OSDN Git Service

android-x86/external-llvm.git
6 years ago[dsymutil] Don't mark forward declarations as canonical.
Jonas Devlieghere [Thu, 31 Aug 2017 18:06:44 +0000 (18:06 +0000)]
[dsymutil] Don't mark forward declarations as canonical.

This patch completes the work done by Frederic Riss to addresses
dsymutil incorrectly considering forward declaration as canonical during
uniquing. This resulted in references to the forward declaration even
after the definition was encountered.

In addition to the test provided by Alexander Shaposhnikov in D29609, I
added another test to cover several scenarios that were mentioned in his
conversation with Fred. We now also check that uniquing still occurs
after the definition was encountered.

For more context please refer to D29609

Differential revision: https://reviews.llvm.org/D37127

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312264 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[lit] Make symlinks in test paths work a different way
Reid Kleckner [Thu, 31 Aug 2017 17:07:35 +0000 (17:07 +0000)]
[lit] Make symlinks in test paths work a different way

Use os.path.normpath instead of realpath to collapse '..' and '.' path
components. Use realpath when caching search results about a path for
good measure.

I considered rigging up a test involving symlinks for this, but I doubt
I can check a symlink into SVN. The test would have to conditionally
create a symlink at runtime if the host OS supports it. This sounds too
fragile and complicated to me to be worth it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312254 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DAGCombiner] Do a better job of ensuring we don't split elements when combining...
Craig Topper [Thu, 31 Aug 2017 17:02:22 +0000 (17:02 +0000)]
[DAGCombiner] Do a better job of ensuring we don't split elements when combining an extract_subvector of a bitcasted build_vector.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312253 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-dwarfdump] Brief mode only dumps debug_info by default
Jonas Devlieghere [Thu, 31 Aug 2017 16:44:47 +0000 (16:44 +0000)]
[llvm-dwarfdump] Brief mode only dumps debug_info by default

This patch changes the default behavior in brief mode to only show the
debug_info section. This is undoubtedly the most popular and likely the
one you'd want in brief mode.

Non-brief mode behavior is not affected and still defaults to all.

Differential revision: https://reviews.llvm.org/D37334

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312252 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "[lit] Don't call realpath on the path used for test suite search"
Reid Kleckner [Thu, 31 Aug 2017 16:42:30 +0000 (16:42 +0000)]
Revert "[lit] Don't call realpath on the path used for test suite search"

This reverts r312250, it breaks the lit test suite.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312251 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[lit] Don't call realpath on the path used for test suite search
Reid Kleckner [Thu, 31 Aug 2017 16:35:08 +0000 (16:35 +0000)]
[lit] Don't call realpath on the path used for test suite search

This preserves symlinks in paths, so that someone can symlink more tests
into a larger test suite. For example, debuginfo-tests is currently
designed to be checked out into clang/test. With this change, it can be
symlinked into place instead, which works better with the monorepo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312250 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstCombine] improve demanded vector elements analysis of insertelement
Sanjay Patel [Thu, 31 Aug 2017 15:57:17 +0000 (15:57 +0000)]
[InstCombine] improve demanded vector elements analysis of insertelement

Recurse instead of returning on the first found optimization. Also, return early in the caller
instead of continuing because that allows another round of simplification before we might
potentially lose undef information from a shuffle mask by eliminating the shuffle.

As noted in the review, we could probably do better and be more efficient by moving all of
demanded elements into a separate pass, but this is yet another quick fix to instcombine.

Differential Revision: https://reviews.llvm.org/D37236

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312248 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[codeview] Generalize DIExpression parsing to handle load chains
Reid Kleckner [Thu, 31 Aug 2017 15:56:49 +0000 (15:56 +0000)]
[codeview] Generalize DIExpression parsing to handle load chains

Summary:
Hopefully this also clarifies exactly when and why we're rewriting
certiain S_LOCALs using reference types: We're using the reference type
to stand in for a zero-offset load.

Reviewers: inglorion

Subscribers: llvm-commits, hiraditya

Differential Revision: https://reviews.llvm.org/D37309

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312247 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert r312240
Alex Lorenz [Thu, 31 Aug 2017 15:51:23 +0000 (15:51 +0000)]
Revert r312240

The buildbots have shown that -Wstrict-prototypes behaves differently in GCC
and Clang so we should keep it disabled until Clang follows GCC's behaviour

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312246 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoUpdate test:
Konstantin Zhuravlyov [Thu, 31 Aug 2017 15:35:33 +0000 (15:35 +0000)]
Update test:

  - REQUIRES: x86_64-linux -> REQUIRES: shell

Differential Revision: https://reviews.llvm.org/D37316

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312245 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert r311525: "[XRay][CodeGen] Use PIC-friendly code in XRay sleds; remove syntheti...
Daniel Jasper [Thu, 31 Aug 2017 15:17:17 +0000 (15:17 +0000)]
Revert r311525: "[XRay][CodeGen] Use PIC-friendly code in XRay sleds; remove synthetic references in .text"

Breaks builds internally. Will forward repo instructions to author.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312243 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SLPVectorizer] Move out Entry->NeedToGather check and assert of inner loop as invari...
Dinar Temirbulatov [Thu, 31 Aug 2017 14:10:07 +0000 (14:10 +0000)]
[SLPVectorizer] Move out Entry->NeedToGather check and assert of inner loop as invariant, NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312242 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Added run line to intrinsics upgrade test. NFC.
Yael Tsafrir [Thu, 31 Aug 2017 13:56:22 +0000 (13:56 +0000)]
[X86] Added run line to intrinsics upgrade test. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312241 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoBuild LLVM with -Wstrict-prototypes enabled
Alex Lorenz [Thu, 31 Aug 2017 13:23:24 +0000 (13:23 +0000)]
Build LLVM with -Wstrict-prototypes enabled

Clang 5 supports -Wstrict-prototypes. We should use it to catch any C
declarations that declare a non-prototype function.

rdar://33705313

Differential Revision: https://reviews.llvm.org/D36669

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312240 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[BinaryFormat] Fix out of bounds read.
Benjamin Kramer [Thu, 31 Aug 2017 12:50:42 +0000 (12:50 +0000)]
[BinaryFormat] Fix out of bounds read.

