oota-llvm.git
9 years ago[mips][ias] Implement macro expansion when bcc has an immediate where a register...
Daniel Sanders [Mon, 12 Oct 2015 14:24:05 +0000 (14:24 +0000)]
[mips][ias] Implement macro expansion when bcc has an immediate where a register belongs.

Summary: Fixes PR24915.

Reviewers: vkalintiris

Subscribers: emaste, seanbruno, llvm-commits

Differential Revision: http://reviews.llvm.org/D13533

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250042 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips] Whitespace cleanup in MIPS16 tests to reduce noise in following changes. NFC.
Daniel Sanders [Mon, 12 Oct 2015 14:16:52 +0000 (14:16 +0000)]
[mips] Whitespace cleanup in MIPS16 tests to reduce noise in following changes. NFC.

Mostly tabs -> spaces and double spacing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250041 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips] Clean up most macro expansions to use the emit*() functions.
Daniel Sanders [Mon, 12 Oct 2015 14:09:12 +0000 (14:09 +0000)]
[mips] Clean up most macro expansions to use the emit*() functions.

Reviewers: vkalintiris

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D13591

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250040 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips] Handle undef when extracting subregs from FP64 registers.
Daniel Sanders [Mon, 12 Oct 2015 13:55:44 +0000 (13:55 +0000)]
[mips] Handle undef when extracting subregs from FP64 registers.

Summary:
This removes unnecessary instructions when extracting from an undefined register
and also fixes a crash for O32 when passing undef to a double argument in
held in integer registers.

Reviewers: vkalintiris

Subscribers: llvm-commits, zoran.jovanovic, petarj

Differential Revision: http://reviews.llvm.org/D13467

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250039 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoGlobalOpt does not treat externally_initialized globals correctly
Oliver Stannard [Mon, 12 Oct 2015 13:20:52 +0000 (13:20 +0000)]
GlobalOpt does not treat externally_initialized globals correctly

GlobalOpt currently merges stores into the initialisers of internal,
externally_initialized globals, but should not do so as the value of the global
may change between the initialiser and any code in the module being run.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250035 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[ARM] Mark Swift MISched model as incomplete
James Molloy [Mon, 12 Oct 2015 12:49:59 +0000 (12:49 +0000)]
[ARM] Mark Swift MISched model as incomplete

The Swift Machine Scheduler Model is incomplete. There are instructions
missing which can trigger the "incomplete machine model" abort. This was
observed when a downstream SchedMachineModel was added to the ARM
target.

Patch by Christof Douma!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250033 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[LoopVectorize] Shrink integer operations into the smallest type possible
James Molloy [Mon, 12 Oct 2015 12:34:45 +0000 (12:34 +0000)]
[LoopVectorize] Shrink integer operations into the smallest type possible

C semantics force sub-int-sized values (e.g. i8, i16) to be promoted to int
type (e.g. i32) whenever arithmetic is performed on them.

For targets with native i8 or i16 operations, usually InstCombine can shrink
the arithmetic type down again. However InstCombine refuses to create illegal
types, so for targets without i8 or i16 registers, the lengthening and
shrinking remains.

Most SIMD ISAs (e.g. NEON) however support vectors of i8 or i16 even when
their scalar equivalents do not, so during vectorization it is important to
remove these lengthens and truncates when deciding the profitability of
vectorization.

The algorithm this uses starts at truncs and icmps, trawling their use-def
chains until they terminate or instructions outside the loop are found (or
unsafe instructions like inttoptr casts are found). If the use-def chains
starting from different root instructions (truncs/icmps) meet, they are
unioned. The demanded bits of each node in the graph are ORed together to form
an overall mask of the demanded bits in the entire graph. The minimum bitwidth
that graph can be truncated to is the bitwidth minus the number of leading
zeroes in the overall mask.

The intention is that this algorithm should "first do no harm", so it will
never insert extra cast instructions. This is why the use-def graphs are
unioned, so that subgraphs with different minimum bitwidths do not need casts
inserted between them.

This algorithm works hard to reduce compile time impact. DemandedBits are only
queried if there are extends of illegal types and if a truncate to an illegal
type is seen. In the general case, this results in a simple linear scan of the
instructions in the loop.

No non-noise compile time impact was seen on a clang bootstrap build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250032 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Add XSAVE intrinsic family
Amjad Aboud [Mon, 12 Oct 2015 11:47:46 +0000 (11:47 +0000)]
[X86] Add XSAVE intrinsic family

Add intrinsics for the
  XSAVE instructions (XSAVE/XSAVE64/XRSTOR/XRSTOR64)
  XSAVEOPT instructions (XSAVEOPT/XSAVEOPT64)
  XSAVEC instructions (XSAVEC/XSAVEC64)
  XSAVES instructions (XSAVES/XSAVES64/XRSTORS/XRSTORS64)

Differential Revision: http://reviews.llvm.org/D13012

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250029 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[x86] PR24562: fix incorrect folding of PSHUFB nodes with a mask where all indices...
Andrea Di Biagio [Mon, 12 Oct 2015 11:25:41 +0000 (11:25 +0000)]
[x86] PR24562: fix incorrect folding of PSHUFB nodes with a mask where all indices have the most significant bit set.

