OSDN Git Service

android-x86/external-llvm.git
5 years ago[llvm-objcopy] test commit
Alex Brachet [Sat, 1 Jun 2019 07:36:57 +0000 (07:36 +0000)]
[llvm-objcopy] test commit

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362289 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add AVX512BF16 and AVX512VP2INTERSECT instructions to the loading folding tables.
Craig Topper [Sat, 1 Jun 2019 06:20:59 +0000 (06:20 +0000)]
[X86] Add AVX512BF16 and AVX512VP2INTERSECT instructions to the loading folding tables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362288 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Make the X86FoldTablesEmitter functional again. Fix the spacing in the output...
Craig Topper [Sat, 1 Jun 2019 06:20:55 +0000 (06:20 +0000)]
[X86] Make the X86FoldTablesEmitter functional again. Fix the spacing in the output to make it easier to diff.

Fix a few other formatting issues in the manual table. And remove some
old FIXMEs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362287 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[RuntimeDyld] fix too-small-bitmask error
Nick Desaulniers [Sat, 1 Jun 2019 04:51:26 +0000 (04:51 +0000)]
[RuntimeDyld] fix too-small-bitmask error

Summary:
This was flagged in https://www.viva64.com/en/b/0629/ under "Snippet No.
33".

It seems that this statement is doing the standard bitwise trick for
adjusting a value to have a specific alignment.

The issue is that getStubAlignment() returns an unsigned, while DataSize
is declared a uint64_t. The right hand side of the expression is not
extended to 64b before bitwise negation, resulting in the top half of
the mask being 0s, which is not correct for realignment.

Reviewers: lhames, MaskRay

Reviewed By: MaskRay

Subscribers: RKSimon, MaskRay, hiraditya, llvm-commits, srhines

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62227

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362286 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoInline variable into assert to fix unused variable warning.
Richard Trieu [Sat, 1 Jun 2019 03:32:20 +0000 (03:32 +0000)]
Inline variable into assert to fix unused variable warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362285 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[LoopPred] Eliminate a redundant/confusing cover function [NFC]
Philip Reames [Sat, 1 Jun 2019 03:09:28 +0000 (03:09 +0000)]
[LoopPred] Eliminate a redundant/confusing cover function [NFC]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362284 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[COFF, ARM64] Fix location of ARM64 CodeView test
Tom Tan [Sat, 1 Jun 2019 02:38:08 +0000 (02:38 +0000)]
[COFF, ARM64] Fix location of ARM64 CodeView test

ARM64 CodeView test was incorrectly put under test/DebugInfo/COFF folder which
runs for all all architectures. This fix moves it to a subfolder AArch64 with
lit.local.cfg which specify it supports AArch64 only.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362283 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[LoopPred] Handle a subset of NE comparison based latches
Philip Reames [Sat, 1 Jun 2019 00:31:58 +0000 (00:31 +0000)]
[LoopPred] Handle a subset of NE comparison based latches

At the moment, LoopPredication completely bails out if it sees a latch of the form:
%cmp = icmp ne %iv, %N
br i1 %cmp, label %loop, label %exit
OR
%cmp = icmp ne %iv.next, %NPlus1
br i1 %cmp, label %loop, label %exit

This is unfortunate since this is exactly the form that LFTR likes to produce. So, go ahead and recognize simple cases where we can.

For pre-increment loops, we leverage the fact that LFTR likes canonical counters (i.e. those starting at zero) and a (presumed) range fact on RHS to discharge the check trivially.

For post-increment forms, the key insight is in remembering that LFTR had to insert a (N+1) for the RHS. CVP can hopefully prove that add nsw/nuw (if there's appropriate range on N to start with). This leaves us both with the post-inc IV and the RHS involving an nsw/nuw add, and SCEV can discharge that with no problem.

This does still need to be extended to handle non-one steps, or other harder patterns of variable (but range restricted) starting values. That'll come later.

Differential Revision: https://reviews.llvm.org/D62748

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362282 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CodeGen] Fix hashing for MO_ExternalSymbol MachineOperands.
Eli Friedman [Sat, 1 Jun 2019 00:08:54 +0000 (00:08 +0000)]
[CodeGen] Fix hashing for MO_ExternalSymbol MachineOperands.

We were hashing the string pointer, not the string, so two instructions
could be identical (isIdenticalTo), but have different hash codes.

This showed up as a very rare, non-deterministic assertion failure
rehashing a DenseMap constructed by MachineOutliner.  So there's no
"real" testcase, just a unittest which checks that the hash function
behaves correctly.

I'm a little scared fixing this is going to cause a regression in
outlining or MachineCSE, but hopefully we won't run into any issues.

Differential Revision: https://reviews.llvm.org/D61975

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362281 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[COFF, ARM64] Add CodeView register mapping
Tom Tan [Fri, 31 May 2019 23:43:31 +0000 (23:43 +0000)]
[COFF, ARM64] Add CodeView register mapping

CodeView has its own register map which is defined in cvconst.h. Missing this
mapping before saving register to CodeView causes debugger to show incorrect
value for all register based variables, like variables in register and local
variables addressed by register (stack pointer + offset).

This change added mapping between LLVM register and CodeView register so the
correct register number will be stored to CodeView/PDB, it aso fixed the
mapping from CodeView register number to register name based on current
CPUType but print PDB to yaml still assumes X86 CPU and needs to be fixed.

Differential Revision: https://reviews.llvm.org/D62608

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362280 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[PowerPC] check for INLINEASM_BR along w/ INLINEASM
Nick Desaulniers [Fri, 31 May 2019 23:02:13 +0000 (23:02 +0000)]
[PowerPC] check for INLINEASM_BR along w/ INLINEASM

Summary:
It looks like since INLINEASM_BR was created off of INLINEASM (r353563),
a few checks for INLINEASM needed to be updated to check for either
case.

pr/41999

Reviewers: hfinkel

Reviewed By: hfinkel

Subscribers: nemanjai, hiraditya, kbarton, jsji, llvm-commits, craig.topper, srhines

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62403

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362278 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[codeview] Revert inline line table change of r362264
Reid Kleckner [Fri, 31 May 2019 22:55:03 +0000 (22:55 +0000)]
[codeview] Revert inline line table change of r362264

Testing with debuggers shows that our previous behavior was correct.
The reason I thought MSVC did things differently is that MSVC prefers to
use the 0xB combined code offset and code length update opcode when
inline sites are discontiguous.

