platform/upstream/llvm.git
4 years ago[libc] Fix benchmarks CMakeLists.txt
Guillaume Chatelet [Mon, 27 Jan 2020 09:55:32 +0000 (10:55 +0100)]
[libc] Fix benchmarks CMakeLists.txt

Summary:
This is a follow up on https://reviews.llvm.org/rGaba80d0734d1#886881.
`target_link_options` requires CMake>=3.13.

Reviewers: abrachet

Subscribers: mgorny, MaskRay, tschuett, libc-commits

Tags: #libc-project

Differential Revision: https://reviews.llvm.org/D73452

4 years ago[mlir][Linalg] Introduce folding patterns to remove certain MemRefCastOp
Nicolas Vasilache [Tue, 28 Jan 2020 18:44:37 +0000 (13:44 -0500)]
[mlir][Linalg] Introduce folding patterns to remove certain MemRefCastOp

Summary:
Canonicalization and folding patterns in StandardOps may interfere with the needs
of Linalg. This revision introduces specific foldings for dynamic memrefs that can
be proven to be static.

Very concretely:

Determines whether it is possible to fold it away in the parent Linalg op:

```mlir
  %1 = memref_cast %0 : memref<8x16xf32> to memref<?x?xf32>
  %2 = linalg.slice %1 ... : memref<?x?xf32> ...
  // or
  %1 = memref_cast %0 : memref<8x16xf32, affine_map<(i, j)->(16 * i + j)>>
         to memref<?x?xf32>
  linalg.generic(%1 ...) : memref<?x?xf32> ...
```

into

```mlir
  %2 = linalg.slice %0 ... : memref<8x16xf32> ...
  // or
  linalg.generic(%0 ... : memref<8x16xf32, affine_map<(i, j)->(16 * i + j)>>
```

Reviewers: ftynse, aartbik, jsetoain, tetuante, asaadaldien

Subscribers: mehdi_amini, rriddle, jpienaar, burmako, shauheen, antiagainst, arpith-jacob, mgester, lucyrfox, liufengdb, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D73565

4 years agoAMDGPU/GlobalISel: Manually select scalar f64 G_FNEG
Matt Arsenault [Sun, 26 Jan 2020 00:34:22 +0000 (19:34 -0500)]
AMDGPU/GlobalISel: Manually select scalar f64 G_FNEG

This should be no problem to support with a pattern, but it turns out
there are just too many yaks to shave. The main problem is in the DAG
emitter, which I have no desire to sink effort into fixing.

If we had a bit to disable patterns in the DAG importer, fixing the
GlobalISelEmitter is more manageable.

4 years agoAnalysis: Add max recursison to isDereferenceableAndAlignedPointer
Matt Arsenault [Sun, 12 Jan 2020 21:50:56 +0000 (16:50 -0500)]
Analysis: Add max recursison to isDereferenceableAndAlignedPointer

Fixes stack overflow in test/CodeGen/X86/large-gep-chain.ll when store
lowering starts adding dereferenceable flags.

4 years agoGlobalISel: Lower G_WRITE_REGISTER
Matt Arsenault [Sun, 12 Jan 2020 18:29:44 +0000 (13:29 -0500)]
GlobalISel: Lower G_WRITE_REGISTER

4 years agoMore fixes of implicit std::string conversions
Mikael Holmén [Wed, 29 Jan 2020 09:32:04 +0000 (10:32 +0100)]
More fixes of implicit std::string conversions

4 years agoAMDGPU: Fix AMDGPUUnifyDivergentExitNodes with no normal returns
Connor Abbott [Mon, 9 Dec 2019 11:04:00 +0000 (12:04 +0100)]
AMDGPU: Fix AMDGPUUnifyDivergentExitNodes with no normal returns

Summary:
The code was assuming in a few places that if there was only one exit
from the function that it was a normal return, which is invalid. It
could be an infinite loop, in which case we still need to insert the
usual fake edge so that the null export happens. This fixes shaders that
end with an infinite loop that discards.

Reviewers: arsenm, nhaehnle, critson

Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D71192

4 years agoAMDGPU: Fix handling of infinite loops in fragment shaders
Connor Abbott [Wed, 27 Nov 2019 13:09:13 +0000 (14:09 +0100)]
AMDGPU: Fix handling of infinite loops in fragment shaders

Summary:
Due to the fact that kill is just a normal intrinsic, even though it's
supposed to terminate the thread, we can end up with provably infinite
loops that are actually supposed to end successfully. The
AMDGPUUnifyDivergentExitNodes pass breaks up these loops, but because
there's no obvious place to make the loop branch to, it just makes it
return immediately, which skips the exports that are supposed to happen
at the end and hangs the GPU if all the threads end up being killed.

While it would be nice if the fact that kill terminates the thread were
modeled in the IR, I think that the structurizer as-is would make a mess if we
did that when the kill is inside control flow. For now, we just add a null
export at the end to make sure that it always exports something, which fixes
the immediate problem without penalizing the more common case. This means that
we sometimes do two "done" exports when only some of the threads enter the
discard loop, but from tests the hardware seems ok with that.

This fixes dEQP-VK.graphicsfuzz.while-inside-switch with radv.

Reviewers: arsenm, nhaehnle

Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70781

4 years ago[clangd] Remove the temporary alias for clangd::DiagnosticConsumer.
Haojian Wu [Wed, 29 Jan 2020 11:46:51 +0000 (12:46 +0100)]
[clangd] Remove the temporary alias for clangd::DiagnosticConsumer.

