OSDN Git Service

android-x86/external-llvm.git
6 years ago[Debugify] Add debug intrinsics before terminating musttail calls
Vedant Kumar [Mon, 4 Jun 2018 03:33:01 +0000 (03:33 +0000)]
[Debugify] Add debug intrinsics before terminating musttail calls

After r333856, opt -debugify would just stop emitting debug value
intrinsics after encountering a musttail call. This wasn't sufficient to
avoid verifier failures.

Debug value intrinicss for all instructions preceding a musttail call
must also be emitted before the musttail call.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333866 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstCombine] Fix div handling
Serguei Katkov [Mon, 4 Jun 2018 02:52:36 +0000 (02:52 +0000)]
[InstCombine] Fix div handling

When we optimize select basing on fact that div by 0 is undef
we should not traverse the instruction which are not guaranteed to
transfer execution to next instruction. Guard intrinsic is an example.

Reviewers: spatel, craig.topper
Reviewed By: spatel
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D47576

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333864 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Debugify] Don't apply DI before the bitcode writer pass
Vedant Kumar [Mon, 4 Jun 2018 00:11:49 +0000 (00:11 +0000)]
[Debugify] Don't apply DI before the bitcode writer pass

Applying synthetic debug info before the bitcode writer pass has no
testing-related purpose. This commit prevents that from happening.

It also adds tests which check that IR produced with/without
-debugify-each enabled is identical after stripping. This makes it
possible to check that individual passes (or full pipelines) are
invariant to debug info.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333861 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[opt] Add a -strip-module-flags option
Vedant Kumar [Mon, 4 Jun 2018 00:11:48 +0000 (00:11 +0000)]
[opt] Add a -strip-module-flags option

The -strip-module-flags option strips llvm.module.flags metadata from a
module at the beginning of the opt pipeline.

This will be used to test whether the output of a pass is debug info
(DI) invariant.

E.g, after applying synthetic debug info to a test case, we'd like to
strip out all DI-related metadata and check that the final IR is
identical to a baseline file without any DI applied, to check that
optimizations aren't inhibited by debug info.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333860 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoReformat overflowing lines, NFC
Vedant Kumar [Mon, 4 Jun 2018 00:11:47 +0000 (00:11 +0000)]
Reformat overflowing lines, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333859 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Remove and autoupgrade masked avx512vnni intrinsics using the unmasked intrinsi...
Craig Topper [Sun, 3 Jun 2018 23:24:17 +0000 (23:24 +0000)]
[X86] Remove and autoupgrade masked avx512vnni intrinsics using the unmasked intrinsics and select instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333857 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Debugify] Skip dbg.value placement for EH pads, musttail
Vedant Kumar [Sun, 3 Jun 2018 22:50:22 +0000 (22:50 +0000)]
[Debugify] Skip dbg.value placement for EH pads, musttail

Placing meta-instructions into EH pads breaks certain IR invariants, as
does placing instructions after a musttail call.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333856 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRe-land: [MachO] Fixing ub in MachO BinaryFormat
Chris Bieneman [Sun, 3 Jun 2018 20:33:42 +0000 (20:33 +0000)]
Re-land: [MachO] Fixing ub in MachO BinaryFormat

This re-lands r333797 with a fix for big endian systems.

Original commit message:

This isn't encountered anywhere inside LLVM, so I wrote a test case to expose the issue and verify that it is fixed.

The basic problem is that the macho_load_command union contains all load comamnd structs. Load command structs in 32-bit macho files can be 32-bit aligned instead of 64-bit aligned.

There are some strange circumstances in which this can be exposed in a 64-bit macho if the load commands are invalid or if a 32-bit aligned load command is used. In the past we've worked around this type of problem with changes like r264232.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333854 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ORC] Add a constructor to create an IRMaterializationUnit from a module and
Lang Hames [Sun, 3 Jun 2018 19:22:48 +0000 (19:22 +0000)]
[ORC] Add a constructor to create an IRMaterializationUnit from a module and
pre-existing SymbolFlags and SymbolToDefinition maps.

This constructor is useful when delegating work from an existing
IRMaterialiaztionUnit to a new one, as it avoids the cost of re-computing these
maps.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333852 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][TBM] Use realistic BEXTR control bits
Simon Pilgrim [Sun, 3 Jun 2018 18:15:06 +0000 (18:15 +0000)]
[X86][TBM] Use realistic BEXTR control bits

Avoid constant values that are guaranteed to give zero

Found while investigating BEXTR optimizations for PR34042.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333849 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstCombine] improve sub with bool folds
Sanjay Patel [Sun, 3 Jun 2018 16:35:26 +0000 (16:35 +0000)]
[InstCombine] improve sub with bool folds

There's a patchwork of existing transforms trying to handle
these cases, but as seen in the changed test, we weren't
catching them all.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333845 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][AVX512] Cleanup intrinsics tests
Simon Pilgrim [Sun, 3 Jun 2018 14:56:04 +0000 (14:56 +0000)]
[X86][AVX512] Cleanup intrinsics tests

Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage.

Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333843 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][AVX512BW] Regenerate arithmetic tests using update_llc_test_checks.py script
Simon Pilgrim [Sun, 3 Jun 2018 14:31:30 +0000 (14:31 +0000)]
[X86][AVX512BW] Regenerate arithmetic tests using update_llc_test_checks.py script

Require manual stripping of existing CHECKs as update_llc_test_checks doesn't remove them if they're outside the function

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333842 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][BMI1] Test i32 intrinsics on 32/64 bits + branch off i64 tests
Simon Pilgrim [Sun, 3 Jun 2018 14:11:34 +0000 (14:11 +0000)]
[X86][BMI1] Test i32 intrinsics on 32/64 bits + branch off i64 tests

Further refactoring will wait until D47452 has landed.

Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333841 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][BMI] Remove CTTZ tests - this is fully covered in clz.ll
Simon Pilgrim [Sun, 3 Jun 2018 13:55:17 +0000 (13:55 +0000)]
[X86][BMI] Remove CTTZ tests - this is fully covered in clz.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333840 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][TBM] Branch off i32 intrinsics and test on 32/64 bits
Simon Pilgrim [Sun, 3 Jun 2018 13:38:52 +0000 (13:38 +0000)]
[X86][TBM] Branch off i32 intrinsics and test on 32/64 bits

Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333839 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRemove SETCCE use from Lanai's backend
Amaury Sechet [Sun, 3 Jun 2018 12:56:24 +0000 (12:56 +0000)]
Remove SETCCE use from Lanai's backend

Summary: This creates a small perf regression, but after talking with Jacques Pienaar, he was good with it to get things moving toward removng SETCCE.

Reviewers: jpienaar, bryant

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D47626

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333838 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ORC] Use JITEvaluatedSymbol for IndirectStubsManager findStub and findPointer.
Lang Hames [Sun, 3 Jun 2018 02:18:03 +0000 (02:18 +0000)]
[ORC] Use JITEvaluatedSymbol for IndirectStubsManager findStub and findPointer.

Existing implementations of these methods do not require lazy materialization,
and switching to JITEvaluatedSymbol allows us to remove error checking on the
client side.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333835 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][SSE] Cleanup AVX1 intrinsics tests
Simon Pilgrim [Sat, 2 Jun 2018 21:35:48 +0000 (21:35 +0000)]
[X86][SSE] Cleanup AVX1 intrinsics tests

Ensure we cover 32/64-bit targets for SSE/AVX/AVX512 cases as necessary, strip -mcpu usage.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333834 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][SSE] Cleanup SSE1 intrinsics tests
Simon Pilgrim [Sat, 2 Jun 2018 20:25:56 +0000 (20:25 +0000)]
[X86][SSE] Cleanup SSE1 intrinsics tests

Ensure we cover 32/64-bit targets for SSE/AVX/AVX512 cases as necessary

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333833 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][SSE] Cleanup SSE2 intrinsics tests
Simon Pilgrim [Sat, 2 Jun 2018 19:43:14 +0000 (19:43 +0000)]
[X86][SSE] Cleanup SSE2 intrinsics tests

Ensure we cover 32/64-bit targets for SSE/AVX/AVX512 cases as necessary

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333832 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][SSE] Cleanup SSE3/SSSE3 intrinsics tests
Simon Pilgrim [Sat, 2 Jun 2018 18:41:46 +0000 (18:41 +0000)]
[X86][SSE] Cleanup SSE3/SSSE3 intrinsics tests

Ensure we cover 32/64-bit targets for SSE/AVX/AVX512 cases as necessary

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333831 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][SSE4] Tweak rL333828 sse41/sse42 cleanup to recover SKX/EVEX2VEX testing
Simon Pilgrim [Sat, 2 Jun 2018 18:01:09 +0000 (18:01 +0000)]
[X86][SSE4] Tweak rL333828 sse41/sse42 cleanup to recover SKX/EVEX2VEX testing

Just testing for avx512f was missing the tests for EVEX TO VEX Compression encoding etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333830 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][SSE] Cleanup SSE4A/SSE41/SSE42 intrinsics tests
Simon Pilgrim [Sat, 2 Jun 2018 17:33:26 +0000 (17:33 +0000)]
[X86][SSE] Cleanup SSE4A/SSE41/SSE42 intrinsics tests

Ensure we cover 32/64-bit targets for SSE/AVX/AVX512 cases as necessary

Added some missing encoding checks to SSE4A tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333828 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][BMI2] Test i32 intrinsics on 32/64 bits + branch off i64 tests
Simon Pilgrim [Sat, 2 Jun 2018 17:22:13 +0000 (17:22 +0000)]
[X86][BMI2] Test i32 intrinsics on 32/64 bits + branch off i64 tests

I had to tweak the i32 tests so we check both reg-reg and reg-mem cases.

I also added i64 load tests.

Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333827 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][BMI1] Remove test for non-existent andn i16 instruction
Simon Pilgrim [Sat, 2 Jun 2018 17:02:27 +0000 (17:02 +0000)]
[X86][BMI1] Remove test for non-existent andn i16 instruction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333826 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[NEON] Support VLD1xN intrinsics in AArch32 mode (LLVM part)
Ivan A. Kosarev [Sat, 2 Jun 2018 16:40:03 +0000 (16:40 +0000)]
[NEON] Support VLD1xN intrinsics in AArch32 mode (LLVM part)

We currently support them only in AArch64. The NEON Reference,
however, says they are 'ARMv7, ARMv8' intrinsics.

Differential Revision: https://reviews.llvm.org/D47120

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333825 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert r333819 "[NEON] Support VLD1xN intrinsics in AArch32 mode (Clang part)"
Ivan A. Kosarev [Sat, 2 Jun 2018 16:38:38 +0000 (16:38 +0000)]
Revert r333819 "[NEON] Support VLD1xN intrinsics in AArch32 mode (Clang part)"

The LLVM part was committed instead of the Clang part.

Differential Revision: https://reviews.llvm.org/D47121

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333824 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MC] Add assembler support for .cg_profile.
Michael J. Spencer [Sat, 2 Jun 2018 16:33:01 +0000 (16:33 +0000)]
[MC]  Add assembler support for .cg_profile.