Keep the test changes, and update the llvm-pdbutil inline line table
dumper to account for this new interpretation of the opcodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362277 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAMDGPU: Fix not adding ImplicitBufferPtr as a live-in
Matt Arsenault [Fri, 31 May 2019 22:47:36 +0000 (22:47 +0000)]
AMDGPU: Fix not adding ImplicitBufferPtr as a live-in

Fixes missing test from r293000.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362275 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[SimplifyLibCalls] Fold more fortified functions into non-fortified variants
Erik Pilkington [Fri, 31 May 2019 22:41:36 +0000 (22:41 +0000)]
[SimplifyLibCalls] Fold more fortified functions into non-fortified variants

When the object size argument is -1, no checking can be done, so calling the
_chk variant is unnecessary. We already did this for a bunch of these
functions.

rdar://50797197

Differential revision: https://reviews.llvm.org/D62358

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362272 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoNFC: Pull out a function to reduce some duplication
Erik Pilkington [Fri, 31 May 2019 22:41:31 +0000 (22:41 +0000)]
NFC: Pull out a function to reduce some duplication

Part of https://reviews.llvm.org/D62358

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362271 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Tests] Better represent the postinc form produced by LFTR in LoopPred tests
Philip Reames [Fri, 31 May 2019 22:22:29 +0000 (22:22 +0000)]
[Tests] Better represent the postinc form produced by LFTR in LoopPred tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362270 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[SelectionDAG] Make the code in mutateStrictFPToFP less aware of how many operands...
Craig Topper [Fri, 31 May 2019 22:18:45 +0000 (22:18 +0000)]
[SelectionDAG] Make the code in mutateStrictFPToFP less aware of how many operands each node has. NFCI

Just copy all of the operands except the chain and call MorphNode on that.
This removes the IsUnary and IsTernary flags.

Also always get the result type from the result type of the original
nodes. Previously we got it from the operand except for two nodes
where that didn't work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362269 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Bugpoint] fix another use-after-move. NFC
Nick Desaulniers [Fri, 31 May 2019 21:36:21 +0000 (21:36 +0000)]
[Bugpoint] fix another use-after-move. NFC

Summary:
This was flagged in https://www.viva64.com/en/b/0629/ under "Snippet No.
7".

These statements are order independent, short of the use-after-move.

Reviewers: echristo, srhines, RKSimon

Reviewed By: RKSimon

Subscribers: dblaikie, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62114

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362267 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[RegisterCoalescer] fix potential use of undef value. NFC
Nick Desaulniers [Fri, 31 May 2019 21:20:13 +0000 (21:20 +0000)]
[RegisterCoalescer] fix potential use of undef value. NFC

Summary:
Fixes a warning produced from scan-build (llvm.org/reports/scan-build/),
further warnings found by annotation isMoveInstr [[nodiscard]].

isMoveInstr potentially does not assign to its parameters, so if they
were uninitialized, they will potentially stay uninitialized.  It seems
most call sites pass references to uninitialized values, then use them
without checking the return value.

Reviewers: wmi

Reviewed By: wmi

Subscribers: MatzeB, qcolombet, hiraditya, tpr, llvm-commits, srhines

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62109

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362265 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[codeview] Fix inline line table accuracy for discontiguous segments
Reid Kleckner [Fri, 31 May 2019 20:55:31 +0000 (20:55 +0000)]
[codeview] Fix inline line table accuracy for discontiguous segments

After improving the inline line table dumper in llvm-pdbutil and looking
at MSVC's inline line tables, it is clear that setting the length of the
inlined code region does not update the code offset. This means that the
delta to the beginning of a new discontiguous inlined code region should
be calculated relative to the last code offset, excluding the length.
Implementing this is a one line fix for MC: simply don't update
LastLabel.

While I'm updating these test cases, switch them to use llvm-objdump -d
and llvm-pdbutil. This allows us to show offsets of each instruction and
correlate the line table offsets to the actual code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362264 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoReapply [CVP] Simplify non-overflowing saturating add/sub
Nikita Popov [Fri, 31 May 2019 20:48:26 +0000 (20:48 +0000)]
Reapply [CVP] Simplify non-overflowing saturating add/sub

If we can determine that a saturating add/sub will not overflow based
on range analysis, convert it into a simple binary operation. This is
a sibling transform to the existing with.overflow handling.

Reapplying this with an additional check that the saturating intrinsic
has integer type, as LVI currently does not support vector types.

Differential Revision: https://reviews.llvm.org/D62703

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362263 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CVP] Add vector saturating add test; NFC
Nikita Popov [Fri, 31 May 2019 20:42:13 +0000 (20:42 +0000)]
[CVP] Add vector saturating add test; NFC

Extra test for the assertion failure from D62703.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362262 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CVP] Fix assertion failure on vector with.overflow
Nikita Popov [Fri, 31 May 2019 20:42:07 +0000 (20:42 +0000)]
[CVP] Fix assertion failure on vector with.overflow

Noticed on D62703. LVI only handles plain integers, not vectors of
integers. This was previously not an issue, because vector support
for with.overflow is only a relatively recent addition.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362261 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Tests] Add ne icmp tests w/preinc forms for LoopPredication
Philip Reames [Fri, 31 May 2019 20:34:57 +0000 (20:34 +0000)]
[Tests] Add ne icmp tests w/preinc forms for LoopPredication

Turns out this is substaintially easier to match then the post increment form, so let's start there.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362260 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Resync Host.cpp with compiler-rt's cpu_model.c to enable 0x55 to be identified...
Craig Topper [Fri, 31 May 2019 19:18:07 +0000 (19:18 +0000)]
[X86] Resync Host.cpp with compiler-rt's cpu_model.c to enable 0x55 to be identified as cascadelake when avx512vnni is detected.