This patch fixes a problem in function 'combineX86ShuffleChain' that causes a
chain of shuffles to be wrongly folded away when the combined shuffle mask has
only one element.

We may end up with a combined shuffle mask of one element as a result of
multiple calls to function 'canWidenShuffleElements()'.
Function canWidenShuffleElements attempts to simplify a shuffle mask by widening
the size of the elements being shuffled.
For every pair of shuffle indices, function canWidenShuffleElements checks if
indices refer to adjacent elements. If all pairs refer to "adjacent" elements
then the shuffle mask is safely widened. As a consequence of widening, we end up
with a new shuffle mask which is half the size of the original shuffle mask.

The byte shuffle (pshufb) from test pr24562.ll has a mask of all SM_SentinelZero
indices. Function canWidenShuffleElements would combine each pair of
SM_SentinelZero indices into a single SM_SentinelZero index. So, in a
logarithmic number of steps (4 in this case), the pshufb mask is simplified to
a mask with only one index which is equal to SM_SentinelZero.

Before this patch, function combineX86ShuffleChain wrongly assumed that a mask
of size one is always equivalent to an identity mask. So, the entire shuffle
chain was just folded away as the combined shuffle mask was treated as a no-op
mask.

With this patch we know check if the only element of a combined shuffle mask is
SM_SentinelZero. In case, we propagate a zero vector.

Differential Revision: http://reviews.llvm.org/D13364

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250027 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoTest commit
Zlatko Buljan [Mon, 12 Oct 2015 11:19:40 +0000 (11:19 +0000)]
Test commit

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250026 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agocmake: Avoid leading space in LLVM_DEFINITIONS.
Pawel Bylica [Mon, 12 Oct 2015 10:42:03 +0000 (10:42 +0000)]
cmake: Avoid leading space in LLVM_DEFINITIONS.

Summary: Unnecessary space at the beginning of LLVM_DEFINITIONS in cmake shared files can break projects that use the variable.

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D13432

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250025 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SystemZ] testcase MC/SystemZ/insn-good-z13.s extended.
Jonas Paulsson [Mon, 12 Oct 2015 10:13:57 +0000 (10:13 +0000)]
[SystemZ] testcase MC/SystemZ/insn-good-z13.s extended.

New instructions using floating point registers have been added, to check
that AsmParser can deal with fp regs in vector instructions.

This tests r249810.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250023 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[MISched] Python script to check coverage of misched info
James Molloy [Mon, 12 Oct 2015 08:50:47 +0000 (08:50 +0000)]
[MISched] Python script to check coverage of misched info

This script prints a CSV of all misched models of a target when given the output of the debug output of subtarget using:

    llvm-tblgen --gen-subtarget --debug-only=subtarget-emitter ...

With thanks to Dave Estes for mentioning the idea at the 2014 LLVM Developers' Meeting.

Patch by Christof Douma!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250020 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSCEV: Allow simple AddRec * Parameter products in delinearization
Tobias Grosser [Mon, 12 Oct 2015 08:02:00 +0000 (08:02 +0000)]
SCEV: Allow simple AddRec * Parameter products in delinearization

This patch also allows the -delinearize pass to delinearize expressions that do
not have an outermost SCEVAddRec expression. The SCEV::delinearize
infrastructure allowed this since r240952, but the -delinearize pass was not
updated yet.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250018 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Use u8imm for the immediate type for all shift and rotate instructions. This...
Craig Topper [Mon, 12 Oct 2015 06:23:10 +0000 (06:23 +0000)]
[X86] Use u8imm for the immediate type for all shift and rotate instructions. This way the assembler will perform range checking. Believe this matches gas behavior.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250016 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Add support to assembler and MCInst lowering to use the other vmovq %xmmX,...
Craig Topper [Mon, 12 Oct 2015 04:57:59 +0000 (04:57 +0000)]
[X86] Add support to assembler and MCInst lowering to use the other vmovq %xmmX, %xmmX encoding if it would be a shorter VEX encoding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250014 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Cleanup formatting a bit. NFC
Craig Topper [Mon, 12 Oct 2015 04:27:17 +0000 (04:27 +0000)]
[X86] Cleanup formatting a bit. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250013 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Change the immediate for IN/OUT instructions to u8imm so the assembly parser...
Craig Topper [Mon, 12 Oct 2015 04:17:55 +0000 (04:17 +0000)]
[X86] Change the immediate for IN/OUT instructions to u8imm so the assembly parser will check the size.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250012 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Add some instruction aliases to get the assembly parser table to favor arithmet...
Craig Topper [Mon, 12 Oct 2015 03:39:57 +0000 (03:39 +0000)]
[X86] Add some instruction aliases to get the assembly parser table to favor arithmetic instructions with 8-bit immediates over the forms that implicitly use the ax/eax/rax.

