OSDN Git Service

android-x86/external-llvm.git
6 years agoAMDGPU: Make isIntrinsicSourceOfDivergence table-driven
Nicolai Haehnle [Sun, 1 Apr 2018 17:09:14 +0000 (17:09 +0000)]
AMDGPU: Make isIntrinsicSourceOfDivergence table-driven

Summary:
This is in preparation for the new dimension-aware image intrinsics,
which I'd rather not have to list here by hand.

Change-Id: Iaa16e3a635a11283918ce0d9e1e618591b0bf6fa

Reviewers: arsenm, rampitec, b-sumner

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

Differential Revision: https://reviews.llvm.org/D44938

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328939 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Make getTgtMemIntrinsic table-driven for resource-based intrinsics
Nicolai Haehnle [Sun, 1 Apr 2018 17:09:07 +0000 (17:09 +0000)]
AMDGPU: Make getTgtMemIntrinsic table-driven for resource-based intrinsics

Summary:
Avoids having to list all intrinsics manually.

This is in preparation for the new dimension-aware image intrinsics,
which I'd rather not have to list here by hand.

Change-Id: If7ced04998397ef68c4cb8f7de66b5050fb767e5

Reviewers: arsenm, rampitec, b-sumner

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

Differential Revision: https://reviews.llvm.org/D44937

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328938 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoTableGen: Support Intrinsic values in SearchableTable
Nicolai Haehnle [Sun, 1 Apr 2018 17:08:58 +0000 (17:08 +0000)]
TableGen: Support Intrinsic values in SearchableTable

Summary:
We will use this in the AMDGPU backend in a subsequent patch
in the stack to lookup target-specific per-intrinsic information.

The generic CodeGenIntrinsic machinery is used to ensure that,
even though we don't calculate actual enum values here, we do
get the intrinsics in the right order for the binary search
index.

Change-Id: If61cd5587963a4c5a1cc53df1e59c5e4dec1f9dc

Reviewers: arsenm, rampitec, b-sumner

Subscribers: wdng, tpr, llvm-commits

Differential Revision: https://reviews.llvm.org/D44935

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328937 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoTableGen: More helpful error messages
Nicolai Haehnle [Sun, 1 Apr 2018 17:08:49 +0000 (17:08 +0000)]
TableGen: More helpful error messages

Summary: Change-Id: I3c23f6f6597912423762780cd8c5315870412bbe

Reviewers: arsenm, rampitec, b-sumner

Subscribers: wdng, llvm-commits

Differential Revision: https://reviews.llvm.org/D44936

Change-Id: Ie62614a3e2d7774f46e4034478b28f57100a2c92

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328936 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DebugInfo] Change std::sort to llvm::sort in response to r327219
Mandeep Singh Grang [Sun, 1 Apr 2018 16:18:49 +0000 (16:18 +0000)]
[DebugInfo] Change std::sort to llvm::sort in response to r327219

Summary:
r327219 added wrappers to std::sort which randomly shuffle the container before sorting.
This will help in uncovering non-determinism caused due to undefined sorting
order of objects having the same key.

To make use of that infrastructure we need to invoke llvm::sort instead of std::sort.

Note: This patch is one of a series of patches to replace *all* std::sort to llvm::sort.
Refer the comments section in D44363 for a list of all the required patches.

Reviewers: echristo, zturner, samsonov

Reviewed By: echristo

Subscribers: JDevlieghere, llvm-commits

Differential Revision: https://reviews.llvm.org/D45134

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328935 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ThinLTO] Add an import cutoff for debugging/triaging
Teresa Johnson [Sun, 1 Apr 2018 15:54:40 +0000 (15:54 +0000)]
[ThinLTO] Add an import cutoff for debugging/triaging

Summary:
Adds -import-cutoff=N which will stop importing during the thin link
after N imports. Default is -1 (no  limit).

Reviewers: wmi

Subscribers: inglorion, llvm-commits

Differential Revision: https://reviews.llvm.org/D45127

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328934 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LoopRotate] Rotate loops with loop exiting latches
David Green [Sun, 1 Apr 2018 12:48:24 +0000 (12:48 +0000)]
[LoopRotate] Rotate loops with loop exiting latches

If a loop has a loop exiting latch, it can be profitable
to rotate the loop if it leads to the simplification of
a phi node. Perform rotation in these cases even if loop
rotate itself didnt simplify the loop to get there.