Some other formatting changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362256 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[NFC][InstCombine] Add unary FNeg tests to AMDGPU/amdgcn-intrinsics.ll
Cameron McInally [Fri, 31 May 2019 19:12:59 +0000 (19:12 +0000)]
[NFC][InstCombine] Add unary FNeg tests to AMDGPU/amdgcn-intrinsics.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362255 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRevert "[CVP] Simplify non-overflowing saturating add/sub"
Nikita Popov [Fri, 31 May 2019 19:04:47 +0000 (19:04 +0000)]
Revert "[CVP] Simplify non-overflowing saturating add/sub"

This reverts commit 1e692d1777ae34dcb93524b5798651a29defae09.

Causes assertion failure in builtins-wasm.c clang test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362254 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[NFC][InstCombine] Add unary FNeg to cos-1.ll cos-2.ll cos-sin-intrinsic.ll
Cameron McInally [Fri, 31 May 2019 18:54:44 +0000 (18:54 +0000)]
[NFC][InstCombine] Add unary FNeg to cos-1.ll cos-2.ll cos-sin-intrinsic.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362253 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MCA] Remove unused fields from BottleneckAnalysis. NFC
Andrea Di Biagio [Fri, 31 May 2019 18:01:42 +0000 (18:01 +0000)]
[MCA] Remove unused fields from BottleneckAnalysis. NFC

This should appease the buildbots.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362251 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CMake] Feed BUNDLE_PATH through llvm target wrappers
Chris Bieneman [Fri, 31 May 2019 17:40:49 +0000 (17:40 +0000)]
[CMake] Feed BUNDLE_PATH through llvm target wrappers

This feeds the new llvm_codsign BUNDLE_PATH option through from the llvm target wrapper functions, so that you can specify the BUNDLE_PATH on the target's codesign.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362248 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MIR-Canon] Don't do vreg skip for independent instructions if there are none.
Puyan Lotfi [Fri, 31 May 2019 17:34:25 +0000 (17:34 +0000)]
[MIR-Canon] Don't do vreg skip for independent instructions if there are none.

We don't want to create vregs if there is nothing to use them for. That causes
verifier errors.

Differential Revision: https://reviews.llvm.org/D62740

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362247 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MCA] Refactor class BottleneckAnalysis. NFCI
Andrea Di Biagio [Fri, 31 May 2019 17:18:34 +0000 (17:18 +0000)]
[MCA] Refactor class BottleneckAnalysis. NFCI

The resource pressure distribution computation is now delegated by class
BottleneckAnalysis to an instance of class PressureTracker.
Class PressureTracker is also responsible for:
 - tracking users of processor resource units.
 - tracking the number of delay cycles caused by increases in backpressure.

BottleneckAnalysis internally initializes a dependency graph. Each nodes
represents an instruction in the input code sequence.  Edges of the dependency
graph are critical register/memory/resource dependencies.  Dependencies are only
added to the graph if they are seen as critical by backend pressure events.

The DependencyGraph is currently unused. It is possible to print the dependency
 graph (see method DependencyGraph::dump()) for debugging purposes.
The long term goal is to use the information stored by the dependency graph in
order to do critical path computation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362246 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Tests] Add tests for loop predication of loops w/ne latch conditions
Philip Reames [Fri, 31 May 2019 16:54:38 +0000 (16:54 +0000)]
[Tests] Add tests for loop predication of loops w/ne latch conditions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362244 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CVP] Simplify non-overflowing saturating add/sub
Nikita Popov [Fri, 31 May 2019 16:46:05 +0000 (16:46 +0000)]
[CVP] Simplify non-overflowing saturating add/sub

If we can determine that a saturating add/sub will not overflow
based on range analysis, convert it into a simple binary operation.
This is a sibling transform to the existing with.overflow handling.

Differential Revision: https://reviews.llvm.org/D62703

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362242 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRevert revert of r362112 with minor SystemZ test file corrections.
Kevin P. Neal [Fri, 31 May 2019 16:32:12 +0000 (16:32 +0000)]
Revert revert of r362112 with minor SystemZ test file corrections.

[FPEnv] Added a special UnrollVectorOp method to deal with the chain on StrictFP opcodes

This change creates UnrollVectorOp_StrictFP. The purpose of this is to address a failure that consistently occurs when calling StrictFP functions on vectors whose number of elements is 3 + 2n on most platforms, such as PowerPC or SystemZ. The old UnrollVectorOp method does not expect that the vector that it will unroll will have a chain, so it has an assert that prevents it from running if this is the case. This new StrictFP version of the method deals with the chain while unrolling the vector. With this new function in place during vector widending, llc can run vector-constrained-fp-intrinsics.ll for SystemZ successfully.

Submitted by: Drew Wock <drew.wock@sas.com>
Reviewed by: Cameron McInally, Kevin P. Neal
Approved by: Cameron McInally
Differential Revision: https://reviews.llvm.org/D62546

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362241 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AMDGPU] Use InliningThresholdMultiplier for inline hint
Stanislav Mekhanoshin [Fri, 31 May 2019 16:19:26 +0000 (16:19 +0000)]
[AMDGPU] Use InliningThresholdMultiplier for inline hint

AMDGPU uses multiplier 9 for the inline cost. It is taken into account
everywhere except for inline hint threshold. As a result we are penalizing
functions with the inline hint making them less probable to be inlined
than those without the hint. Defaults are 225 for a normal function and
325 for a function with an inline hint. Currently we have effective
threshold 225 * 9 = 2025 for normal functions and just 325 for those with
the hint. That is fixed by this patch.

Differential Revision: https://reviews.llvm.org/D62707

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362239 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[NFC][InstCombine] Add unary FNeg tests to fabs.ll
Cameron McInally [Fri, 31 May 2019 16:17:04 +0000 (16:17 +0000)]
[NFC][InstCombine] Add unary FNeg tests to fabs.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362238 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[PPC] Correctly adjust branch probability in PPCReduceCRLogicals
Guozhi Wei [Fri, 31 May 2019 16:11:17 +0000 (16:11 +0000)]
[PPC] Correctly adjust branch probability in PPCReduceCRLogicals

In PPCReduceCRLogicals after splitting the original MBB into 2, the 2 impacted branches still use original branch probability. This is unreasonable. Suppose we have following code, and the probability of each successor is 50%.

    condc = conda || condb
    br condc, label %target, label %fallthrough

It can be transformed to following,

    br conda, label %target, label %newbb
  newbb:
    br condb, label %target, label %fallthrough

Since each branch has a probability of 50% to each successor, the total probability to %fallthrough is 25% now, and the total probability to %target is 75%. This actually changed the original profiling data. A more reasonable probability can be set to 70% to the false side for each branch instruction, so the total probability to %fallthrough is close to 50%.