Object FIle Representation
At codegen time this is emitted into the ELF file a pair of symbol indices and a weight. In assembly it looks like:

.cg_profile a, b, 32
.cg_profile freq, a, 11
.cg_profile freq, b, 20

When writing an ELF file these are put into a SHT_LLVM_CALL_GRAPH_PROFILE (0x6fff4c02) section as (uint32_t, uint32_t, uint64_t) tuples as (from symbol index, to symbol index, weight).

Differential Revision: https://reviews.llvm.org/D44965

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333823 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add tied source operand to AVX5124FMAPS and AVX5124VNNIW instructions.
Craig Topper [Sat, 2 Jun 2018 16:30:39 +0000 (16:30 +0000)]
[X86] Add tied source operand to AVX5124FMAPS and AVX5124VNNIW instructions.

This doesn't affect the assembly or disassembly, but is more accurate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333822 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Fix warning message for AVX5124FMAPS and AVX5124VNNIW instructions in the assem...
Craig Topper [Sat, 2 Jun 2018 16:30:36 +0000 (16:30 +0000)]
[X86] Fix warning message for AVX5124FMAPS and AVX5124VNNIW instructions in the assembly parser.

The caret was positioned on the wrong operand. It's too hard to get right so just put the caret at the beginning of the instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333821 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstCombine] call simplify before trying vector folds
Sanjay Patel [Sat, 2 Jun 2018 16:27:44 +0000 (16:27 +0000)]
[InstCombine] call simplify before trying vector folds

As noted in the review thread for rL333782, we could have
made a bug harder to hit if we were simplifying instructions
before trying other folds.

The shuffle transform in question isn't ever a simplification;
it's just a canonicalization. So I've renamed that to make that
clearer.

This is NFCI at this point, but I've regenerated the test file
to show the cosmetic value naming difference of using
instcombine's RAUW vs. the builder.

Possible follow-ups:
1. Move reassociation folds after simplifies too.
2. Refactor common code; we shouldn't have so much repetition.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333820 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[NEON] Support VLD1xN intrinsics in AArch32 mode (Clang part)
Ivan A. Kosarev [Sat, 2 Jun 2018 16:26:42 +0000 (16:26 +0000)]
[NEON] Support VLD1xN intrinsics in AArch32 mode (Clang part)

We currently support them only in AArch64. The NEON Reference,
however, says they are 'ARMv7, ARMv8' intrinsics.

Differential Revision: https://reviews.llvm.org/D47121

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333819 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstCombine] add more tests for shuffle-binop; NFC
Sanjay Patel [Sat, 2 Jun 2018 16:16:42 +0000 (16:16 +0000)]
[InstCombine] add more tests for shuffle-binop; NFC

As noted in the review thread for rL333782, we're lacking coverage
for this transform, so add tests for each binop opcode with constant
operand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333818 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Support] Remove unused raw_ostream::handle whose anchor role was superseded by anchor()
Fangrui Song [Sat, 2 Jun 2018 06:00:35 +0000 (06:00 +0000)]
[Support] Remove unused raw_ostream::handle whose anchor role was superseded by anchor()

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333817 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add encoding information for the AVX5124FMAPS and AVX5124VNNIW instructions...
Craig Topper [Sat, 2 Jun 2018 02:15:10 +0000 (02:15 +0000)]
[X86] Add encoding information for the AVX5124FMAPS and AVX5124VNNIW instructions so they can be assembled and disassembled.

These instructions are unusual in that they operate on 4 consecutive registers so supporting them in codegen will be more difficult than normal.

Includes an assembler check to warn if the source register is not the first register of a 4 register group.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333812 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PM/LoopUnswitch] Fix how the cloned loops are handled when updating analyses.
Chandler Carruth [Sat, 2 Jun 2018 01:29:01 +0000 (01:29 +0000)]
[PM/LoopUnswitch] Fix how the cloned loops are handled when updating analyses.

Summary:
I noticed this issue because we didn't put the primary cloned loop into
the `NonChildClonedLoops` vector and so never iterated on it. Once
I fixed that, it made it clear why I had to do a really complicated and
unnecesasry dance when updating the loops to remain in canonical form --
I was unwittingly working around the fact that the primary cloned loop
wasn't in the expected list of cloned loops. Doh!

Now that we include it in this vector, we don't need to return it and we
can consolidate the update logic as we correctly have a single place
where it can be handled.

I've just added a test for the iteration order aspect as every time
I changed the update logic partially or incorrectly here, an existing
test failed and caught it so that seems well covered (which is also
evidenced by the extensive working around of this missing update).

Reviewers: asbirlea, sanjoy

Subscribers: mcrosier, hiraditya, llvm-commits

Differential Revision: https://reviews.llvm.org/D47647

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333811 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRemove the test from r333801
Vedant Kumar [Sat, 2 Jun 2018 00:05:17 +0000 (00:05 +0000)]
Remove the test from r333801

In r333801 I added a test for a dump method that, for reasons I don't
understand, fails on an msvc bot:

http://lab.llvm.org:8011/builders/clang-x86-windows-msvc2015/builds/12306/

I'll remove the test for now to unblock the bot and try to look into why
there's a discrepancy on this platform later.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333807 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DebugInfo] Refactoring DIType::setFlags to DIType::cloneWithFlags, NFC
Roman Tereshin [Fri, 1 Jun 2018 23:15:09 +0000 (23:15 +0000)]
[DebugInfo] Refactoring DIType::setFlags to DIType::cloneWithFlags, NFC

and using the latter in DIBuilder::createArtificialType and
DIBuilder::createObjectPointerType methods as well as introducing
mirroring DISubprogram::cloneWithFlags and
DIBuilder::createArtificialSubprogram methods.

The primary goal here is to add createArtificialSubprogram to support
a pass downstream while keeping the method consistent with the
existing ones and making sure we don't encourage changing already
created DI-nodes.