Reviewers: sammccall

Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D73619

4 years ago[InstCombine] canonicalize splat shuffle after cmp
Sanjay Patel [Tue, 28 Jan 2020 22:06:22 +0000 (17:06 -0500)]
[InstCombine] canonicalize splat shuffle after cmp

cmp (splat V1, M), SplatC --> splat (cmp V1, SplatC'), M

As discussed in PR44588:
https://bugs.llvm.org/show_bug.cgi?id=44588
...we try harder to push shuffles after binops than after compares.

This patch handles the special (but presumably most common case) of
splat shuffles. If both operands are splats, then we can do the
comparison on the non-splat inputs followed by splat of the compare.
That should take care of the regression noted in D73411.

There's another potential fold requested in PR37463 to scalarize the
compare, but that's another patch (and it's not clear if we can do
that without the ability to undo it later):
https://bugs.llvm.org/show_bug.cgi?id=37463

Differential Revision: https://reviews.llvm.org/D73575

4 years ago[AArch64] Add IR intrinsics for sq(r)dmulh_lane(q)
Sanne Wouda [Wed, 29 Jan 2020 13:07:15 +0000 (13:07 +0000)]
[AArch64] Add IR intrinsics for sq(r)dmulh_lane(q)

Summary:
Currently, sqdmulh_lane and friends from the ACLE (implemented in arm_neon.h),
are represented in LLVM IR as a (by vector) sqdmulh and a vector of (repeated)
indices, like so:

   %shuffle = shufflevector <4 x i16> %v, <4 x i16> undef, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
   %vqdmulh2.i = tail call <4 x i16> @llvm.aarch64.neon.sqdmulh.v4i16(<4 x i16> %a, <4 x i16> %shuffle)

When %v's values are known, the shufflevector is optimized away and we are no
longer able to select the lane variant of sqdmulh in the backend.

This defeats a (hand-coded) optimization that packs several constants into a
single vector and uses the lane intrinsics to reduce register pressure and
trade-off materialising several constants for a single vector load from the
constant pool, like so:

   int16x8_t v = {2,3,4,5,6,7,8,9};
   a = vqdmulh_laneq_s16(a, v, 0);
   b = vqdmulh_laneq_s16(b, v, 1);
   c = vqdmulh_laneq_s16(c, v, 2);
   d = vqdmulh_laneq_s16(d, v, 3);
   [...]

In one microbenchmark from libjpeg-turbo this accounts for a 2.5% to 4%
performance difference.

We could teach the compiler to recover the lane variants, but this would likely
require its own pass.  (Alternatively, "volatile" could be used on the constants
vector, but this is a bit ugly.)

This patch instead implements the following LLVM IR intrinsics for AArch64 to
maintain the original structure through IR optmization and into instruction
selection:
- sqdmulh_lane
- sqdmulh_laneq
- sqrdmulh_lane
- sqrdmulh_laneq.

These 'lane' variants need an additional register class.  The second argument
must be in the lower half of the 64-bit NEON register file, but only when
operating on i16 elements.

Note that the existing patterns for shufflevector and sqdmulh into sqdmulh_lane
(etc.) remain, so code that does not rely on NEON intrinsics to generate these
instructions is not affected.

This patch also changes clang to emit these IR intrinsics for the corresponding
NEON intrinsics (AArch64 only).

Reviewers: SjoerdMeijer, dmgreen, t.p.northover, rovka, rengolin, efriedma

Reviewed By: efriedma

Subscribers: kristof.beyls, hiraditya, jdoerfert, cfe-commits, llvm-commits

Tags: #clang, #llvm

Differential Revision: https://reviews.llvm.org/D71469

4 years ago[MVE][MC] evaluateBranch: add missing MVE opcode
Sjoerd Meijer [Wed, 29 Jan 2020 13:19:45 +0000 (13:19 +0000)]
[MVE][MC] evaluateBranch: add missing MVE opcode

This adds some missing MVE opcodes to evaluateBranch, which results in
llvm-objdump being able to print the PC relative branch target as an
annotation.

Differential Revision: https://reviews.llvm.org/D73553

4 years ago[VE] Isel patterns for fp32/64 and i32/64 conversion
Kazushi (Jam) Marukawa [Wed, 29 Jan 2020 12:58:29 +0000 (13:58 +0100)]
[VE] Isel patterns for fp32/64 and i32/64 conversion

Summary:
fp32/64 <> signed/unsigned i32/64 conversion isel patterns and tests

(This patch depends on `fsub` implemented by https://reviews.llvm.org/D73540 )

Reviewers: arsenm, craig.topper, rengolin, k-ishizaka

Reviewed By: arsenm

Subscribers: merge_guards_bot, wdng, hiraditya, llvm-commits

Tags: #ve, #llvm

Differential Revision: https://reviews.llvm.org/D73544

4 years agoRegenerate aarch64-neon-2velem.c CHECK lines
Sanne Wouda [Fri, 13 Dec 2019 10:46:39 +0000 (10:46 +0000)]
Regenerate aarch64-neon-2velem.c CHECK lines

4 years agoFix clang test build
Sanne Wouda [Wed, 29 Jan 2020 13:01:24 +0000 (13:01 +0000)]
Fix clang test build

4 years agoAdd TagDecl AST matcher
Karasev Nikita [Wed, 29 Jan 2020 12:57:27 +0000 (07:57 -0500)]
Add TagDecl AST matcher

4 years ago[yaml2obj][obj2yaml] - Add lost test cases.
Georgii Rymar [Wed, 29 Jan 2020 12:37:50 +0000 (15:37 +0300)]
[yaml2obj][obj2yaml] - Add lost test cases.

It is a part of https://reviews.llvm.org/D71872 which
was lost somehow during relanding after being reverted:

https://reviews.llvm.org/rG7570d387c21935b58afa67cb9ee17250e38721fa

4 years agoclang-format: insert trailing commas into containers.
Martin Probst [Fri, 24 Jan 2020 15:13:51 +0000 (16:13 +0100)]
clang-format: insert trailing commas into containers.

