]> granicus.if.org Git - llvm/log
llvm
5 years ago[InstCombine] make fold for icmp with sext more efficient; NFC
Sanjay Patel [Tue, 20 Aug 2019 17:03:22 +0000 (17:03 +0000)]
[InstCombine] make fold for icmp with sext more efficient; NFC

We were creating 2 instructions and relying on a subsequent fold
to invert a not(icmp). Create the final icmp directly instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369411 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Use isNullConstant instead of getConstantOperandVal == 0. NFC
Craig Topper [Tue, 20 Aug 2019 16:55:12 +0000 (16:55 +0000)]
[X86] Use isNullConstant instead of getConstantOperandVal == 0. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369410 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CodeGen] Add EarlyIfConvert test missed in previous commit
Thomas Raoux [Tue, 20 Aug 2019 16:34:47 +0000 (16:34 +0000)]
[CodeGen] Add EarlyIfConvert test missed in previous commit

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369405 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ARM] Select vaddva
Sam Tebbs [Tue, 20 Aug 2019 16:33:34 +0000 (16:33 +0000)]
[ARM] Select vaddva

This patch adds vaddva selection.

Differential revision: https://reviews.llvm.org/D66410

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369404 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[GlobalISel] Handle multiple registers in dbg.value intrinsic
Aditya Nandakumar [Tue, 20 Aug 2019 16:28:37 +0000 (16:28 +0000)]
[GlobalISel] Handle multiple registers in dbg.value intrinsic

https://reviews.llvm.org/D66077

The value passed into dbg.value may relate to multiple registers,
each of which need a DBG_VALUE.

This fix calls MIRBuilder.buildDirectDbgValue for each register.

Without this, IR passed in from flang-compiler/flang may fail an
assertion in getOrCreateVReg.

Patch by : peterwaller-arm.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369403 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agogn build: Merge r369298
Nico Weber [Tue, 20 Aug 2019 16:19:50 +0000 (16:19 +0000)]
gn build: Merge r369298

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369401 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRegex: Add isValid() with no parameter
Jan Kratochvil [Tue, 20 Aug 2019 16:05:23 +0000 (16:05 +0000)]
Regex: Add isValid() with no parameter

There will be some performance (only a little) improvement for LLDB's
RegularExpression::Execute.

Differential Revision: https://reviews.llvm.org/D66463

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369396 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago [CodeGen] Add a pass to do block predication on SSA machine IR.
Thomas Raoux [Tue, 20 Aug 2019 15:54:59 +0000 (15:54 +0000)]
[CodeGen] Add a pass to do block predication on SSA machine IR.

    For targets requiring aggressive scheduling and/or software pipeline we need to
    apply predication before preRA scheduling. This adds a pass re-using the early
    if-cvt infrastructure but generating predicated instructions instead of
    speculatively executing instructions. It allows doing if conversion on blocks
    containing instructions with side-effects. The pass re-use the target hook from
    postRA if-conversion to let the target decide on the heuristic to apply.

    Differential Revision: https://reviews.llvm.org/D66190

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369395 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-objcopy][test] Add a test to show that argv[0] is included in error/warning...
Fangrui Song [Tue, 20 Aug 2019 15:34:07 +0000 (15:34 +0000)]
[llvm-objcopy][test] Add a test to show that argv[0] is included in error/warning messages

test/llvm-objcopy/ELF/error-format.test is similar to test/llvm-readobj/error-format.test added in D66425.

Reviewed By: grimar, jhenderson

Differential Revision: https://reviews.llvm.org/D66476

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369392 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-objcopy] Append '\n' to warning messages
Fangrui Song [Tue, 20 Aug 2019 15:00:07 +0000 (15:00 +0000)]
[llvm-objcopy] Append '\n' to warning messages

Currently the warning message of `llvm-strip %t.o %t.o` does not include
the trailing newline. Fix this by appending a '\n'.

This is the only warning llvm-objcopy and llvm-strip can issue.

Reviewed By: jhenderson

Differential Revision: https://reviews.llvm.org/D66475

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369391 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] improve readability for icmp with cast folds; NFC
Sanjay Patel [Tue, 20 Aug 2019 14:56:44 +0000 (14:56 +0000)]
[InstCombine] improve readability for icmp with cast folds; NFC

1. Update function name and stale code comments.
2. Use variable names that are less ambiguous.
3. Move operand checks into the function as early exits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369390 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[BlockExtractor] Avoid assert with wrong line format
Jinsong Ji [Tue, 20 Aug 2019 14:46:02 +0000 (14:46 +0000)]
[BlockExtractor] Avoid assert with wrong line format

Summary:
When the line format is wrong, we may end up accessing out of bound
memory. eg: the test with invalide line will cause assert.
Assertion `idx < size()' failed

The fix is to report fatal when we found mismatched line format.

Reviewers: qcolombet, volkan

Reviewed By: qcolombet

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66444

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369389 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86][BtVer2] Fix latency and throughput of atomic INC/DEC/NEG/NOT.
Andrea Di Biagio [Tue, 20 Aug 2019 14:31:27 +0000 (14:31 +0000)]
[X86][BtVer2] Fix latency and throughput of atomic INC/DEC/NEG/NOT.

Latency and throughput of LOCK INC/DEC/NEG/NOT is always 19cy.
Number of uOPs is still 1.

Differential Revision: https://reviews.llvm.org/D66469

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369388 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] simplify min/max of min/max with same operands (PR35607)
Sanjay Patel [Tue, 20 Aug 2019 13:39:17 +0000 (13:39 +0000)]
[InstCombine] simplify min/max of min/max with same operands (PR35607)

This is the original integer variant requested in:
https://bugs.llvm.org/show_bug.cgi?id=35607

As noted in the TODO and several similar TODOs around this block,
we could do this in instsimplify, but then it would cost more
because we would be trying to match min/max via ValueTracking
in 2 different places.