Reviewed By: aprantl

Differential Revision: https://reviews.llvm.org/D47615

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333806 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "Re-land: [MachO] Fixing ub in MachO BinaryFormat"
Chris Bieneman [Fri, 1 Jun 2018 23:09:37 +0000 (23:09 +0000)]
Revert "Re-land: [MachO] Fixing ub in MachO BinaryFormat"

This reverts commit r333803.

Still breaking on big endian. Will sort this out later.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333805 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Do something sensible when an expand load intrinsic is passed a 0 mask.
Craig Topper [Fri, 1 Jun 2018 22:59:07 +0000 (22:59 +0000)]
[X86] Do something sensible when an expand load intrinsic is passed a 0 mask.

Previously we just returned undef, but really we should be returning the pass thru input. We also need to make sure we preserve the chain output that the original intrinsic node had to maintain connectivity in the DAG. So we should just return the incoming chain as the output chain.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333804 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRe-land: [MachO] Fixing ub in MachO BinaryFormat
Chris Bieneman [Fri, 1 Jun 2018 22:52:59 +0000 (22:52 +0000)]
Re-land: [MachO] Fixing ub in MachO BinaryFormat

This re-lands r333797 with a fix for big endian systems.

Original commit message:

This isn't encountered anywhere inside LLVM, so I wrote a test case to expose the issue and verify that it is fixed.

The basic problem is that the macho_load_command union contains all load comamnd structs. Load command structs in 32-bit macho files can be 32-bit aligned instead of 64-bit aligned.

There are some strange circumstances in which this can be exposed in a 64-bit macho if the load commands are invalid or if a 32-bit aligned load command is used. In the past we've worked around this type of problem with changes like r264232.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333803 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAdd a debug dump for DbgValueHistoryMap
Vedant Kumar [Fri, 1 Jun 2018 22:33:15 +0000 (22:33 +0000)]
Add a debug dump for DbgValueHistoryMap

This makes it easier to inspect the results of
DbgValueHistoryCalculator.

Differential Revision: https://reviews.llvm.org/D47663

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333801 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add isel patterns to use vexpand with zero masking when the passthru value...
Craig Topper [Fri, 1 Jun 2018 22:28:28 +0000 (22:28 +0000)]
[X86] Add isel patterns to use vexpand with zero masking when the passthru value is a zero vector.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333800 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "[MachO] Fixing ub in MachO BinaryFormat"
Chris Bieneman [Fri, 1 Jun 2018 22:28:23 +0000 (22:28 +0000)]
Revert "[MachO] Fixing ub in MachO BinaryFormat"

This reverts commit r333797.

This patch is failing on BigEndian bots. I will fix and re-land:

http://lab.llvm.org:8011/builders/clang-ppc64be-linux/builds/19505/

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333799 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoMove some function declarations out of WindowsSupport.h
Zachary Turner [Fri, 1 Jun 2018 22:23:46 +0000 (22:23 +0000)]
Move some function declarations out of WindowsSupport.h

The idea behind WindowsSupport.h is that it's in the source directory so
that windows.h'isms don't leak out into the larger LLVM project. To that
end, any symbol that references a symbol from windows.h must be in this
private header, and not in a public header.

However, we had some useful utility functions in WindowsSupport.h which
have no dependency on the Windows API, but still only make sense on
Windows. Those functions should be usable outside of Support since there
is no risk of causing a windows.h leak. Although this introduces some
preprocessor logic in some header files, It's not too egregious and it's
better than the alternative of duplicating a ton of code.

Differential Revision: https://reviews.llvm.org/D47662

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333798 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MachO] Fixing ub in MachO BinaryFormat
Chris Bieneman [Fri, 1 Jun 2018 22:07:36 +0000 (22:07 +0000)]
[MachO] Fixing ub in MachO BinaryFormat

This isn't encountered anywhere inside LLVM, so I wrote a test case to expose the issue and verify that it is fixed.

The basic problem is that the macho_load_command union contains all load comamnd structs. Load command structs in 32-bit macho files can be 32-bit aligned instead of 64-bit aligned.

There are some strange circumstances in which this can be exposed in a 64-bit macho if the load commands are invalid or if a 32-bit aligned load command is used. In the past we've worked around this type of problem with changes like r264232.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333797 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Expand the testing of expand and compress intrinsics
Craig Topper [Fri, 1 Jun 2018 21:59:24 +0000 (21:59 +0000)]
[X86] Expand the testing of expand and compress intrinsics

The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking.

This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333795 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add fast-isel tests for avx512vbmi2 instructions.
Craig Topper [Fri, 1 Jun 2018 21:59:22 +0000 (21:59 +0000)]
[X86] Add fast-isel tests for avx512vbmi2 instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333794 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ConstantFold] Disallow folding vector geps into bitcasts
Karl-Johan Karlsson [Fri, 1 Jun 2018 19:34:35 +0000 (19:34 +0000)]
[ConstantFold] Disallow folding vector geps into bitcasts

Summary:
Getelementptr returns a vector of pointers, instead of a single address,
when one or more of its arguments is a vector. In such case it is not
possible to simplify the expression by inserting a bitcast of operand(0)
into the destination type, as it will create a bitcast between different
sizes.

Reviewers: majnemer, mkuper, mssimpso, spatel

Reviewed By: spatel

Subscribers: lebedev.ri, llvm-commits

Differential Revision: https://reviews.llvm.org/D46379

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333783 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstCombine] fix vector shuffle transform to replace undef elements (PR37648)
Sanjay Patel [Fri, 1 Jun 2018 19:23:18 +0000 (19:23 +0000)]
[InstCombine] fix vector shuffle transform to replace undef elements (PR37648)

This bug:
https://bugs.llvm.org/show_bug.cgi?id=37648
...was created with the enhancement to this transform with rL332479.