This patch assumes the branch target with two incoming edges have same edge frequency and computes new probability fore each target, and keep the total probability to original targets unchanged.

Differential Revision: https://reviews.llvm.org/D62430

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362237 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[NFC][InstCombine] Add unary FNeg tests to fcmp.ll
Cameron McInally [Fri, 31 May 2019 15:40:03 +0000 (15:40 +0000)]
[NFC][InstCombine] Add unary FNeg tests to fcmp.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362234 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MachinePipeliner][NFC] Add some debug log and statistics
Jinsong Ji [Fri, 31 May 2019 15:35:19 +0000 (15:35 +0000)]
[MachinePipeliner][NFC] Add some debug log and statistics

This is to add some log and statistics for debugging

Differential Revision: https://reviews.llvm.org/D62165

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362233 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[NFC][InstCombine] Add unary FNeg tests to fdiv.ll
Cameron McInally [Fri, 31 May 2019 15:10:34 +0000 (15:10 +0000)]
[NFC][InstCombine] Add unary FNeg tests to fdiv.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362231 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AMDGPU] Regenerate add/sub shrink constant tests for an upcoming patch
Simon Pilgrim [Fri, 31 May 2019 15:06:51 +0000 (15:06 +0000)]
[AMDGPU] Regenerate add/sub shrink constant tests for an upcoming patch

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362230 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AMDGPU] Regenerate CTLZ tests for an upcoming patch
Simon Pilgrim [Fri, 31 May 2019 15:06:14 +0000 (15:06 +0000)]
[AMDGPU] Regenerate CTLZ tests for an upcoming patch

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362229 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[UpdateTestChecks] Add support for -march=r600 to match existing -march=amdgcn support
Simon Pilgrim [Fri, 31 May 2019 15:05:06 +0000 (15:05 +0000)]
[UpdateTestChecks] Add support for -march=r600 to match existing -march=amdgcn support

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362228 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[NFC][InstCombine] Add unary FNeg tests to fma.ll
Cameron McInally [Fri, 31 May 2019 14:49:31 +0000 (14:49 +0000)]
[NFC][InstCombine] Add unary FNeg tests to fma.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362227 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-readobj] - Remove excessive `dynamic.test`
George Rimar [Fri, 31 May 2019 13:16:21 +0000 (13:16 +0000)]
[llvm-readobj] - Remove excessive `dynamic.test`

dynamic.test is a test that checks dumping of
dynamic tags. It uses precompiled objects as inputs
and it is completely excessive nowadays:

Now we have elf-dynamic-tags-machine-specific.test
and elf-dynamic-tags.test.
(https://github.com/llvm-mirror/llvm/blob/master/test/tools/llvm-readobj/elf-dynamic-tags-machine-specific.test)
(https://github.com/llvm-mirror/llvm/blob/master/test/tools/llvm-readobj/elf-dynamic-tags.test)

First is used to check target specific tags and second tests the common flags.
These tests use YAML, which is much better than using precompiled binaries.

Note that new reviews tend to update the YAML based
tests to add new tags, e.g. see D62596.

With this patch it became possible to remove
dynamic-table-so.aarch64 binary from the inputs folder.
(other binaries are still used in other tests).

Differential revision: https://reviews.llvm.org/D62728

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362224 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agogn build: Merge r362160
Nico Weber [Fri, 31 May 2019 12:07:05 +0000 (12:07 +0000)]
gn build: Merge r362160

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362223 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agogn build: Merge r362196
Nico Weber [Fri, 31 May 2019 11:52:59 +0000 (11:52 +0000)]
gn build: Merge r362196

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362222 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agogn build: Merge r362190
Nico Weber [Fri, 31 May 2019 11:51:42 +0000 (11:51 +0000)]
gn build: Merge r362190

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362221 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoftime-trace: Trace loop passes
Russell Gallop [Fri, 31 May 2019 10:14:04 +0000 (10:14 +0000)]
ftime-trace: Trace loop passes

These can take a significant amount of time in some builds.

Suggested by Andrea Di Biagio.

Differential Revision: https://reviews.llvm.org/D62666

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362219 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] 'C-(C2-X) --> X+(C-C2)' constant-fold
Roman Lebedev [Fri, 31 May 2019 09:47:16 +0000 (09:47 +0000)]
[InstCombine] 'C-(C2-X) --> X+(C-C2)' constant-fold

It looks this fold was already partially happening, indirectly
via some other folds, but with one-use limitation.
No other fold here has that restriction.

https://rise4fun.com/Alive/ftR

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362217 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] 'add (sub C1, X), C2 --> sub (add C1, C2), X' constant-fold
Roman Lebedev [Fri, 31 May 2019 09:47:04 +0000 (09:47 +0000)]
[InstCombine] 'add (sub C1, X), C2 --> sub (add C1, C2), X' constant-fold

https://rise4fun.com/Alive/qJQ

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362216 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AArch64][SVE2] Asm: support WHILE instructions
Cullen Rhodes [Fri, 31 May 2019 09:13:55 +0000 (09:13 +0000)]
[AArch64][SVE2] Asm: support WHILE instructions

Summary:
Patch adds support for the following instructions:
    * WHILEGE, WHILEGT, WHILEHS, WHILEHI, WHILEWR, WHILERW

The specification can be found here:
https://developer.arm.com/docs/ddi0602/latest

Reviewed By: chill

Differential Revision: https://reviews.llvm.org/D62601

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362215 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AArch64][SVE2] Asm: support TBL/TBX instructions
Cullen Rhodes [Fri, 31 May 2019 09:06:53 +0000 (09:06 +0000)]
[AArch64][SVE2] Asm: support TBL/TBX instructions

Summary:
A three sources variant of the TBL instruction is added to the existing
SVE instruction in SVE2. This is implemented with minor changes to the
existing TableGen class. TBX is a new instruction with its own
definition.