Differential Revision: https://reviews.llvm.org/D44199

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328933 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Don't check for folding into a store when deciding if we can promote an i16...
Craig Topper [Sun, 1 Apr 2018 06:29:32 +0000 (06:29 +0000)]
[X86] Don't check for folding into a store when deciding if we can promote an i16 mul.

There's no RMW mul operation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328931 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Check if the load and store are to the same pointer before preventing i16 RMW...
Craig Topper [Sun, 1 Apr 2018 06:29:28 +0000 (06:29 +0000)]
[X86] Check if the load and store are to the same pointer before preventing i16 RMW shifts and subtracts from being promoted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328930 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add test case to show failure to promote i16 subtract when the LHS is a load...
Craig Topper [Sun, 1 Apr 2018 06:29:27 +0000 (06:29 +0000)]
[X86] Add test case to show failure to promote i16 subtract when the LHS is a load and the result is stored to a different address.

We mistakenly believe we might be able to fold this as a RMW operation, but that doesn't end up happening.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328929 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Allow i16 subtracts to be promoted if the load is on the LHS and its not being...
Craig Topper [Sun, 1 Apr 2018 06:29:25 +0000 (06:29 +0000)]
[X86] Allow i16 subtracts to be promoted if the load is on the LHS and its not being stored.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328928 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add test case to show failure to promote i16 subtract because we mistakenly...
Craig Topper [Sun, 1 Apr 2018 06:29:23 +0000 (06:29 +0000)]
[X86] Add test case to show failure to promote i16 subtract because we mistakenly believe the load can be folded. NFC

The left hand side of the subtract is a load, but we cna't fold those unless we also have a store.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328927 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Remove unneeded temporary variable. NFC
Craig Topper [Sun, 1 Apr 2018 06:29:21 +0000 (06:29 +0000)]
[X86] Remove unneeded temporary variable. NFC

This Promote flag was alwasys set to true except in the default case. But in the default case we don't need to set PVT and can just return false.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328926 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Analysis] Change std::sort to llvm::sort in response to r327219
Mandeep Singh Grang [Sun, 1 Apr 2018 01:46:51 +0000 (01:46 +0000)]
[Analysis] Change std::sort to llvm::sort in response to r327219

Summary:
r327219 added wrappers to std::sort which randomly shuffle the container before sorting.
This will help in uncovering non-determinism caused due to undefined sorting
order of objects having the same key.

To make use of that infrastructure we need to invoke llvm::sort instead of std::sort.

Note: This patch is one of a series of patches to replace *all* std::sort to llvm::sort.
Refer D44363 for a list of all the required patches.

Reviewers: sanjoy, dexonsmith, hfinkel, RKSimon

Reviewed By: dexonsmith

Subscribers: david2050, llvm-commits

Differential Revision: https://reviews.llvm.org/D44944

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328925 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DAGCombine] (float)((int) f) --> ftrunc (PR36617)
Sanjay Patel [Sat, 31 Mar 2018 17:55:44 +0000 (17:55 +0000)]
[DAGCombine] (float)((int) f) --> ftrunc (PR36617)

fptosi / fptoui round towards zero, and that's the same behavior as ISD::FTRUNC,
so replace a pair of casts with the equivalent node. We don't have to account for
special cases (NaN, INF) because out-of-range casts are undefined.

Differential Revision: https://reviews.llvm.org/D44909

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328921 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-rtdyld] Fix the InputFileList cl::opt description: it accepts multiple
Lang Hames [Sat, 31 Mar 2018 16:01:01 +0000 (16:01 +0000)]
[llvm-rtdyld] Fix the InputFileList cl::opt description: it accepts multiple
input files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328920 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][Btver2] Add MMX_PSHUFB to the JWritePSHUFB InstRW entries
Simon Pilgrim [Sat, 31 Mar 2018 09:15:54 +0000 (09:15 +0000)]
[X86][Btver2] Add MMX_PSHUFB to the JWritePSHUFB InstRW entries