This allows us to remove the explicit code for working around the existing priority

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250011 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[llvm-rtdyld] General modernization/cleanup in preparation for (bigger) changes.
Davide Italiano [Mon, 12 Oct 2015 00:57:29 +0000 (00:57 +0000)]
[llvm-rtdyld] General modernization/cleanup in preparation for (bigger) changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250004 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Bugpoint] Get rid of dead code. No functional change.
Davide Italiano [Sun, 11 Oct 2015 21:36:11 +0000 (21:36 +0000)]
[Bugpoint] Get rid of dead code. No functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249999 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Fix CMP and TEST with al/ax/eax/rax to not mark EFLAGS as a use or al/ax/eax...
Craig Topper [Sun, 11 Oct 2015 19:54:02 +0000 (19:54 +0000)]
[X86] Fix CMP and TEST with al/ax/eax/rax to not mark EFLAGS as a use or al/ax/eax/rax as a def. Probably doesn't have a functional affect since these aren't used in isel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249994 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[DAGCombiner] Improved FMA combine support for vectors
Simon Pilgrim [Sun, 11 Oct 2015 19:48:12 +0000 (19:48 +0000)]
[DAGCombiner] Improved FMA combine support for vectors

Enabled constant canonicalization for all constants.

Improved combining of constant vectors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249993 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Completed SHL cost model tests
Simon Pilgrim [Sun, 11 Oct 2015 18:33:48 +0000 (18:33 +0000)]
[X86] Completed SHL cost model tests

As discussed in D8690.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249990 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Remove special validation for INT immediate operand from AsmParser. Instead...
Craig Topper [Sun, 11 Oct 2015 18:27:24 +0000 (18:27 +0000)]
[X86] Remove special validation for INT immediate operand from AsmParser. Instead mark its operand type as u8imm which will cause it to fail to match. This is more consistent with other instruction behavior.

This also fixes a bug where negative immediates below -128 were not being reported as errors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249989 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Renamed SHL cost model tests
Simon Pilgrim [Sun, 11 Oct 2015 17:34:32 +0000 (17:34 +0000)]
[X86] Renamed SHL cost model tests

Matches naming conventions for ASHR/LSHR cost tests

As discussed in D8690.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249984 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Added LSHR cost model tests
Simon Pilgrim [Sun, 11 Oct 2015 17:29:26 +0000 (17:29 +0000)]
[X86] Added LSHR cost model tests

There are several dodgy costings due to AVX1 legalizing 256-bit integer vectors that need fixing.

As discussed in D8690.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249983 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Added ASHR cost model tests
Simon Pilgrim [Sun, 11 Oct 2015 17:08:05 +0000 (17:08 +0000)]
[X86] Added ASHR cost model tests

There are several dodgy costings due to AVX1 legalizing 256-bit integer vectors that need fixing.

As discussed in D8690.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249981 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[TableGen] Add a space between type and '*' in front of a variable name in output...
Craig Topper [Sun, 11 Oct 2015 16:59:29 +0000 (16:59 +0000)]
[TableGen] Add a space between type and '*' in front of a variable name in output file. While there replace type with 'auto' since there's a cast on the right side of the assignment. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249980 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Simplify immediate range checking code.
Craig Topper [Sun, 11 Oct 2015 16:38:14 +0000 (16:38 +0000)]
[X86] Simplify immediate range checking code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249979 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[DAGCombiner] Tidyup FMINNUM/FMAXNUM constant folding
Simon Pilgrim [Sun, 11 Oct 2015 16:02:28 +0000 (16:02 +0000)]
[DAGCombiner] Tidyup FMINNUM/FMAXNUM constant folding

Enable constant folding for vector splats as well as scalars.

Enable constant canonicalization for all scalar and vector constants.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249978 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[InstCombine][X86][XOP] Combine XOP integer vector comparisons to native IR
Simon Pilgrim [Sun, 11 Oct 2015 14:38:34 +0000 (14:38 +0000)]
[InstCombine][X86][XOP] Combine XOP integer vector comparisons to native IR

We now have lowering support for XOP PCOM/PCOMU instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249977 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to...
Simon Pilgrim [Sun, 11 Oct 2015 14:15:17 +0000 (14:15 +0000)]
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions.

The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249976 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[ProfileData] Test commit for slingn
Nathan Slingerland [Sun, 11 Oct 2015 13:30:56 +0000 (13:30 +0000)]
[ProfileData] Test commit for slingn

This is a test of the LLVM commit system. In the event of a real commit there would be some useful code changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249972 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86][SSE] Vector signed/unsigned integer compare tests.
Simon Pilgrim [Sat, 10 Oct 2015 22:21:05 +0000 (22:21 +0000)]
[X86][SSE] Vector signed/unsigned integer compare tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249954 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoChange isUIntN/isIntN calls with constant N to use the template version. NFC
Craig Topper [Sat, 10 Oct 2015 20:17:07 +0000 (20:17 +0000)]
Change isUIntN/isIntN calls with constant N to use the template version. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249952 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoIn isUIntN, make sure N is less than 64 before using in a shift to avoid undefined...
Craig Topper [Sat, 10 Oct 2015 18:54:26 +0000 (18:54 +0000)]
In isUIntN, make sure N is less than 64 before using in a shift to avoid undefined behavior. Also change it to use the same formula as the template version which I think results in less math in compiled code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249951 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix PR25101 - Handle anonymous functions without VST entries
Teresa Johnson [Sat, 10 Oct 2015 14:18:36 +0000 (14:18 +0000)]
Fix PR25101 - Handle anonymous functions without VST entries

Summary:
The change to use the VST function entries for lazy deserialization did
not handle the case of anonymous functions without aliases. In that case
we must fall back to scanning the function blocks as there is no VST
entry.