Found by OSS-FUZZ!
https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=3220

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312238 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMD family 17h (znver1) scheduler model update.
Ashutosh Nema [Thu, 31 Aug 2017 12:38:35 +0000 (12:38 +0000)]
AMD family 17h (znver1) scheduler model update.

Summary:
This patch enables the following:
1) Regex based Instruction itineraries for integer instructions.
2) The instructions are grouped as per the nature of the instructions
   (move, arithmetic, logic, Misc, Control Transfer).
3) FP instructions and their itineraries are added which includes values
   for SSE4A, BMI, BMI2 and SHA instructions.

Patch by Ganesh Gopalasubramanian

Reviewers: RKSimon, craig.topper

Subscribers: vprasad, shivaram, ddibyend, andreadb, javed.absar, llvm-commits

Differential Revision: https://reviews.llvm.org/D36617

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312237 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Docs] Update CodingStandards to recommend range-based for loops
Alex Bradbury [Thu, 31 Aug 2017 12:34:20 +0000 (12:34 +0000)]
[Docs] Update CodingStandards to recommend range-based for loops

The CodingStandards section on avoiding the re-evaluation of end() hasn't been
updated since range-based for loops were adopted in the LLVM codebase. This
patch adds a very brief section that documents how range-based for loops
should be used wherever possible. It also moves example code in
CodingStandards to use range-based for loops and auto when appropriate.

Differential Revision: https://reviews.llvm.org/D37264

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312236 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Object] Verify object sizes before handing out StringRefs pointing out
Benjamin Kramer [Thu, 31 Aug 2017 12:27:10 +0000 (12:27 +0000)]
[Object] Verify object sizes before handing out StringRefs pointing out
of bounds.

This can only happen on corrupt input. Found by OSS-FUZZ!
https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=3228

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312235 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[GlobalISel][X86] Refactor X86LegalizerInfo. NFC.
Igor Breger [Thu, 31 Aug 2017 11:40:03 +0000 (11:40 +0000)]
[GlobalISel][X86] Refactor X86LegalizerInfo. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312234 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64] v8.3-a complex number support
Sam Parker [Thu, 31 Aug 2017 09:27:04 +0000 (09:27 +0000)]
[AArch64] v8.3-a complex number support

New instructions are added to AArch32 and AArch64 to aid
floating-point multiplication and addition of complex numbers,
where the complex numbers are packed in a vector register as a
pair of elements. The Imaginary part of the number is placed in the
more significant element, and the Real part of the number is placed
in the less significant element.

Differential Revision: https://reviews.llvm.org/D36792

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312228 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-cov] Read in function names for filtering from a text file.
Sean Eveson [Thu, 31 Aug 2017 09:11:31 +0000 (09:11 +0000)]
[llvm-cov] Read in function names for filtering from a text file.

Summary: Add a -name-whitelist option, which behaves in the same way as -name, but it reads in multiple function names from the given input file(s).

Reviewers: vsk

Reviewed By: vsk

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D37111

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312227 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ARM] Reverse PostRASched subtarget feature logic
Sam Parker [Thu, 31 Aug 2017 08:57:51 +0000 (08:57 +0000)]
[ARM] Reverse PostRASched subtarget feature logic

Replace the UsePostRAScheduler SubtargetFeature with
DisablePostRAScheduler, which is then used by Swift and Cyclone.
This patch maintains enabling PostRA scheduling for other Thumb2
capable cores and/or for functions which are being compiled in Arm
mode.

Differential Revision: https://reviews.llvm.org/D37055

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312226 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64] IDSAR6 register assembler support
Sam Parker [Thu, 31 Aug 2017 08:36:45 +0000 (08:36 +0000)]
[AArch64] IDSAR6 register assembler support

The IDSAR6 system register has been introduced to identify the
v8.3-a Javascript data type conversion and v8.2-a dot product
support.

Differential Revision: https://reviews.llvm.org/D37068

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312225 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64] Support COFF linker directives
Martin Storsjo [Thu, 31 Aug 2017 08:28:48 +0000 (08:28 +0000)]
[AArch64] Support COFF linker directives

This is similar to what was done for ARM in SVN r269574; the code
and the test are straight copypaste to the corresponding AArch64
code and test directory.

Differential revision: https://reviews.llvm.org/D37204

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312223 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[IRCE] Identify loops with latch comparison against current IV value
Max Kazantsev [Thu, 31 Aug 2017 07:04:20 +0000 (07:04 +0000)]
[IRCE] Identify loops with latch comparison against current IV value

Current implementation of parseLoopStructure interprets the latch comparison as a
comarison against `iv.next`. If the actual comparison is made against the `iv` current value
then the loop may be rejected, because this misinterpretation leads to incorrect evaluation
of the latch start value.

This patch teaches the IRCE to distinguish this kind of loops and perform the optimization
for them. Now we use `IndVarBase` variable which can be either next or current value of the
induction variable (previously we used `IndVarNext` which was always the value on next iteration).

Differential Revision: https://reviews.llvm.org/D36215

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312221 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert r312194: "[MachineOutliner] Add missed optimization remarks for the outliner."
Daniel Jasper [Thu, 31 Aug 2017 06:22:35 +0000 (06:22 +0000)]
Revert r312194: "[MachineOutliner] Add missed optimization remarks for the outliner."

Breaks on buildbot:
http://bb.pgr.jp/builders/test-llvm-i686-linux-RA/builds/6026/steps/test_llvm/logs/LLVM%20%3A%3A%20CodeGen__AArch64__machine-outliner-remarks.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312219 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[IRCE][NFC] Rename IndVarNext to IndVarBase
Max Kazantsev [Thu, 31 Aug 2017 05:58:15 +0000 (05:58 +0000)]
[IRCE][NFC] Rename IndVarNext to IndVarBase

Renaming as a preparation step to generalizing IRCE for comparison not only against
the next value of an indvar, but also against the current.