The urem test shows the disaster potential: any undef divisor lane makes
the whole op undef.

The test diffs show that vector demanded elements turns some of the potential,
but not all, unused binop operands back into undef already.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333782 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstCombine] add tests for broken shuffle transform (PR37648)
Sanjay Patel [Fri, 1 Jun 2018 18:52:38 +0000 (18:52 +0000)]
[InstCombine] add tests for broken shuffle transform (PR37648)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333779 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips] Support 64-bit offsets for lb/sb/ld/sd/lld ... instructions
Simon Atanasyan [Fri, 1 Jun 2018 16:37:53 +0000 (16:37 +0000)]
[mips] Support 64-bit offsets for lb/sb/ld/sd/lld ... instructions

The `MipsAsmParser::loadImmediate` can load immediates of various sizes
into a register. Idea of this change is to use `loadImmediate` in the
`MipsAsmParser::expandMemInst` method to load offset into a register and
then call required load/store instruction.

The patch removes separate `expandLoadInst` and `expandStoreInst`
methods and does everything in the `expandMemInst` method to escape code
duplication.

Differential Revision: https://reviews.llvm.org/D47316

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333774 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips] Extend list of relocations supported by the `.reloc` directive
Simon Atanasyan [Fri, 1 Jun 2018 16:37:42 +0000 (16:37 +0000)]
[mips] Extend list of relocations supported by the `.reloc` directive

Supporting GOT and TLS related relocations by the `.reloc` directive is
useful for purpose of testing various tools like a linker, for example.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333773 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-objcopy] Fix null symbol handling
Paul Semel [Fri, 1 Jun 2018 16:19:46 +0000 (16:19 +0000)]
[llvm-objcopy] Fix null symbol handling

This fixes the bug where strip-all option was
leading to a malformed outputted ELF file.

Differential Revision: https://reviews.llvm.org/D47414

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333772 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Avoid UB when shifting unsigned integer left by 32
Krzysztof Parzyszek [Fri, 1 Jun 2018 15:39:10 +0000 (15:39 +0000)]
[Hexagon] Avoid UB when shifting unsigned integer left by 32

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333771 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LangRef] fix typo; NFC
Sanjay Patel [Fri, 1 Jun 2018 15:21:14 +0000 (15:21 +0000)]
[LangRef] fix typo; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333770 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ThinLTOBitcodeWriter] Emit summaries for regular LTO modules
Vlad Tsyrklevich [Fri, 1 Jun 2018 15:20:47 +0000 (15:20 +0000)]
[ThinLTOBitcodeWriter] Emit summaries for regular LTO modules

Summary:
Emit summaries for bitcode modules that are only destined for the
regular LTO portion of the build so they can participate in
summary-based dead stripping.

This change reduces the size of a nacl_helper build with cfi-icall
enabled by 7%, removing the majority of the overhead due to enabling
cfi-icall. The cfi-icall size increase was caused by compiling in lots
of unused code and cfi-icall generating jumptable references to unused
symbols that could no longer be removed by -Wl,-gc-sections. Increasing
the visibility of summary-based dead stripping prevented jumptable
entries being created for unused symbols from the regular LTO portion
of the build.

Reviewers: pcc

Reviewed By: pcc

Subscribers: dschuff, mehdi_amini, inglorion, eraman, llvm-commits, kcc

Differential Revision: https://reviews.llvm.org/D47594

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333768 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ConstantFold] Add lit testcase for bitcast problem. NFC
Karl-Johan Karlsson [Fri, 1 Jun 2018 15:08:14 +0000 (15:08 +0000)]
[ConstantFold] Add lit testcase for bitcast problem. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333767 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DAG] Avoid checking for consecutive stores in store merge. NFCI.
Nirav Dave [Fri, 1 Jun 2018 15:05:55 +0000 (15:05 +0000)]
[DAG] Avoid checking for consecutive stores in store merge. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333766 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DAG] Simplify Expression. NFC.
Nirav Dave [Fri, 1 Jun 2018 15:05:30 +0000 (15:05 +0000)]
[DAG] Simplify Expression. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333765 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DAG] Remove untriggerable check. NFCI.
Nirav Dave [Fri, 1 Jun 2018 15:05:05 +0000 (15:05 +0000)]
[DAG] Remove untriggerable check. NFCI.

Candidate check precludes this check.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333764 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DAG] Prune store merge legal store check to stop invalid size. NFCI.
Nirav Dave [Fri, 1 Jun 2018 15:04:40 +0000 (15:04 +0000)]
[DAG] Prune store merge legal store check to stop invalid size. NFCI.

Do not consider store sizes large than the maximum legal store size.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333763 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Select HVX code for vector CTPOP, CTLZ, and CTTZ
Krzysztof Parzyszek [Fri, 1 Jun 2018 14:52:58 +0000 (14:52 +0000)]
[Hexagon] Select HVX code for vector CTPOP, CTLZ, and CTTZ

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333760 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-exegesis] Fix off-by-one in llvm-exegesis documentation.
Clement Courbet [Fri, 1 Jun 2018 14:49:06 +0000 (14:49 +0000)]
[llvm-exegesis] Fix off-by-one in llvm-exegesis documentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333759 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstCombine] add baseline test for bug with div+select transform (D47576)
Sanjay Patel [Fri, 1 Jun 2018 14:39:05 +0000 (14:39 +0000)]
[InstCombine] add baseline test for bug with div+select transform (D47576)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333756 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-mca] Move the logic that computes the block throughput into Support.h. NFC
Andrea Di Biagio [Fri, 1 Jun 2018 14:35:21 +0000 (14:35 +0000)]
[llvm-mca] Move the logic that computes the block throughput into Support.h. NFC

This will allow us to share the logic that computes the block throughput with
other views.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333755 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[NFC] Zero initialize local variables
Hiroshi Inoue [Fri, 1 Jun 2018 14:23:15 +0000 (14:23 +0000)]
[NFC] Zero initialize local variables

This patch makes local variables zero initialized to avoid broken values in debug output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333754 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-exegesis] Analysis: Display idealized sched class port pressure.
Clement Courbet [Fri, 1 Jun 2018 14:18:02 +0000 (14:18 +0000)]
[llvm-exegesis] Analysis: Display idealized sched class port pressure.