Summary:
This change adds an option to insert trailing commas into container
literals. For example, in JavaScript:

    const x = [
      a,
      b,
       ^~~~~ inserted if missing.
    ]

This is implemented as a seperate post-processing pass after formatting
(because formatting might change whether the container literal does or
does not wrap). This keeps the code relatively simple and orthogonal,
though it has the notable drawback that the newly inserted comma is not
taken into account for formatting decisions (e.g. it might exceed the 80
char limit). To avoid exceeding the ColumnLimit, a comma is only
inserted if it fits into the limit.

Trailing comma insertion conceptually conflicts with argument
bin-packing: inserting a comma disables bin-packing, so we cannot do
both. clang-format rejects FormatStyle configurations that do both with
this change.

Reviewers: krasimir, MyDeveloperDay

Subscribers: cfe-commits

Tags: #clang

4 years ago[AArch64][SVE] Add SVE2 intrinsics for uniform DSP operations
Kerry McLaughlin [Wed, 29 Jan 2020 10:40:48 +0000 (10:40 +0000)]
[AArch64][SVE] Add SVE2 intrinsics for uniform DSP operations

Summary:
Implements the following intrinsics:
 - sqrdmlah, sqrdmlsh, sqrdmulh & sqdmulh
 - [s|u]hadd, [s|u]hsub, [s|u]rhadd & [s|u]hsubr
 - urecpe, ursqrte, sqabs & sqneg

Reviewers: sdesmalen, efriedma, dancgr, cameron.mcinally

Reviewed By: efriedma

Subscribers: tschuett, kristof.beyls, hiraditya, rkruppe, psnobl, cfe-commits, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D73493

4 years ago[NFC][ARM] Add test
Sam Parker [Wed, 29 Jan 2020 11:58:41 +0000 (06:58 -0500)]
[NFC][ARM] Add test

4 years ago[clangd][vscode] Update lsp dependencies to pickup the progress support in LSP 3.15
Haojian Wu [Wed, 29 Jan 2020 10:15:47 +0000 (11:15 +0100)]
[clangd][vscode] Update lsp dependencies to pickup the progress support in LSP 3.15

Reviewers: sammccall

Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D73612

4 years ago[clangd] Replace raw lexer code with token buffer in prepare rename.
Haojian Wu [Wed, 29 Jan 2020 09:12:56 +0000 (10:12 +0100)]
[clangd] Replace raw lexer code with token buffer in prepare rename.

Summary:
there is a slight behavior change in this patch:

- before: `in^t a;`, returns our internal error message (no symbol at given location)
- after: `in^t a, returns null, and client displays their message (e.g.
   e.g. "the element can't be renamed" in vscode).

both are sensible according LSP, and we'd save one `rename` call in the later case.

Reviewers: sammccall

Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D73610

4 years ago[clangd] Go-to-definition on 'override' jumps to overridden method(s)
Sam McCall [Fri, 24 Jan 2020 17:21:47 +0000 (18:21 +0100)]
[clangd] Go-to-definition on 'override' jumps to overridden method(s)

Reviewers: kadircet

Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D73367

4 years ago[clangd] add CODE_OWNERS
Sam McCall [Tue, 28 Jan 2020 11:10:50 +0000 (12:10 +0100)]
[clangd] add CODE_OWNERS

Reviewers: klimek

Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D73537

4 years ago[LLD][ELF][ARM] Do not substitute BL/BLX for non STT_FUNC symbols.
Peter Smith [Mon, 27 Jan 2020 19:45:28 +0000 (19:45 +0000)]
[LLD][ELF][ARM] Do not substitute BL/BLX for non STT_FUNC symbols.

D73474 disabled the generation of interworking thunks for branch
relocations to non STT_FUNC symbols. This patch handles the case of BL and
BLX instructions to non STT_FUNC symbols. LLD would normally look at the
state of the caller and the callee and write a BL if the states are the
same and a BLX if the states are different.

This patch disables BL/BLX substitution when the destination symbol does
not have type STT_FUNC. This brings our behavior in line with GNU ld which
may prevent difficult to diagnose runtime errors when switching to lld.

Differential Revision: https://reviews.llvm.org/D73542

4 years ago[MLIR] Add OpenMP dialect with barrier operation
David Truby [Wed, 29 Jan 2020 11:31:25 +0000 (11:31 +0000)]
[MLIR] Add OpenMP dialect with barrier operation

Summary:
Barrier is a simple operation that takes no arguments and returns
nothing, but implies a side effect (synchronization of all threads)

Reviewers: jdoerfert

Subscribers: mgorny, guansong, mehdi_amini, rriddle, jpienaar, burmako, shauheen, antiagainst, nicolasvasilache, arpith-jacob, mgester, lucyrfox, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D72400

4 years ago[clangd] Get rid of delayed template parsing
Kadir Cetinkaya [Wed, 29 Jan 2020 11:01:46 +0000 (12:01 +0100)]
[clangd] Get rid of delayed template parsing

Summary:
No need to pass fno-delayed-template-parsing as the opposite flag is
only passed to cc1 when abi is set to msvc. Sending as a follow-up to D73613 to
keep changes in the release branch minimal.

Reviewers: sammccall

Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D73615

4 years ago[clangd][Hover] Make tests hermetic by setting target triplet
Kadir Cetinkaya [Wed, 29 Jan 2020 10:17:45 +0000 (11:17 +0100)]
[clangd][Hover] Make tests hermetic by setting target triplet

Summary: Fixes https://bugs.llvm.org/show_bug.cgi?id=44696

Reviewers: sammccall

Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D73613

4 years agoFix an implicit conversion in clang-tidy. GCC 5 complains about it.
Benjamin Kramer [Wed, 29 Jan 2020 11:05:01 +0000 (12:05 +0100)]
Fix an implicit conversion in clang-tidy. GCC 5 complains about it.