The specification can be found here:
https://developer.arm.com/docs/ddi0602/latest

Reviewed By: chill

Differential Revision: https://reviews.llvm.org/D62600

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362214 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AArch64][SVE2] Asm: support SVE2 store instructions
Cullen Rhodes [Fri, 31 May 2019 08:59:40 +0000 (08:59 +0000)]
[AArch64][SVE2] Asm: support SVE2 store instructions

Summary:
Patch adds support for the following instructions:
    * STNT1B, STNT1H, STNT1S, STNT1D

The specification can be found here:
https://developer.arm.com/docs/ddi0602/latest

Reviewed By: chill

Differential Revision: https://reviews.llvm.org/D62599

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362213 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MIPS GlobalISel] Add detailed tests for lower call
Petar Avramovic [Fri, 31 May 2019 08:40:08 +0000 (08:40 +0000)]
[MIPS GlobalISel] Add detailed tests for lower call

Test different operand types of callee and their behavior whether
relocation model is pic or not.
Possible operand types are:
Register (function pointer),
External symbol (used for libcalls e.g. __udivdi3 or memcpy),
Global address.

Global address has different handling depending on relocation model
and linkage type. Register and external symbol do not.

Differential Revision: https://reviews.llvm.org/D62590

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362212 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoFollow up and fix for rL362064
Sjoerd Meijer [Fri, 31 May 2019 08:39:34 +0000 (08:39 +0000)]
Follow up and fix for rL362064

Fix the misleadingly indentation introduced in rL362064. This will get rid of
the compiler warning, and it was actually a bug. This change will be used and
tested in D62669.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362211 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MIPS GlobalISel] Handle position independent code
Petar Avramovic [Fri, 31 May 2019 08:27:06 +0000 (08:27 +0000)]
[MIPS GlobalISel] Handle position independent code

Handle position independent code for MIPS32.
When callee is global address, lower call will emit callee
as G_GLOBAL_VALUE and add target flag if needed.
Support $gp in getRegBankFromRegClass().
Select G_GLOBAL_VALUE, specially handle case when
there are target flags attached by lowerCall.

Differential Revision: https://reviews.llvm.org/D62589

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362210 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[NFC][InstCombine] Copy add/sub constant-folding tests from codegen
Roman Lebedev [Fri, 31 May 2019 08:24:07 +0000 (08:24 +0000)]
[NFC][InstCombine] Copy add/sub constant-folding tests from codegen

Last three patterns are missed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362209 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[NFC][Codegen] Add/sub constant-folding: add scalar tests too
Roman Lebedev [Fri, 31 May 2019 08:23:48 +0000 (08:23 +0000)]
[NFC][Codegen] Add/sub constant-folding: add scalar tests too

Just for completeness.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362208 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[mips] Move initGlobalBaseReg to MipsFunctionInfo. NFC
Petar Avramovic [Fri, 31 May 2019 08:15:28 +0000 (08:15 +0000)]
[mips] Move initGlobalBaseReg to MipsFunctionInfo. NFC

Move initGlobalBaseReg from MipsSEDAGToDAGISel to MipsFunctionInfo.
This way functions used for handling position independent code during
instruction selection, getGlobalBaseReg and initGlobalBaseReg,
end up in same class.

Differential Revision: https://reviews.llvm.org/D62586

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362206 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstructionSimplify] Add missing implementation of llvm::SimplifyUnOp. NFC
Craig Topper [Fri, 31 May 2019 08:10:23 +0000 (08:10 +0000)]
[InstructionSimplify] Add missing implementation of llvm::SimplifyUnOp. NFC

There are no callers currently, but the function is declared so we should at
least implement it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362205 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MIPS GlobalISel] Lower call for callee that is register
Petar Avramovic [Fri, 31 May 2019 08:06:17 +0000 (08:06 +0000)]
[MIPS GlobalISel] Lower call for callee that is register

Lower call for callee that is register for MIPS32.
Register should contain callee function address.

Differential Revision: https://reviews.llvm.org/D62585

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362204 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Remove patterns for X86VSintToFP/X86VUintToFP+loadv4f32 to v2f64.
Craig Topper [Fri, 31 May 2019 07:38:26 +0000 (07:38 +0000)]
[X86] Remove patterns for X86VSintToFP/X86VUintToFP+loadv4f32 to v2f64.

These patterns can incorrectly narrow a volatile load from 128-bits to 64-bits.
Similar to PR42079.

Switch to using (v4i32 (bitcast (v2i64 (scalar_to_vector (loadi64))))) as the
load pattern used in the instructions.

This probably still has issues in 32-bit mode where loadi64 isn't legal. Maybe
we should use VZMOVL for widened loads even when we don't need the upper bits
as zeroes?

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362203 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add test cases for failure to use 128-bit masked vcvtdq2pd when load starts...
Craig Topper [Fri, 31 May 2019 07:38:22 +0000 (07:38 +0000)]
[X86] Add test cases for failure to use 128-bit masked vcvtdq2pd when load starts as v2i32.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362202 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add test cases for a volatile load shrinking bug involving cvtdq2pd. NFC
Craig Topper [Fri, 31 May 2019 07:38:18 +0000 (07:38 +0000)]
[X86] Add test cases for a volatile load shrinking bug involving cvtdq2pd. NFC

Similar to PR42079

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362201 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Copy a test case from avx512-cvt.ll to avx512-cvt-widen.ll. NFC
Craig Topper [Fri, 31 May 2019 07:38:14 +0000 (07:38 +0000)]
[X86] Copy a test case from avx512-cvt.ll to avx512-cvt-widen.ll. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362200 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Remove avx512 isel patterns for fpextend+load. Prefer to only match fp extloads...
Craig Topper [Fri, 31 May 2019 06:21:53 +0000 (06:21 +0000)]
[X86] Remove avx512 isel patterns for fpextend+load. Prefer to only match fp extloads instead.

DAG combine will usually fold fpextend+load to an fp extload anyway. So the
256 and 512 patterns were probably unnecessary. The 128 bit pattern was special
in that it looked for a v4f32 load, but then used it in an instruction that
only loads 64-bits. This is bad if the load happens to be volatile. We could
probably make the patterns volatile aware, but that's more work for something
that's probably rare. The peephole pass might kick in and save us anyway. We
might also be able to fix this with some additional DAG combines.

