Commit Graph

36048 Commits

Author SHA1 Message Date
Mitch Bodart e60465ddf7 [X86] Enable the post-RA-scheduler for clang's default 32-bit cpu.
For compilations with no explicit cpu specified, this exhibits
nice gains on Silvermont, with neutral performance on big cores.

Differential Revision: http://reviews.llvm.org/D19138

llvm-svn: 267809
2016-04-27 22:52:35 +00:00
Kevin Enderby c493085c8d Fix a bug in llvm-objdump printing of 32-bit addresses for -section in non i386 and x86 files.
rdar://25896202

llvm-svn: 267807
2016-04-27 22:36:18 +00:00
Quentin Colombet bf200688de [X86][FastISel] Make sure we use the right register class when we select stores.
llvm-svn: 267806
2016-04-27 22:33:42 +00:00
Rong Xu 4b1dc5d60b Fix buildbot failure due to r267792
Relax the test check as some targets do not have name compression.

llvm-svn: 267803
2016-04-27 22:06:35 +00:00
Colin LeMahieu a3782da3e3 [Hexagon] Merging nops in to previous packet rather than always creating a new one.
llvm-svn: 267798
2016-04-27 21:37:44 +00:00
Quentin Colombet d6dbec4c6f [X86] Fix the lowering of TLS calls.
The callseq_end node must be glued with the TLS calls, otherwise,
the generic code will miss the uses of the returned value and will
mark it dead.
Moreover, TLSCall 64-bit pseudo must not set an implicit-use on RDI,
the pseudo uses the symbol address at this point not RDI and the
lowering will do the right thing.

llvm-svn: 267797
2016-04-27 21:37:37 +00:00
Rong Xu af5aebaa32 [PGO] Prohibit address recording if the function is both internal and COMDAT
Differential Revision: http://reviews.llvm.org/D19515

llvm-svn: 267792
2016-04-27 21:17:30 +00:00
Matt Arsenault 0547b016b1 AMDGPU: Account for globals in AMDGPUPromoteAlloca pass
Patch by Bas Nieuwenhuizen

llvm-svn: 267791
2016-04-27 21:05:08 +00:00
Kevin Enderby 60c4e6aafa Add a test case for the crash fixed with r267037. David Blaikie said it would be nice to have!
This was crashing llvm-objdump with -macho -objc-meta-data when trying dump a non-existent section.
So the test binary is simply created from an empty .s file compiled with: clang -arch armv7 empty.s -c

llvm-svn: 267782
2016-04-27 20:37:06 +00:00
Ahmed Bougacha 9e71425f54 [AArch64] Set correct successors in CMPXCHG pseudo expansion.
transferSuccessors() would LoadCmpBB a successor of DoneBB,
whereas it should be a successor of the original MBB.

Follow-up to r266339.

Unfortunately, it's tricky to catch this in the verifier.

llvm-svn: 267779
2016-04-27 20:33:02 +00:00
Ahmed Bougacha b4af107239 [ARM] Set correct successors in CMPXCHG pseudo expansion.
transferSuccessors() would LoadCmpBB a successor of DoneBB, whereas
it should be a successor of the original MBB.

The testcase changes are caused by Thumb2SizeReduction, which
was previously confused by the broken CFG.

Follow-up to r266679.

Unfortunately, it's tricky to catch this in the verifier.

llvm-svn: 267778
2016-04-27 20:32:54 +00:00
Simon Pilgrim 3f595aabe2 [InstCombine][AVX2] Add AVX2 per-element vector shift tests
At the moment we don't simplify PSRAV/PSRLV/PSLLV intrinsics to generic IR for constant shift amounts, but we could.

llvm-svn: 267777
2016-04-27 20:25:34 +00:00
Kevin B. Smith c378a99ba5 [X86]: Quit promoting 16 bit loads to 32 bit.
Differential Revision: http://reviews.llvm.org/D19592

llvm-svn: 267773
2016-04-27 19:58:03 +00:00
David Majnemer 0c80e2eac6 [CodeGenPrepare] Don't sink a cast past its user
The sink cast machinery is supposed to sink casts as close to their user
as possible.  However, an EH pad is the first instruction in it's basic
block.  Don't sink if the user is an EH pad.

This fixes PR27536.

llvm-svn: 267767
2016-04-27 19:36:38 +00:00
Ahmed Bougacha ace97c1f7d [LIR] Set attributes on memset_pattern16.
"inferattrs" will deduce the attribute, but it will be too late for
many optimizations. Set it ourselves when creating the call.