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328918 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix trailing whitespace. NFCI.
Simon Pilgrim [Sat, 31 Mar 2018 09:14:14 +0000 (09:14 +0000)]
Fix trailing whitespace. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328917 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoUnbreak the build of the go bindings after r328839.
Benjamin Kramer [Sat, 31 Mar 2018 07:41:25 +0000 (07:41 +0000)]
Unbreak the build of the go bindings after r328839.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328916 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MIR-Canon] Adding support for local idempotent instruction hoisting.
Puyan Lotfi [Sat, 31 Mar 2018 05:48:51 +0000 (05:48 +0000)]
[MIR-Canon] Adding support for local idempotent instruction hoisting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328915 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add SchedRW for PMULLD
Craig Topper [Sat, 31 Mar 2018 04:54:32 +0000 (04:54 +0000)]
[X86] Add SchedRW for PMULLD

Summary:
It seems many CPUs don't implement this instruction as well as the other vector multiplies. Often using a multi uop flow. Silvermont in particular has a 7 uop flow with 11 cycle throughput. Sandy Bridge implements it as a single uop with 5 cycle latency and 1 cycle throughput. But Haswell and later use 2 uops with 10 cycle latency and 2 cycle throughput.

This patch adds a new X86SchedWritePair we can use to tag this instruction separately. I've provided correct information for Silvermont, Btver2, and Sandy Bridge. I've removed the InstRWs for SandyBridge. I've left Haswell/Broadwell/Skylake InstRWs in place because I wasn't sure how to account for the different load latency between 128 and 256 bits. I also left Znver1 InstRWs in place because the existing values don't match Agner's spreadsheet.

I also left a FIXME in the SandyBridge model because it being used for the "generic" model is too optimistic for the 256/512-bit versions since those are multiple uops on all known CPUs.

Reviewers: RKSimon, GGanesh, courbet

Reviewed By: RKSimon

Subscribers: gchatelet, gbedwell, andreadb, llvm-commits

Differential Revision: https://reviews.llvm.org/D44972

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328914 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ThinLTO] Add an option to force summary call edges cold for debugging
Teresa Johnson [Sat, 31 Mar 2018 00:18:08 +0000 (00:18 +0000)]
[ThinLTO] Add an option to force summary call edges cold for debugging

Summary:
Useful to selectively disable importing into specific modules for
debugging/triaging/workarounds.

Reviewers: eraman

Subscribers: inglorion, llvm-commits

Differential Revision: https://reviews.llvm.org/D45062

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328909 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix a bunch of typoes. NFC
Fangrui Song [Fri, 30 Mar 2018 22:22:31 +0000 (22:22 +0000)]
Fix a bunch of typoes. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328907 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoPrevent data races in concurrent ThinLTO processes.
Ekaterina Romanova [Fri, 30 Mar 2018 21:35:42 +0000 (21:35 +0000)]
Prevent data races in concurrent ThinLTO processes.

Make sure ThinLTO with caching doesn't use non-atomic writes to the cache file (to prevent data races and cache files corruption).

1. Place temp file to the same place where the caching directory is (instead of creating it the directory pointed to by TMP/TEMP variable). This will help to prevent using non-atomic rename and falling back to non-atomic "direct" write to the cache file.
2. if rename failed do not write to the cache file directly (direct write to the file is non-atomic and could cause data race conditions).
3. if cache file doesn't exist (e.g., because 'rename' failed or because some other reasons), bypass using the cache altogether.

Differential Revision:  https://reviews.llvm.org/D45076

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328904 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Register wasm passes with the PassRegistry
Jacob Gravelle [Fri, 30 Mar 2018 20:36:58 +0000 (20:36 +0000)]
[WebAssembly] Register wasm passes with the PassRegistry

Summary:
This exposes WebAssembly passes for use on the command line (as
arguments to -print-before and the like).

Reviewers: dschuff, sunfish

Subscribers: MatzeB, jfb, sbc100, llvm-commits, aheejin

Differential Revision: https://reviews.llvm.org/D45103

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328901 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Fix testcase
Krzysztof Parzyszek [Fri, 30 Mar 2018 19:46:28 +0000 (19:46 +0000)]
[Hexagon] Fix testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328899 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Reduce excessive indentation in .s output
Krzysztof Parzyszek [Fri, 30 Mar 2018 19:30:28 +0000 (19:30 +0000)]
[Hexagon] Reduce excessive indentation in .s output

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328898 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Avoid creating invalid offsets in packetizer
Krzysztof Parzyszek [Fri, 30 Mar 2018 19:28:37 +0000 (19:28 +0000)]
[Hexagon] Avoid creating invalid offsets in packetizer

Two memory instructions with a dependency only on the address register
between the two (the first one of them being post-incrememnt) can be
packetized together after the offset on the second was updated to the
incremement value. Make sure that the new offset is valid for the
instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328897 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][BtVer2] Fixed the number of micro opcodes for AVX vector converts and
Andrea Di Biagio [Fri, 30 Mar 2018 18:53:47 +0000 (18:53 +0000)]
[X86][BtVer2] Fixed the number of micro opcodes for AVX vector converts and
VSQRT instructions.