Reviewers: dexonsmith, joker.eph, davidxl

Subscribers: tstellarAMD, llvm-commits

Differential Revision: http://reviews.llvm.org/D13596

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249947 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SystemZ] CodeGen/SystemZ/asm-18.ll run with -verify-machineinstrs
Jonas Paulsson [Sat, 10 Oct 2015 07:20:23 +0000 (07:20 +0000)]
[SystemZ] CodeGen/SystemZ/asm-18.ll run with -verify-machineinstrs

Relates to the fixes of r249811.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249946 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SystemZ] Fixes in the backend I/R.
Jonas Paulsson [Sat, 10 Oct 2015 07:14:24 +0000 (07:14 +0000)]
[SystemZ] Fixes in the backend I/R.

expandPostRAPseudo():
STX -> 2 * STD: The first STD should not have the kill flag set for the address.

SystemZElimCompare:
BRC -> BRCT conversion: Don't forget to remove the CC<use,kill> operand.

Needed to make SystemZ/asm-17.ll pass with -verify-machineinstrs, which
now runs with this flag.

Reviewed by Ulrich Weigand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249945 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[IndVars] Use `auto`; NFC
Sanjoy Das [Sat, 10 Oct 2015 06:33:33 +0000 (06:33 +0000)]
[IndVars] Use `auto`; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249944 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse range-based for loops. NFC
Craig Topper [Sat, 10 Oct 2015 05:38:14 +0000 (05:38 +0000)]
Use range-based for loops. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249943 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[RuntimeDyld] Fix performance problem in resolveRelocations with many sections
Keno Fischer [Sat, 10 Oct 2015 05:37:02 +0000 (05:37 +0000)]
[RuntimeDyld] Fix performance problem in resolveRelocations with many sections

Summary:
Rather than just iterating over all sections and checking whether we have relocations for them, iterate over the relocation map instead. This showed up heavily in an artificial julia benchmark that does lots of compilation. On that particular benchmark, this patch gives
~15% performance improvements. As far as I can tell the primary reason why the original
loop was so expensive is that Relocations[i] actually constructs a relocationList (allocating memory & doing lots of other unnecessary computing) if none is found.

Reviewers: lhames

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D13545

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249942 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse range-based for loops. NFC.
Craig Topper [Sat, 10 Oct 2015 05:25:06 +0000 (05:25 +0000)]
Use range-based for loops. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249941 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse emplace_back instead of a constructor call and push_back. NFC
Craig Topper [Sat, 10 Oct 2015 05:25:02 +0000 (05:25 +0000)]
Use emplace_back instead of a constructor call and push_back. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249940 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSuppress LLVM::tools/llvm-symbolizer/coff-dwarf.test for mingw, for now.
NAKAMURA Takumi [Sat, 10 Oct 2015 02:57:02 +0000 (02:57 +0000)]
Suppress LLVM::tools/llvm-symbolizer/coff-dwarf.test for mingw, for now.

FIXME: Improve llvm-symbolizer, or rename the feature "system-windows".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249937 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[libFuzzer] document more trophies
Kostya Serebryany [Sat, 10 Oct 2015 02:14:18 +0000 (02:14 +0000)]
[libFuzzer] document more trophies

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249933 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMove llvm-objdump malformed Mach-O tests to X86 test directory.
Kevin Enderby [Sat, 10 Oct 2015 01:06:20 +0000 (01:06 +0000)]
Move llvm-objdump malformed Mach-O tests to X86 test directory.

rdar://22983603

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249927 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAnalysis: Remove implicit ilist iterator conversions
Duncan P. N. Exon Smith [Sat, 10 Oct 2015 00:53:03 +0000 (00:53 +0000)]
Analysis: Remove implicit ilist iterator conversions

Remove implicit ilist iterator conversions from LLVMAnalysis.

I came across something really scary in `llvm::isKnownNotFullPoison()`
which relied on `Instruction::getNextNode()` being completely broken
(not surprising, but scary nevertheless).  This function is documented
(and coded to) return `nullptr` when it gets to the sentinel, but with
an `ilist_half_node` as a sentinel, the sentinel check looks into some
other memory and we don't recognize we've hit the end.

Rooting out these scary cases is the reason I'm removing the implicit
conversions before doing anything else with `ilist`; I'm not at all
surprised that clients rely on badness.