Differential Revision: http://reviews.llvm.org/D17598

llvm-svn: 267762
2016-04-27 19:04:50 +00:00
Ahmed Bougacha 44c19876c7 [InferAttrs] Mark memset_pattern16 params nocapture.
Differential Revision: http://reviews.llvm.org/D19471

llvm-svn: 267760
2016-04-27 19:04:43 +00:00
Chad Rosier 03e1647d19 Revert "[AMDGPU][llvm-mc] Add support of TTMP quads. Rework M0 exclusion for SMRD."
This reverts commit r267733 due to a -Werror,-Wunused-function error.

llvm-svn: 267752
2016-04-27 18:29:11 +00:00
Matthew Simpson 622b95be7b [LV] Reallow positive-stride interleaved load groups with gaps
We previously disallowed interleaved load groups that may cause us to
speculatively access memory out-of-bounds (r261331). We did this by ensuring
each load group had an access corresponding to the first and last member.
Instead of bailing out for these interleaved groups, this patch enables us to
peel off the last vector iteration, ensuring that we execute at least one
iteration of the scalar remainder loop. This solution was proposed in the
review of the previous patch.

Differential Revision: http://reviews.llvm.org/D19487

llvm-svn: 267751
2016-04-27 18:21:36 +00:00
Marcin Koscielnicki 7efdca5622 [Mips] Add support for llvm.thread.pointer intrinsic.
This will be used to implement __builtin_thread_pointer in clang.

Differential Revision: http://reviews.llvm.org/D19569

llvm-svn: 267743
2016-04-27 17:21:49 +00:00
Gerolf Hoflehner 88017c08a6 [InstCombine] Sharpended test case in pr21210.ll
llvm-svn: 267742
2016-04-27 17:19:54 +00:00
Artem Tamazov 3896f8f83d [AMDGPU][llvm-mc] Add support of TTMP quads. Rework M0 exclusion for SMRD.
Added support of TTMP quads.
Reworked M0 exclusion machinery for SMRD and similar instructions
to enable usage of TTMP registers in those instructions as destinations.
Tests added.

Differential Revision: http://reviews.llvm.org/D19342

llvm-svn: 267733
2016-04-27 16:20:23 +00:00
Reid Kleckner 0336cc05e7 [PDB] Fix function names for private symbols in PDBs
Summary:
llvm-symbolizer wants to get linkage names of functions for historical
reasons. Linkage names are only recorded in the PDB for public symbols,
and the linkage name is apparently stored separately in some "public
symbol" record. We had a workaround in PDBContext which would look for
such symbols when the user requested linkage names.

However, when given an address that was truly in a private function and
public funciton, we would accidentally find nearby public symbols and
return those function names. The fix is to look for both function
symbols and public symbols and only prefer the public symbol name if the
addresses of the symbols agree.

Fixes PR27492

Reviewers: zturner

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D19571

llvm-svn: 267732
2016-04-27 16:10:29 +00:00
Nicolai Haehnle f66bdb5ea8 AMDGPU/SI: Add llvm.amdgcn.s.waitcnt.all intrinsic
Summary:
So it appears that to guarantee some of the ordering requirements of a GLSL
memoryBarrier() executed in the shader, we need to emit an s_waitcnt.

(We can't use an s_barrier, because memoryBarrier() may appear anywhere in
the shader, in particular it may appear in non-uniform control flow.)

Reviewers: arsenm, mareko, tstellarAMD

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D19203

llvm-svn: 267729
2016-04-27 15:46:01 +00:00
Matthew Simpson e5dfb08fcb [TTI] Add hook for vector extract with extension
This change adds a new hook for estimating the cost of vector extracts followed
by zero- and sign-extensions. The motivating example for this change is the
SMOV and UMOV instructions on AArch64. These instructions move data from vector
to general purpose registers while performing the corresponding extension
(sign-extend for SMOV and zero-extend for UMOV) at the same time. For these
operations, TargetTransformInfo can assume the extensions are free and only
report the cost of the vector extract. The SLP vectorizer has been updated to
make use of the new hook.

Differential Revision: http://reviews.llvm.org/D18523

llvm-svn: 267725
2016-04-27 15:20:21 +00:00
Artem Tamazov 5cd55b1784 [AMDGPU][llvm-mc] s_getreg/setreg* - Support symbolic names of hardware registers.
Possibility to specify code of hardware register kept.
Disassemble to symbolic name, if name is known.
Tests updated/added.

Differential Revision: http://reviews.llvm.org/D19335

llvm-svn: 267724
2016-04-27 15:17:03 +00:00
Nico Weber e69b9548b8 Revert r267649, it caused PR27539.
llvm-svn: 267723
2016-04-27 15:16:54 +00:00
Kristof Beyls a14e9a5576 Remove size 1 from check as that isn't part of what the test is meant to be testing.
This test also runs on e.g. ARM-native builds when the X86 backend is also
built.  This test produces code for the default instruction set, even though it
is in a "X86" sub-directory. Given that this test doesn't seem to be testing
anything architecture-specific, it seems it's best to adapt the check to not
check for an architecture-dependent value (the size of the function), rather
than hard-code the test to target x86.

llvm-svn: 267722
2016-04-27 15:03:09 +00:00
Teresa Johnson 02e98331c0 [ThinLTO] Use valueid instead of bitcode offsets in combined index file
Summary:
With the removal of support for lazy parsing of combined index summary
records (e.g. r267344), we no longer need to include the summary record
bitcode offset in the VST entries for definitions. Change the combined
index format to be similar to the per-module index format in using value
ids to cross-reference from the summary record to the VST entry (rather
than the summary record bitcode offset to cross-reference in the other
direction).