There were still a few AVX instructions with an incorrect number of opcodes.
These should be fixed now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328892 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDataFlowSanitizer: wrappers of functions with local linkage should have the same...
Peter Collingbourne [Fri, 30 Mar 2018 18:37:55 +0000 (18:37 +0000)]
DataFlowSanitizer: wrappers of functions with local linkage should have the same linkage as the function being wrapped

This patch resolves link errors when the address of a static function is taken, and that function is uninstrumented by DFSan.

This change resolves bug 36314.

Patch by Sam Kerner!

Differential Revision: https://reviews.llvm.org/D44784

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328890 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MIR] Adding support for Named Virtual Registers in MIR.
Puyan Lotfi [Fri, 30 Mar 2018 18:15:54 +0000 (18:15 +0000)]
[MIR] Adding support for Named Virtual Registers in MIR.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328887 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][BtVer2] Fix the number of uOps for horizontal operations.
Andrea Di Biagio [Fri, 30 Mar 2018 18:15:30 +0000 (18:15 +0000)]
[X86][BtVer2] Fix the number of uOps for horizontal operations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328886 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[NVPTX] Enable StructuredCFG for NVPTX
Tim Shen [Fri, 30 Mar 2018 17:51:03 +0000 (17:51 +0000)]
[NVPTX] Enable StructuredCFG for NVPTX

Summary:
Make NVPTX require structured CFG. Added a temporary flag to
"roll back" the behavior for easy deployment.

Combined with D45008, this fixes several internal Nvidia GPU test
failures that we suspect to be ptxas miscompiles (PR27738).

Reviewers: jlebar

Subscribers: jholewinski, sanjoy, llvm-commits, hiraditya

Differential Revision: https://reviews.llvm.org/D45070

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328885 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[BlockPlacement] Disable block placement tail duplciation in structured CFG.
Tim Shen [Fri, 30 Mar 2018 17:51:00 +0000 (17:51 +0000)]
[BlockPlacement] Disable block placement tail duplciation in structured CFG.

Summary:
Tail duplication easily breaks the structure of CFG, e.g. duplicating on
a region entry. If the structure is intended to be preserved, then we
may want to configure tail duplication, or disable it for structured
CFG. From our benchmark results disabling it doesn't cause performance
regression.

Notice that this currently affects AMDGPU backend. In the next patch, I
also plan to turn on requiresStructuredCFG for NVPTX.

All unit tests still pass.

Reviewers: jlebar, arsenm

Subscribers: jholewinski, sanjoy, wdng, tpr, hiraditya, llvm-commits

Differential Revision: https://reviews.llvm.org/D45008

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328884 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LLVM-C] Finish exception instruction bindings - Round 2
Robert Widmann [Fri, 30 Mar 2018 17:49:53 +0000 (17:49 +0000)]
[LLVM-C] Finish exception instruction bindings - Round 2

Summary:
Previous revision caused a leak in the echo test that got caught by the ASAN bots because of missing free of the handlers array and was reverted in r328759.  Resubmitting the patch with that correction.

Add support for cleanupret, catchret, catchpad, cleanuppad and catchswitch and their associated accessors.

Test is modified from SimplifyCFG because it contains many diverse usages of these instructions.

Reviewers: whitequark, deadalnix

Reviewed By: whitequark

Subscribers: llvm-commits, vlad.tsyrklevich

Differential Revision: https://reviews.llvm.org/D45100

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328883 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix some signed / unsigned conversion problems.
Zachary Turner [Fri, 30 Mar 2018 17:28:35 +0000 (17:28 +0000)]
Fix some signed / unsigned conversion problems.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328881 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-pdbutil] Dig deeper into the PDB and DBI streams when explaining.
Zachary Turner [Fri, 30 Mar 2018 17:16:50 +0000 (17:16 +0000)]
[llvm-pdbutil] Dig deeper into the PDB and DBI streams when explaining.