I found another scary case -- this time, not relying on badness, just
bad (but I guess getting lucky so far) -- in
`ObjectSizeOffsetEvaluator::compute_()`.  Here, we save out the
insertion point, do some things, and then restore it.  Previously, we
let the iterator auto-convert to `Instruction*`, and then set it back
using the `Instruction*` version:

    Instruction *PrevInsertPoint = Builder.GetInsertPoint();

    /* Logic that may change insert point */

    if (PrevInsertPoint)
      Builder.SetInsertPoint(PrevInsertPoint);

The check for `PrevInsertPoint` doesn't protect correctly against bad
accesses.  If the insertion point has been set to the end of a basic
block (i.e., `SetInsertPoint(SomeBB)`), then `GetInsertPoint()` returns
an iterator pointing at the list sentinel.  The version of
`SetInsertPoint()` that's getting called will then call
`PrevInsertPoint->getParent()`, which explodes horribly.  The only
reason this hasn't blown up is that it's fairly unlikely the builder is
adding to the end of the block; usually, we're adding instructions
somewhere before the terminator.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249925 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[llvm-rtdyld] Use range-based loop. NFC.
Davide Italiano [Sat, 10 Oct 2015 00:45:24 +0000 (00:45 +0000)]
[llvm-rtdyld] Use range-based loop. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249923 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMC: Remove implicit ilist iterator conversions, NFC
Duncan P. N. Exon Smith [Sat, 10 Oct 2015 00:13:11 +0000 (00:13 +0000)]
MC: Remove implicit ilist iterator conversions, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249922 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix a bugs in the Mach-O disassembler when disassembling from a
Kevin Enderby [Sat, 10 Oct 2015 00:05:01 +0000 (00:05 +0000)]
Fix a bugs in the Mach-O disassembler when disassembling from a
malformed Mach-O file that caused a crash.  This was because of an
assert where the code was incorrectly attempting to parse relocation
entries off of the sections and the filetype was not an MH_OBJECT.

rdar://22983603

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249921 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[WinEH] Remove more dead code
David Majnemer [Sat, 10 Oct 2015 00:04:29 +0000 (00:04 +0000)]
[WinEH] Remove more dead code

wineh-parent is dead, so is ValueOrMBB.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249920 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[WinEH] Delete the old landingpad implementation of Windows EH
Reid Kleckner [Fri, 9 Oct 2015 23:34:53 +0000 (23:34 +0000)]
[WinEH] Delete the old landingpad implementation of Windows EH

The new implementation works at least as well as the old implementation
did.

Also delete the associated preparation tests. They don't exercise
interesting corner cases of the new implementation. All the codegen
tests of the EH tables have already been ported.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249918 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SEH] Update SEH codegen tests to use the new IR
Reid Kleckner [Fri, 9 Oct 2015 23:05:54 +0000 (23:05 +0000)]
[SEH] Update SEH codegen tests to use the new IR

Also Fix a buglet where SEH tables had ranges that spanned funclets.

The remaining tests using the old landingpad IR are preparation tests,
and will be deleted along with the old preparation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249917 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: Remove implicit ilist iterator conversions, NFC
Duncan P. N. Exon Smith [Fri, 9 Oct 2015 22:56:24 +0000 (22:56 +0000)]
CodeGen: Remove implicit ilist iterator conversions, NFC

Finish removing implicit ilist iterator conversions from LLVMCodeGen.
I'm sure there are lots more of these in lib/CodeGen/*/.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249915 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[CMake] Parallel make breaks on native tablegen
Chris Bieneman [Fri, 9 Oct 2015 22:26:04 +0000 (22:26 +0000)]
[CMake] Parallel make breaks on native tablegen

Patch by Alex Wang

This patch resolves a parallelization issue that occurs when native tablegen targets are built at the same time. They both try to build libSupport and clobber each other causing the builds to fail.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249911 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[WinEH] Insert the catchpad return before CSR restoration
David Majnemer [Fri, 9 Oct 2015 22:18:45 +0000 (22:18 +0000)]
[WinEH] Insert the catchpad return before CSR restoration

x64 catchpads use rax to inform the unwinder where control should go
next.  However, we must initialize rax before the epilogue sequence so
as to not perturb the unwinder.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249910 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix use of uninitialized bool, found by ubsan in portion of
Richard Smith [Fri, 9 Oct 2015 22:09:56 +0000 (22:09 +0000)]
Fix use of uninitialized bool, found by ubsan in portion of
test/tools/llvm-objdump/malformed-machos.test added in r249845.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249909 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix assert when emitting llvm.pow.f86.
James Y Knight [Fri, 9 Oct 2015 21:36:19 +0000 (21:36 +0000)]
Fix assert when emitting llvm.pow.f86.

This occurred due to introducing the invalid i64 type after type
legalization had already finished, in an attempt to workaround bitcast
f64 -> v2i32 not doing constant folding.

The *right* thing is to actually fix bitcast, but that has other
complications. So, for now, just get rid of the broken workaround, and
check in a test-case showing that it doesn't crash, with TODOs for
emitting proper code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249908 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove unused function in sample profile writer API - NFC.
Diego Novillo [Fri, 9 Oct 2015 21:33:13 +0000 (21:33 +0000)]
Remove unused function in sample profile writer API - NFC.