This also adds patterns for vselect+extload to enabled masked vcvtps2pd to be
used. Previously we looked for the unlikely vselect+fpextend+load.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362199 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add test to show missed opportunity to use masked vcvtps2pd for vselect+extload.
Craig Topper [Fri, 31 May 2019 06:21:49 +0000 (06:21 +0000)]
[X86] Add test to show missed opportunity to use masked vcvtps2pd for vselect+extload.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362198 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add test case for PR42079. NFC
Craig Topper [Fri, 31 May 2019 06:21:45 +0000 (06:21 +0000)]
[X86] Add test case for PR42079. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362197 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MIR-Canon] Skip the first N vreg names lazily.
Puyan Lotfi [Fri, 31 May 2019 06:02:38 +0000 (06:02 +0000)]
[MIR-Canon] Skip the first N vreg names lazily.

This consolidates the vreg skip code into one function (SkipVRegs()).
SkipVRegs() now knows if it should skip as if it is the first initialization or
subsequent skips.

The first skip is also done the first time createVirtualRegister is called by
the cursor instead of by the cursor's constructor. This prevents verifier
errors on machine functions that have no vregs (where the verifier will
complain that there are vregs when the function uses none).

Differential Revision: https://reviews.llvm.org/D62717

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362195 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Correct the ins operand order for MASKPAIR16STORE to match other store instruct...
Craig Topper [Fri, 31 May 2019 05:20:27 +0000 (05:20 +0000)]
[X86] Correct the ins operand order for MASKPAIR16STORE to match other store instructions.

This makes the 5 address operands come first. And the data operand comes last.

This matches the operand order the instruction is created with. It's also the
expected order in X86MCInstLower. So everything appeared to work, but the
operands didn't match their declared type.

Fixes a -verify-machineinstrs failure.

Also remove the isel patterns from these instructions since they should only
be used for stack spills and reloads. I'm not even sure what types the patterns
were looking for to match.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362193 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MIR-Canon] Hardening propagateLocalCopies.
Puyan Lotfi [Fri, 31 May 2019 04:49:58 +0000 (04:49 +0000)]
[MIR-Canon] Hardening propagateLocalCopies.

This is am almost NFC, it does the following:
- If there is no register class for a COPY's src or dst, bail.
- Fixes uses iterator invalidation bug.

Differential Revision: https://reviews.llvm.org/D62713

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362191 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoFix bad go bindings test.
Richard Trieu [Fri, 31 May 2019 03:45:11 +0000 (03:45 +0000)]
Fix bad go bindings test.

After r362128, the "byval" attribute has a stricter check and will cause an
assertion.  Remove the "byval" test case for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362189 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add VP2INTERSECT instructions
Pengfei Wang [Fri, 31 May 2019 02:50:41 +0000 (02:50 +0000)]
[X86] Add VP2INTERSECT instructions

Support Intel AVX512 VP2INTERSECT instructions in llvm

Patch by Xiang Zhang (xiangzhangllvm)

Differential Revision: https://reviews.llvm.org/D62366

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362188 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CMake] Provide an option to use relative paths in debug info
Petr Hosek [Fri, 31 May 2019 01:34:51 +0000 (01:34 +0000)]
[CMake] Provide an option to use relative paths in debug info

CMake always uses absolute file paths in the generated compiler
invocation which results in absolute file paths being embedded in debug
info. This is undesirable when building a toolchain e.g. on bots as the
debug info may embed the bot source checkout path which is meaningless
anywhere else.

This change introduces the LLVM_USE_RELATIVE_PATHS_IN_DEBUG_INFO which uses
-fdebug-prefix-map (where supported) options to rewrite paths embedded
into debug info with relative ones. Additionally, LLVM_SOURCE_PREFIX can
be used to override the path to source directory with a different one.

Differential Revision: https://reviews.llvm.org/D62622

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362185 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoFix -DBUILD_SHARED_LIBS=ON build after rL362160
Sam Clegg [Fri, 31 May 2019 01:04:00 +0000 (01:04 +0000)]
Fix -DBUILD_SHARED_LIBS=ON build after rL362160

Differential Revision: https://reviews.llvm.org/D62709

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362180 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Remove result type constraints from the extloadv2f32/extloadv4f32/extloadv8f32...
Craig Topper [Thu, 30 May 2019 23:35:24 +0000 (23:35 +0000)]
[X86] Remove result type constraints from the extloadv2f32/extloadv4f32/extloadv8f32 PatFrags. NFC

The result types aren't mentioned in the pattern name so really shouldn't be in the PatFrags.

The users of these either have their own type constraint or rely on the type constranit system to realize the only legal extend would be to f64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362175 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoMISched: Fix -misched-regpressure=0 if subreg liveness enabled
Matt Arsenault [Thu, 30 May 2019 23:31:36 +0000 (23:31 +0000)]
MISched: Fix -misched-regpressure=0 if subreg liveness enabled

Test is waiting on fixing several more crashes in the AMDGPU scheduler
implementation with this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362174 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Remove code that unnecessarily sets EXTLOAD with src type of v2f32/v4f32/v8f32...
Craig Topper [Thu, 30 May 2019 22:29:06 +0000 (22:29 +0000)]
[X86] Remove code that unnecessarily sets EXTLOAD with src type of v2f32/v4f32/v8f32 as Legal for SSE2/AVX/AVX512 respectively. NFC

The LoadExt table defaults to all combinations being Legal. For
vector types, only src VTs with an i1 element type were ever changed.
So we don't need to mark them legal manually.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362170 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoSupport codesigning bundles and forcing
Chris Bieneman [Thu, 30 May 2019 22:25:48 +0000 (22:25 +0000)]
Support codesigning bundles and forcing

Summary:
Clangd's framework is assembled by copying binaries from the lib and bin directories into a bundle shape. This results in an invalid bundle code signature because the signature only applies to the binaries not the resources.

This patch adds two new options to `llvm_codesign` to enable re-signing the library and XPC service as bundles.

The `BUNDLE_PATH` option allow specifying an explicit path to codesign, which enables signing bundles which aren't generated using CMake's `FRAMEWORK` or `BUNDLE` target properties.