This will show more detail when using `llvm-pdbutil explain` on an
offset in the DBI or PDB streams.  Specifically, it will dig into
individual header fields and substreams to give a more precise
description of what the byte represents.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328878 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Refactor tablegen for store instructions (NFC)
Derek Schuff [Fri, 30 Mar 2018 17:02:50 +0000 (17:02 +0000)]
[WebAssembly] Refactor tablegen for store instructions (NFC)

Summary: Add patterns similar to loads.

Differential Revision: https://reviews.llvm.org/D45064

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328876 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "peel loops with runtime small trip counts"
Krzysztof Parzyszek [Fri, 30 Mar 2018 16:55:44 +0000 (16:55 +0000)]
Revert "peel loops with runtime small trip counts"

This reverts commit r328854, it breaks some Hexagon tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328875 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU] Fixed some instructions latencies
Stanislav Mekhanoshin [Fri, 30 Mar 2018 16:19:13 +0000 (16:19 +0000)]
[AMDGPU] Fixed some instructions latencies

Differential Revision: https://reviews.llvm.org/D45073

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328874 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SelectionDAG] Removing FABS folding from DAGCombiner
Sanjay Patel [Fri, 30 Mar 2018 15:42:52 +0000 (15:42 +0000)]
[SelectionDAG] Removing FABS folding from DAGCombiner

The code has bugs dealing with -0.0.

Since D44550 introduced FABS pattern folding in InstCombine,
this patch removes the now-redundant code that causes
https://bugs.llvm.org/show_bug.cgi?id=36600.

Patch by Mikhail Dvoretckii!

Differential Revision: https://reviews.llvm.org/D44683

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328872 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Recognize and handle :endloop01
Krzysztof Parzyszek [Fri, 30 Mar 2018 15:29:47 +0000 (15:29 +0000)]
[Hexagon] Recognize and handle :endloop01

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328870 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Fix printing :mem_noshuf on compiler-generated packets
Krzysztof Parzyszek [Fri, 30 Mar 2018 15:09:05 +0000 (15:09 +0000)]
[Hexagon] Fix printing :mem_noshuf on compiler-generated packets

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328869 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Fix flags for store-related intrinsics
Krzysztof Parzyszek [Fri, 30 Mar 2018 14:57:01 +0000 (14:57 +0000)]
[Hexagon] Fix flags for store-related intrinsics

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328868 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][BtVer2] Add missing ReadAfterLd to RM variants of AVX horizontal adds and
Andrea Di Biagio [Fri, 30 Mar 2018 14:48:08 +0000 (14:48 +0000)]
[X86][BtVer2] Add missing ReadAfterLd to RM variants of AVX horizontal adds and
most vector logic instructions.

Fixed a few InstRW that forgot to specify a ReadAfterLd for the register input
operand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328867 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Remove unused scheduling classes
Krzysztof Parzyszek [Fri, 30 Mar 2018 14:34:32 +0000 (14:34 +0000)]
[Hexagon] Remove unused scheduling classes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328866 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][BtVer2] Add tests that show how ReadAfterLd is missing for some
Andrea Di Biagio [Fri, 30 Mar 2018 14:29:33 +0000 (14:29 +0000)]
[X86][BtVer2] Add tests that show how ReadAfterLd is missing for some
instructions.

In the Btver2 model, there are a few InstRW overrides that don't specify a
ReadAfterLd for the register input operand.

As a result, a few AVX variants of horizontal operations and most vector logic
operations with a folded memory operand don't have a ReadAdvance info associated
to their input register operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328865 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Pass pointer to SelectionDAG to dump functions
Krzysztof Parzyszek [Fri, 30 Mar 2018 14:29:15 +0000 (14:29 +0000)]
[Hexagon] Pass pointer to SelectionDAG to dump functions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328864 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add llvm-mca tests for r328834.
Andrea Di Biagio [Fri, 30 Mar 2018 13:38:37 +0000 (13:38 +0000)]
[X86] Add llvm-mca tests for r328834.

Verify that the ReadAfterLd is correctly applied to FMA and 4-ops variable blend
instructions.

As Craig pointed out in D44726, some Intel models still have to be fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328861 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add tests to verify the presence of "ReadAfterLd" after r328823.
Andrea Di Biagio [Fri, 30 Mar 2018 11:44:48 +0000 (11:44 +0000)]
[X86] Add tests to verify the presence of "ReadAfterLd" after r328823.