These functions are not needed and are getting in the way of changes
for implementing a table of contents for the binary format.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249907 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SEH] Fix _except_handler4 table base states
Reid Kleckner [Fri, 9 Oct 2015 21:27:28 +0000 (21:27 +0000)]
[SEH] Fix _except_handler4 table base states

We got them right for the old IR, but not with funclets.  Port the old
test to the new IR and fix the code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249906 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: Avoid more ilist iterator implicit conversions, NFC
Duncan P. N. Exon Smith [Fri, 9 Oct 2015 21:08:19 +0000 (21:08 +0000)]
CodeGen: Avoid more ilist iterator implicit conversions, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249903 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: Use range-based for in PostRAScheduler, NFC
Duncan P. N. Exon Smith [Fri, 9 Oct 2015 21:05:00 +0000 (21:05 +0000)]
CodeGen: Use range-based for in PostRAScheduler, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249901 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SEH] Remember to emit the last invoke range for SEH
Reid Kleckner [Fri, 9 Oct 2015 20:39:39 +0000 (20:39 +0000)]
[SEH] Remember to emit the last invoke range for SEH

This wasn't very observable in execution tests, because usually there is
an invoke in the catchpad that unwinds the the catchendpad but never
actually throws.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249898 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoGeneralize convergent check to handle invokes as well as calls.
Owen Anderson [Fri, 9 Oct 2015 20:17:46 +0000 (20:17 +0000)]
Generalize convergent check to handle invokes as well as calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249892 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix assert in X86 backend.
James Y Knight [Fri, 9 Oct 2015 20:10:14 +0000 (20:10 +0000)]
Fix assert in X86 backend.

When running combine on an extract_vector_elt, it wants to look through
a bitcast to check if the argument to the bitcast was itself an
extract_vector_elt with particular operands.

However, it called getOperand() on the argument to the bitcast *before*
checking that the opcode was EXTRACT_VECTOR_ELT, assert-failing if there
were zero operands for the actual opcode.

Fix, and add trivial test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249891 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert "Simplify code. NFC."
Chad Rosier [Fri, 9 Oct 2015 19:48:48 +0000 (19:48 +0000)]
Revert "Simplify code.  NFC."

This reverts commit r248610.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249887 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: Continue removing ilist iterator implicit conversions
Duncan P. N. Exon Smith [Fri, 9 Oct 2015 19:40:45 +0000 (19:40 +0000)]
CodeGen: Continue removing ilist iterator implicit conversions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249884 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: Remove implicit iterator conversions from MBB.cpp
Duncan P. N. Exon Smith [Fri, 9 Oct 2015 19:36:12 +0000 (19:36 +0000)]
CodeGen: Remove implicit iterator conversions from MBB.cpp

Remove implicit ilist iterator conversions from MachineBasicBlock.cpp.

I've also added an overload of `splice()` that takes a pointer, since
it's a natural API.  This is similar to the overloads I added for
`remove()` and `erase()` in r249867.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249883 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: Avoid ilist iterator implicit conversions in a few more places, NFC
Duncan P. N. Exon Smith [Fri, 9 Oct 2015 19:23:20 +0000 (19:23 +0000)]
CodeGen: Avoid ilist iterator implicit conversions in a few more places, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249880 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: Remove more ilist iterator implicit conversions, NFC
Duncan P. N. Exon Smith [Fri, 9 Oct 2015 19:13:58 +0000 (19:13 +0000)]
CodeGen: Remove more ilist iterator implicit conversions, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249879 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: Use range-based for in IntrinsicLowering::AddPrototypes, NFC
Duncan P. N. Exon Smith [Fri, 9 Oct 2015 19:07:41 +0000 (19:07 +0000)]
CodeGen: Use range-based for in IntrinsicLowering::AddPrototypes, NFC

This happens to avoid a host of implicit ilist iterator conversions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249877 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: Use range-based for in GlobalMerge, NFC
Duncan P. N. Exon Smith [Fri, 9 Oct 2015 18:57:47 +0000 (18:57 +0000)]
CodeGen: Use range-based for in GlobalMerge, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249876 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: Remove a few more ilist iterator implicit conversions, NFC
Duncan P. N. Exon Smith [Fri, 9 Oct 2015 18:44:40 +0000 (18:44 +0000)]
CodeGen: Remove a few more ilist iterator implicit conversions, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249875 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoTeach LoopUnswitch not to perform non-trivial unswitching on loops containing converg...
Owen Anderson [Fri, 9 Oct 2015 18:40:20 +0000 (18:40 +0000)]
Teach LoopUnswitch not to perform non-trivial unswitching on loops containing convergent operations.

Doing so could cause the post-unswitching convergent ops to be
control-dependent on the unswitch condition where they were not before.
This check could be refined to allow unswitching where the convergent
operation was already control-dependent on the unswitch condition.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249874 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd iterator ranges for blocks in a Loop.
Owen Anderson [Fri, 9 Oct 2015 18:40:15 +0000 (18:40 +0000)]
Add iterator ranges for blocks in a Loop.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249873 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: Remove implicit iterator conversions from SlotIndexes.h, NFC
Duncan P. N. Exon Smith [Fri, 9 Oct 2015 18:35:09 +0000 (18:35 +0000)]
CodeGen: Remove implicit iterator conversions from SlotIndexes.h, NFC

Be explicit about changes between pointers and iterators, as with other
recent commits.  This transitively removes implicit ilist iterator
conversions from about 20 source files in CodeGen.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249869 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: Remove implicit conversions from Analysis and BranchFolding
Duncan P. N. Exon Smith [Fri, 9 Oct 2015 18:23:49 +0000 (18:23 +0000)]
CodeGen: Remove implicit conversions from Analysis and BranchFolding

Remove a few more implicit ilist iterator conversions, this time from
Analysis.cpp and BranchFolding.cpp.