4 years ago[ARM] Add documentation for -march= and -mfpu= command line options
Momchil Velikov [Wed, 29 Jan 2020 10:38:10 +0000 (10:38 +0000)]
[ARM] Add documentation for -march= and -mfpu= command line options

Differential Revision: https://reviews.llvm.org/D73459

4 years ago[AArch64][SVE] Add SVE2 intrinsics for pairwise arithmetic
Kerry McLaughlin [Wed, 29 Jan 2020 09:57:30 +0000 (09:57 +0000)]
[AArch64][SVE] Add SVE2 intrinsics for pairwise arithmetic

Summary:
Implements the following intrinsics:
 - addp
 - smaxp, sminp, umaxp & uminp
 - sadalp & uadalp

Reviewers: dancgr, efriedma, sdesmalen, c-rhodes

Reviewed By: c-rhodes

Subscribers: tschuett, kristof.beyls, hiraditya, rkruppe, psnobl, cameron.mcinally, cfe-commits, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D73347

4 years ago[DebugInfo] Make most debug line prologue errors non-fatal to parsing
James Henderson [Tue, 28 Jan 2020 14:04:42 +0000 (14:04 +0000)]
[DebugInfo] Make most debug line prologue errors non-fatal to parsing

Many of the debug line prologue errors are not inherently fatal. In most
cases, we can make reasonable assumptions and carry on. This patch does
exactly that. In the case of length problems, the approach of "assume
stated length is correct" is taken which means the offset might need
adjusting.

This is a relanding of b94191fe, fixing an LLD test and the LLDB build.

Reviewed by: dblaikie, labath

Differential Revision: https://reviews.llvm.org/D72158

4 years ago[lldb] More windows StringRef fixes
Pavel Labath [Wed, 29 Jan 2020 10:15:20 +0000 (11:15 +0100)]
[lldb] More windows StringRef fixes

I don't have a windows build around, so I am just going by the buildbot
messages.

4 years ago[VE] fp32/64 fadd/fsub/fdiv/fmul isel patterns
Kazushi (Jam) Marukawa [Wed, 29 Jan 2020 09:16:42 +0000 (10:16 +0100)]
[VE] fp32/64 fadd/fsub/fdiv/fmul isel patterns

Summary: fp32/64 fadd/fsub/fdiv/fmul isel patterns and tests.

Reviewers: arsenm, craig.topper, rengolin, k-ishizaka

Subscribers: merge_guards_bot, wdng, hiraditya, llvm-commits

Differential Revision: https://reviews.llvm.org/D73540

4 years ago[ARM64] Debug info for structure argument missing DW_AT_location
David Stenberg [Wed, 29 Jan 2020 08:44:02 +0000 (09:44 +0100)]
[ARM64] Debug info for structure argument missing DW_AT_location

Summary:
Prevent eliminating dbg_val due to COPY.

Fixes this
https://bugs.llvm.org/show_bug.cgi?id=40709

Patch by: Kamlesh Kumar (kamleshbhalui)

Reviewers: aprantl, dblaikie, vsk, dsanders

Reviewed By: dsanders

Subscribers: dstenb, kristof.beyls, hiraditya, llvm-commits

Tags: #debug-info, #llvm

Differential Revision: https://reviews.llvm.org/D73159

4 years ago[ASTMatchers] StringRef'ify hasName
Benjamin Kramer [Wed, 29 Jan 2020 09:52:25 +0000 (10:52 +0100)]
[ASTMatchers] StringRef'ify hasName

This was just inconvenient, and we make a copy anyways.

4 years ago[VE][fix] (more) explicit StringRef to std::string
Simon Moll [Wed, 29 Jan 2020 09:42:00 +0000 (10:42 +0100)]
[VE][fix] (more) explicit StringRef to std::string

4 years ago[AMDGPU] Simplify DS and SM cases in getMemOperandsWithOffset
Jay Foad [Mon, 6 Jan 2020 14:56:24 +0000 (14:56 +0000)]
[AMDGPU] Simplify DS and SM cases in getMemOperandsWithOffset

Summary:
This removes a couple of unnecessary isReg checks, now that
memOpsHaveSameBasePtr can handle FI operands, but is otherwise NFC.

Reviewers: arsenm, rampitec

Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, hiraditya, kerbowa, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D73485

4 years ago[VE][fix] Explicit StringRef to std::string conversion
Simon Moll [Wed, 29 Jan 2020 09:29:43 +0000 (10:29 +0100)]
[VE][fix] Explicit StringRef to std::string conversion

Adapt to changes of "[ADT] Make StringRef's std::string conversion
operator explicit" (777180a32).

4 years ago[clangd] Add a symbol-name-based blacklist for rename.
Haojian Wu [Mon, 27 Jan 2020 09:39:55 +0000 (10:39 +0100)]
[clangd] Add a symbol-name-based blacklist for rename.

Summary:
This patch adds a simple mechanism to disallow global rename
on std symbols. We might extend it to other symbols, e.g. protobuf.

Reviewers: kadircet

Subscribers: mgorny, ilya-biryukov, MaskRay, jkorous, arphaman, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D73450

4 years agoFix clang unnittest build with GCC 5
Benjamin Kramer [Wed, 29 Jan 2020 09:30:36 +0000 (10:30 +0100)]
Fix clang unnittest build with GCC 5

4 years ago[lldb] Fix windows build for the StringRef conversion operator change
Pavel Labath [Wed, 29 Jan 2020 09:08:40 +0000 (10:08 +0100)]
[lldb] Fix windows build for the StringRef conversion operator change

"operator std::string()" is now explicit.