Summary: Screenshot in phabricator diff.

Reviewers: gchatelet

Subscribers: mgorny, tschuett, mgrang, llvm-commits

Differential Revision: https://reviews.llvm.org/D47329

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333753 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SelectionDAG] Expand UADDO/USUBO into ADD/SUBCARRY if legal for target
Krzysztof Parzyszek [Fri, 1 Jun 2018 14:00:32 +0000 (14:00 +0000)]
[SelectionDAG] Expand UADDO/USUBO into ADD/SUBCARRY if legal for target

Additionally, implement handling of ADD/SUBCARRY on Hexagon, utilizing
the UADDO/USUBO expansion.

Differential Revision: https://reviews.llvm.org/D47559

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333751 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[x86] NFC. Reautogenerate test/CodeGen/X86/vector-half-conversions.ll
Alexander Ivchenko [Fri, 1 Jun 2018 13:51:53 +0000 (13:51 +0000)]
[x86] NFC. Reautogenerate test/CodeGen/X86/vector-half-conversions.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333750 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Utils][X86] Help update_llc_test_checks.py to recognise retl/retq to reduce CHECK...
Simon Pilgrim [Fri, 1 Jun 2018 13:37:01 +0000 (13:37 +0000)]
[Utils][X86] Help update_llc_test_checks.py to recognise retl/retq to reduce CHECK duplication (PR35003)

This patch replaces the --x86_extra_scrub command line argument to automatically support a second level of regex-scrubbing if it improves the matching of nearly-identical code patterns. The argument '--extra_scrub' is there now to force extra matching if required.

This is mostly useful to help us share 32-bit/64-bit x86 vector tests which only differs by retl/retq instructions, but any scrubber can now technically support this, meaning test checks don't have to be needlessly obfuscated.

I've updated some of the existing checks that had been manually run with --x86_extra_scrub, to demonstrate the extra "ret{{[l|q]}}" scrub now only happens when useful, and re-run the sse42-intrinsics file to show extra matches - most sse/avx intrinsics files should be able to now share 32/64 checks.

Tested with the opt/analysis scripts as well which share common code - AFAICT the other update scripts use their own versions.

Differential Revision: https://reviews.llvm.org/D47485

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333749 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoSet ADDE/ADDC/SUBE/SUBC to expand by default
Amaury Sechet [Fri, 1 Jun 2018 13:21:33 +0000 (13:21 +0000)]
Set ADDE/ADDC/SUBE/SUBC to expand by default

Summary:
They've been deprecated in favor of UADDO/ADDCARRY or USUBO/SUBCARRY for a while.

Target that uses these opcodes are changed in order to ensure their behavior doesn't change.

Reviewers: efriedma, craig.topper, dblaikie, bkramer

Subscribers: jholewinski, arsenm, jyknight, sdardis, nemanjai, nhaehnle, kbarton, fedor.sergeev, asb, rbar, johnrusso, simoncook, jordy.potman.lists, apazos, sabuasal, niosHD, jrtc27, zzheng, edward-jones, mgrang, atanasyan, llvm-commits

Differential Revision: https://reviews.llvm.org/D47422

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333748 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64][GlobalISel] Zero-extend s1 values when returning.
Amara Emerson [Fri, 1 Jun 2018 13:20:32 +0000 (13:20 +0000)]
[AArch64][GlobalISel] Zero-extend s1 values when returning.

Before we were relying on the any extend of the s1 to s32, but
for AAPCS we need to zero-extend it to at least s8.

Fixes PR36719

Differential Revision: https://reviews.llvm.org/D47425

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333747 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert r333740: IPSCCP] Use PredicateInfo to propagate facts from cmp.
Florian Hahn [Fri, 1 Jun 2018 12:58:43 +0000 (12:58 +0000)]
Revert r333740: IPSCCP] Use PredicateInfo to propagate facts from cmp.

This is breaking the clang-with-thin-lto-ubuntu bot.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333745 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64][SVE] Asm: Support for FDUP_ZI (copy fp immediate) instruction.
Sander de Smalen [Fri, 1 Jun 2018 12:54:46 +0000 (12:54 +0000)]
[AArch64][SVE] Asm: Support for FDUP_ZI (copy fp immediate) instruction.

Unpredicated copy of floating-point immediate value into SVE vector,
along with MOV-aliases.

Reviewers: rengolin, fhahn, samparker, SjoerdMeijer, javed.absar

Reviewed By: fhahn

Differential Revision: https://reviews.llvm.org/D47482

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333744 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips] Guard more aliases correctly.
Simon Dardis [Fri, 1 Jun 2018 10:57:13 +0000 (10:57 +0000)]
[mips] Guard more aliases correctly.

Also, duplicate an alias for microMIPS.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333741 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRecommit r333268: [IPSCCP] Use PredicateInfo to propagate facts from cmp instructions.
Florian Hahn [Fri, 1 Jun 2018 10:48:54 +0000 (10:48 +0000)]
Recommit r333268: [IPSCCP] Use PredicateInfo to propagate facts from cmp instructions.