The visible changes are:
1) Add the value id to the combined summary records
2) Remove the summary offset from the combined VST records, which has
the following effects:
- No longer need the VST_CODE_COMBINED_GVDEFENTRY record, as all
  combined index VST entries now only contain the value id and
  corresponding GUID.
- No longer have duplicate VST entries in the case where there are
  multiple definitions of a symbol (e.g. weak/linkonce), as they all
  have the same value id and GUID.

An implication of #2 above is that in order to hook up an alias to the
correct aliasee based on the value id of the aliasee recorded in the
combined index alias record, we need to scan the entries in the index
for that GUID to find the one from the same module (i.e. the case where
there are multiple entries for the aliasee). But the reader no longer
has to maintain a special map to hook up the alias/aliasee.

Reviewers: joker.eph

Subscribers: joker.eph, llvm-commits

Differential Revision: http://reviews.llvm.org/D19481

llvm-svn: 267712
2016-04-27 13:28:35 +00:00
Simon Pilgrim f23aa2a9c9 [InstCombine][SSE] Regenerated vector shift tests
llvm-svn: 267699
2016-04-27 12:04:44 +00:00
Zlatko Buljan de0bbe6d1c [mips][microMIPS] Add CodeGen support for SUBU16, SUB, SUBU, DSUB and DSUBU instructions
Differential Revision: http://reviews.llvm.org/D16676

llvm-svn: 267694
2016-04-27 11:31:44 +00:00
Zlatko Buljan 29813620bc [mips][microMIPS] Add CodeGen support for SLL16, SRL16, SLL, SLLV, SRA, SRAV, SRL and SRLV instructions
Differential Revision: http://reviews.llvm.org/D17989

llvm-svn: 267693
2016-04-27 11:02:23 +00:00
Artur Pilipenko c97eac6555 Use DL preferred alignment for alloca in Value::getPointerAlignment
Teach Value::getPointerAlignment that allocas with no explicit alignment are aligned to preferred alignment of the allocated type.

Reviewed By: hfinkel

Differential Revision: http://reviews.llvm.org/D17569

llvm-svn: 267689
2016-04-27 10:42:29 +00:00
Simon Pilgrim d2ea708739 [InstCombine][SSE] Added DemandedBits tests for MOVMSK instructions
MOVMSK zeros the upper bits of the gpr - we should be able to use this.

llvm-svn: 267686
2016-04-27 09:53:09 +00:00
Adam Nemet d2fa414718 [LoopDist] Add llvm.loop.distribute.enable loop metadata
Summary:
D19403 adds a new pragma for loop distribution.  This change adds
support for the corresponding metadata that the pragma is translated to
by the FE.

As part of this I had to rethink the flag -enable-loop-distribute.  My
goal was to be backward compatible with the existing behavior:

  A1. pass is off by default from the optimization pipeline
  unless -enable-loop-distribute is specified

  A2. pass is on when invoked directly from opt (e.g. for unit-testing)

The new pragma/metadata overrides these defaults so the new behavior is:

  B1. A1 + enable distribution for individual loop with the pragma/metadata

  B2. A2 + disable distribution for individual loop with the pragma/metadata

The default value whether the pass is on or off comes from the initiator
of the pass.  From the PassManagerBuilder the default is off, from opt
it's on.

I moved -enable-loop-distribute under the pass.  If the flag is
specified it overrides the default from above.

Then the pragma/metadata can further modifies this per loop.

As a side-effect, we can now also use -enable-loop-distribute=0 from opt
to emulate the default from the optimization pipeline.  So to be precise
this is the new behavior:

  C1. pass is off by default from the optimization pipeline
  unless -enable-loop-distribute or the pragma/metadata enables it

  C2. pass is on when invoked directly from opt
  unless -enable-loop-distribute=0 or the pragma/metadata disables it

Reviewers: hfinkel

Subscribers: joker.eph, mzolotukhin, llvm-commits

Differential Revision: http://reviews.llvm.org/D19431

llvm-svn: 267672
2016-04-27 05:28:18 +00:00
Mehdi Amini c7b950171d Revert "Support "preserving" the summary information when using setModule() API in LTOCodeGenerator"
This reverts commit r267665.
ASAN shows that there is a use of undefined value.