There are 4 commuted variants for each of smin/smax/umin/umax
that are not matched here. There are also icmp predicate variants
that are not included in the affected test file because they are
already handled by instsimplify by folding the final icmp to
true/false.

https://rise4fun.com/Alive/3KVc

  Name: smax(smax, smin)
  %c1 = icmp slt i32 %x, %y
  %c2 = icmp slt i32 %y, %x
  %min = select i1 %c1, i32 %x, i32 %y
  %max = select i1 %c2, i32 %x, i32 %y
  %c3 = icmp sgt i32 %max, %min
  %r = select i1 %c3, i32 %max, i32 %min
  =>
  %r = %max

  Name: smin(smax, smin)
  %c1 = icmp slt i32 %x, %y
  %c2 = icmp slt i32 %y, %x
  %min = select i1 %c1, i32 %x, i32 %y
  %max = select i1 %c2, i32 %x, i32 %y
  %c3 = icmp sgt i32 %max, %min
  %r = select i1 %c3, i32 %min, i32 %max
  =>
  %r = %min

  Name: umax(umax, umin)
  %c1 = icmp ult i32 %x, %y
  %c2 = icmp ult i32 %y, %x
  %min = select i1 %c1, i32 %x, i32 %y
  %max = select i1 %c2, i32 %x, i32 %y
  %c3 = icmp ult i32 %min, %max
  %r = select i1 %c3, i32 %max, i32 %min
  =>
  %r = %max

  Name: umin(umax, umin)
  %c1 = icmp ult i32 %x, %y
  %c2 = icmp ult i32 %y, %x
  %min = select i1 %c1, i32 %x, i32 %y
  %max = select i1 %c2, i32 %x, i32 %y
  %c3 = icmp ult i32 %min, %max
  %r = select i1 %c3, i32 %min, i32 %max
  =>
  %r = %min

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369386 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86][FMA] Add FMA 'negated expression' combine tests for D63141
Simon Pilgrim [Tue, 20 Aug 2019 13:25:55 +0000 (13:25 +0000)]
[X86][FMA] Add FMA 'negated expression' combine tests for D63141

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369384 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRegex: +regex string lifetime comment
Jan Kratochvil [Tue, 20 Aug 2019 13:25:19 +0000 (13:25 +0000)]
Regex: +regex string lifetime comment

Differential Revision: https://reviews.llvm.org/D66464

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369383 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-objdump] - Remove one of `report_error` functions and improve the error reporting.
George Rimar [Tue, 20 Aug 2019 13:19:16 +0000 (13:19 +0000)]
[llvm-objdump] - Remove one of `report_error` functions and improve the error reporting.

One of the report_error functions was taking object::Archive::Child as an
argument. It feels excessive, this patch removes it and introduce a helper
function instead. Also I fixed a "TODO" in this patch what improved the message printed.

Differential revision: https://reviews.llvm.org/D66468

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369382 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DWARF] Fix reading 64-bit DWARF type units.
Igor Kudrin [Tue, 20 Aug 2019 12:52:32 +0000 (12:52 +0000)]
[DWARF] Fix reading 64-bit DWARF type units.

The type_offset field is 8 bytes long in DWARF64. The patch extends
TypeOffset to uint64_t and fixes its reading. The patch also fixes
checking of TypeOffset bounds as it was inaccurate in DWARF64 case.

Differential Revision: https://reviews.llvm.org/D66465

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369378 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-readobj] Prepend argv[0] to error/warning messages
Fangrui Song [Tue, 20 Aug 2019 12:49:15 +0000 (12:49 +0000)]
[llvm-readobj] Prepend argv[0] to error/warning messages

Summary:
Currently, we report:

    error: ...

Prepend argv[0] (tool name):

    llvm-readobj: error: ...

This is consistent with most GNU binutils/clang/lld, and gives a bit
more context in a long build log.

Reviewed By: grimar, jhenderson, rupprecht

Differential Revision: https://reviews.llvm.org/D66425

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369377 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] add tests for min/max with min/max of same operands; NFC
Sanjay Patel [Tue, 20 Aug 2019 12:49:03 +0000 (12:49 +0000)]
[InstCombine] add tests for min/max with min/max of same operands; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369376 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[RISCV] Implement getExprForFDESymbol to ensure RISCV_32_PCREL is used for the FDE...
Alex Bradbury [Tue, 20 Aug 2019 12:32:31 +0000 (12:32 +0000)]
[RISCV] Implement getExprForFDESymbol to ensure RISCV_32_PCREL is used for the FDE location

Follow binutils in using RISCV_32_PCREL for the FDE initial location. As
explained in the relevant binutils commit
<https://github.com/riscv/riscv-binutils-gdb/commit/a6cbf936e3dce68114d28cdf60d510a3f78a6d40>,
the ADD/SUB pair of relocations is problematic in the presence of linker
relaxation.

This patch has the same end goal as D64715 but includes test changes and
avoids adding a new global VariantKind to MCExpr.h (preferring
RISCVMCExpr VKs like the rest of the RISC-V backend).

Differential Revision: https://reviews.llvm.org/D66419

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369375 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRecommit "MemoryBuffer: Add a missing error-check to getOpenFileImpl"
Pavel Labath [Tue, 20 Aug 2019 12:08:52 +0000 (12:08 +0000)]
Recommit "MemoryBuffer: Add a missing error-check to getOpenFileImpl"

This recommits r368977, which was reverted in r369027 due to test
failures in lldb. The cause of this was different behavior of
readNativeFileSlice on windows and unix. These have been addressed in
r369269.

The original commit message was:
In case the function was called with a desired read size *and* the file
was not an "mmap()" candidate, the function was falling back to a
"pread()", but it was failing to check the result of that system call.
This meant that the function would return "success" even though the read
operation failed, and it returned a buffer full of uninitialized memory.

Reviewers: rnk, dblaikie

Subscribers: kristina, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66224

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369370 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CMake] Update C4324 MSVC warning comment to explain its still broken at VS2019
Simon Pilgrim [Tue, 20 Aug 2019 11:20:05 +0000 (11:20 +0000)]
[CMake] Update C4324 MSVC warning comment to explain its still broken at VS2019

As promised, I've updated the comment for the C4324 MSVC warning that was re-disabled at rL367409 / rG8f823e63e3edf87ab029ba32b68f3eb5d2f392b5 to put it in terms of currently supported VS versions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369368 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MCA][X86] Add tests for LOCK variants of standard X86 arithmetic ops
Simon Pilgrim [Tue, 20 Aug 2019 11:13:20 +0000 (11:13 +0000)]
[MCA][X86] Add tests for LOCK variants of standard X86 arithmetic ops

D66424 adds the base support for LOCK so we should be able to add special case support for all these cases in future patches

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369367 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoFix MSVC "result of 32-bit shift implicitly converted to 64 bits" warning. NFCI.
Simon Pilgrim [Tue, 20 Aug 2019 10:25:57 +0000 (10:25 +0000)]
Fix MSVC "result of 32-bit shift implicitly converted to 64 bits" warning. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369366 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86][Btver2] Fix latency and throughput of CMPXCHG instructions.
Andrea Di Biagio [Tue, 20 Aug 2019 10:23:55 +0000 (10:23 +0000)]
[X86][Btver2] Fix latency and throughput of CMPXCHG instructions.

On Jaguar, CMPXCHG has a latency of 11cy, and a maximum throughput of 0.33 IPC.
Throughput is superiorly limited to 0.33 because of the implicit in/out
dependency on register EAX. In the case of repeated non-atomic CMPXCHG with the
same memory location, store-to-load forwarding occurs and values for sequent
loads are quickly forwarded from the store buffer.