This change adds a couple of tests to verify the change introduced by revision
328823 ([X86] Correct the placement of ReadAfterLd in BEXTR and BZHI).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328859 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "[LLVM-C] Finish exception instruction bindings"
Vlad Tsyrklevich [Fri, 30 Mar 2018 06:21:28 +0000 (06:21 +0000)]
Revert "[LLVM-C] Finish exception instruction bindings"

This reverts commit r328759. It was causing LSan failures on sanitizer-x86_64-linux-bootstrap

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328858 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU] Fix the SDWA Peephole phase to handle src for dst:UNUSED_PRESERVE.
Michael Bedy [Fri, 30 Mar 2018 05:03:36 +0000 (05:03 +0000)]
[AMDGPU] Fix the SDWA Peephole phase to handle src for dst:UNUSED_PRESERVE.

Summary:
The phase attempts to transform operations that extract a portion of a value
into an SDWA src operand in cases where that value is used only once. It
was not prepared for this use to be the preserved portion of a value for
dst:UNUSED_PRESERVE, resulting in a crash or assert.

This change either rejects the illegal SDWA attempt, or in the case where
dst:WORD_1 and the src_sel would be WORD_0, removes the unneeded
extract instruction.

Reviewers: arsenm, #amdgpu

Reviewed By: arsenm, #amdgpu

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

Differential Revision: https://reviews.llvm.org/D44364

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328856 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] add missing lit config file
Ikhlas Ajbar [Fri, 30 Mar 2018 03:32:24 +0000 (03:32 +0000)]
[Hexagon] add missing lit config file

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328855 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agopeel loops with runtime small trip counts
Ikhlas Ajbar [Fri, 30 Mar 2018 03:05:34 +0000 (03:05 +0000)]
peel loops with runtime small trip counts

For Hexagon, peeling loops with small runtime trip count is beneficial for our
benchmarks. We set PeelCount in HexagonTargetInfo.cpp and we use PeelCount set
by the target for computing the desired peel count.

Differential Revision: https://reviews.llvm.org/D44880

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328854 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MachineCopyPropagation] Handle COPY with overlapping source/dest.
Eli Friedman [Fri, 30 Mar 2018 00:56:03 +0000 (00:56 +0000)]
[MachineCopyPropagation] Handle COPY with overlapping source/dest.

MachineCopyPropagation::CopyPropagateBlock has a bunch of special
handling for COPY instructions. This handling assumes that COPY
instructions do not modify the source of the copy; this is wrong if
the COPY destination overlaps the source.

To fix the bug, check explicitly for this situation, and fall back to
the generic instruction handling.

This bug can't happen for most register classes because they don't
have this sort of overlap, but there are a few register classes
where this is possible. The testcase uses the AArch64 QQQQ register
class.

Differential Revision: https://reviews.llvm.org/D44911

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328851 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[IR] Fix some Clang-tidy modernize-use-auto warnings; other minor fixes (NFC).
Eugene Zelenko [Fri, 30 Mar 2018 00:47:31 +0000 (00:47 +0000)]
[IR] Fix some Clang-tidy modernize-use-auto warnings; other minor fixes (NFC).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328850 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoStyle update. NFC.
Rafael Espindola [Thu, 29 Mar 2018 23:32:54 +0000 (23:32 +0000)]
Style update. NFC.

Rename 3 functions to start with lowercase letters. Don't repeat the
name in the comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328848 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix some layering in StripNonLineTableDebugInfo, moving its declaration from IPO...
David Blaikie [Thu, 29 Mar 2018 22:42:08 +0000 (22:42 +0000)]
Fix some layering in StripNonLineTableDebugInfo, moving its declaration from IPO.h to Utils.h to match its implementation

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328844 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRemove unused header to fix layering.
David Blaikie [Thu, 29 Mar 2018 22:35:59 +0000 (22:35 +0000)]
Remove unused header to fix layering.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328842 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRemove unused headers to fix layering
David Blaikie [Thu, 29 Mar 2018 22:31:39 +0000 (22:31 +0000)]
Remove unused headers to fix layering

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328840 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agollvm-c: Split Utils out of Scalar.h
David Blaikie [Thu, 29 Mar 2018 22:31:38 +0000 (22:31 +0000)]
llvm-c: Split Utils out of Scalar.h

To fix layering (so that Scalar.h, a libScalarOpts header, isn't
included from Utils - which libScalarOpts depends on).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328839 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRemove some unneeded #includes to fix layering
David Blaikie [Thu, 29 Mar 2018 22:31:36 +0000 (22:31 +0000)]
Remove some unneeded #includes to fix layering

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328838 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add ReadAfterLds to some 3 src instructions
Craig Topper [Thu, 29 Mar 2018 22:03:05 +0000 (22:03 +0000)]
[X86] Add ReadAfterLds to some 3 src instructions

Sometimes the operand comes after the memory operand so we need 5 ReadDefaults first.