Differential Revision: https://reviews.llvm.org/D36509

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312215 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoTemporarily revert "Update branch coalescing to be a PowerPC specific pass"
Eric Christopher [Thu, 31 Aug 2017 05:56:16 +0000 (05:56 +0000)]
Temporarily revert "Update branch coalescing to be a PowerPC specific pass"
From comments and code review it wasn't intended to be enabled by default yet.

This reverts commit r311588.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312214 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Don't assert in TTI with fp32 denorms enabled
Matt Arsenault [Thu, 31 Aug 2017 05:47:00 +0000 (05:47 +0000)]
AMDGPU: Don't assert in TTI with fp32 denorms enabled

Also refine for f16 and rcp cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312213 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "llvm-mt: Fix release of OutputDoc"
Vitaly Buka [Thu, 31 Aug 2017 05:32:36 +0000 (05:32 +0000)]
Revert "llvm-mt: Fix release of OutputDoc"

Multiple bots are broken.

This reverts commit r312207.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312212 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agollvm-mt: Fix release of OutputDoc
Vitaly Buka [Thu, 31 Aug 2017 02:38:58 +0000 (02:38 +0000)]
llvm-mt: Fix release of OutputDoc

Summary:
xmlDoc needs to be released with xmlFreeDoc.
Reset root element before release to avoid release of CombinedRoot owned
by CombinedDoc,

Reviewers: ecbeckmann, rnk, zturner, ruiu

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D37321

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312207 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Use set for tracked registers
Matt Arsenault [Thu, 31 Aug 2017 01:53:09 +0000 (01:53 +0000)]
AMDGPU: Use set for tracked registers

The majority of the time spent in the pass checking
for the register reads. Rather than searching all of
the defined registers for uses in each instruction,
use a set of defined registers and check the operands
of the instruction.

This process still is algorithmically not great,
but with the additional trick of skipping the analysis
for addresses with one use, this brings one slow
testcase into a reasonable range.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312206 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Orc] Add a comment about member variable dependencies to OrcMCJITReplacement.
Lang Hames [Thu, 31 Aug 2017 01:09:56 +0000 (01:09 +0000)]
[Orc] Add a comment about member variable dependencies to OrcMCJITReplacement.

The comment explains the reason behind the change in member variable order in
r312086.

Thanks to Philip Reames for the suggestion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312205 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[XRay][tools] Fix an accounting bug in llvm-xray account
Dean Michael Berris [Thu, 31 Aug 2017 01:07:24 +0000 (01:07 +0000)]
[XRay][tools] Fix an accounting bug in llvm-xray account

Summary:
Before this patch, llvm-xray account will assume that thread stacks will
not be empty. Unfortunately there are cases where an instrumented
function will see a call to `fork()` which will cause the child process
to not see the start of the function, but only see the end of the
function. The tooling cannot assume that threads will always have
perfect stacks, and so we change it to support this reality.

Reviewers: dblaikie

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D31870

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312204 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agocmake: Invent add_llvm_fuzzer to set up fuzzer targets
Justin Bogner [Thu, 31 Aug 2017 00:36:33 +0000 (00:36 +0000)]
cmake: Invent add_llvm_fuzzer to set up fuzzer targets

This moves the cmake configuration for fuzzers in LLVM to a new macro,
add_llvm_fuzzer. This will make it easier to keep things consistent
while implementing llvm.org/pr34314.

I've also made a couple of minor functional changes here:

- the fuzzers now use add_llvm_executable rather than add_llvm_tool.
  This means they won't create install targets and stuff like that,
  because those made little sense for these fuzzers.
- I've grouped these under "Fuzzers" rather than in with "Tools" for
  people who build with IDEs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312200 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "Revert r312139 "Verifier: Verify the correctness of fragment expressions...
Adrian Prantl [Thu, 31 Aug 2017 00:07:33 +0000 (00:07 +0000)]
Revert "Revert r312139 "Verifier: Verify the correctness of fragment expressions attached to globals.""

This reverts commit r312182 after fixing PR34390.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312197 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDon't add a fragment expression when GlobalSRA splits up a single-member struct
Adrian Prantl [Thu, 31 Aug 2017 00:06:18 +0000 (00:06 +0000)]
Don't add a fragment expression when GlobalSRA splits up a single-member struct

Fixes PR34390.

https://bugs.llvm.org/show_bug.cgi?id=34390

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312196 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agollvm-isel-fuzzer: Stop including FuzzerInterface.h
Justin Bogner [Thu, 31 Aug 2017 00:01:28 +0000 (00:01 +0000)]
llvm-isel-fuzzer: Stop including FuzzerInterface.h

All this does is forward declare the interface functions (and make
sure that they're `extern "C"`), but since we're using libFuzzer from
the toolchain it doesn't make sense to include the local copy of the
interface.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312195 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MachineOutliner] Add missed optimization remarks for the outliner.
Jessica Paquette [Wed, 30 Aug 2017 23:31:49 +0000 (23:31 +0000)]
[MachineOutliner] Add missed optimization remarks for the outliner.

This adds missed optimization remarks which report viable candidates that
were not outlined because they would increase code size.

Other remarks will come in separate commits.

This will help to diagnose code size regressions and changes in outliner
behaviour in projects using the outliner.

https://reviews.llvm.org/D37085

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312194 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[yaml2obj][ELF] Make symbols optional for relocations
Petr Hosek [Wed, 30 Aug 2017 23:13:31 +0000 (23:13 +0000)]
[yaml2obj][ELF] Make symbols optional for relocations

Some kinds of relocations do not have symbols, like R_X86_64_RELATIVE
for instance. I would like to test this case in D36554 but currently
can't because symbols are required by yaml2obj. The other option is
using the empty symbol but that doesn't seem quite right to me.