From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 267668
2016-04-27 05:11:44 +00:00
Mehdi Amini 360ed847bc Support "preserving" the summary information when using setModule() API in LTOCodeGenerator
Another attempt at r267655...

From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 267665
2016-04-27 04:24:10 +00:00
Mehdi Amini a1b8b6cd56 Revert "Support "preserving" the summary information when using setModule() API in LTOCodeGenerator"
This reverts commit r267657, r267656, and r267655.
The test does not pass on multiple bots, I'm unsure why yet but let's unbreak them.

From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 267664
2016-04-27 03:34:28 +00:00
Evgeny Stupachenko 23ce61b663 The patch fixes PR27392.
Summary:
 It is incorrect to compare TripCount (which is BECount + 1)
  with extraiters (or Count) to check if we should enter unrolled
  loop or not, because TripCount can potentially overflow
  (when BECount is max unsigned integer).
 While comparing BECount with (Count - 1) is overflow safe and
  therefore correct.

Reviewer: hfinkel

Differential Revision: http://reviews.llvm.org/D19256

From: Evgeny Stupachenko <evstupac@gmail.com>
llvm-svn: 267662
2016-04-27 03:04:54 +00:00
Chuang-Yu Cheng 8676c3d599 [ppc64] fix bug in prologue that mfocrf's cr operand should be explict state instead of implicit
This fixes PR27414

Reviewers: kbarton mgrang tjablin

http://reviews.llvm.org/D19255

llvm-svn: 267660
2016-04-27 02:59:28 +00:00
Mehdi Amini 4960919723 Fix the test from r267656: Support "preserving" the summary information when using setModule() API in LTOCodeGenerator
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 267657
2016-04-27 01:49:11 +00:00
Mehdi Amini 1f65039f06 Add a test for r267655: Support "preserving" the summary information when using setModule() API in LTOCodeGenerator
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 267656
2016-04-27 01:47:46 +00:00
Ahmed Bougacha 19a2ee591a [X86] Don't assume that MMX extractelts are from index 0.
It's probably the case for all 3 MMX users out there, but with
hand-crafted IR, you can trigger selection failures. Fix that.

llvm-svn: 267652
2016-04-27 01:35:29 +00:00
Ahmed Bougacha e68363a03c [X86] Re-enable MMX i32 extractelt combine.
This effectively adds back the extractelt combine removed by r262358:
the direct case can still occur (because x86_mmx is special, see
r262446), but it's the indirect case that's now superseded by the
generic combine.

llvm-svn: 267651
2016-04-27 01:35:25 +00:00
Cong Hou 6f879d9eb1 Detects the SAD pattern on X86 so that much better code will be emitted once the pattern is matched.
Differential revision: http://reviews.llvm.org/D14840

llvm-svn: 267649
2016-04-27 01:29:18 +00:00
Mehdi Amini b4e1e8297b ThinLTO: do not promote GlobalVariable that have a specific section.
Differential Revision: http://reviews.llvm.org/D18298

From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 267646
2016-04-27 00:32:13 +00:00
Mehdi Amini da168fbc2e LTOCodeGenerator: turns linkonce(_odr) into weak_(odr) when present "MustPreserve" set
Summary:
If the linker requested to preserve a linkonce function, we should
honor this even if we drop all uses.

Reviewers: dexonsmith

Subscribers: llvm-commits, joker.eph

Differential Revision: http://reviews.llvm.org/D19527

From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 267644
2016-04-27 00:32:02 +00:00
Philip Reames 3f83dbeed9 [LVI] Reduce compile time by lazily scanning blocks if needed
When encountering a non-local pointer, LVI would eagerly scan the block for dereferences of the given object to prove the pointer to be non null.  That's all well and good, but *then* we'd go recurse through our input blocks.  As a result, we could end up scanning each and every block we traverse, even if the final definition was obviously non null or we found a constant value somewhere up the chain.  The previous code papered over this by using the isKnownNonNull routine from value tracking.  This made the duplication less painful in the common case.

Instead, we know do the block scan only *after* we've gotten the recursive results back.  This lets us stop scanning individual blocks as soon as we've determined it to be non-null in any predecessor block and use our usual merge rules to propagate that information cheaply through successor blocks.  For a pointer which can be found non-null, this does strictly less work and sometimes substaintially so.

Note that the case where we *can't* prove something non-null is still the really expensive case.  We end up scanning each and every block looking for a dereference and never end up finding one.

llvm-svn: 267642
2016-04-27 00:30:55 +00:00
Quentin Colombet 4ff3cfb673 [X86] Make sure it is safe to clobber EFLAGS, if need be, when choosing
the prologue.

Do not use basic blocks that have EFLAGS live-in as prologue if we need
to realign the stack. Realigning the stack uses AND instruction and this
clobbers EFLAGS.