I suspect we also need to do something for the mask operand for masked avx512 instructions? I'm not sure if the mask should be ReadAfterLd or not since it can mask faults. If it shouldn't be ReadAfterLd then we're probably wrong for zero masking instructions already.

Differential Revision: https://reviews.llvm.org/D44726

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328834 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoTypo fix: epilouge->epilogue. NFC.
Eric Christopher [Thu, 29 Mar 2018 21:59:04 +0000 (21:59 +0000)]
Typo fix: epilouge->epilogue. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328833 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Fix build warning in release
Matt Arsenault [Thu, 29 Mar 2018 21:44:44 +0000 (21:44 +0000)]
AMDGPU: Fix build warning in release

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328832 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Support realigning stack
Matt Arsenault [Thu, 29 Mar 2018 21:30:06 +0000 (21:30 +0000)]
AMDGPU: Support realigning stack

While the stack access instructions don't care about
alignment > 4, some transformations on the pointer calculation
do make assumptions based on knowing the low bits of a pointer
are 0. If a stack object ends up being accessed through its
absolute address (relative to the kernel scratch wave offset),
the addressing expression may depend on the stack frame being
properly aligned. This was breaking in a testcase due to the
add->or combine.

I think some of the SP/FP handling logic is still backwards,
and overly simplistic to support all of the stack features.
Code which tries to modify the SP with inline asm for example
or variable sized objects will probably require redoing this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328831 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAdd msan custom mapping options.
Evgeniy Stepanov [Thu, 29 Mar 2018 21:18:17 +0000 (21:18 +0000)]
Add msan custom mapping options.

Similarly to https://reviews.llvm.org/D18865 this adds options to provide custom mapping for msan.
As discussed in http://lists.llvm.org/pipermail/llvm-dev/2018-February/121339.html

Patch by vit9696(at)avp.su.

Differential Revision: https://reviews.llvm.org/D44926

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328830 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Remove ReadAfterLd from BMI and TBM instructions that don't have a register...
Craig Topper [Thu, 29 Mar 2018 21:03:53 +0000 (21:03 +0000)]
[X86] Remove ReadAfterLd from BMI and TBM instructions that don't have a register operand in their memory form

The memory form of these instructions only read an input from memory. They don't have any register operands.

Differential Revision: https://reviews.llvm.org/D44836

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328828 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoTry to fix sanitizer-x86_64-linux-fast bot due to change in r328820.
Kevin Enderby [Thu, 29 Mar 2018 20:49:24 +0000 (20:49 +0000)]
Try to fix sanitizer-x86_64-linux-fast bot due to change in r328820.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328824 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Correct the placement of ReadAfterLd in BEXTR and BZHI. Add dedicated SchedRW...
Craig Topper [Thu, 29 Mar 2018 20:41:39 +0000 (20:41 +0000)]
[X86] Correct the placement of ReadAfterLd in BEXTR and BZHI. Add dedicated SchedRW for BEXTR/BZHI.

These instructions have the memory operand before the register operand. So we need to put ReadDefault for all the load ops first. Then the ReadAfterLd

Differential Revision: https://reviews.llvm.org/D44838

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328823 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[NFC][LICM] Rearrange checks to have the cheap bail out first
Philip Reames [Thu, 29 Mar 2018 20:32:15 +0000 (20:32 +0000)]
[NFC][LICM] Rearrange checks to have the cheap bail out first

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328822 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Increase default stack alignment
Matt Arsenault [Thu, 29 Mar 2018 20:22:04 +0000 (20:22 +0000)]
AMDGPU: Increase default stack alignment

8 and 16-byte values are common, so increase the default
alignment to avoid realigning the stack in most functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328821 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFor llvm-nm and Mach-O files that are fully stripped, special case a redacted LC_MAIN
Kevin Enderby [Thu, 29 Mar 2018 20:04:29 +0000 (20:04 +0000)]
For llvm-nm and Mach-O files that are fully stripped, special case a redacted LC_MAIN

