rather than as a
(custom?) legalization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279074 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[GlobalISel] Add support for DIV/REM.
Ahmed Bougacha [Thu, 18 Aug 2016 15:17:01 +0000 (15:17 +0000)]
[GlobalISel] Add support for DIV/REM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279073 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstCombine] use APInt in isSignTest instead of ConstantInt; NFC
Sanjay Patel [Thu, 18 Aug 2016 14:59:14 +0000 (14:59 +0000)]
[InstCombine] use APInt in isSignTest instead of ConstantInt; NFC

This will enable vector splat folding, but NFC until the callers
have their ConstantInt restrictions removed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279072 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agollvm-readobj: handle import libraries with -coff-exports
Saleem Abdulrasool [Thu, 18 Aug 2016 14:32:11 +0000 (14:32 +0000)]
llvm-readobj: handle import libraries with -coff-exports

`link -dump -exports` lists exported symbols from import libraries as well as
normal dlls.  Ensure that we can handle import libraries as well in
llvm-readobj.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279069 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agofix typo; NFC
Sanjay Patel [Thu, 18 Aug 2016 14:17:34 +0000 (14:17 +0000)]
fix typo; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279068 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Hexagon] Create vcombine in HexagonCopyToCombine
Krzysztof Parzyszek [Thu, 18 Aug 2016 14:12:34 +0000 (14:12 +0000)]
[Hexagon] Create vcombine in HexagonCopyToCombine

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279067 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstCombine] use m_APInt to allow icmp (xor X, Y), C folds for splat constant vectors
Sanjay Patel [Thu, 18 Aug 2016 14:10:48 +0000 (14:10 +0000)]
[InstCombine] use m_APInt to allow icmp (xor X, Y), C folds for splat constant vectors

This is a sibling of:
https://reviews.llvm.org/rL278859
https://reviews.llvm.org/rL278935
https://reviews.llvm.org/rL278945

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279066 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][SSE] Add SSE1 tests to make sure we don't merge loads on illegal types
Simon Pilgrim [Thu, 18 Aug 2016 13:41:26 +0000 (13:41 +0000)]
[X86][SSE] Add SSE1 tests to make sure we don't merge loads on illegal types

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279065 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[mips] Correct tail call encoding for MIPSR6
Simon Dardis [Thu, 18 Aug 2016 13:22:43 +0000 (13:22 +0000)]
[mips] Correct tail call encoding for MIPSR6

r277708 enabled tails calls for MIPS but used the 'jr' instruction when the
jump target was held in a register. For MIPSR6, 'jalr $zero, $reg' should
have been used. Additionally, add missing patterns for external and global
symbols for tail calls.

Reviewers: dsanders, vkalintiris

Differential Review: https://reviews.llvm.org/D23301

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279064 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Reassociate] Add test for PR28367.
Chad Rosier [Thu, 18 Aug 2016 13:22:37 +0000 (13:22 +0000)]
[Reassociate] Add test for PR28367.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279063 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago(Trivial) TargetPassConfig: assert when TargetMachine has no MCAsmInfo
Alex Bradbury [Thu, 18 Aug 2016 13:08:58 +0000 (13:08 +0000)]
(Trivial) TargetPassConfig: assert when TargetMachine has no MCAsmInfo

Summary:
This is a pretty trivial, but I thought it was worth just checking that nobody feels it's completely the wrong thing to be doing.

The motivation is that when starting a new backend, you often start with a minimal stub, pretty much just FooTargetMachine and FooTargetInfo. Once that's built, you might naturally try `llc -march=foo myinput.ll` and it seems more developer-friendly if this ends up asserting due to the lack of MCAsmInfo with an informative message rather than just segfaulting.

Reviewers: MatzeB, chandlerc

Subscribers: bogner, llvm-commits

Differential Revision: https://reviews.llvm.org/D23443

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279061 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRemove trailing whitespace
Simon Pilgrim [Thu, 18 Aug 2016 11:22:22 +0000 (11:22 +0000)]
Remove trailing whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279054 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "ADT: Remove UB in ilist (and use a circular linked list)"
Diana Picus [Thu, 18 Aug 2016 11:17:53 +0000 (11:17 +0000)]
Revert "ADT: Remove UB in ilist (and use a circular linked list)"

This reverts commit r278974 which broke some of our bots (e.g.
clang-cmake-aarch64-42vma, clang-cmake-aarch64-full).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279053 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "ADT: Tidy up ilist_traits static asserts, NFC"
Diana Picus [Thu, 18 Aug 2016 11:17:47 +0000 (11:17 +0000)]
Revert "ADT: Tidy up ilist_traits static asserts, NFC"

This reverts commit r279012.
r278974 broke some bots, I have to revert this to get to it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279052 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Docs] More MemorySSA doc updates!
George Burgess IV [Thu, 18 Aug 2016 02:56:05 +0000 (02:56 +0000)]
[Docs] More MemorySSA doc updates!

Thanks again to Michael Kuperstein for the feedback.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279032 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoTestcase for r279022
Matthias Braun [Thu, 18 Aug 2016 02:21:54 +0000 (02:21 +0000)]
Testcase for r279022

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279031 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert r279016 -- it breaks win32-elf JIT tests.
Lang Hames [Thu, 18 Aug 2016 01:33:28 +0000 (01:33 +0000)]
Revert r279016 -- it breaks win32-elf JIT tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279029 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[sanitizer-coverage/libFuzzer] instrument comparisons with __sanitizer_cov_trace_cmp...
Kostya Serebryany [Thu, 18 Aug 2016 01:25:28 +0000 (01:25 +0000)]
[sanitizer-coverage/libFuzzer] instrument comparisons with  __sanitizer_cov_trace_cmp[1248] instead of  __sanitizer_cov_trace_cmp, don't pass the comparison type to save a bit performance. Use these new callbacks in libFuzzer

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279027 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix bot failure due to new test
Teresa Johnson [Thu, 18 Aug 2016 01:18:15 +0000 (01:18 +0000)]
Fix bot failure due to new test

I had updated the output file name but not the corresponding nm based check
before submitting as r279023. This should fix the bot failures

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279025 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ThinLTO] Keep common symbols in ThinLTO modules
Teresa Johnson [Thu, 18 Aug 2016 01:08:50 +0000 (01:08 +0000)]
[ThinLTO] Keep common symbols in ThinLTO modules

Summary:
Skip the merging of common symbols for ThinLTO modules, they will be
merged by the final native object link. Trying to merge the symbols and
add to a combined module will incorrectly enable the common symbol to be
internalized in the ThinLTO module. Additionally, we will not want to
create a combined module for ThinLTO distributed builds.

This fixes failures in 7 cpu2006 benchmarks from the new LTO API in
ThinLTO mode.

Reviewers: mehdi_amini

Subscribers: pcc, llvm-commits, mehdi_amini

Differential Revision: https://reviews.llvm.org/D23637

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279023 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoTailDuplicator: Fix crash after r278974
Matthias Braun [Thu, 18 Aug 2016 00:59:32 +0000 (00:59 +0000)]
TailDuplicator: Fix crash after r278974

Some inputs would after r278974 without this fix (see
http://lab.llvm.org:8080/green/job/clang-stage2-cmake-RgSan_build/2733/console
for an example)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279022 91177308-0d34-0410-b5e6-96231b3b80d8