An other alternative would have been to save and restore EFLAGS around
the stack realignment code, but this is likely inefficient.

Fixes PR27531.

llvm-svn: 267634
2016-04-26 23:44:14 +00:00
Justin Bogner c2bf63d29d PM: Port Reassociate to the new pass manager
llvm-svn: 267631
2016-04-26 23:39:29 +00:00
Mitch Bodart 807e13379b [X86] Replace -mcpu with -mattr in several tests
Differential Revision: http://reviews.llvm.org/D19568

llvm-svn: 267629
2016-04-26 23:36:38 +00:00
Sanjay Patel 29dea0d230 [SimplifyCFG] propagate branch metadata when creating select
llvm-svn: 267624
2016-04-26 23:15:48 +00:00
Quentin Colombet 08e79990a0 [MachineBasicBlock] Take advantage of the partially dead information.
Thanks to that information we wouldn't lie on a register being live whereas it
is not.

llvm-svn: 267622
2016-04-26 23:14:29 +00:00
Quentin Colombet 3f19245015 [MachineInstrBundle] Improvement the recognition of dead definitions.
Now, it is possible to know that partial definitions are dead definitions and
recognize that clobbered registers are also dead.

llvm-svn: 267621
2016-04-26 23:14:24 +00:00
Philip Reames 053c2a6f25 [LVI] Apply transfer rule for overdefine inputs for binary operators
As pointed out by John Regehr over in http://reviews.llvm.org/D19485, LVI was being incredibly stupid about applying its transfer rules.  Rather than gathering local facts from the expression itself, it was simply giving up entirely if one of the inputs was overdefined.  This greatly impacts the precision of the overall analysis and makes it far more fragile as well.

This patch builds on 267609 which did the same thing for unary casts.

llvm-svn: 267620
2016-04-26 23:10:35 +00:00
Philip Reames e5030e85ea [LVI] A better fix for the assertion error introduced by 267609
Essentially, I was using the wrong size function.  For types which were sized, but not primitive, I wasn't getting a useful size for the operand and failed an assert.  I fixed this, and also added a guard that the input is a sized type.  Test case is for the original mistake.  I'm not sure how to actually exercise the sized type check.

llvm-svn: 267618
2016-04-26 22:52:30 +00:00
Sanjay Patel d2d2aa52cd [LowerExpectIntrinsic] make default likely/unlikely ratio bigger
We need the default ratio to be sufficiently large that it triggers transforms 
based on block frequency info (BFI) and plays well with the recently introduced
BranchProbability used by CGP.

Differential Revision: http://reviews.llvm.org/D19435

llvm-svn: 267615
2016-04-26 22:23:38 +00:00
Philip Reames 38c87c2e50 [LVI] Infer local facts from unary expressions
As pointed out by John Regehr over in http://reviews.llvm.org/D19485, LVI was being incredibly stupid about applying its transfer rules. Rather than gathering local facts from the expression itself, it was simply giving up entirely if one of the inputs was overdefined. This greatly impacts the precision of the overall analysis and makes it far more fragile as well.

This patch implements only the unary operation case. Once this is in, I'll implement the same for the binary operations.

Differential Revision: http://reviews.llvm.org/D19492

llvm-svn: 267609
2016-04-26 21:48:16 +00:00
David Majnemer abb9f55c80 Revert "[SimplifyLibCalls] sprintf doesn't copy null bytes"
The destination buffer that sprintf uses is restrict qualified, we do
not need to worry about derived pointers referenced via format
specifiers.

This reverts commit r267580.

llvm-svn: 267605
2016-04-26 21:04:47 +00:00
Nico Weber 2f1459cbb7 Try to get ResponseFile.ll passing on Windows after r267556.
llvm-svn: 267599
2016-04-26 20:32:51 +00:00
Elena Demikhovsky 308a7eb0d2 Masked Store in Loop Vectorizer - bugfix
Fixed a bug in loop vectorization with conditional store.

Differential Revision: http://reviews.llvm.org/D19532

llvm-svn: 267597
2016-04-26 20:18:04 +00:00
Justin Bogner 4563a06cee PM: Port Internalize to the new pass manager
llvm-svn: 267596
2016-04-26 20:15:52 +00:00
Zachary Turner 53a65ba5c9 Parse and dump PDB DBI Stream Header Information
The DBI stream contains a lot of bookkeeping information for other
streams. In particular it contains information about section contributions
and linked modules. This patch is a first attempt at parsing some of the
information out of the DBI stream. It currently only parses and dumps the
headers of the DBI stream, so none of the module data or section
contribution data is pulled out.

This is just a proof of concept that we understand the basic properties of
the DBI stream's metadata, and followup patches will try to extract more
detailed information out.