4 years ago[ARC] Fix ARCTargetMachine after 777180a32b6107
Fangrui Song [Wed, 29 Jan 2020 08:59:16 +0000 (00:59 -0800)]
[ARC] Fix ARCTargetMachine after 777180a32b6107

4 years ago[RDA][ARM] Move functionality into RDA
Sam Parker [Wed, 29 Jan 2020 08:26:11 +0000 (03:26 -0500)]
[RDA][ARM] Move functionality into RDA

Add several new helpers to RDA:
- hasLocalDefBefore
- isRegDefinedAfter
- isSafeToDefRegAt

And move two bits of logic from ARMLowOverheadLoops into RDA:
- isSafeToMove
- isSafeToRemove

Both of these have some wrappers too to make them more convienent to
use.

Differential Revision: https://reviews.llvm.org/D73460

4 years ago[lldb] Don't create duplicate declarations when completing a forward declaration...
Raphael Isemann [Wed, 29 Jan 2020 08:20:06 +0000 (09:20 +0100)]
[lldb] Don't create duplicate declarations when completing a forward declaration with a definition from another source

Summary:
I noticed this strange line in `ASTImporterDelegate::ImportDefinitionTo` which doesn't make a lot of sense:
```
to_tag->setCompleteDefinition(from_tag->isCompleteDefinition());
```

It forcibly sets the imported TagDecl to be defined if the source TagDecl was defined. This doesn't make any
sense as in this code we already forced the ASTImporter to import the definition so this should always be
a no-op.

Turns out this is hiding two bugs:
1. The way we handle forward declarations in the debug info that might be completed later is that we
  import them and then mark them as having external lexical storage. This makes Clang ask for the definition
  later when it needs it (at which point we hopefully have the definition around and can complete it). However,
  this is currently not completing the forward decls with external storage but instead creates a duplicated decl
  in the target AST which is then defined. The forward decl is kept incomplete after the import and we just
  forcibly make it a definition of the record without any content with our workaround. The TestSharedLib* tests
  is only passing because of this.
2. Minimal import of lambdas is broken and never imports the definition it seems. That appears to be a bug
  in the ASTImporter which gives the definition of lambda's some special treatment. TestLambdas.py is actually
  broken but is passing because of this workaround.

This patch fixes the first bug by forcing the ASTImporter to import to the target forward declaration. We can't
delete the workaround as the second bug is still around but that will be a follow up review for the ASTImporter.
However it will get rid of all the duplicated RecordDecls that are in our expression AST that are strangely defined
but don't have any of the fields they are supposed to have.

Reviewers: shafik, labath

Reviewed By: shafik

Subscribers: aprantl, abidh, JDevlieghere, lldb-commits

Tags: #lldb

Differential Revision: https://reviews.llvm.org/D73345

4 years ago[lldb] Complete return types of CXXMethodDecls to prevent crashing due to covariant...
Raphael Isemann [Wed, 29 Jan 2020 08:07:55 +0000 (09:07 +0100)]
[lldb] Complete return types of CXXMethodDecls to prevent crashing due to covariant return types

Summary:
Currently we crash in Clang's CodeGen when we call functions with covariant return types with this assert:
```
Assertion failed: (DD && "queried property of class with no definition"), function data, file clang/include/clang/AST/DeclCXX.h, line 433.
```
when calling `clang::CXXRecordDecl::isDerivedFrom` from the `ItaniumVTableBuilder`.

Clang seems to assume that the underlying record decls of covariant return types are already completed.
This is true during a normal Clang invocation as there the type checker will complete both decls when
checking if the overloaded function is valid (i.e., the return types are covariant).

When we minimally import our AST into the expression in LLDB we don't do this type checking (which
would complete the record decls) and we end up trying to access the invalid record decls from CodeGen
which makes us trigger the assert.

This patch just completes the underlying types of ptr/ref return types of virtual function so that the
underlying records are complete and we behave as Clang expects us to do.

Fixes rdar://38048657

Reviewers: lhames, shafik

Reviewed By: shafik

Subscribers: abidh, JDevlieghere, lldb-commits

Tags: #lldb

Differential Revision: https://reviews.llvm.org/D73024

4 years ago[lldb] Use CompletionRequest in REPL::CompleteCode and remove translation code to...
Raphael Isemann [Mon, 27 Jan 2020 20:24:39 +0000 (21:24 +0100)]
[lldb] Use CompletionRequest in REPL::CompleteCode and remove translation code to old API

Any REPL client should just move to CompletionRequest instead of relying on
the translation code from the old API, so let's remove that translation code.

4 years ago[X86] matchAdd: don't fold a large offset into a %rip relative address
Fangrui Song [Wed, 29 Jan 2020 05:18:29 +0000 (21:18 -0800)]
[X86] matchAdd: don't fold a large offset into a %rip relative address

For `ret i64 add (i64 ptrtoint (i32* @foo to i64), i64 1701208431)`,

```
X86DAGToDAGISel::matchAdd
  ...
// AM.setBaseReg(CurDAG->getRegister(X86::RIP, MVT::i64));
  if (!matchAddressRecursively(N.getOperand(0), AM, Depth+1) &&
// Try folding offset but fail; there is a symbolic displacement, so offset cannot be too large
      !matchAddressRecursively(Handle.getValue().getOperand(1), AM, Depth+1))
    return false;
  ...
  // Try again after commuting the operands.
// AM.Disp = Val; foldOffsetIntoAddress() does not know there will be a symbolic displacement
  if (!matchAddressRecursively(Handle.getValue().getOperand(1), AM, Depth+1) &&
// AM.setBaseReg(CurDAG->getRegister(X86::RIP, MVT::i64));
      !matchAddressRecursively(Handle.getValue().getOperand(0), AM, Depth+1))
// Succeeded! Produced leaq sym+disp(%rip),...
    return false;
```

`foldOffsetIntoAddress()` currently does not know there is a symbolic
displacement and can fold a large offset.