Interestingly, the functionality in LLVM that computes the reciprocal throughput
doesn't seem to know about RMW instructions. That functionality only looks at
the "consumed resource cycles" for the throughput computation. It should be
fixed/improved by a future patch. In particular, for RMW instructions, that
logic should also take into account for the write latency of in/out register
operands.

An atomic CMPXCHG has a latency of ~17cy. Throughput is also limited to
~17cy/inst due to cache locking, which prevents other memory uOPs to start
executing before the "lock releasing" store uOP.

CMPXCHG8rr and CMPXCHG8rm are treated specially because they decode to one less
macro opcode. Their latency tend to be the same as the other RR/RM variants. RR
variants are relatively fast 3cy (but still microcoded - 5 macro opcodes).

CMPXCHG8B is 11cy and unfortunately doesn't seem to benefit from store-to-load
forwarding. That means, throughput is clearly limited by the in/out dependency
on GPR registers. The uOP composition is sadly unknown (due to the lack of PMCs
for the Integer pipes). I have reused the same mix of consumed resource from the
other CMPXCHG instructions for CMPXCHG8B too.
LOCK CMPXCHG8B is instead 18cycles.

CMPXCHG16B is 32cycles. Up to 38cycles when the LOCK prefix is specified. Due to
the in/out dependencies, throughput is limited to 1 instruction every 32 (or 38)
cycles dependeing on whether the LOCK prefix is specified or not.
I wouldn't be surprised if the microcode for CMPXCHG16B is similar to 2x
microcode from CMPXCHG8B. So, I have speculatively set the JALU01 consumption to
2x the resource cycles used for CMPXCHG8B.

The two new hasLockPrefix() functions are used by the btver2 scheduling model
check if a MCInst/MachineInst has a LOCK prefix. Calls to hasLockPrefix() have
been encoded in predicates of variant scheduling classes that describe lat/thr
of CMPXCHG.

Differential Revision: https://reviews.llvm.org/D66424

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369365 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[yaml2obj/obj2yaml][MachO] Fix a test failure in big endian hosts
Seiya Nuta [Tue, 20 Aug 2019 09:58:31 +0000 (09:58 +0000)]
[yaml2obj/obj2yaml][MachO] Fix a test failure in big endian hosts

These section contents are dummy data (0xdeadbeef) and it's endianess
does not matter.

- http://lab.llvm.org:8011/builders/clang-ppc64be-linux/builds/37265

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369360 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DWARF] Fix DWARFUnit::getDebugInfoSize() for 64-bit DWARF.
Igor Kudrin [Tue, 20 Aug 2019 09:50:44 +0000 (09:50 +0000)]
[DWARF] Fix DWARFUnit::getDebugInfoSize() for 64-bit DWARF.

The calculation there was correct only for DWARF32.

Differential Revision: https://reviews.llvm.org/D66421

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369356 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[yaml2obj/obj2yaml][MachO] Allow setting custom section data
Seiya Nuta [Tue, 20 Aug 2019 08:49:07 +0000 (08:49 +0000)]
[yaml2obj/obj2yaml][MachO] Allow setting custom section data

Reviewers: alexshap, jhenderson, rupprecht

Reviewed By: alexshap, jhenderson

Subscribers: abrachet, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65799

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369348 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-objcopy][MachO] Fix method names. NFC.
Seiya Nuta [Tue, 20 Aug 2019 08:36:47 +0000 (08:36 +0000)]
[llvm-objcopy][MachO] Fix method names. NFC.

Reviewers: alexshap, rupprecht, jhenderson

Reviewed By: alexshap, rupprecht

Subscribers: jakehehrlich, abrachet, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65540

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369346 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[test/Object] - Move/rewrite 2 more test cases.
George Rimar [Tue, 20 Aug 2019 08:23:57 +0000 (08:23 +0000)]
[test/Object] - Move/rewrite 2 more test cases.

This patch makes a change for test/Object tests responsible
for relocations.

* 2 tests were moved to llvm-readobj/llvm-objdump folders:
Object/elf-reloc-no-sym.test -> tools/llvm-readobj/elf-reloc-no-sym.test
Object/objdump-reloc-shared.test -> tools/llvm-objdump/relocations-in-nonreloc.test

* A prerecompiled binary was removed and these tests were refactored.

Differential revision: https://reviews.llvm.org/D66291

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369342 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MC] Delete an overload of MCExpr::evaluateKnownAbsolute and its associated hack
Fangrui Song [Tue, 20 Aug 2019 07:42:04 +0000 (07:42 +0000)]
[MC] Delete an overload of MCExpr::evaluateKnownAbsolute and its associated hack

The hack dated back to 2010 (r121076) and was documented by r122144:

  // FIXME: The use if InSet = Addrs is a hack. Setting InSet causes us
  // absolutize differences across sections and that is what the MachO writer
  // uses Addrs for.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369337 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Attributor] Fix -Wunused-variable in -DLLVM_ENABLE_ASSERTIONS=off builds after r369331
Fangrui Song [Tue, 20 Aug 2019 07:21:43 +0000 (07:21 +0000)]
[Attributor] Fix -Wunused-variable in -DLLVM_ENABLE_ASSERTIONS=off builds after r369331

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369334 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add back the -x86-experimental-vector-widening-legalization comand line flag...
Craig Topper [Tue, 20 Aug 2019 06:58:00 +0000 (06:58 +0000)]
[X86] Add back the -x86-experimental-vector-widening-legalization comand line flag and all associated code, but leave it enabled by default

Google is reporting performance issues with the new default behavior
and have asked for a way to switch back to the old behavior while we
investigate and make fixes.

I've restored all of the code that had since been removed and added
additional checks of the command flag onto code paths that are
not otherwise guarded by a check of getTypeAction.

I've also modified the cost model tables to hopefully get us back
to the previous costs.

Hopefully we won't need to support this for very long since we
have no test coverage of the old behavior so we can very easily
break it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369332 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Attributor] Create abstract attributes on-demand
Johannes Doerfert [Tue, 20 Aug 2019 06:15:50 +0000 (06:15 +0000)]
[Attributor] Create abstract attributes on-demand

Before, we create the set of abstract attributes initially and then
dealt with the fact hat a lookup could fail, e.g., return a nullptr.
This patch will ensure we always return a valid object from a lookup,
allowing us not only to remove the nullptr checks but also to grow the
set of abstract attributes "in-flight" on-demand.

One can now start from those that have the best chance of improving
performance without the need to specify all they might depend on.

While this introduces some boilerplate, the usage of attributes is much
easier and cleaner now.