Differential Revision: http://reviews.llvm.org/D19500
Reviewed By: majnemer, ruiu

llvm-svn: 267585
2016-04-26 18:42:34 +00:00
Krzysztof Parzyszek 4773f647bd [Tail duplication] Handle source registers with subregisters
When a block is tail-duplicated, the PHI nodes from that block are
replaced with appropriate COPY instructions. When those PHI nodes
contained use operands with subregisters, the subregisters were
dropped from the COPY instructions, resulting in incorrect code.

Keep track of the subregister information and use this information
when remapping instructions from the duplicated block.

Differential Revision: http://reviews.llvm.org/D19337

llvm-svn: 267583
2016-04-26 18:36:34 +00:00
Tim Northover 4397837be2 Reapply: "ARM: put correct symbol index on indirect pointers in __thread_ptr.""
A latent bug in llvm-objdump used the wrong format specifier on 32-bit
targets, causing the test to fail. This fixes the issue.

llvm-svn: 267582
2016-04-26 18:29:16 +00:00
David Majnemer 8cd77baebc [SimplifyLibCalls] sprintf doesn't copy null bytes
sprintf doesn't read or copy the terminating null byte from it's string
operands.  sprintf will append it's own after processing all of the
format specifiers.

This fixes PR27526.

llvm-svn: 267580
2016-04-26 18:16:49 +00:00
Manman Ren 1c3f65a18c Swift Calling Convention: use %RAX for sret.
We don't need to copy the sret argument into %rax upon return.
rdar://25671494

llvm-svn: 267579
2016-04-26 18:08:06 +00:00
Saleem Abdulrasool 4c6c4e2bbb tests: tweak MIR for ARM tests to correct MI issues
The Machine Instruction Verifier flagged some issues in the serialized MIR.
Adjust the input to correct them.

Fixes the remaining portion of PR27480.

llvm-svn: 267578
2016-04-26 17:54:21 +00:00
Saleem Abdulrasool 601e029ba3 test: remove some bleeding whitespace
Kill bleeding whitespace.  NFC

llvm-svn: 267577
2016-04-26 17:54:16 +00:00
Sanjay Patel d66607bd8c [CodeGenPrepare] use branch weight metadata to decide if a select should be turned into a branch
This is part of solving PR27344:
https://llvm.org/bugs/show_bug.cgi?id=27344

CGP should undo the SimplifyCFG transform for the same reason that earlier patches have used this
same mechanism: it's possible that passes between SimplifyCFG and CGP may be able to optimize the
IR further with a select in place.

For the TLI hook default, >99% taken or not taken is chosen as the default threshold for a highly
predictable branch. Even the most limited HW branch predictors will be correct on this branch almost
all the time, so even a massive mispredict penalty perf loss would be overcome by the win from all
the times the branch was predicted correctly.

As a follow-up, we could make the default target hook less conservative by using the SchedMachineModel's
MispredictPenalty. Or we could just let targets override the default by implementing the hook with that
and other target-specific options. Note that trying to statically determine mispredict rates for 
close-to-balanced profile weight data is generally impossible if the HW is sufficiently advanced. Ie, 
50/50 taken/not-taken might still be 100% predictable.

Finally, note that this patch as-is will not solve PR27344 because the current __builtin_unpredictable()
branch weight default values are 4 and 64. A proposal to change that is in D19435.

Differential Revision: http://reviews.llvm.org/D19488

llvm-svn: 267572
2016-04-26 17:11:17 +00:00
Zachary Turner f34e01624a Refactor some more PDB reading code into DebugInfoPDB.
Differential Revision: http://reviews.llvm.org/D19445
Reviewed By: David Majnemer

llvm-svn: 267564
2016-04-26 16:20:00 +00:00
Konstantin Zhuravlyov 1d99c4d03c [AMDGPU] Reserve VGPRs for trap handler usage if instructed
Differential Revision: http://reviews.llvm.org/D19235

llvm-svn: 267563
2016-04-26 15:43:14 +00:00
Sam Kolton 3025e7f25f [AMDGPU] Assembler: basic support for SDWA instructions
Support for SDWA instructions for VOP1 and VOP2 encoding.
Not done yet:
  - converters for support optional operands and modifiers
  - VOPC
  - sext() modifier
  - intrinsics
  - VOP2b (see vop_dpp.s)
  - V_MAC_F32 (see vop_dpp.s)

Differential Revision: http://reviews.llvm.org/D19360

llvm-svn: 267553
2016-04-26 13:33:56 +00:00
Andrey Turetskiy b405606432 [X86] PR27502: Fix the LEA optimization pass.
Handle MachineBasicBlock as a memory displacement operand in the LEA optimization pass.