The produced `leaq sym+disp(%rip), %rax` instruction is relocated by
an R_X86_64_PC32. If disp is large and sym+disp-rip>=2**31, there
will be a relocation overflow.

This approach is still not elegant. Unfortunately the isRIPRelative
interface is a bit clumsy. I tried several solutions and eventually
picked this one.

Differential Revision: https://reviews.llvm.org/D73606

4 years ago[Attributor][Fix] Initialize unused but loaded variable
Johannes Doerfert [Wed, 29 Jan 2020 05:51:25 +0000 (23:51 -0600)]
[Attributor][Fix] Initialize unused but loaded variable

This hopefully un-breaks:
  http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast/builds/38333

4 years ago[Attributor] Reuse existing logic to avoid duplication
Johannes Doerfert [Tue, 28 Jan 2020 15:46:15 +0000 (09:46 -0600)]
[Attributor] Reuse existing logic to avoid duplication

There was a TODO in AAValueConstantRangeArgument to reuse
AAArgumentFromCallSiteArguments. We now do this by allowing new States
to be build from the bestState.

4 years ago[Attributor][FIX] Treat invalidated attributes as changed
Johannes Doerfert [Tue, 28 Jan 2020 15:38:07 +0000 (09:38 -0600)]
[Attributor][FIX] Treat invalidated attributes as changed

If we invalidate an attribute we need to inform all dependent ones even
if the fixpoint state is not invalid. Before we only continued
invalidation if the fixpoint state was invalid, now we signal a change
in case the fixpoint state is valid.

The test case was already included in D71620 but the problem was hiding
because it only manifested with the old PM (for that input).

4 years ago[Attributor] Modularize AANoAliasCallSiteArgument to simplify extensions
Johannes Doerfert [Tue, 28 Jan 2020 04:24:32 +0000 (22:24 -0600)]
[Attributor] Modularize AANoAliasCallSiteArgument to simplify extensions

This patch modularizes the way we check for no-alias call site arguments
by putting the existing logic into helper functions. The reasoning was
not changed but special cases for readonly/readnone were added.

4 years ago[Attributor] Mark a non-defined `null` pointer as `noalias`
Johannes Doerfert [Tue, 28 Jan 2020 04:19:11 +0000 (22:19 -0600)]
[Attributor] Mark a non-defined `null` pointer as `noalias`

If `null` is not defined we cannot access it, hence the pointer is
`noalias`. While this is not helpful on it's own it simplifies later
deductions that can skip over already known `noalias` pointers in
certain situations.

4 years ago[Attributor][NFC] Remove ugly and unneeded cast
Johannes Doerfert [Tue, 28 Jan 2020 17:49:59 +0000 (11:49 -0600)]
[Attributor][NFC] Remove ugly and unneeded cast

4 years ago[Attributor][NFC] Improve debug messages
Johannes Doerfert [Tue, 28 Jan 2020 17:49:35 +0000 (11:49 -0600)]
[Attributor][NFC] Improve debug messages

4 years ago[Attributor][NFC] Internalize helper function
Johannes Doerfert [Sun, 26 Jan 2020 08:49:58 +0000 (02:49 -0600)]
[Attributor][NFC] Internalize helper function

4 years agoFix polly build after StringRef change.
Eli Friedman [Wed, 29 Jan 2020 03:44:20 +0000 (19:44 -0800)]
Fix polly build after StringRef change.

4 years agoOne more bugpoitn fix for GCC5
Benjamin Kramer [Wed, 29 Jan 2020 02:42:02 +0000 (03:42 +0100)]
One more bugpoitn fix for GCC5

4 years agoTry harder to fix bugpoint with GCC5
Benjamin Kramer [Wed, 29 Jan 2020 02:30:05 +0000 (03:30 +0100)]
Try harder to fix bugpoint with GCC5

4 years ago[driver][Darwin] Add an -ibuiltininc flag that lets Darwin driver
Alex Lorenz [Mon, 27 Jan 2020 21:01:06 +0000 (13:01 -0800)]
[driver][Darwin] Add an -ibuiltininc flag that lets Darwin driver
include Clang builtin headers even with -nostdinc

Some projects use -nostdinc, but need to access some intrinsics files when building specific files.
The new -ibuiltininc flag lets them use this flag when compiling these files to ensure they can
find Clang's builtin headers.

The use of -nobuiltininc after the -ibuiltininc flag does not add the builtin header
search path to the list of header search paths.

Differential Revision: https://reviews.llvm.org/D73500

4 years agoMake bugpoint work with gcc5 again.
Benjamin Kramer [Wed, 29 Jan 2020 02:11:00 +0000 (03:11 +0100)]
Make bugpoint work with gcc5 again.

4 years agoFix more implicit conversions. Getting closer to having clang working with gcc 5...
Benjamin Kramer [Wed, 29 Jan 2020 01:57:59 +0000 (02:57 +0100)]
Fix more implicit conversions. Getting closer to having clang working with gcc 5 again

4 years agoFix conversions in clang and examples
Benjamin Kramer [Wed, 29 Jan 2020 01:48:15 +0000 (02:48 +0100)]
Fix conversions in clang and examples

4 years agoGCC5 buildbot made it to clang. Fix implicit conversions it found.
Benjamin Kramer [Wed, 29 Jan 2020 01:19:22 +0000 (02:19 +0100)]
GCC5 buildbot made it to clang. Fix implicit conversions it found.

4 years ago[NFC] Fix unused variable warning.
Nate Voorhies [Wed, 29 Jan 2020 00:07:48 +0000 (16:07 -0800)]
[NFC] Fix unused variable warning.