Reviewers: uenoku, sstefan1

Subscribers: hiraditya, bollu, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66276

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369331 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Attributor][NFC] Cleanup statistics code
Johannes Doerfert [Tue, 20 Aug 2019 06:09:56 +0000 (06:09 +0000)]
[Attributor][NFC] Cleanup statistics code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369330 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Attributor] Use structured deduction for AADereferenceable
Johannes Doerfert [Tue, 20 Aug 2019 06:08:35 +0000 (06:08 +0000)]
[Attributor] Use structured deduction for AADereferenceable

Summary:
This is analogous to D66128 but for AADereferenceable. We have the logic
concentrated in the floating value updateImpl and we use the combiner
helper classes for arguments and return values.

The regressions will go away with "on-demand" attribute creation.
Improvements are already visible in the existing tests.

Reviewers: uenoku, sstefan1

Subscribers: hiraditya, bollu, jfb, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66272

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369329 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Attributor] Use structured deduction for AANonNull
Johannes Doerfert [Tue, 20 Aug 2019 06:02:39 +0000 (06:02 +0000)]
[Attributor] Use structured deduction for AANonNull

Summary:
What D66126 did for AAAlign, this patch does for AANonNull. Agian, the
logic becomes more concise and localized. Again, returned poiners are
not annotated properly but that will not be an issue if this lands with
the "on-demand" generation of attributes. First improvements due to the
genericValueTraversal are already visible.

Reviewers: sstefan1, uenoku

Subscribers: hiraditya, bollu, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66128

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369328 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Attributor] Fix the "clamp" operator
Johannes Doerfert [Tue, 20 Aug 2019 05:57:01 +0000 (05:57 +0000)]
[Attributor] Fix the "clamp" operator

The clamp operator should not take the known of the given state as the
known is potentially based on assumed information. This also adds TODOs
to guide improvements.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369327 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[NFC] Test commit, fix some comment spelling.
Thomas Raoux [Tue, 20 Aug 2019 05:21:27 +0000 (05:21 +0000)]
[NFC] Test commit, fix some comment spelling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369326 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AsmPrinter] Remove const qualifier from EmitBasicBlockStart.
Karl-Johan Karlsson [Tue, 20 Aug 2019 05:13:57 +0000 (05:13 +0000)]
[AsmPrinter] Remove const qualifier from EmitBasicBlockStart.

Overriders may want to modify state in it. AMDGPU wants
to, but has to make its members mutable in order to do so.

Besides, EmitBasicBlockEnd is not const, so why should
Start be?

Patch by Bevin Hansson.

Reviewed By: nickdesaulniers

Differential Revision: https://reviews.llvm.org/D66341

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369325 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoMCAsmMacro: add `#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)` to some dump...
Fangrui Song [Tue, 20 Aug 2019 04:14:43 +0000 (04:14 +0000)]
MCAsmMacro: add `#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)` to some dump() declarations

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369324 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[WebAssembly][MC] Fix -Wunused-variable in -DLLVM_ENABLE_ASSERTIONS=off builds after...
Fangrui Song [Tue, 20 Aug 2019 02:02:57 +0000 (02:02 +0000)]
[WebAssembly][MC] Fix -Wunused-variable in -DLLVM_ENABLE_ASSERTIONS=off builds after r369317

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369318 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[WebAssembly][MC] Simplify WasmObjectWriter::recordRelocation. NFC.
Sam Clegg [Tue, 20 Aug 2019 00:33:50 +0000 (00:33 +0000)]
[WebAssembly][MC] Simplify WasmObjectWriter::recordRelocation. NFC.

WebAssembly doesn't support PC relative relocation or relocation
expressions that can't be reduced to single symbol.

The only support for we have for fixups involving two symbols are when
both symbols are defined and withing the same section.  In this case
evaluateFixup will already have evaluated to the expression before
calling recordRelocation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369317 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[SLP][NFC] Avoid repetitive calls to getSameOpcode()
Dinar Temirbulatov [Tue, 20 Aug 2019 00:22:04 +0000 (00:22 +0000)]
[SLP][NFC] Avoid repetitive calls to getSameOpcode()

We can avoid repetitive calls getSameOpcode() for already known tree elements by keeping MainOp and AltOp in TreeEntry.

Differential Revision: https://reviews.llvm.org/D64700

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369315 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[cmake] Link in LLVMPasses due to dependency by LLVMOrcJIT; NFC
Hubert Tong [Mon, 19 Aug 2019 23:12:48 +0000 (23:12 +0000)]
[cmake] Link in LLVMPasses due to dependency by LLVMOrcJIT; NFC

Summary:
rL367756 (f5c40cb) increases the dependency of LLVMOrcJIT on LLVMPasses.
In particular, symbols defined in LLVMPasses that are referenced by the
destructor of `PassBuilder` are now referenced by LLVMOrcJIT through
`Speculation.cpp.o`.

We believe that referencing symbols defined in LLVMPasses in the
destructor of `PassBuilder` is valid, and that adding to the set of such
symbols is legitimate. To support such cases, this patch adds LLVMPasses
to the set of libraries being linked when linking in LLVMOrcJIT causes
such symbols from LLVMPasses to be referenced.

Reviewers: Whitney, anhtuyen, pree-jackie

Reviewed By: pree-jackie

Subscribers: mgorny, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66441

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369310 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[lit] Check for accidental external command calls
Joel E. Denny [Mon, 19 Aug 2019 22:59:37 +0000 (22:59 +0000)]
[lit] Check for accidental external command calls

This patch extends lit's test suite to check that lit's internal shell
doesn't accidentally execute internal commands as external commands.
It does so by putting fake failing versions of those commands in
`PATH` while the entire lit test suite is running.  Without the fixes
in D65697 but with its tests, this approach catches accidental
external `env` calls.

Reviewed By: probinson

Differential Revision: https://reviews.llvm.org/D66293

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369309 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Support][Time profiler] Make FE codegen blocks to be inside frontend blocks
Anton Afanasyev [Mon, 19 Aug 2019 22:58:26 +0000 (22:58 +0000)]
[Support][Time profiler] Make FE codegen blocks to be inside frontend blocks

Summary:
Add `Frontend` time trace entry to `HandleTranslationUnit()` function.
Add test to check all codegen blocks are inside frontend blocks.
Also, change `--time-trace-granularity` option a bit to make sure very small
time blocks are outputed to json-file when using `--time-trace-granularity=0`.