Differential Revision: http://reviews.llvm.org/D19409

llvm-svn: 267551
2016-04-26 12:18:12 +00:00
Marcin Koscielnicki 0cfb612413 [PowerPC] Add support for llvm.thread.pointer
Differential Revision: http://reviews.llvm.org/D19304

llvm-svn: 267546
2016-04-26 10:37:22 +00:00
Marcin Koscielnicki 33571e2c41 [SPARC] [SSP] Add support for LOAD_STACK_GUARD.
This fixes PR22248 on sparc.

Differential Revision: http://reviews.llvm.org/D19386

llvm-svn: 267545
2016-04-26 10:37:14 +00:00
Marcin Koscielnicki fafb44951a [SPARC] Add support for llvm.thread.pointer.
Differential Revision: http://reviews.llvm.org/D19387

llvm-svn: 267544
2016-04-26 10:37:01 +00:00
Mehdi Amini aa309b1a81 ThinLTOCodeGenerator: preserve linkonce when in "MustPreserved" set
If the linker specifically requested for a linkonce to be preserved,
we need to make sure we won't drop it even if all the uses in the
current module disappear.

From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 267543
2016-04-26 10:35:01 +00:00
Renato Golin 5a55a029c0 Revert "ARM: put correct symbol index on indirect pointers in __thread_ptr."
This reverts commit r267488, as it broke some ARM buildbots.

llvm-svn: 267541
2016-04-26 10:02:02 +00:00
Sanjoy Das 51df5fae4a Symbolize operand bundle blocks for bcanalyzer
Reviewers: joker.eph

Subscribers: mcrosier, llvm-commits

Differential Revision: http://reviews.llvm.org/D19523

llvm-svn: 267524
2016-04-26 05:59:08 +00:00
Craig Topper c5551bfc26 [AArch64] Expand v1i64 and v2i64 ctlz.
The default is legal, which results in 'Cannot select' errors.

llvm-svn: 267522
2016-04-26 05:26:51 +00:00
Craig Topper d8d6be4f99 [ARM] Expand vector ctlz_zero_undef so it becomes ctlz.
The default is Legal, which results in 'Cannot select' errors.

llvm-svn: 267521
2016-04-26 05:04:37 +00:00
Craig Topper edb4a6ba98 [ARM] Expand v1i64 and v2i64 ctlz.
The default is legal, which results in 'Cannot select' errors.

llvm-svn: 267520
2016-04-26 05:04:33 +00:00
Dehao Chen 5d6d4841ed Tune basic block annotation algorithm.
Summary:
Instead of using maximum IR weight as the basic block weight, this patch uses the voting algorithm to find the most likely weight for the basic block. This can effectively avoid the cases when some IRs are annotated incorrectly due to code motion of the profiled binary.

This patch also updates propagate.ll unittest to include discriminator in the input file so that it is testing something meaningful.

Reviewers: davidxl, dnovillo

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D19301

llvm-svn: 267519
2016-04-26 04:59:11 +00:00
Bill Seurer d6e92135bd [powerpc] mark JIT tests as UNSUPPORTED on powerpc64 big endian
Some of the JIT tests began failing with "[llvm] r266663 - [Orc] Re-commit 
r266581 with fixes for MSVC, and format cleanups." on powerpc64 big endian.  
To get the buildbots running I am marking these as UNSUPPORTED for now.

If this is fixed remove the UNSUPPORTED flag "powerpc64-unknown-linux-gnu".

In r267516 I marked these as XFAIL but they succeed on some of the bots
on stage1.

llvm-svn: 267518
2016-04-26 03:59:19 +00:00
Richard Trieu d7f31a31d1 Pass the test file in through stdin instead of by filename.
When passed in via filename, this test will fail if the path to the test
has the strings "f1" and "f2" in somewhere.  Pass the file through stdin
to prevent test failures due to coincidences in path names.

llvm-svn: 267517
2016-04-26 03:43:49 +00:00
Bill Seurer ab5171f988 [powerpc] mark JIT tests as XFAIL on powerpc64 big endian
Some of the JIT tests began failing with "[llvm] r266663 - [Orc] Re-commit 
r266581 with fixes for MSVC, and format cleanups." on powerpc64 big endian.  
To get the buildbots running I am marking these as XFAIL for now.

If this is fixed remove the XFAIL flag "powerpc64-unknown-linux-gnu".

llvm-svn: 267516
2016-04-26 02:33:22 +00:00
Hal Finkel e4c0c1679b [SimplifyCFG] Preserve !llvm.mem.parallel_loop_access when merging
When SimplifyCFG merges identical instructions from both sides of a diamond, it
can preserve !llvm.mem.parallel_loop_access (as it does with most of the other
metadata). There's no real data or control dependency change in this case.

llvm-svn: 267515
2016-04-26 02:06:06 +00:00
Hal Finkel 411d31ad72 [LoopVectorize] Don't consider conditional-load dereferenceability for marked parallel loops
I really thought we were doing this already, but we were not. Given this input:

void Test(int *res, int *c, int *d, int *p) {
  for (int i = 0; i < 16; i++)
    res[i] = (p[i] == 0) ? res[i] : res[i] + d[i];
}

we did not vectorize the loop. Even with "assume_safety" the check that we
don't if-convert conditionally-executed loads (to protect against
data-dependent deferenceability) was not elided.