Reviewers: dschuff

Reviewed By: dschuff

Subscribers: hiraditya, aheejin, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D73591

4 years ago[CodeExtractor] Remove stale llvm.assume calls from extracted region
Vedant Kumar [Wed, 29 Jan 2020 01:03:39 +0000 (17:03 -0800)]
[CodeExtractor] Remove stale llvm.assume calls from extracted region

During extraction, stale llvm.assume handles may be retained in the
original function. The setup is:

1) CodeExtractor unregisters assumptions in the blocks that are to be
   extracted.

2) Extraction happens. There are now two functions: f1 and f1.extracted.

3) Leftover assumptions in f1 (/not/ removed as they were not in the set of
   blocks to be extracted) now have affected-value llvm.assume handles in
   f1.extracted.

When assumptions for a value used in f1 are looked up, ValueTracking can assert
as some of the handles are in the wrong function. To fix this, simply erase the
llvm.assume calls in the extracted function.

Alternatives include flushing the assumption cache in the original function, or
walking all values used in the original function to prune stale affected-value
handles. Both seem more expensive.

Testing: check-llvm, LNT run with -mllvm -hot-cold-split enabled

rdar://58460728

4 years agoAnother stab at making the gold plugin compile again
Benjamin Kramer [Wed, 29 Jan 2020 01:12:23 +0000 (02:12 +0100)]
Another stab at making the gold plugin compile again

4 years ago[lldb/API] Fix bogus copy assignment operator
Jonas Devlieghere [Wed, 29 Jan 2020 01:09:49 +0000 (17:09 -0800)]
[lldb/API] Fix bogus copy assignment operator

The copy assignment operator is supposed to return the class and not
void. Fix the methods and the reproducer instrumentation macros.

4 years agoAnother round of GCC5 fixes.
Benjamin Kramer [Wed, 29 Jan 2020 01:09:24 +0000 (02:09 +0100)]
Another round of GCC5 fixes.

4 years ago[clangd] Fix null check in FindTarget.
Sam McCall [Wed, 29 Jan 2020 01:01:42 +0000 (02:01 +0100)]
[clangd] Fix null check in FindTarget.

I've hit this stack trace a few times but don't have a good reproducer.
The code is unsafe by inspection, though.

4 years ago[WebAssembly] Preserve debug frame base information through register coloring
Derek Schuff [Tue, 28 Jan 2020 21:46:03 +0000 (13:46 -0800)]
[WebAssembly] Preserve debug frame base information through register coloring

2 fixes:

Register coloring can re-assign virtual registers. When the frame base register
is colored, update the DwarfFrameBase accordingly When the frame base register
is stackified, do not attempt to encode DW_AT_frame_base as a local In the
future we will presumably want to handle this case better but for now we can
emit worse debug info rather than crashing.

Differential Revision: https://reviews.llvm.org/D73581

4 years agoFix one round of implicit conversions found by g++5.
Benjamin Kramer [Wed, 29 Jan 2020 00:49:54 +0000 (01:49 +0100)]
Fix one round of implicit conversions found by g++5.

4 years agoFix clangd-xpc-test-client build after 777180a32b6107
Nico Weber [Wed, 29 Jan 2020 00:48:31 +0000 (19:48 -0500)]
Fix clangd-xpc-test-client build after 777180a32b6107

4 years ago[lldb/API] Implement the copy (assignment) constructor for SBLaunchInfo
Jonas Devlieghere [Wed, 29 Jan 2020 00:44:30 +0000 (16:44 -0800)]
[lldb/API] Implement the copy (assignment) constructor for SBLaunchInfo

Currently the constructor is compiler generated which means it doesn't
get instrumented for the reproducers.

4 years agoFix xpc build after 777180a32b6107
Nico Weber [Wed, 29 Jan 2020 00:35:11 +0000 (19:35 -0500)]
Fix xpc build after 777180a32b6107

4 years ago[X86] Use SelectionDAG::getZExtOrTrunc to simplify some code. NFCI
Craig Topper [Wed, 29 Jan 2020 00:13:51 +0000 (16:13 -0800)]
[X86] Use SelectionDAG::getZExtOrTrunc to simplify some code. NFCI

4 years ago[X86] Add test case for llvm.flt.rounds
Craig Topper [Tue, 28 Jan 2020 23:37:22 +0000 (15:37 -0800)]
[X86] Add test case for llvm.flt.rounds

4 years ago[lldb/API] Implement the copy (assignment) constructor for SBPlatform
Jonas Devlieghere [Wed, 29 Jan 2020 00:13:15 +0000 (16:13 -0800)]
[lldb/API] Implement the copy (assignment) constructor for SBPlatform

Currently the constructor is compiler generated which means it doesn't
get instrumented for the reproducers.

4 years agoFix AVR build after 777180a32b6107
Nico Weber [Wed, 29 Jan 2020 00:22:22 +0000 (19:22 -0500)]
Fix AVR build after 777180a32b6107

4 years ago[lldb] Remove unused header from ValueObject.cpp
Alex Langford [Tue, 28 Jan 2020 22:44:32 +0000 (14:44 -0800)]
[lldb] Remove unused header from ValueObject.cpp

In commit 5eaf44f99f0a0a3bdfa892892b8aaca841c8dbe0 I removed the last
instance of TypeSystemClang from ValueObject, so the header is no longer
needed.

4 years agoAddress implicit conversions detected by g++ 5 only.
Benjamin Kramer [Wed, 29 Jan 2020 00:01:09 +0000 (01:01 +0100)]
Address implicit conversions detected by g++ 5 only.