This fixes http://llvm.org/pr41969

Reviewers: russell.gallop, lebedev.ri, thakis

Reviewed By: russell.gallop

Subscribers: vsapsai, aras-p, lebedev.ri, hiraditya, cfe-commits, llvm-commits

Tags: #clang, #llvm

Differential Revision: https://reviews.llvm.org/D63325

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369308 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ORC] fix use-after-free detected by -Wreturn-stack-address
Matthias Gehre [Mon, 19 Aug 2019 21:59:44 +0000 (21:59 +0000)]
[ORC] fix use-after-free detected by -Wreturn-stack-address

Summary:
llvm/lib/ExecutionEngine/Orc/Layer.cpp:53:12: warning: returning address of local temporary object [-Wreturn-stack-address]

In
```
StringRef IRMaterializationUnit::getName() const {
[...]
     return TSM.withModuleDo(
        [](const Module &M) { return M.getModuleIdentifier(); });
```
`getModuleIdentifier()` returns a `const std::string &`, but the implicit return type
of the lambda is `std::string` by value, and thus the returned `StringRef` refers
to a temporary `std::string`.

Detect by annotating `llvm::StringRef` with `[[gsl::Pointer]]`.

Reviewers: lhames, sgraenitz

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66440

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369306 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CaptureTracker] Let subclasses provide dereferenceability information
Johannes Doerfert [Mon, 19 Aug 2019 21:56:38 +0000 (21:56 +0000)]
[CaptureTracker] Let subclasses provide dereferenceability information

Summary:
CaptureTracker subclasses might have better dereferenceability
information which allows null pointer checks to be no-capturing.
The first user will be D59922.

Reviewers: sanjoy, hfinkel, aykevl, sstefan1, uenoku, xbolva00

Subscribers: hiraditya, bollu, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66371

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369305 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRecommit "[Attributor] Fix: Do not partially resolve returned calls."
Johannes Doerfert [Mon, 19 Aug 2019 21:35:31 +0000 (21:35 +0000)]
Recommit "[Attributor] Fix: Do not partially resolve returned calls."

This reverts commit b1752f670f3d6393306dd5d37546b6e23384d8a2.

Fixed the issue with a different commit, reapply this one as it was,
afaik, not broken.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369303 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRecommit "[llvm-objcopy][MachO] Implement a layout algorithm for executables"
Seiya Nuta [Mon, 19 Aug 2019 21:12:02 +0000 (21:12 +0000)]
Recommit "[llvm-objcopy][MachO] Implement a layout algorithm for executables"

Summary: The layout algorithm for relocatable objects and for executable are somewhat different. This patch implements the latter one based on the algorithm in LLD (MachOFileLayout).

Reviewers: alexshap, rupprecht, jhenderson

Reviewed By: alexshap

Subscribers: jakehehrlich, abrachet, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65539

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369301 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRefactor isPointerOffset (NFC).
Evgeniy Stepanov [Mon, 19 Aug 2019 21:08:04 +0000 (21:08 +0000)]
Refactor isPointerOffset (NFC).

Summary:
Simplify the API using Optional<> and address comments in
         https://reviews.llvm.org/D66165

Reviewers: vitalybuka

Subscribers: hiraditya, llvm-commits, ostannard, pcc

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66317

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369300 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoFixed placement of llvm.global_dtors on Windows.
Vyacheslav Zakharin [Mon, 19 Aug 2019 21:07:03 +0000 (21:07 +0000)]
Fixed placement of llvm.global_dtors on Windows.

Differential revision: https://reviews.llvm.org/D66373

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369299 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRecommit "[llvm-objcopy][MachO] Support load commands used in executables/shared...
Seiya Nuta [Mon, 19 Aug 2019 21:05:31 +0000 (21:05 +0000)]
Recommit "[llvm-objcopy][MachO] Support load commands used in executables/shared libraries"

Summary:
This patch implements copying some load commands that appear in executables/shared libraries such as the indirect symbol table.

I don't add tests intentionally because this patch is incomplete: we need a layout algorithm for executables/shared libraries. I'll submit it as a separate patch with tests.

Reviewers: alexshap, rupprecht, jhenderson, compnerd

Reviewed By: alexshap

Subscribers: abrachet, mgorny, mgrang, MaskRay, mtrent, jakehehrlich, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D63395

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369298 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoMemTag: stack initializer merging.
Evgeniy Stepanov [Mon, 19 Aug 2019 20:47:09 +0000 (20:47 +0000)]
MemTag: stack initializer merging.

Summary:
MTE provides instructions to update memory tags and data at the same
time. This change makes use of those to generate more compact code for
stack variable tagging + initialization.

We collect memory store and memset instructions following an alloca or a
lifetime.start call, and replace them with the corresponding MTE
intrinsics. Since the intrinsics work on 16-byte aligned chunks, the
stored values are combined as necessary.

Reviewers: pcc, vitalybuka, ostannard

Subscribers: srhines, javed.absar, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66167

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369297 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Support] Replace sys::Mutex with their standard equivalents.
Benjamin Kramer [Mon, 19 Aug 2019 19:49:57 +0000 (19:49 +0000)]
[Support] Replace sys::Mutex with their standard equivalents.

Only use a recursive mutex if it can be locked recursively.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369295 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRe-apply fixed "[Attributor] Fix: Make sure we set the changed flag"
Johannes Doerfert [Mon, 19 Aug 2019 19:14:10 +0000 (19:14 +0000)]
Re-apply fixed "[Attributor] Fix: Make sure we set the changed flag"

This reverts commit cedd0d9a6e4b433e1cd6585d1d4d152eb5e60b11.

Re-apply the original commit but make sure the variables are initialized
(even if they are not used) so UBSan is not complaining.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369294 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[WebAssembly][MC] Allow empty assembly functions
Sam Clegg [Mon, 19 Aug 2019 19:04:54 +0000 (19:04 +0000)]
[WebAssembly][MC] Allow empty assembly functions

Differential Revision: https://reviews.llvm.org/D66434

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369292 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MemorySSA] Rename uses when inserting memory uses.
Alina Sbirlea [Mon, 19 Aug 2019 18:57:40 +0000 (18:57 +0000)]
[MemorySSA] Rename uses when inserting memory uses.

Summary:
When inserting uses from outside the MemorySSA creation, we don't
normally need to rename uses, based on the assumption that there will be
no inserted Phis (if  Def existed that required a Phi, that Phi already
exists). However, when dealing with unreachable blocks, MemorySSA will
optimize away Phis whose incoming blocks are unreachable, and these Phis end
up being re-added when inserting a Use.
There are two potential solutions here:
1. Analyze the inserted Phis and clean them up if they are unneeded
(current method for cleaning up trivial phis does not cover this)
2. Leave the Phi in place and rename uses, the same way as whe inserting
defs.
This patch use approach 2.

Resolves first test in PR42940.