The `FORCE` option allows re-signing binaries that have already been signed. This is required for how clangd exposes the clangd library and tools as both XPC and non-XPC services using the same binary.

Reviewers: jkorous, bogner

Reviewed By: bogner

Subscribers: mgorny, ilya-biryukov, dexonsmith, arphaman, kadircet, cfe-commits, llvm-commits

Tags: #clang, #llvm

Differential Revision: https://reviews.llvm.org/D62693

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362169 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoFix test to add missing '|' to regex.
Douglas Yung [Thu, 30 May 2019 22:20:31 +0000 (22:20 +0000)]
Fix test to add missing '|' to regex.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362168 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoUpdate the tests in r362121 / r362141 to allow for Windows-specific error
Michael Trent [Thu, 30 May 2019 22:11:29 +0000 (22:11 +0000)]
Update the tests in r362121 / r362141 to allow for Windows-specific error
messages: "Is a directory" instead of "is a directory"

This should resolve the errors being reported on clang-x64-windows-msvc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362167 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAdd enums as global variables in the IR metadata.
Amy Huang [Thu, 30 May 2019 22:04:11 +0000 (22:04 +0000)]
Add enums as global variables in the IR metadata.

Summary:
Keeps track of the enums that were used by saving them as DIGlobalVariables,
since CodeView emits debug info for global constants.

Reviewers: rnk

Subscribers: aprantl, cfe-commits, llvm-commits

Tags: #clang, #llvm

Differential Revision: https://reviews.llvm.org/D62635

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362166 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Remarks] Fix usage of enum class
Francis Visoiu Mistrih [Thu, 30 May 2019 22:01:56 +0000 (22:01 +0000)]
[Remarks] Fix usage of enum class

Breaks the build on some compilers:

http://lab.llvm.org:8011/builders/clang-cmake-x86_64-avx2-linux/builds/9720/steps/build%20stage%201/logs/stdio

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362165 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Remarks][NFC] Move the serialization to lib/Remarks
Francis Visoiu Mistrih [Thu, 30 May 2019 21:45:59 +0000 (21:45 +0000)]
[Remarks][NFC] Move the serialization to lib/Remarks

Separate the remark serialization to YAML from the LLVM Diagnostics.

This adds a new serialization abstraction: remarks::Serializer. It's
completely independent from lib/IR and it provides an easy way to
replace YAML by providing a new remarks::Serializer.

Differential Revision: https://reviews.llvm.org/D62632

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362160 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[NFC][ARM] Add a test that potentially causes endless combine loop with D62266
Roman Lebedev [Thu, 30 May 2019 21:41:21 +0000 (21:41 +0000)]
[NFC][ARM] Add a test that potentially causes endless combine loop with D62266

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362159 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MIR-Canon] Fixing case where MachineFunction is empty.
Puyan Lotfi [Thu, 30 May 2019 21:37:25 +0000 (21:37 +0000)]
[MIR-Canon] Fixing case where MachineFunction is empty.

In cases where the machine function is empty: bail on the RPO traversal.

Differential Revision: https://reviews.llvm.org/D62617

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362158 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DAGCombine] Limit 'hoist add/sub binop w/ constant op' to non-opaque consts
Roman Lebedev [Thu, 30 May 2019 21:10:37 +0000 (21:10 +0000)]
[DAGCombine] Limit 'hoist add/sub binop w/ constant op' to non-opaque consts

I don't have a test case for these, but there is a test case for D62266
where, even after all the constant-folding patches, we still end up
with endless combine loop. Which makes sense, since we don't constant
fold for opaque constants.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362156 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CVP] Add tests for non-overflowing saturating math; NFC
Nikita Popov [Thu, 30 May 2019 21:03:17 +0000 (21:03 +0000)]
[CVP] Add tests for non-overflowing saturating math; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362153 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CVP] Generalize willNotOverflow(); NFC
Nikita Popov [Thu, 30 May 2019 21:03:10 +0000 (21:03 +0000)]
[CVP] Generalize willNotOverflow(); NFC

Change argument from WithOverflowInst to BinaryOpIntrinsic, so this
function can also be used for saturating math intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362152 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[RuntimeDyld] Update reserveAllocationSpace to account for stub padding.
Lang Hames [Thu, 30 May 2019 20:58:28 +0000 (20:58 +0000)]
[RuntimeDyld] Update reserveAllocationSpace to account for stub padding.

This should fix the buildbot failures caused by r362139.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362151 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] Avoid use after free in DenseMap, when built with GCC
Martin Storsjo [Thu, 30 May 2019 20:53:21 +0000 (20:53 +0000)]
[InstCombine] Avoid use after free in DenseMap, when built with GCC

Previously, this used a statement like this:
    Map[A] = Map[B];

This is equivalent to the following:
    const auto &Src = Map[B];
    auto &Dest = Map[A];
    Dest = Src;

The second statement, "auto &Dest = Map[A];" can insert a new
element into the DenseMap, which can potentially grow and reallocate
the DenseMap's internal storage, which will invalidate the existing
reference to the source. When doing the actual assignment,
the Src reference is dereferenced, accessing memory that was
freed when the DenseMap grew.

This issue hasn't shown up when LLVM was built with Clang, because
the right hand side ended up dereferenced before evaulating the
left hand side. (If the value type is a larger data type, Clang doesn't
do this but behaves like GCC.)

With GCC, a cast to Value* isn't enough to make it dereference the
right hand side reference before invoking operator[] (while that is
enough to make Clang/LLVM do the right thing for larger types), but
storing it in an intermediate variable in a separate statement works.

This fixes PR42065.

Differential Revision: https://reviews.llvm.org/D62624

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362150 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DAGCombiner][X86][AArch64] (x - C) + y -> (x + y) - C fold. Try 2
Roman Lebedev [Thu, 30 May 2019 20:37:49 +0000 (20:37 +0000)]
[DAGCombiner][X86][AArch64] (x - C) + y  ->  (x + y) - C  fold. Try 2

Summary:
Only vector tests are being affected here,
since subtraction by scalar constant is rewritten
as addition by negated constant.

No surprising test changes.

https://rise4fun.com/Alive/pbT

This is a recommit, originally committed in rL361852, but reverted
to investigate test-suite compile-time hangs.