This change makes the Symbol field of Relocation optional and in the
case where the user does not specify a symbol name the Symbol index is 0.

Patch by Jake Ehrlich

Differential Revision: https://reviews.llvm.org/D37276

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312192 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Remove some code from fast isel that is no longer needed with i1 being an illeg...
Craig Topper [Wed, 30 Aug 2017 23:05:54 +0000 (23:05 +0000)]
[X86] Remove some code from fast isel that is no longer needed with i1 being an illegal type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312190 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SanitizeCoverage] Enable stack-depth coverage for -fsanitize=fuzzer
Matt Morehouse [Wed, 30 Aug 2017 22:49:31 +0000 (22:49 +0000)]
[SanitizeCoverage] Enable stack-depth coverage for -fsanitize=fuzzer

Summary:
- Don't sanitize __sancov_lowest_stack.
- Don't instrument leaf functions.
- Add CoverageStackDepth to Fuzzer and FuzzerNoLink.
- Only enable on Linux.

Reviewers: vitalybuka, kcc, george.karpenkov

Reviewed By: kcc

Subscribers: kubamracek, cfe-commits, llvm-commits, hiraditya

Differential Revision: https://reviews.llvm.org/D37156

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312185 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert r312139 "Verifier: Verify the correctness of fragment expressions attached...
Hans Wennborg [Wed, 30 Aug 2017 22:41:27 +0000 (22:41 +0000)]
Revert r312139 "Verifier: Verify the correctness of fragment expressions attached to globals."

This caused PR34390.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312182 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ARM] Replace fixed-size SmallSet with a bitset.
Benjamin Kramer [Wed, 30 Aug 2017 22:28:30 +0000 (22:28 +0000)]
[ARM] Replace fixed-size SmallSet with a bitset.

It's smaller. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312180 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Correct operand types for v_mad_mix*
Matt Arsenault [Wed, 30 Aug 2017 22:18:40 +0000 (22:18 +0000)]
AMDGPU: Correct operand types for v_mad_mix*

These aren't really packed instructions, so the default
op_sel_hi should be 0 since this indicates a conversion.
The operand types are scalar values that behave similar
to an f16 scalar that may be converted to f32.

Doesn't change the default printing for op_sel_hi, just
the parsing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312179 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert r312154 "Re-enable "[MachineCopyPropagation] Extend pass to do COPY source...
Hans Wennborg [Wed, 30 Aug 2017 22:11:37 +0000 (22:11 +0000)]
Revert r312154 "Re-enable "[MachineCopyPropagation] Extend pass to do COPY source forwarding""

It caused PR34387: Assertion failed: (RegNo < NumRegs && "Attempting to access record for invalid register number!")

> Issues identified by buildbots addressed since original review:
> - Fixed ARMLoadStoreOptimizer bug exposed by this change in r311907.
> - The pass no longer forwards COPYs to physical register uses, since
>   doing so can break code that implicitly relies on the physical
>   register number of the use.
> - The pass no longer forwards COPYs to undef uses, since doing so
>   can break the machine verifier by creating LiveRanges that don't
>   end on a use (since the undef operand is not considered a use).
>
>   [MachineCopyPropagation] Extend pass to do COPY source forwarding
>
>   This change extends MachineCopyPropagation to do COPY source forwarding.
>
>   This change also extends the MachineCopyPropagation pass to be able to
>   be run during register allocation, after physical registers have been
>   assigned, but before the virtual registers have been re-written, which
>   allows it to remove virtual register COPY LiveIntervals that become dead
>   through the forwarding of all of their uses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312178 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoSimplify writeArchive return type.
Rui Ueyama [Wed, 30 Aug 2017 22:11:03 +0000 (22:11 +0000)]
Simplify writeArchive return type.

writeArchive returned a pair, but the first element of the pair is always
its first argument on failure, so it doesn't make sense to return it from
the function. This patch change the return type so that it does't return it.

Differential Revision: https://reviews.llvm.org/D37313

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312177 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix test after rL312144
Konstantin Zhuravlyov [Wed, 30 Aug 2017 21:59:14 +0000 (21:59 +0000)]
Fix test after rL312144

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312176 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstCombine] add more vector demand examples; NFC
Sanjay Patel [Wed, 30 Aug 2017 21:11:46 +0000 (21:11 +0000)]
[InstCombine] add more vector demand examples; NFC

See D37236 for discussion. It seems unlikely that we actually want/need
to do this kind of folding in InstCombine in the long run, but moving
everything will be a bigger follow-up step.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312172 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoSelectionDAG: Emit correct debug info for multi-register function arguments.
Adrian Prantl [Wed, 30 Aug 2017 20:51:20 +0000 (20:51 +0000)]
SelectionDAG: Emit correct debug info for multi-register function arguments.

Previously we would just describe the first register and then call it
quits. This patch emits fragment expressions for each register.

<rdar://problem/34075307>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312169 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[IR] Don't print "!DIExpression() = !DIExpression()" when dumping
Reid Kleckner [Wed, 30 Aug 2017 20:40:36 +0000 (20:40 +0000)]
[IR] Don't print "!DIExpression() = !DIExpression()" when dumping

Now that we print DIExpressions inline everywhere, we don't need to
print them once as an operand and again as a value. This is only really
visible when calling dump() or print() directly on a DIExpression during
debugging.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312168 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRefactor DIBuilder::createFragmentExpression into a static DIExpression member
Adrian Prantl [Wed, 30 Aug 2017 20:04:17 +0000 (20:04 +0000)]
Refactor DIBuilder::createFragmentExpression into a static DIExpression member

NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312165 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ARM] Use Swift error registers on non-Darwin targets
Brian Gesiak [Wed, 30 Aug 2017 20:03:54 +0000 (20:03 +0000)]
[ARM] Use Swift error registers on non-Darwin targets

Summary:
Remove a check for `ARMSubtarget::isTargetDarwin` when determining
whether to use Swift error registers, so that Swift errors work
properly on non-Darwin ARM32 targets (specifically Android).