Reviewers: george.burgess.iv

Subscribers: Prazek, sanjoy.google, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66033

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369291 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Move scheduling tests for CMPXCHG to the corresponding resources-x86_64.s files...
Andrea Di Biagio [Mon, 19 Aug 2019 18:20:30 +0000 (18:20 +0000)]
[X86] Move scheduling tests for CMPXCHG to the corresponding resources-x86_64.s files. NFC

In D66424 it has been requested to move all the new tests added by r369278 into
resources-x86_64.s. That is because only the 8b/16 ops should be tested by
resources-cmpxchg.s. This partially reverts r369278.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369288 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Teach lowerV4I32Shuffle to only use broadcasts if the mask has more than one...
Craig Topper [Mon, 19 Aug 2019 18:15:50 +0000 (18:15 +0000)]
[X86] Teach lowerV4I32Shuffle to only use broadcasts if the mask has more than one undef element. Prioritize shifts over broadcast in lowerV8I16Shuffle.

The motivating case are the changes in vector-reduce-add.ll where
we were doing extra work in the scalar domain instead of shuffling.
There may be some one use check that needs to be looked into there,
but this patch sidesteps the issue by avoiding broadcasts that
aren't really broadcasting.

Differential Revision: https://reviews.llvm.org/D66071

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369287 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CGP] Remove ModifiedDT from the makeBitReverse loop
Craig Topper [Mon, 19 Aug 2019 18:02:24 +0000 (18:02 +0000)]
[CGP] Remove ModifiedDT from the makeBitReverse loop

I don't think anything in this loop modifies the control flow and we don't restart any iteration after setting the flag.

This code was added in http://reviews.llvm.org/D16893 but looking at the test case added there the code that caused the dominator tree to change was merging blocks with their predecessor not the bitreverse optimization.

Differential Revision: https://reviews.llvm.org/D66366

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369283 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Added extensive scheduling model tests for all the CMPXCHG variants. NFC
Andrea Di Biagio [Mon, 19 Aug 2019 17:07:26 +0000 (17:07 +0000)]
[X86] Added extensive scheduling model tests for all the CMPXCHG variants. NFC

Addresses a review comment in D66424

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369279 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Attributor] -attributor-verify has false positives. Temporarily remove
Stefan Stipanovic [Mon, 19 Aug 2019 16:39:32 +0000 (16:39 +0000)]
[Attributor] -attributor-verify has false positives. Temporarily remove
it, until resolved.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369277 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoFilesystem/Windows: fix inconsistency in readNativeFileSlice API
Pavel Labath [Mon, 19 Aug 2019 15:40:49 +0000 (15:40 +0000)]
Filesystem/Windows: fix inconsistency in readNativeFileSlice API

Summary:
The windows version implementation of readNativeFileSlice, was trying to
match the POSIX behavior of not treating EOF as an error, but it was
only handling the case of reading from a pipe. Attempting to read past
the end of a regular file returns a slightly different error code, which
needs to be handled too. This patch adds ERROR_HANDLE_EOF to the list of
error codes to be treated as an end of file, and adds some unit tests
for the API.

This issue was found while attempting to land D66224, which caused a bunch of
lldb tests to start failing on windows.

Reviewers: rnk, aganea

Subscribers: kristina, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66344

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369269 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[TargetLowering] x s% C == 0 fold: vector divisor with INT_MIN handling
Roman Lebedev [Mon, 19 Aug 2019 15:01:42 +0000 (15:01 +0000)]
[TargetLowering] x s% C == 0 fold: vector divisor with INT_MIN handling

Summary:
The general fold is only valid for positive divisors.
Which effectively means, it is invalid for `INT_MIN` divisors,
and we currently bailout if we see them.

But that is too strict, we can just fix-up the results.
For that, let's do a second computation 'in parallel':
```
Name: srem -> and
Pre: isPowerOf2(C)
%o = srem i8 %X, C
%r = icmp eq %o, 0
  =>
%n = and i8 %X, C-1
%r = icmp eq %n, 0
```
https://rise4fun.com/Alive/Sup

And then just blend results: if the divisor was `INT_MIN`,
pick the value we got via bit-test,
else pick the value from general fold.

There's interesting observation - `ISD::ROTR` is set to
`LegalizeAction::Expand` before AVX512, so we should not
treat `INT_MIN` divisor as even; and as it can be seen
while `@test_srem_odd_even_one` improves on all run-lines,
`@test_srem_odd_even_INT_MIN` only improves for AVX512.

Reviewers: RKSimon, craig.topper, spatel

Reviewed By: RKSimon

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66300

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369268 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[nfc] Silent gcc warning
Serge Guelton [Mon, 19 Aug 2019 14:40:33 +0000 (14:40 +0000)]
[nfc] Silent gcc warning

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369266 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-objdump] - Restore lost comment.
George Rimar [Mon, 19 Aug 2019 14:35:08 +0000 (14:35 +0000)]
[llvm-objdump] - Restore lost comment.

It was removed by mistake in r369263 (D66417).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369265 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAutomatically generate AVX512 test cases. NFC
Amaury Sechet [Mon, 19 Aug 2019 14:34:08 +0000 (14:34 +0000)]
Automatically generate AVX512 test cases. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369264 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Object/COFF.h] - Stop returning std::error_code in a few methods. NFCI.
George Rimar [Mon, 19 Aug 2019 14:32:23 +0000 (14:32 +0000)]
[Object/COFF.h] - Stop returning std::error_code in a few methods. NFCI.

There are 4 methods that return std::error_code now,
though they do not have to because they are always succeed.
I refactored them.

This allows to simplify the code in tools a bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369263 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[PeepholeOptimizer] Don't assume bitcast def always has input
Jinsong Ji [Mon, 19 Aug 2019 14:19:04 +0000 (14:19 +0000)]
[PeepholeOptimizer] Don't assume bitcast def always has input

Summary:
If we have a MI marked with bitcast bits, but without input operands,
PeepholeOptimizer might crash with assert.

eg:
If we apply the changes in PPCInstrVSX.td as in this patch:

[(set v4i32:$XT, (bitconvert (v16i8 immAllOnesV)))]>;

We will get assert in PeepholeOptimizer.

```
llvm-lit llvm-project/llvm/test/CodeGen/PowerPC/build-vector-tests.ll -v

llvm-project/llvm/include/llvm/CodeGen/MachineInstr.h:417: const
llvm::MachineOperand &llvm::MachineInstr::getOperand(unsigned int)
const: Assertion `i < getNumOperands() && "getOperand() out of range!"'
failed.
```

The fix is to abort if we found out of bound access.