Reviewers: RKSimon, craig.topper, spatel

Reviewed By: RKSimon

Subscribers: javed.absar, kristof.beyls, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62257

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362146 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DAGCombine] (x - C) - y -> (x - y) - C fold. Try 3
Roman Lebedev [Thu, 30 May 2019 20:37:39 +0000 (20:37 +0000)]
[DAGCombine] (x - C) - y  ->  (x - y) - C  fold. Try 3

Summary:
Again only vectors affected. Frustrating. Let me take a look into that..

https://rise4fun.com/Alive/AAq

This is a recommit, originally committed in rL361852, but reverted
to investigate test-suite compile-time hangs, and then reverted in
rL362109 to fix missing constant folds that were causing
endless combine loops.

Reviewers: RKSimon, craig.topper, spatel

Reviewed By: RKSimon

Subscribers: javed.absar, JDevlieghere, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62294

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362145 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DAGCombine][X86][AArch64][AMDGPU] (x - y) + -1 -> add (xor y, -1), x fold. Try 3
Roman Lebedev [Thu, 30 May 2019 20:37:29 +0000 (20:37 +0000)]
[DAGCombine][X86][AArch64][AMDGPU] (x - y) + -1  ->  add (xor y, -1), x  fold. Try 3

Summary:
This prevents regressions in next patch,
and somewhat recovers from the regression to AMDGPU test in D62223.

It is indeed not great that we leave vector decrement,
don't transform it into vector add all-ones..

https://rise4fun.com/Alive/ZRl

This is a recommit, originally committed in rL361852, but reverted
to investigate test-suite compile-time hangs, and then reverted in
rL362109 to fix missing constant folds that were causing
endless combine loops.

Reviewers: RKSimon, craig.topper, spatel, arsenm

Reviewed By: RKSimon, arsenm

Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, javed.absar, dstuttard, tpr, t-tye, kristof.beyls, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62263

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362144 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DAGCombiner][X86][AArch64][SPARC][SystemZ] y - (x + C) -> (y - x) - C fold. Try 3
Roman Lebedev [Thu, 30 May 2019 20:37:18 +0000 (20:37 +0000)]
[DAGCombiner][X86][AArch64][SPARC][SystemZ] y - (x + C)  ->  (y - x) - C  fold. Try 3

Summary:
Direct sibling of D62223 patch.
While i don't have a direct motivational pattern for this,
it would seem to make sense to handle both patterns (or none),
for symmetry?

The aarch64 changes look neutral;
sparc and systemz look like improvement (one less instruction each);
x86 changes - 32bit case improves, 64bit case shows that LEA no longer
gets constructed, which may be because that whole test is `-mattr=+slow-lea,+slow-3ops-lea`

https://rise4fun.com/Alive/ffh

This is a recommit, originally committed in rL361852, but reverted
to investigate test-suite compile-time hangs, and then reverted in
rL362109 to fix missing constant folds that were causing
endless combine loops.

Reviewers: RKSimon, craig.topper, spatel, t.p.northover

Reviewed By: t.p.northover

Subscribers: t.p.northover, jyknight, javed.absar, kristof.beyls, fedor.sergeev, jrtc27, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62252

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362143 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DAGCombiner][X86][AArch64][AMDGPU] (x + C) - y -> (x - y) + C fold. Try 3
Roman Lebedev [Thu, 30 May 2019 20:36:54 +0000 (20:36 +0000)]
[DAGCombiner][X86][AArch64][AMDGPU] (x + C) - y  ->  (x - y) + C  fold. Try 3

Summary:
The main motivation is shown by all these `neg` instructions that are now created.
In particular, the `@reg32_lshr_by_negated_unfolded_sub_b` test.

AArch64 test changes all look good (`neg` created), or neutral.

X86 changes look neutral (vectors), or good (`neg` / `xor eax, eax` created).

I'm not sure about `X86/ragreedy-hoist-spill.ll`, it looks like the spill
is now hoisted into preheader (which should still be good?),
2 4-byte reloads become 1 8-byte reload, and are elsewhere,
but i'm not sure how that affects that loop.

I'm unable to interpret AMDGPU change, looks neutral-ish?

This is hopefully a step towards solving [[ https://bugs.llvm.org/show_bug.cgi?id=41952 | PR41952 ]].

https://rise4fun.com/Alive/pkdq (we are missing more patterns, i'll submit them later)

This is a recommit, originally committed in rL361852, but reverted
to investigate test-suite compile-time hangs, and then reverted in
rL362109 to fix missing constant folds that were causing
endless combine loops.

Reviewers: craig.topper, RKSimon, spatel, arsenm

Reviewed By: RKSimon

Subscribers: bjope, qcolombet, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, javed.absar, dstuttard, tpr, t-tye, kristof.beyls, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62223

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362142 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoWrite new tests for r362121
Michael Trent [Thu, 30 May 2019 20:09:09 +0000 (20:09 +0000)]
Write new tests for r362121

Summary:
The tests for r362121 ran dsymutil against a test binary every time.
This caused problems on lld-x86_64-ubuntu-fast as dsymutil required
a lipo tool be available to process those binaries.

This change rewrites the new test cases in macho-disassemble-g-dsym
to use bespoke test binaries (exe and dwarf) simplifying the test's
runtime dependencies.

The changes to tools/llvm-objdump/MachODump.cpp are unchanged from
r362121

Reviewers: pete, lhames, JDevlieghere

Reviewed By: pete

Subscribers: smeenai, aprantl, rupprecht, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62694

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362141 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[RuntimeDyld] Apply padding and alignment bumps to all sections with stubs, and
Lang Hames [Thu, 30 May 2019 19:59:20 +0000 (19:59 +0000)]
[RuntimeDyld] Apply padding and alignment bumps to all sections with stubs, and
increase the MachO/x86-64 stub alignment to 8.

Stub alignment should be guaranteed for any section containing RuntimeDyld
stubs/GOT-entries. To do this we should pad and align all sections containing
stubs, not just code sections.

This commit also bumps the MachO/x86-64 stub alignment to 8, so that GOT entries
will be aligned.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362139 91177308-0d34-0410-b5e6-96231b3b80d8