Before this patch, generated code would save and restores ARM register r8 at
the entry and returns of a function that throws. As r8 is used as a virtual
return value for the object being thrown, this gets overwritten by the restore,
and calling code is unable to catch the error. In turn this caused Swift code
that used `do`/`try`/`catch` to work improperly on Android ARM32 targets.

Addresses Swift bug report https://bugs.swift.org/browse/SR-5438.

Patch by John Holdsworth.

Reviewers: manmanren, rjmccall, aschwaighofer

Reviewed By: aschwaighofer

Subscribers: srhines, aschwaighofer, aemerson, javed.absar, kristof.beyls, llvm-commits

Differential Revision: https://reviews.llvm.org/D35835

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312164 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Update debug info test after r312144
Derek Schuff [Wed, 30 Aug 2017 19:54:08 +0000 (19:54 +0000)]
[WebAssembly] Update debug info test after r312144

Add the expr field to the DIGlobalVariableExpression

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312163 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoNewGVN: Make sure we add the correct user if we swapped the comparison operands
Daniel Berlin [Wed, 30 Aug 2017 19:53:23 +0000 (19:53 +0000)]
NewGVN: Make sure we add the correct user if we swapped the comparison operands

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312162 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoNewGVN: Allow simplification into variables
Daniel Berlin [Wed, 30 Aug 2017 19:52:39 +0000 (19:52 +0000)]
NewGVN: Allow simplification into variables

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312161 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[GISel]: Add a clean up combiner during legalization.
Aditya Nandakumar [Wed, 30 Aug 2017 19:32:59 +0000 (19:32 +0000)]
[GISel]: Add a clean up combiner during legalization.

Added a combiner which can clean up truncs/extends that are created in
order to make the types work during legalization.

Also moved the combineMerges to the LegalizeCombiner.

https://reviews.llvm.org/D36880

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312158 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[GVNSink] Remove dependency on SmallPtrSet iteration order.
Benjamin Kramer [Wed, 30 Aug 2017 18:46:37 +0000 (18:46 +0000)]
[GVNSink] Remove dependency on SmallPtrSet iteration order.

Found by LLVM_ENABLE_REVERSE_ITERATION.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312156 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRe-enable "[MachineCopyPropagation] Extend pass to do COPY source forwarding"
Geoff Berry [Wed, 30 Aug 2017 18:41:07 +0000 (18:41 +0000)]
Re-enable "[MachineCopyPropagation] Extend pass to do COPY source forwarding"

Issues identified by buildbots addressed since original review:
- Fixed ARMLoadStoreOptimizer bug exposed by this change in r311907.
- The pass no longer forwards COPYs to physical register uses, since
  doing so can break code that implicitly relies on the physical
  register number of the use.
- The pass no longer forwards COPYs to undef uses, since doing so
  can break the machine verifier by creating LiveRanges that don't
  end on a use (since the undef operand is not considered a use).

  [MachineCopyPropagation] Extend pass to do COPY source forwarding

  This change extends MachineCopyPropagation to do COPY source forwarding.

  This change also extends the MachineCopyPropagation pass to be able to
  be run during register allocation, after physical registers have been
  assigned, but before the virtual registers have been re-written, which
  allows it to remove virtual register COPY LiveIntervals that become dead
  through the forwarding of all of their uses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312154 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Remove unneed AVX512 check from fast isel.
Craig Topper [Wed, 30 Aug 2017 18:08:58 +0000 (18:08 +0000)]
[X86] Remove unneed AVX512 check from fast isel.

This is no longer necessary now that i1 is illegal.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312146 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Add target feature for atomics
Derek Schuff [Wed, 30 Aug 2017 18:07:45 +0000 (18:07 +0000)]
[WebAssembly] Add target feature for atomics

Summary:
This tracks the WebAssembly threads feature proposal at
https://github.com/WebAssembly/threads/blob/master/proposals/threads/Overview.md

Differential Revision: https://reviews.llvm.org/D37300

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312145 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoCanonicalize the representation of empty an expression in DIGlobalVariableExpression
Adrian Prantl [Wed, 30 Aug 2017 18:06:51 +0000 (18:06 +0000)]
Canonicalize the representation of empty an expression in DIGlobalVariableExpression

This change simplifies code that has to deal with
DIGlobalVariableExpression and mirrors how we treat DIExpressions in
debug info intrinsics. Before this change there were two ways of
representing empty expressions on globals, a nullptr and an empty
!DIExpression().

If someone needs to upgrade out-of-tree testcases:
  perl -pi -e 's/(!DIGlobalVariableExpression\(var: ![0-9]*)\)/\1, expr: !DIExpression())/g' <MYTEST.ll>
will catch 95%.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312144 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[codeview] make DbgVariableLocation::extractFromMachineInstruction use Optional
Bob Haarman [Wed, 30 Aug 2017 17:50:21 +0000 (17:50 +0000)]
[codeview] make DbgVariableLocation::extractFromMachineInstruction use Optional

Summary:
DbgVariableLocation::extractFromMachineInstruction originally
returned a boolean indicating success. This change makes it return
an Optional<DbgVariableLocation> so we cannot try to access the fields
of the struct if they aren't valid.

Reviewers: aprantl, rnk, zturner

Subscribers: llvm-commits, hiraditya

Differential Revision: https://reviews.llvm.org/D37279

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312143 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agodocs: remove dead doxygen tarball link
Hans Wennborg [Wed, 30 Aug 2017 17:40:28 +0000 (17:40 +0000)]
docs: remove dead doxygen tarball link

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312142 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoVerifier: Verify the correctness of fragment expressions attached to globals.
Adrian Prantl [Wed, 30 Aug 2017 16:49:21 +0000 (16:49 +0000)]
Verifier: Verify the correctness of fragment expressions attached to globals.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312139 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AVX512] Don't use 32-bit elements version of AND/OR/XOR/ANDN during isel unless...
Craig Topper [Wed, 30 Aug 2017 16:38:33 +0000 (16:38 +0000)]
[AVX512] Don't use 32-bit elements version of AND/OR/XOR/ANDN during isel unless we're matching a masked op or broadcast

Selecting 32-bit element logical ops without a select or broadcast requires matching a bitconvert on the inputs to the and. But that's a weird thing to rely on. It's entirely possible that one of the inputs doesn't have a bitcast and one does.