Reviewers: qcolombet, MatzeB, hfinkel, arsenm

Reviewed By: qcolombet

Subscribers: wdng, arsenm, steven.zhang, wuzish, nemanjai, hiraditya, kbarton, MaskRay, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65542

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369261 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[RISCV] Don't force absolute FK_Data_X fixups to relocs
Alex Bradbury [Mon, 19 Aug 2019 13:23:02 +0000 (13:23 +0000)]
[RISCV] Don't force absolute FK_Data_X fixups to relocs

The current behavior of shouldForceRelocation forces relocations for the
majority of fixups when relaxation is enabled. This makes sense for
fixups which incorporate symbols but is unnecessary for simple data
fixups where the fixup target is already resolved to an absolute value.

Differential Revision: https://reviews.llvm.org/D63404
Patch by Edward Jones.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369257 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DebugInfo] Allow bundled calls in the MIR's call site info
David Stenberg [Mon, 19 Aug 2019 12:41:22 +0000 (12:41 +0000)]
[DebugInfo] Allow bundled calls in the MIR's call site info

Summary:
Extend the MIR parser and writer so that the call site information can
refer to calls that are bundled.

Reviewers: aprantl, asowda, NikolaPrica, djtodoro, ivanbaev, vsk

Reviewed By: aprantl

Subscribers: arsenm, hiraditya, llvm-commits

Tags: #debug-info, #llvm

Differential Revision: https://reviews.llvm.org/D66145

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369256 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[SLP] add test that requires shuffle of scalars; NFC
Sanjay Patel [Mon, 19 Aug 2019 12:41:09 +0000 (12:41 +0000)]
[SLP] add test that requires shuffle of scalars; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369255 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-readobj] unwrapOrError: remove unnecessary llvm_unreachable after r369194
Fangrui Song [Mon, 19 Aug 2019 12:18:18 +0000 (12:18 +0000)]
[llvm-readobj] unwrapOrError: remove unnecessary llvm_unreachable after r369194

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369254 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[SLP] reduce duplicated code; NFC
Sanjay Patel [Mon, 19 Aug 2019 11:39:56 +0000 (11:39 +0000)]
[SLP] reduce duplicated code; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369250 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MC] Simplify ELFObjectWriter::recordRelocation. NFC
Fangrui Song [Mon, 19 Aug 2019 10:05:59 +0000 (10:05 +0000)]
[MC] Simplify ELFObjectWriter::recordRelocation. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369248 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DebugInfo] Make postra sinking of DBG_VALUEs subregister-safe
Jeremy Morse [Mon, 19 Aug 2019 09:53:07 +0000 (09:53 +0000)]
[DebugInfo] Make postra sinking of DBG_VALUEs subregister-safe

Currently the machine instruction sinker identifies DBG_VALUE insts that
also need to sink by comparing register numbers. Unfortunately this isn't
safe, because (after register allocation) a DBG_VALUE may read a register
that aliases what's being sunk. To fix this, identify the DBG_VALUEs that
need to sink by recording & examining their register units. Register units
gives us the following guarantee:

  "Two registers overlap if and only if they have a common register unit"
  [MCRegisterInfo.h]

Thus we can always identify aliasing DBG_VALUEs if the set of register
units read by the DBG_VALUE, and the register units of the instruction
being sunk, intersect. (MachineSink already uses classes like
"LiveRegUnits" for determining sinking validity anyway).

The test added checks for super and subregister DBG_VALUE reads of a sunk
copy being sunk as well.

Differential Revision: https://reviews.llvm.org/D58191

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369247 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ARM] Add support for MVE vaddv
Sam Tebbs [Mon, 19 Aug 2019 09:38:28 +0000 (09:38 +0000)]
[ARM] Add support for MVE vaddv

This patch adds vecreduce_add and the relevant instruction selection for
vaddv.

Differential revision: https://reviews.llvm.org/D66085

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369245 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ARM] MVE sext costs
David Green [Mon, 19 Aug 2019 09:13:22 +0000 (09:13 +0000)]
[ARM] MVE sext costs

This adds some sext costs for MVE, taken from the length of assembly sequences
that we currently generate.

Differential Revision: https://reviews.llvm.org/D66010

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369244 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DebugInfo] Test for variable range un-coalescing
Jeremy Morse [Mon, 19 Aug 2019 09:02:18 +0000 (09:02 +0000)]
[DebugInfo] Test for variable range un-coalescing

LiveDebugVariables can coalesce ranges of variable locations across
multiple basic blocks. However when it recreates DBG_VALUE instructions,
it has to recreate one DBG_VALUE per block, otherwise it doesn't
represent the pre-regalloc layout and variable assignments can go missing.

This feature works -- however while mucking around with LiveDebugVariables,
I commented the relevant code it out and no tests failed. Thus, here's a
test that checks LiveDebugVariables preserves DBG_VALUEs across block
boundaries.

Differential Revision: https://reviews.llvm.org/D66347

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369243 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRevert [Attributor] Fix: Make sure we set the changed flag
David L. Jones [Mon, 19 Aug 2019 08:00:08 +0000 (08:00 +0000)]
Revert [Attributor] Fix: Make sure we set the changed flag

This reverts r369159 (git commit cbaf1fdea2de891bdbc49cdec89ae2077e6b9ed0)

r369160 caused a test to fail under UBSAN. See thread on llvm-commits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369241 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MC] Delete unnecessary diagnostic: "No relocation available to represent this relati...
Fangrui Song [Mon, 19 Aug 2019 07:59:35 +0000 (07:59 +0000)]
[MC] Delete unnecessary diagnostic: "No relocation available to represent this relative expression"

Replace

- error: No relocation available to represent this relative expression

with

+ error: symbol 'undef' can not be undefined in a subtraction expression

or

+ error: Cannot represent a difference across sections

Keep !IsPcRel as an assertion after the two diagnostic checks are done.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369239 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRevert [Attributor] Fix: Do not partially resolve returned calls.
David L. Jones [Mon, 19 Aug 2019 07:16:24 +0000 (07:16 +0000)]
Revert [Attributor] Fix: Do not partially resolve returned calls.

This reverts r369160 (git commit f72d9b1c97b41fff48ad1eecbba59a29c171bff4)

r369160 caused some tests to fail under UBSAN. See thread on llvm-commits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369236 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRevert r369230 and r369231
Seiya Nuta [Mon, 19 Aug 2019 06:45:48 +0000 (06:45 +0000)]
Revert r369230 and r369231

Looks these commits break CI builds:

- http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-gn/builds/4159

This commit reverts r369230 and r369231 (git coommit: 4a198a7 and dee9546).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369234 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MC] Don't emit .symver redirected symbols to the symbol table
Fangrui Song [Mon, 19 Aug 2019 06:17:30 +0000 (06:17 +0000)]
[MC] Don't emit .symver redirected symbols to the symbol table

GNU as keeps the original symbol in the symbol table for defined @ and
@@, but suppresses it in other cases (@@@ or undefined). The original
symbol is usually undesired:
In a shared object, the original symbol can be localized with a version
script, but it is hard to remove/localize in an archive:

1) a post-processing step removes the undesired original symbol
2) consumers (executable) of the archive are built with the
   version script

Moreover, it can cause linker issues like binutils PR/18703 if the
original symbol name and the base name of the versioned symbol is the
same (both ld.bfd and gold have some code to work around defined @ and
@@). In lld, if it sees f and f@v1:

  --version-script =(printf 'v1 {};') => f and f@v1
  --version-script =(printf 'v1 { f; };') => f@v1 and f@@v1

It can be argued that @@@ added on 2000-11-13 corrected the @ and @@ mistake.