I added a few overloads for `remove()` and `erase()`, which quite
naturally take pointers as well as iterators as parameters.  This will
reduce the churn at least in the short term, but I don't really have a
problem with these existing for longer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249867 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRefine the definition of convergent to only disallow the addition of new control...
Owen Anderson [Fri, 9 Oct 2015 18:06:13 +0000 (18:06 +0000)]
Refine the definition of convergent to only disallow the addition of new control dependencies.

This covers the common case of operations that cannot be sunk.
Operations that cannot be hoisted should already be handled properly via
the safe-to-speculate rules and mechanisms.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249865 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agofix typos; NFC
Sanjay Patel [Fri, 9 Oct 2015 18:01:03 +0000 (18:01 +0000)]
fix typos; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249863 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[CMake] If LLVM_DYLIB_EXPORT_ALL is On don't generate an export list at all, just...
Chris Bieneman [Fri, 9 Oct 2015 17:55:21 +0000 (17:55 +0000)]
[CMake] If LLVM_DYLIB_EXPORT_ALL is On don't generate an export list at all, just export the world.

This should resolve Bug 24157 - CMake built shared library does not export all public symbols

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249862 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd inline stack streaming to binary sample profiles.
Diego Novillo [Fri, 9 Oct 2015 17:54:24 +0000 (17:54 +0000)]
Add inline stack streaming to binary sample profiles.

With this patch we can now read and write inline stacks in sample
profiles to the binary encoded profiles.

In a subsequent patch, I will add a string table to the binary encoding.
Right now function names are emitted as strings every time we find them.
This is too bloated and will produce large files in applications with
lots of inlining.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249861 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[WebAssembly] Rename floating-point operators to match their spec names.
Dan Gohman [Fri, 9 Oct 2015 17:50:00 +0000 (17:50 +0000)]
[WebAssembly] Rename floating-point operators to match their spec names.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249859 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd verification for align, dereferenceable, dereferenceable_or_null load metadata
Artur Pilipenko [Fri, 9 Oct 2015 17:41:29 +0000 (17:41 +0000)]
Add verification for align, dereferenceable, dereferenceable_or_null load metadata

Reviewed By: reames

Differential Revision: http://reviews.llvm.org/D13428

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249856 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoClear SectionSymbols in MCContext::Reset
Keno Fischer [Fri, 9 Oct 2015 17:24:54 +0000 (17:24 +0000)]
Clear SectionSymbols in MCContext::Reset

This was just forgotten when SectionSymbols was introduced and could cause
corruption if the MCContext was reused after Reset.

Reviewers: rafael

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D13547

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249854 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAArch64: Make getNextNode() cleanup in r249764 more clear
Duncan P. N. Exon Smith [Fri, 9 Oct 2015 16:54:54 +0000 (16:54 +0000)]
AArch64: Make getNextNode() cleanup in r249764 more clear

After r249764, if you didn't see the full context, it looked like
`std::next(I)` would get the same result as
`++MachineBasicBlock::iterator(I)`.  However, `I` is a `MachineInstr*`
(not a `MachineBasicBlock::iterator`).

Use the `getIterator()` helper I added later (r249782) to make this code
more clear.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249852 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: Start removing implicit conversions to/from list iterators, NFC
Duncan P. N. Exon Smith [Fri, 9 Oct 2015 16:54:49 +0000 (16:54 +0000)]
CodeGen: Start removing implicit conversions to/from list iterators, NFC

Start removing implicit conversions to/from list iterators in CodeGen,
ala r249782 for IR.  A lot more to go after this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249851 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert "Support: Partially revert r249782 to unbreak clang build"
Duncan P. N. Exon Smith [Fri, 9 Oct 2015 16:51:23 +0000 (16:51 +0000)]
Revert "Support: Partially revert r249782 to unbreak clang build"

This reverts commit r249783, fully reinstating r249782.  I've fixed the
bug in clang: it was a non-const iterator that dereferenced to const
(but had an implicit conversion to non-const).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249850 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMake HeaderLineno a local variable.
Dehao Chen [Fri, 9 Oct 2015 16:50:16 +0000 (16:50 +0000)]
Make HeaderLineno a local variable.

http://reviews.llvm.org/D13576

As we are using hierarchical profile, there is no need to keep HeaderLineno a member variable. This is because each level of the inline stack will have its own header lineno. One should use the head lineno of its own inline stack level instead of the actual symbol.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249848 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix pdb.test when python is not on PATH
Reid Kleckner [Fri, 9 Oct 2015 16:49:56 +0000 (16:49 +0000)]
Fix pdb.test when python is not on PATH

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249847 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFixed two bugs in llvm-objdump’s printing of Objective-C meta data
Kevin Enderby [Fri, 9 Oct 2015 16:48:44 +0000 (16:48 +0000)]
Fixed two bugs in llvm-objdump’s printing of Objective-C meta data
from malformed Mach-O files that caused crashes.  The first because the
offset in a dyld bind table entry was out of range.  The second because their
was no image info section and the routine printing it did not have the
need check to see the section did not exist.