4 years agoOne more batch of things found by g++ 6
Benjamin Kramer [Tue, 28 Jan 2020 23:49:35 +0000 (00:49 +0100)]
One more batch of things found by g++ 6

4 years ago[AliasAnalysis] Add missing FMRB_* enums.
Eli Friedman [Wed, 22 Jan 2020 00:51:17 +0000 (16:51 -0800)]
[AliasAnalysis] Add missing FMRB_* enums.

Previously, the enums didn't account for all the possible cases, which
could cause misleading results (particularly for a "switch" on
FunctionModRefBehavior).

Fixes regression in polly from recent patch to add writeonly to memset.

While I'm here, also fix a few dubious uses of the FMRB_* enum values.

Differential Revision: https://reviews.llvm.org/D73154

4 years agoFix a couple more implicit conversions that Clang doesn't diagnose.
Benjamin Kramer [Tue, 28 Jan 2020 23:42:56 +0000 (00:42 +0100)]
Fix a couple more implicit conversions that Clang doesn't diagnose.

4 years agoA bunch more implicit string conversions that my Clang didn't detect.
Benjamin Kramer [Tue, 28 Jan 2020 23:29:48 +0000 (00:29 +0100)]
A bunch more implicit string conversions that my Clang didn't detect.

4 years ago[NFC] Removing experimental designation for ninja in docs.
Nate Voorhies [Tue, 28 Jan 2020 19:11:04 +0000 (11:11 -0800)]
[NFC] Removing experimental designation for ninja in docs.

Summary:
Ninja is no longer an experimental tool, documentation changed to
reflect this.

Reviewers: nikola

Reviewed By: nikola

Subscribers: cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D73567

4 years agoFix another implicit conversion in the directory watcher
Jonas Devlieghere [Tue, 28 Jan 2020 23:28:13 +0000 (15:28 -0800)]
Fix another implicit conversion in the directory watcher

4 years ago[tblgen] Fix implicit conversion only diagnosed by g++ 6
Benjamin Kramer [Tue, 28 Jan 2020 23:25:44 +0000 (00:25 +0100)]
[tblgen] Fix implicit conversion only diagnosed by g++ 6

4 years ago[NFC] Fix comment typo
Francis Visoiu Mistrih [Tue, 28 Jan 2020 23:23:28 +0000 (15:23 -0800)]
[NFC] Fix comment typo

4 years agoFix more implicit conversions
Jonas Devlieghere [Tue, 28 Jan 2020 23:19:17 +0000 (15:19 -0800)]
Fix more implicit conversions

4 years ago[Driver] Fix implicit conversion guarded by #ifdef _WIN32
Benjamin Kramer [Tue, 28 Jan 2020 23:18:16 +0000 (00:18 +0100)]
[Driver] Fix implicit conversion guarded by #ifdef _WIN32

4 years agoFix implicit conversions in example code.
Benjamin Kramer [Tue, 28 Jan 2020 23:17:07 +0000 (00:17 +0100)]
Fix implicit conversions in example code.

4 years ago[lldb/Plugin] Fix implicit conversion in GDBRemote
Jonas Devlieghere [Tue, 28 Jan 2020 23:16:56 +0000 (15:16 -0800)]
[lldb/Plugin] Fix implicit conversion in GDBRemote

4 years ago[lldb/Reproducer] s/nullptr_t/std::nullptr_t/
Jonas Devlieghere [Tue, 28 Jan 2020 23:14:40 +0000 (15:14 -0800)]
[lldb/Reproducer] s/nullptr_t/std::nullptr_t/

Fixes error: unknown type name 'nullptr_t'; did you mean
'std::nullptr_t'.

4 years agoFix implicit conversion in the lldb Python plugin
Benjamin Kramer [Tue, 28 Jan 2020 23:06:33 +0000 (00:06 +0100)]
Fix implicit conversion in the lldb Python plugin

4 years ago[lldb/Reproducer] Include result in recording statements
Jonas Devlieghere [Tue, 28 Jan 2020 22:47:29 +0000 (14:47 -0800)]
[lldb/Reproducer] Include result in recording statements

Include the return value in the recording log statements. This helps
diagnose uninstrumented (copy assignment) constructors.

4 years ago[Support] Fix implicit std::string conversions on Win32.
Benjamin Kramer [Tue, 28 Jan 2020 23:02:26 +0000 (00:02 +0100)]
[Support] Fix implicit std::string conversions on Win32.

4 years ago[ADT] Make StringRef's std::string conversion operator explicit
Benjamin Kramer [Tue, 28 Jan 2020 22:30:02 +0000 (23:30 +0100)]
[ADT] Make StringRef's std::string conversion operator explicit

This has the same behavior as converting std::string_view to
std::string. This is an expensive conversion, so explicit conversions
are helpful for avoiding unneccessary string copies.

4 years ago[libcxx] Link against android_support when needed
Shoaib Meenai [Tue, 28 Jan 2020 01:29:41 +0000 (17:29 -0800)]
[libcxx] Link against android_support when needed

libc++ on Android needs to be linked against libandroid_support on API
levels less than 21 to provide needed functions that aren't in the libc
on those platforms (e.g. posix_memalign for libcxxabi). libc++ from the
NDK is a linker script that pulls in libandroid_support, but for
building libc++ itself, we need to explicitly add libandroid_support as
a dependency. Moreover, libc++ headers reference the functions provided
by libandroid_support, so it needs to be added as a public dependency.

Differential Revision: https://reviews.llvm.org/D73516

4 years ago[asan] Fix test compilation on Android API <= 17
Shoaib Meenai [Tue, 28 Jan 2020 01:25:38 +0000 (17:25 -0800)]
[asan] Fix test compilation on Android API <= 17

mlockall and munlockall were introduced in Android API 17, so avoid
referencing them on prior versions.

Differential Revision: https://reviews.llvm.org/D73515