This patch catches some more multiple version errors (defined @ and @@),
and consistently suppress the original symbol. This addresses all the
problems listed above.

If the user wants other aliases to the versioned symbol, they can copy
the original symbol to other symbol names with .set directive, e.g.

    .symver f, f@v1  # emit f@v1 but not f into .symtab
    .set f_impl, f   # emit f_impl into .symtab

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369233 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Teach lower1BitShuffle to match right shifts with upper zero elements on types...
Craig Topper [Mon, 19 Aug 2019 05:45:39 +0000 (05:45 +0000)]
[X86] Teach lower1BitShuffle to match right shifts with upper zero elements on types that don't natively support KSHIFT.

We can support these by widening to a supported type,
then shifting all the way to the left and then
back to the right to ensure that we shift in zeroes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369232 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-objcopy][MachO] Implement a layout algorithm for executables
Seiya Nuta [Mon, 19 Aug 2019 05:41:33 +0000 (05:41 +0000)]
[llvm-objcopy][MachO] Implement a layout algorithm for executables

Summary: The layout algorithm for relocatable objects and for executable are somewhat different. This patch implements the latter one based on the algorithm in LLD (MachOFileLayout).

Reviewers: alexshap, rupprecht, jhenderson

Reviewed By: alexshap

Subscribers: jakehehrlich, abrachet, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65539

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369231 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-objcopy][MachO] Support load commands used in executables/shared libraries
Seiya Nuta [Mon, 19 Aug 2019 05:37:38 +0000 (05:37 +0000)]
[llvm-objcopy][MachO] Support load commands used in executables/shared libraries

Summary:
This patch implements copying some load commands that appear in executables/shared libraries such as the indirect symbol table.

I don't add tests intentionally because this patch is incomplete: we need a layout algorithm for executables/shared libraries. I'll submit it as a separate patch with tests.

Reviewers: alexshap, rupprecht, jhenderson, compnerd

Reviewed By: alexshap

Subscribers: abrachet, mgorny, mgrang, MaskRay, mtrent, jakehehrlich, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D63395

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369230 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Fix the lower1BitShuffle code added in r369215 to correctly pass the widened...
Craig Topper [Mon, 19 Aug 2019 04:08:44 +0000 (04:08 +0000)]
[X86] Fix the lower1BitShuffle code added in r369215 to correctly pass the widened vector to the KSHIFT node.

Not sure how to test this as we have tests that exercise this code,
but nothing failed for the types not matching. Since all the k-registers
use equivalent register classes everything just ends up working.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369228 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Teach lower1BitShuffle to match KSHIFTR that doesn't use Zeroable and only...
Craig Topper [Mon, 19 Aug 2019 04:08:40 +0000 (04:08 +0000)]
[X86] Teach lower1BitShuffle to match KSHIFTR that doesn't use Zeroable and only relies on undef.

This allows us to widen the type when the KSHIFTR instruction
doesn't exist for the type. If we need to shift in zeroes into
the upper elements we would need more work to guarantee zeroes
when widening.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369227 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Teach lower1BitShuffle to recognize padding a subvector with zeros with V2...
Craig Topper [Mon, 19 Aug 2019 00:39:22 +0000 (00:39 +0000)]
[X86] Teach lower1BitShuffle to recognize padding a subvector with zeros with V2 as the source and V1 as the zero vector.

Shuffle canonicalization can swap the sources so the zero vector
might be V1 and the subvector that's being padded can be V2.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369226 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add test case for missed opportunity to recognize a vXi1 shuffle as an insert...
Craig Topper [Mon, 19 Aug 2019 00:39:18 +0000 (00:39 +0000)]
[X86] Add test case for missed opportunity to recognize a vXi1 shuffle as an insert into a zero vector.

We are currently missing this because shuffle canonicalization
puts the zero vector as V1 and the subvector as V2. Our current
code doesn't recognize this case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369225 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add a special case to LowerCONCAT_VECTORSvXi1 to handle concatenating zero...
Craig Topper [Sun, 18 Aug 2019 23:30:11 +0000 (23:30 +0000)]
[X86] Add a special case to LowerCONCAT_VECTORSvXi1 to handle concatenating zero vectors followed by one non-zero vector followed by undef vectors.

For such a case we should only need a KSHIFTL, but we were
previously generating a KSHIFTL followed by a KSHIFTR because
we mistakenly believed we need to zero the undef elements.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369224 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add test cases for suboptimal insertion of a vXi1 vector into a larger vector...
Craig Topper [Sun, 18 Aug 2019 23:30:07 +0000 (23:30 +0000)]
[X86] Add test cases for suboptimal insertion of a vXi1 vector into a larger vector with zeros in the lower elements and undef upper elements.

Currently we generate kshifts to clear both the upper and lower
elements, but we only need one kshift.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369223 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Replace uses of getZeroVector for vXi1 vectors with DAG.getConstant.
Craig Topper [Sun, 18 Aug 2019 23:30:03 +0000 (23:30 +0000)]
[X86] Replace uses of getZeroVector for vXi1 vectors with DAG.getConstant.

vXi1 vectors don't need special handling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369222 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[cmake] Move blocks out of redundant else( MSVC ); NFC
Hubert Tong [Sun, 18 Aug 2019 22:02:24 +0000 (22:02 +0000)]
[cmake] Move blocks out of redundant else( MSVC ); NFC

Address post-commit comment on D66256 regarding the `else( MSVC )` block
containing only blocks guarded with `LLVM_COMPILER_IS_GCC_COMPATIBLE`,
which would imply `NOT MSVC`.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369221 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ORC] Make sure we linker-mangle symbol names in the SpeculationLayer.
Lang Hames [Sun, 18 Aug 2019 21:29:57 +0000 (21:29 +0000)]
[ORC] Make sure we linker-mangle symbol names in the SpeculationLayer.

If mangling is not performed then speculative lookups will fail.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369219 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ORC] Remove some dead code.
Lang Hames [Sun, 18 Aug 2019 21:22:14 +0000 (21:22 +0000)]
[ORC] Remove some dead code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369218 91177308-0d34-0410-b5e6-96231b3b80d8