Since there's no functional difference, just remove the extra patterns and save some isel table size.

Differential Revision: https://reviews.llvm.org/D36854

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312138 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Remove GCCBuiltin for some intrinsics that aren't used by clang. Add TODO to...
Craig Topper [Wed, 30 Aug 2017 16:28:09 +0000 (16:28 +0000)]
[X86] Remove GCCBuiltin for some intrinsics that aren't used by clang. Add TODO to remove them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312136 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[GlobalISel][X86] Support variadic function call.
Igor Breger [Wed, 30 Aug 2017 15:10:15 +0000 (15:10 +0000)]
[GlobalISel][X86] Support variadic function call.

Summary: Support variadic function call. Port the implementation from X86FastISel.

Reviewers: zvi, guyblank, oren_ben_simhon

Reviewed By: guyblank

Subscribers: rovka, kristof.beyls, llvm-commits

Differential Revision: https://reviews.llvm.org/D37261

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312130 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRe-land MachineInstr: Reason locally about some memory objects before going to AA.
Balaram Makam [Wed, 30 Aug 2017 14:57:12 +0000 (14:57 +0000)]
Re-land MachineInstr: Reason locally about some memory objects before going to AA.

Summary:
Reverts r311008 to reinstate r310825 with a fix.

Refine alias checking for pseudo vs value to be conservative.
This fixes the original failure in builtbot unittest SingleSource/UnitTests/2003-07-09-SignedArgs.

Reviewers: hfinkel, nemanjai, efriedma

Reviewed By: efriedma

Subscribers: bjope, mcrosier, nhaehnle, javed.absar, llvm-commits

Differential Revision: https://reviews.llvm.org/D36900

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312126 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstCombine] remove unnecessary vector select fold; NFCI
Sanjay Patel [Wed, 30 Aug 2017 14:04:57 +0000 (14:04 +0000)]
[InstCombine] remove unnecessary vector select fold; NFCI

This code is double-dead:
1. We simplify all selects with constant true/false condition in InstSimplify.
   I've minimized/moved the tests to show that works as expected.
2. All remaining vector selects with a constant condition are canonicalized to
   shufflevector, so we really can't see this pattern.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312123 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agofix more typos; NFC
Sanjay Patel [Wed, 30 Aug 2017 13:19:23 +0000 (13:19 +0000)]
fix more typos; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312120 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agofix typos; NFC
Sanjay Patel [Wed, 30 Aug 2017 13:16:25 +0000 (13:16 +0000)]
fix typos; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312119 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MIPS] Add support to match more patterns for BBIT instruction
Strahinja Petrovic [Wed, 30 Aug 2017 11:25:38 +0000 (11:25 +0000)]
[MIPS] Add support to match more patterns for BBIT instruction

This patch supports one more pattern for bbit0 and bbit1
instructions, CBranchBitNum class is expanded  so it can
take 32 bit immidate.

Differential Revision: https://reviews.llvm.org/D36222

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312111 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstCombine] Fold insert sequence if first ins has multiple users.
Florian Hahn [Wed, 30 Aug 2017 10:54:21 +0000 (10:54 +0000)]
[InstCombine] Fold insert sequence if first ins has multiple users.

Summary:
If the first insertelement instruction has multiple users and inserts at
position 0, we can re-use this instruction when folding a chain of
insertelement instructions. As we need to generate the first
insertelement instruction anyways, this should be a strict improvement.

We could get rid of the restriction of inserting at position 0 by
creating a different shufflemask, but it is probably worth to keep the
first insertelement instruction with position 0, as this is easier to do
efficiently than at other positions I think.

Reviewers: grosser, mkuper, fpetrogalli, efriedma

Reviewed By: fpetrogalli

Subscribers: gareevroman, llvm-commits

Differential Revision: https://reviews.llvm.org/D37064

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312110 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64] allow v4f16 types when FullFP16 is supported
Sjoerd Meijer [Wed, 30 Aug 2017 08:38:13 +0000 (08:38 +0000)]
[AArch64] allow v4f16 types when FullFP16 is supported

Support for scalars was committed in r311154, this adds support for allowing
v4f16 vector types (thus avoiding conversions from/to single precision for
these types).

Differential Revision: https://reviews.llvm.org/D37145

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312104 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][Skylake] Fixing duplicated prefixes in the run command of Code Gen regression...
Gadi Haber [Wed, 30 Aug 2017 08:08:50 +0000 (08:08 +0000)]
[X86][Skylake] Fixing duplicated prefixes in the run command of Code Gen regression tests

NFC.
Replaced duplicated HASWELL prefixes in run commands in the X86 Code Gen regression tests by the SKYLAKE prefix when the -mcpu is set to skylake.
The fix is needed in preparation of an upcoming patch containing the Skylake scheduling info.

Reviewers: zvi, RKSimon, aymanmus, igorb

Differential Revision: https://reviews.llvm.org/D37258

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312103 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AVX512] Correct isel patterns to support selecting masked vbroadcastf32x2/vbroadcast...
Craig Topper [Wed, 30 Aug 2017 07:48:39 +0000 (07:48 +0000)]
[AVX512] Correct isel patterns to support selecting masked vbroadcastf32x2/vbroadcasti32x2

Summary:
This patch adjusts the patterns to make the result type of the broadcast node vXf64/vXi64. Then adds a bitcast to vXi32 after that. Intrinsic lowering was also adjusted to generate this new pattern.

Fixes PR34357

We should probably just drop the intrinsic entirely and use native IR, but I'll leave that for a future patch.