One subtlety: As implemented, it will still prefer to use a masked-load
instrinsic (given target support) over the speculated load. The choice here
seems architecture specific; the best option depends on how expensive the
masked load is compared to a regular load. Ideally, using the masked load still
reduces unnecessary memory traffic, and so should be preferred. If we'd rather
do it the other way, flipping the order of the checks is easy.

The LangRef is updated to make explicit that llvm.mem.parallel_loop_access also
implies that if conversion is okay.

Differential Revision: http://reviews.llvm.org/D19512

llvm-svn: 267514
2016-04-26 02:00:36 +00:00
Dan Gohman f456290fca [WebAssembly] Account for implicit operands when computing operand indices.
llvm-svn: 267511
2016-04-26 01:40:56 +00:00
Sanjay Patel a31b0c0ece [CodeGenPrepare] don't convert an unpredictable select into control flow
Suggested in the review of D19488:
http://reviews.llvm.org/D19488

llvm-svn: 267504
2016-04-26 00:47:39 +00:00
Justin Bogner 1a07501379 PM: Port GlobalOpt to the new pass manager
llvm-svn: 267499
2016-04-26 00:28:01 +00:00
Ahmed Bougacha 5cf735a5b1 [X86] Use LivePhysRegs in X86FixupBWInsts.
Kill-flags, which computeRegisterLiveness uses, are not reliable.
LivePhysRegs is.

Differential Revision: http://reviews.llvm.org/D19472

llvm-svn: 267495
2016-04-26 00:00:48 +00:00
Justin Bogner 6f6c5f2a02 GlobalOpt: Convert a bunch of tests from grep to FileCheck
llvm-svn: 267493
2016-04-25 23:36:50 +00:00
Sanjay Patel 82059090d3 Add check for "branch_weights" with prof metadata
While we're here, fix the comment and variable names to make it
clear that these are raw weights, not percentages.

llvm-svn: 267491
2016-04-25 23:15:16 +00:00
James Y Knight 51208eaccc [Sparc] Fix double-float fabs and fneg on little endian CPUs.
The SparcV8 fneg and fabs instructions interestingly come only in a
single-float variant. Since the sign bit is always the topmost bit no
matter what size float it is, you simply operate on the high
subregister, as if it were a single float.

However, the layout of double-floats in the float registers is reversed
on little-endian CPUs, so that the high bits are in the second
subregister, rather than the first.

Thus, this expansion must check the endianness to use the correct
subregister.

llvm-svn: 267489
2016-04-25 22:54:09 +00:00
Tim Northover cbba0aba16 ARM: put correct symbol index on indirect pointers in __thread_ptr.
Otherwise the linker has no idea what should be resolved.

llvm-svn: 267488
2016-04-25 22:36:07 +00:00
Arch D. Robison be0490a6e8 Optimize store of "bitcast" from vector to aggregate.
This patch is what was the "instcombine" portion of D14185, with an additional 
test added (see julia_pseudovec in test/Transforms/InstCombine/insert-val-extract-elem.ll). 
The patch causes instcombine to replace sequences of extractelement-insertvalue-store 
that act essentially like a bitcast followed by a store.

Differential review: http://reviews.llvm.org/D14260

llvm-svn: 267482
2016-04-25 22:22:39 +00:00
Tim Northover 5c3140f745 ARM: put extern __thread stubs in a special section.
The linker needs to know that the symbols are thread-local to do its job
properly.

llvm-svn: 267473
2016-04-25 21:12:04 +00:00
Quentin Colombet abe2d016cf Re-apply r267206 with a fix for the encoding problem: when the immediate of
log2(Mask) is smaller than 32, we must use the 32-bit variant because the 64-bit
variant cannot encode it. Therefore, set the subreg part accordingly.

[AArch64] Fix optimizeCondBranch logic.

The opcode for the optimized branch does not depend on the size
of the activate bits in the AND masks, but the AND opcode itself.
Indeed, we need to use a X or W variant based on the AND variant
not based on whether the mask fits into the related variant.
Otherwise, we may end up using the W variant of the optimized branch
for 64-bit register inputs!

This fixes the last make check verifier issues for AArch64: PR27479.

llvm-svn: 267465
2016-04-25 20:54:08 +00:00
Matt Arsenault 99c14524ec AMDGPU: Implement addrspacecast
llvm-svn: 267452
2016-04-25 19:27:24 +00:00