This patch updates IPSCCP to use PredicateInfo to propagate
facts to true branches predicated by EQ and to false branches
predicated by NE.

As a follow up, we should be able to extend it to also propagate additional
facts about nonnull.

Reviewers: davide, mssimpso, dberlin, efriedma

Reviewed By: davide, dberlin

Differential Revision: https://reviews.llvm.org/D45330

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333740 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips] Guard 'nop' properly and add mips16's nop instruction
Simon Dardis [Fri, 1 Jun 2018 10:46:00 +0000 (10:46 +0000)]
[mips] Guard 'nop' properly and add mips16's nop instruction

Reviewers: smaksimovic, atanasyan, abeserminji

Differential Revision: https://reviews.llvm.org/D47583

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333739 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDWARFAcceleratorTable: Add an iterator-based api for accessing names in the index
Pavel Labath [Fri, 1 Jun 2018 10:33:11 +0000 (10:33 +0000)]
DWARFAcceleratorTable: Add an iterator-based api for accessing names in the index

Summary:
Back when we were introducing the DWARF v5 name index, there was a
short discussion whether we shouldn't have a nicer api for iterating
over the index. At that time, I did not find it necessary since the
iteration over names was done only from within the index itself (and I
figured the internal implementation can deal with a slightly rough
interface).

However, now I ran into a use for this kind of API in LLDB (for finding
all names matching a regular expression), so it looked like a nice
opportunity to introduce one. To make the API more useful, I've made the
NameTableEntry class a bit smarter: it now stores the string section
reference (so it can return its name) and its position in the name index
(mainly useful for dumping/logging).

I also convert the internal users to use the new API, which also gives
test coverage for the added code.

Reviewers: JDevlieghere, aprantl, dblaikie

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D47590

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333738 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips] Select the correct instruction for computing frameindexes
Simon Dardis [Fri, 1 Jun 2018 10:07:10 +0000 (10:07 +0000)]
[mips] Select the correct instruction for computing frameindexes

Reviewers: smaksimovic, atanasyan, abeserminji

Differential Revision: https://reviews.llvm.org/D47582

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333736 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoNFC Avoid a warning in WasmEHPrepare.cpp
Gabor Buella [Fri, 1 Jun 2018 07:47:46 +0000 (07:47 +0000)]
NFC Avoid a warning in WasmEHPrepare.cpp

```
../lib/CodeGen/WasmEHPrepare.cpp:166:30: warning: extra ‘;’ [-Wpedantic]
                 false, false);
                              ^
```

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333732 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64][SVE] Asm: Support for DUPM (masked immediate) instruction.
Sander de Smalen [Fri, 1 Jun 2018 07:25:46 +0000 (07:25 +0000)]
[AArch64][SVE] Asm: Support for DUPM (masked immediate) instruction.

Unpredicated copy of repeating immediate pattern to SVE vector, along
with MOV-aliases.

Reviewers: rengolin, fhahn, samparker, SjoerdMeijer, javed.absar

Reviewed By: SjoerdMeijer

Differential Revision: https://reviews.llvm.org/D47328

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333731 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Switch some half using-tests to use amdhsa
Matt Arsenault [Fri, 1 Jun 2018 07:06:03 +0000 (07:06 +0000)]
AMDGPU: Switch some half using-tests to use amdhsa

The default clover ABI weirdly promotes half to float,
which should probably be fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333730 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][Disassembler] Make it an error to set EVEX.R' to 0 when modrm.reg encodes a...
Craig Topper [Fri, 1 Jun 2018 06:11:29 +0000 (06:11 +0000)]
[X86][Disassembler] Make it an error to set EVEX.R' to 0 when modrm.reg encodes a GPR.

This is different than the behavior of EVEX.X extending modrm.rm to 5 bits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333728 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][Disassembler] Ignore EVEX.X extension of modrm.rm to 5-bits when modrm.rm encod...
Craig Topper [Fri, 1 Jun 2018 05:36:08 +0000 (05:36 +0000)]
[X86][Disassembler] Ignore EVEX.X extension of modrm.rm to 5-bits when modrm.rm encodes a k-register.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333727 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoImplemented sane default for llvm-objdump's relocation Value format
Daniel Cederman [Fri, 1 Jun 2018 05:31:58 +0000 (05:31 +0000)]
Implemented sane default for llvm-objdump's relocation Value format

Summary:
"Unknown" for platforms that were not manually added into the switch
did not make sense at all. Now it prints Target + addend for all
elf-machines that were not explicitly mentioned.

Addresses PR21059 and PR25124.

Original author: fedor.sergeev

Reviewers: jyknight, espindola, fedor.sergeev

Reviewed By: jyknight

Subscribers: eraman, dcederman, jfb, dschuff, aheejin, llvm-commits

Differential Revision: https://reviews.llvm.org/D36464

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333726 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][Disassembler] Clamp index to 4-bits when decoding GPR registers.
Craig Topper [Fri, 1 Jun 2018 05:12:44 +0000 (05:12 +0000)]
[X86][Disassembler] Clamp index to 4-bits when decoding GPR registers.

A 5-bit value can occur when EVEX.X is 0 due to it being used to extend modrm.rm to encode XMM16-31. But if modrm.rm instead encodes a GPR, the Intel documentation says EVEX.X should be ignored so just mask it to 4 bits once we know its a GPR.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333725 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add a test case showing a bad disassembling of an EVEX instruction with EVEX...
Craig Topper [Fri, 1 Jun 2018 05:12:43 +0000 (05:12 +0000)]
[X86] Add a test case showing a bad disassembling of an EVEX instruction with EVEX.X=0 and a GPR encoded in modrm.rm.