Any idea what instruction we should be lowering the floating point 128-bit result version of this pattern to?  There's a 128-bit v2i32 integer broadcast but not an fp one.

Reviewers: aymanmus, zvi, igorb

Reviewed By: aymanmus

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D37286

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312101 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AVX512] Use 256-bit extract instructions for extracting bits [255:128] from a 512...
Craig Topper [Wed, 30 Aug 2017 07:26:12 +0000 (07:26 +0000)]
[AVX512] Use 256-bit extract instructions for extracting bits [255:128] from a 512-bit register

This enables the use of a smaller encoding by using a VEX instruction when possible.

Differential Revision: https://reviews.llvm.org/D37092

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312100 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Apply SlowIncDec feature to Sandybridge/Ivybridge CPUs as well
Craig Topper [Wed, 30 Aug 2017 05:00:35 +0000 (05:00 +0000)]
[X86] Apply SlowIncDec feature to Sandybridge/Ivybridge CPUs as well

Currently we start applying this on Haswell and newer. I don't believe anything changed in the Haswell architecture to make this the right cutoff point. The partial flag handling around this has been roughly the same since Sandybridge.

Differential Revision: https://reviews.llvm.org/D37250

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312099 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[cfi] Fixed non-determinism in codegen due to DenseSet iteration order
Mandeep Singh Grang [Wed, 30 Aug 2017 04:47:21 +0000 (04:47 +0000)]
[cfi] Fixed non-determinism in codegen due to DenseSet iteration order

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312098 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Provide a separate feature bit for macro fusion support instead of basing it...
Craig Topper [Wed, 30 Aug 2017 04:34:48 +0000 (04:34 +0000)]
[X86] Provide a separate feature bit for macro fusion support instead of basing it on the AVX flag

Summary:
Currently we determine if macro fusion is supported based on the AVX flag as a proxy for the processor being Sandy Bridge".

This is really strange as now AMD supports AVX. It also means if user explicitly disables AVX we disable macro fusion.

This patch adds an explicit macro fusion feature. I've also enabled for the generic 64-bit CPU (which doesn't have AVX)

This is probably another candidate for being in the MI layer, but for now I at least wanted to correct the overloading of the AVX feature.

Reviewers: spatel, chandlerc, RKSimon, zvi

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D37280

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312097 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Don't look for DS merge candidates with one use address
Matt Arsenault [Wed, 30 Aug 2017 03:26:18 +0000 (03:26 +0000)]
AMDGPU: Don't look for DS merge candidates with one use address

The merge is only possible if the base address register is the
same for the two instructions. If there is only the one use,
there's no point in doing an expensive forward scan checking
for memory interference looking for a merge candidate.

This gives a signficant improvement in one extreme testcase.
The code to do the scan is still algorithmically terrible,
so this is still the slowest pass in that example.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312096 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU] Use v_max_f* for fcanonicalize
Stanislav Mekhanoshin [Wed, 30 Aug 2017 03:03:38 +0000 (03:03 +0000)]
[AMDGPU] Use v_max_f* for fcanonicalize

If denorms are not flushed we can use max instead of multiplication
by 1. For double that is simply faster, while for float and half
it is shorter, because mul uses constant bus and VOP3.

Differential Revision: https://reviews.llvm.org/D36856

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312095 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[TableGen] Fix a range based for loop to take the value by reference so that Simplify...
Craig Topper [Wed, 30 Aug 2017 02:05:03 +0000 (02:05 +0000)]
[TableGen] Fix a range based for loop to take the value by reference so that SimplifyTree can modify the copy in the array if its needs to.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312088 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Select clamp pattern with v2f16
Matt Arsenault [Wed, 30 Aug 2017 01:20:17 +0000 (01:20 +0000)]
AMDGPU: Select clamp pattern with v2f16

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312087 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Orc] Fix member variable ordering issue in OrcMCJITReplacement.
Lang Hames [Wed, 30 Aug 2017 00:47:42 +0000 (00:47 +0000)]
[Orc] Fix member variable ordering issue in OrcMCJITReplacement.

https://reviews.llvm.org/D36888

From that review description:

When an OrcMCJITReplacement object gets destructed, LazyEmitLayer may still
contain a shared_ptr of a module, which requires ShouldDelete in the deleter.
But ShouldDelete gets destructed before LazyEmitLayer due to the order of
declaration in OrcMCJITReplacement, which leads to a crash, when the destructor
of LazyEmitLayer is executed.  Changing the order of declaration fixes this.

Patch by Moritz Kroll. Thanks Moritz!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312086 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Error] Add an optional error message to cantFail.
Lang Hames [Tue, 29 Aug 2017 23:29:09 +0000 (23:29 +0000)]
[Error] Add an optional error message to cantFail.

cantFail is the moral equivalent of an assertion that the wrapped call must
return a success value. This patch allows clients to include an associated
error message (the same way they would for an assertion for llvm_unreachable).

If the error message is not specified it will default to: "Failure value
returned from cantFail wrapped call".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312066 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Finish the subtarget and predicate implementation of CLWB.
Craig Topper [Tue, 29 Aug 2017 23:13:36 +0000 (23:13 +0000)]
[X86] Finish the subtarget and predicate implementation of CLWB.

We don't have an intrinsic implemented for this instruction yet, but it looked odd that we were missing the accessor method from the subtarget.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312064 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[codeview] add missing break in CodeGen/AsmPrinter/DebugHandlerBase.cpp
Bob Haarman [Tue, 29 Aug 2017 22:54:31 +0000 (22:54 +0000)]
[codeview] add missing break in CodeGen/AsmPrinter/DebugHandlerBase.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312055 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[cfi] Avoid branch veneers in jump tables when possible.
Evgeniy Stepanov [Tue, 29 Aug 2017 22:40:19 +0000 (22:40 +0000)]
[cfi] Avoid branch veneers in jump tables when possible.