rdar://22983603

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249845 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoValueTracking: use getAlignment in isAligned
Artur Pilipenko [Fri, 9 Oct 2015 15:58:26 +0000 (15:58 +0000)]
ValueTracking: use getAlignment in isAligned

Reviewed By: reames

Differential Revision: http://reviews.llvm.org/D13517

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249841 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[dsymutil] Prevent warning
Frederic Riss [Fri, 9 Oct 2015 15:04:05 +0000 (15:04 +0000)]
[dsymutil] Prevent warning

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249836 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoImprove ISel across lane float min/max reduction
Jun Bum Lim [Fri, 9 Oct 2015 14:11:25 +0000 (14:11 +0000)]
Improve ISel across lane float min/max reduction

In vectorized float min/max reduction code, the final "reduce" step
is sub-optimal. In AArch64, this change wll combine :

  svn0 = vector_shuffle t0, undef<2,3,u,u>
  fmin = fminnum t0,svn0
  svn1 = vector_shuffle fmin, undef<1,u,u,u>
  cc = setcc fmin, svn1, ole
  n0 = extract_vector_elt cc, #0
  n1 = extract_vector_elt fmin, #0
  n2 = extract_vector_elt fmin, #1
  result = select n0, n1,n2
into :
  result = llvm.aarch64.neon.fminnmv t0

This change extends r247575.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249834 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SystemZ] Remove unused code in SystemZElimCompare.cpp
Jonas Paulsson [Fri, 9 Oct 2015 11:27:44 +0000 (11:27 +0000)]
[SystemZ] Remove unused code in SystemZElimCompare.cpp

The Reference IndirectDef and IndirectUse members were unused and therefore
removed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249824 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoVector element extraction without stack operations on Power 8
Nemanja Ivanovic [Fri, 9 Oct 2015 11:12:18 +0000 (11:12 +0000)]
Vector element extraction without stack operations on Power 8

This patch corresponds to review:
http://reviews.llvm.org/D12032

This patch builds onto the patch that provided scalar to vector conversions
without stack operations (D11471).
Included in this patch:

    - Vector element extraction for all vector types with constant element number
    - Vector element extraction for v16i8 and v8i16 with variable element number
    - Removal of some unnecessary COPY_TO_REGCLASS operations that ended up
      unnecessarily moving things around between registers

Not included in this patch (will be in upcoming patch):

    - Vector element extraction for v4i32, v4f32, v2i64 and v2f64 with
      variable element number
    - Vector element insertion for variable/constant element number

Testing is provided for all extractions. The extractions that are not
implemented yet are just placeholders.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249822 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[MemCpyOpt] Fix wrong merging adjacent nontemporal stores into memset calls.
Andrea Di Biagio [Fri, 9 Oct 2015 10:53:41 +0000 (10:53 +0000)]
[MemCpyOpt] Fix wrong merging adjacent nontemporal stores into memset calls.

Pass MemCpyOpt doesn't check if a store instruction is nontemporal.
As a consequence, adjacent nontemporal stores are always merged into a
memset call.

Example:

;;;
define void @foo(<4 x float>* nocapture %p) {
entry:
  store <4 x float> zeroinitializer, <4 x float>* %p, align 16, !nontemporal !0
  %p1 = getelementptr inbounds <4 x float>, <4 x float>* %dst, i64 1
  store <4 x float> zeroinitializer, <4 x float>* %p1, align 16, !nontemporal !0
  ret void
}

!0 = !{i32 1}
;;;

In this example, the two nontemporal stores are combined to a memset of zero
which does not preserve the nontemporal hint. Later on the backend (tested on a
x86-64 corei7) expands that memset call into a sequence of two normal 16-byte
aligned vector stores.

opt -memcpyopt example.ll -S -o - | llc -mcpu=corei7 -o -

Before:
  xorps  %xmm0, %xmm0
  movaps  %xmm0, 16(%rdi)
  movaps  %xmm0, (%rdi)

With this patch, we no longer merge nontemporal stores into calls to memset.
In this example, llc correctly expands the two stores into two movntps:
  xorps  %xmm0, %xmm0
  movntps %xmm0, 16(%rdi)
  movntps  %xmm0, (%rdi)

In theory, we could extend the usage of !nontemporal metadata to memcpy/memset
calls. However a change like that would only have the effect of forcing the
backend to expand !nontemporal memsets back to sequences of store instructions.
A memset library call would not have exactly the same semantic of a builtin
!nontemporal memset call. So, SelectionDAG will have to conservatively expand
it back to a sequence of !nontemporal stores (effectively undoing the merging).

Differential Revision: http://reviews.llvm.org/D13519

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249820 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[EarlyCSE] Address post commit review for r249523.
Arnaud A. de Grandmaison [Fri, 9 Oct 2015 09:23:01 +0000 (09:23 +0000)]
[EarlyCSE] Address post commit review for r249523.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249814 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SystemZ] Remove superfluous braces in SystemZShortenInst.cpp
Jonas Paulsson [Fri, 9 Oct 2015 07:19:20 +0000 (07:19 +0000)]
[SystemZ] Remove superfluous braces in SystemZShortenInst.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@249812 91177308-0d34-0410-b5e6-96231b3b80d8