EVEX.X is used to extended modrm.rm when the instruction encodes a XMM/YMM/ZMM register. But we aren't properly ignoring it when it encodes a GPR and we end up printing whatever registers exist in X86 register enum after the GPRs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333724 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][Disassembler] Make sure EVEX.X is not used to extend base registers of memory...
Craig Topper [Fri, 1 Jun 2018 04:29:34 +0000 (04:29 +0000)]
[X86][Disassembler] Make sure EVEX.X is not used to extend base registers of memory operations.

This was an accidental side effect of EVEX.X being used to encode XMM16-XMM31 using modrm.rm with modrm.mod==0x3.

I think there are still more bugs related to this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333722 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][Disassembler] Use a local variable instead of using a field in the instruction...
Craig Topper [Fri, 1 Jun 2018 04:29:30 +0000 (04:29 +0000)]
[X86][Disassembler] Use a local variable instead of using a field in the instruction object. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333721 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU/R600: Move intrinsics to IntrinsicsAMDGPU.td
Tom Stellard [Fri, 1 Jun 2018 02:19:46 +0000 (02:19 +0000)]
AMDGPU/R600: Move intrinsics to IntrinsicsAMDGPU.td

Reviewers: arsenm, nhaehnle, jvesely

Reviewed By: arsenm

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

Differential Revision: https://reviews.llvm.org/D47487

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333720 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Make sure the check for VEX.vvvv being all ones on instructions that don't...
Craig Topper [Fri, 1 Jun 2018 01:23:52 +0000 (01:23 +0000)]
[X86] Make sure the check for VEX.vvvv being all ones on instructions that don't use it doesn't ignore a bit in 32-bit mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333717 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][Disassembler] Suppress reading of EVEX.V' and EVEX.R' in 32-bit mode.
Craig Topper [Fri, 1 Jun 2018 00:10:36 +0000 (00:10 +0000)]
[X86][Disassembler] Suppress reading of EVEX.V' and EVEX.R' in 32-bit mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333714 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add test cases showing the disassembler producing an xmm16-xmm31 register in...
Craig Topper [Fri, 1 Jun 2018 00:10:32 +0000 (00:10 +0000)]
[X86] Add test cases showing the disassembler producing an xmm16-xmm31 register in 32-bit mode.

We aren't properly suppressing the reading of VEX.R' and VEX.V' in 32-bit mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333713 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoChange ambiguous uses of term 'funclet' to 'EH scopes'. NFC.
Heejin Ahn [Fri, 1 Jun 2018 00:03:21 +0000 (00:03 +0000)]
Change ambiguous uses of term 'funclet' to 'EH scopes'. NFC.

Summary:
`getEHScopeMembership()` function is used not only for funclet-based
EHs; they apply to all EH schemes that use the scoped IR
(catchpad/cleanuppad/...). D47005 (rL333045) changed some of the uses of
the term 'funclet' to 'EH scopes' in case they apply to all scoped EH,
and this fixes more of them. For `FuncletLayout` pass, I left it as is
because the pass is only used for funclet-based EH.

Reviewers: majnemer

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D47611

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333711 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Update to the new names for the memory intrinsics.
Dan Gohman [Thu, 31 May 2018 22:35:25 +0000 (22:35 +0000)]
[WebAssembly] Update to the new names for the memory intrinsics.

The WebAssembly committee has decided on the names `memory.size` and
`memory.grow` for the memory intrinsics, so update the LLVM intrinsics to
follow those names, keeping both sets of old names in place for
compatibility.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333708 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LoopVectorize, x86] add tests to show missing SVML transforms; NFC
Sanjay Patel [Thu, 31 May 2018 22:31:02 +0000 (22:31 +0000)]
[LoopVectorize, x86] add tests to show missing SVML transforms; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333707 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Fix the signatures for the __mulo* libcalls.
Dan Gohman [Thu, 31 May 2018 22:27:24 +0000 (22:27 +0000)]
[WebAssembly] Fix the signatures for the __mulo* libcalls.

The __mulo* libcalls have an extra i32* to return the overflow value.

Fixes PR37401.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333706 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Support instruction selection for catching exceptions
Heejin Ahn [Thu, 31 May 2018 22:25:54 +0000 (22:25 +0000)]
[WebAssembly] Support instruction selection for catching exceptions

Summary:
This lowers exception catching-related instructions:
1. Lowers `wasm.catch` intrinsic to `catch` instruction
2. Removes `catchpad` and `cleanuppad` instructions; they are not
necessary after isel phase. (`MachineBasicBlock::isEHFuncletEntry()` or
`MachineBasicBlock::isEHPad()` can be used instead.)
3. Lowers `catchret` and `cleanupret` instructions to pseudo `catchret`
and `cleanupret` instructions in isel, which will be replaced with other
instructions in `WebAssemblyExceptionPrepare` pass.
4. Adds 'WebAssemblyExceptionPrepare` pass, which is for running various
transformation for EH. Currently this pass only replaces `catchret` and
`cleanupret` instructions into appropriate wasm instructions to make
this patch successfully run until the end.

Currently this does not handle lowering of intrinsics related to LSDA
info generation (`wasm.landingpad.index` and `wasm.lsda`), because they
cannot be tested without implementing `EHStreamer`'s wasm-specific
handlers. They are marked as TODO, which is needed to make isel pass.
Also this does not generate `try` and `end_try` markers yet, which will
be handled in later patches.

This patch is based on the first wasm EH proposal.
(https://github.com/WebAssembly/exception-handling/blob/master/proposals/Exceptions.md)

Reviewers: dschuff, majnemer

Subscribers: jfb, sbc100, jgravelle-google, sunfish, llvm-commits

Differential Revision: https://reviews.llvm.org/D44090

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@333705 91177308-0d34-0410-b5e6-96231b3b80d8