Summary:
When jumptable encoding does not match target code encoding (arm vs
thumb), a veneer is inserted by the linker. We can not avoid this
in all cases, because entries within one jumptable must have the same
encoding, but we can make it less common by selecting the jumptable
encoding to match the majority of its targets.

This change only covers FullLTO, and not ThinLTO.

Reviewers: pcc

Subscribers: aemerson, mehdi_amini, javed.absar, kristof.beyls, llvm-commits, hiraditya

Differential Revision: https://reviews.llvm.org/D37171

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312054 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[CodeGen] Fix some Clang-tidy modernize-use-using and Include What You Use warnings...
Eugene Zelenko [Tue, 29 Aug 2017 22:32:07 +0000 (22:32 +0000)]
[CodeGen] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312053 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[cfi] Build __cfi_check as Thumb when applicable.
Evgeniy Stepanov [Tue, 29 Aug 2017 22:29:15 +0000 (22:29 +0000)]
[cfi] Build __cfi_check as Thumb when applicable.

Summary:
Cross-DSO CFI needs all __cfi_check exports to use the same encoding
(ARM vs Thumb).

Reviewers: pcc

Subscribers: aemerson, srhines, kristof.beyls, hiraditya, llvm-commits

Differential Revision: https://reviews.llvm.org/D37243

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312052 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix the dwarfdump test so that it passes in its new location
Reid Kleckner [Tue, 29 Aug 2017 22:25:16 +0000 (22:25 +0000)]
Fix the dwarfdump test so that it passes in its new location

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312051 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoMove dwarfdump test to DebugInfo/X86 now that it looks for x86 register names
Reid Kleckner [Tue, 29 Aug 2017 22:16:11 +0000 (22:16 +0000)]
Move dwarfdump test to DebugInfo/X86 now that it looks for x86 register names

Otherwise this test will fail on bots (like the hexagon bot) that don't
enable the x86 backend.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312050 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "[SanitizeCoverage] Enable stack-depth coverage for -fsanitize=fuzzer"
Matt Morehouse [Tue, 29 Aug 2017 21:56:56 +0000 (21:56 +0000)]
Revert "[SanitizeCoverage] Enable stack-depth coverage for -fsanitize=fuzzer"

This reverts r312026 due to bot breakage.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312047 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LoopUnswitch] Fix a simple bug which disables loop unswitch for select statement
Wei Mi [Tue, 29 Aug 2017 21:45:11 +0000 (21:45 +0000)]
[LoopUnswitch] Fix a simple bug which disables loop unswitch for select statement

This is to fix PR34257. rL309059 takes an early return when FindLIVLoopCondition
fails to find a loop invariant condition. This is wrong and it will disable loop
unswitch for select. The patch fixes the bug.

Differential Revision: https://reviews.llvm.org/D36985

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312045 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix indentation of find_first_existing_vc_file
Reid Kleckner [Tue, 29 Aug 2017 21:44:46 +0000 (21:44 +0000)]
Fix indentation of find_first_existing_vc_file

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312044 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[cmake] Stop putting the revision info in LLVM_VERSION_STRING
Reid Kleckner [Tue, 29 Aug 2017 21:44:21 +0000 (21:44 +0000)]
[cmake] Stop putting the revision info in LLVM_VERSION_STRING

Summary:
This reduces the number of build actions after a no-op commit from
thousands to about six, which should be acceptable. If six actions is
still too many, developers can disable the LLVM_APPEND_VC_REV cmake
option.

llvm-config.h is a widely included header that should rarely change.
Before this patch, it would change after every re-configure. Very few
users of llvm-config.h need to know the precise version, and those that
do can migrate to incorporating LLVM_REVISION as provided by
llvm/Support/VCSRevision.h.

This should bring LLVM back to the behavior that it had before r306858
from June 30 2017. Most LLVM tools will now print a version string like
"6.0.0svn" instead of "6.0.0-git-c40c2a23de4".

Fixes PR34308

Reviewers: pcc, rafael, hans

Subscribers: mgorny, llvm-commits

Differential Revision: https://reviews.llvm.org/D37272

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312043 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[dwarfdump] Pretty print location expressions and location lists
Reid Kleckner [Tue, 29 Aug 2017 21:41:21 +0000 (21:41 +0000)]
[dwarfdump] Pretty print location expressions and location lists

Summary:
Based on Fred's patch here: https://reviews.llvm.org/D6771

I can't seem to commandeer the old review, so I'm creating a new one.

With that change the locations exrpessions are pretty printed inline in the
DIE tree. The output looks like this for debug_loc entries:

    DW_AT_location [DW_FORM_data4]        (0x00000000
       0x0000000000000001 - 0x000000000000000b: DW_OP_consts +3
       0x000000000000000b - 0x0000000000000012: DW_OP_consts +7
       0x0000000000000012 - 0x000000000000001b: DW_OP_reg0 RAX, DW_OP_piece 0x4
       0x000000000000001b - 0x0000000000000024: DW_OP_breg5 RDI+0)

And like this for debug_loc.dwo entries:
    DW_AT_location [DW_FORM_sec_offset]   (0x00000000
      Addr idx 2 (w/ length 190): DW_OP_consts +0, DW_OP_stack_value
      Addr idx 3 (w/ length 23): DW_OP_reg0 RAX, DW_OP_piece 0x4)

Simple locations without ranges are printed inline:

   DW_AT_location [DW_FORM_block1]       (DW_OP_reg4 RSI, DW_OP_piece 0x4, DW_OP_bit_piece 0x20 0x0)

The debug_loc(.dwo) dumping in changed accordingly to factor the code.

Reviewers: dblaikie, aprantl, friss

Subscribers: mgorny, javed.absar, hiraditya, llvm-commits, JDevlieghere

Differential Revision: https://reviews.llvm.org/D37123

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312042 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Fix typo
Matt Arsenault [Tue, 29 Aug 2017 21:25:51 +0000 (21:25 +0000)]
AMDGPU: Fix typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312040 91177308-0d34-0410-b5e6-96231b3b80d8