As a further refinement on:

r328274 - For llvm-nm and Mach-O files also use function starts info in some cases when printing symbols

we want to special case a redacted LC_MAIN so it is easier to find.

rdar://38978929

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328820 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Fix selection error on constant loads with < 4 byte alignment
Matt Arsenault [Thu, 29 Mar 2018 19:59:28 +0000 (19:59 +0000)]
AMDGPU: Fix selection error on constant loads with < 4 byte alignment

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328818 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix an accidental circular dependence
Philip Reames [Thu, 29 Mar 2018 19:22:12 +0000 (19:22 +0000)]
Fix an accidental circular dependence

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328816 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Mips] Change std::sort to llvm::sort in response to r327219
Mandeep Singh Grang [Thu, 29 Mar 2018 19:05:26 +0000 (19:05 +0000)]
[Mips] Change std::sort to llvm::sort in response to r327219

Summary:
r327219 added wrappers to std::sort which randomly shuffle the container before sorting.
This will help in uncovering non-determinism caused due to undefined sorting
order of objects having the same key.

To make use of that infrastructure we need to invoke llvm::sort instead of std::sort.

Note: This patch is one of a series of patches to replace *all* std::sort to llvm::sort.
Refer the comments section in D44363 for a list of all the required patches.

Reviewers: sdardis, RKSimon, dsanders, atanasyan

Reviewed By: atanasyan

Subscribers: atanasyan, arichardson, llvm-commits

Differential Revision: https://reviews.llvm.org/D44869

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328815 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoTry to fix a couple tests for Windows.
Paul Robinson [Thu, 29 Mar 2018 18:59:33 +0000 (18:59 +0000)]
Try to fix a couple tests for Windows.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328814 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SLPVectorizer] Add tests related to PR30787, NFCI.
Dinar Temirbulatov [Thu, 29 Mar 2018 18:57:03 +0000 (18:57 +0000)]
[SLPVectorizer] Add tests related to PR30787, NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328813 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MSF] Default to FPM2, and always mark FPM pages allocated.
Zachary Turner [Thu, 29 Mar 2018 18:34:15 +0000 (18:34 +0000)]
[MSF] Default to FPM2, and always mark FPM pages allocated.

There are two FPMs in an MSF file, the idea being that for
incremental updates you can write to the alternate one and then
atomically swap them on commit.  LLVM defaulted to using FPM1
on the first commit, but this differs from Microsoft's behavior
which is to default to using FPM2 on the first commit.  To
eliminate some byte-level file differences, this patch changes
LLVM's default to also be FPM2.

Additionally, LLVM was trying to be "smart" about marking FPM
pages allocated.  In addition to marking every page belonging
to the alternate FPM as unallocated, LLVM also marked pages at
the end of the main FPM which were not needed as unallocated.

In order to match the behavior of Microsoft-generated PDBs, we
now always mark every FPM block as allocated, regardless of
whether it is in the main FPM or the alt FPM, and regardless of
whether or not it describes blocks which are actually in the file.

This has the side benefit of simplifying our code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328812 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PDB] Print some more details when explaining MSF fields.
Zachary Turner [Thu, 29 Mar 2018 17:45:34 +0000 (17:45 +0000)]
[PDB] Print some more details when explaining MSF fields.

When we determine that a field belongs to an MSF super block or
the free page map, we wouldn't print any additional information.

With this patch, we now print the value of the field (for super
block fields) or the allocation status of the specified byte (in
the case of offsets in the FPM).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328808 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[IR][CodeGen] Remove dependency on EVT from IR/Function.cpp. Move EVT to CodeGen...
Craig Topper [Thu, 29 Mar 2018 17:21:10 +0000 (17:21 +0000)]
[IR][CodeGen] Remove dependency on EVT from IR/Function.cpp. Move EVT to CodeGen layer.

Currently EVT is in the IR layer only because of Function.cpp needing a very small piece of the functionality of EVT::getEVTString(). The rest of EVT is used in codegen making CodeGen a better place for it.

The previous code converted a Type* to EVT and then called getEVTString. This was only expected to handle the primitive types from Type*. Since there only a few primitive types, we can just print them as strings directly.

Differential Revision: https://reviews.llvm.org/D45017

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328806 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoReapply "[DWARFv5] Emit file 0 to the line table."
Paul Robinson [Thu, 29 Mar 2018 17:16:41 +0000 (17:16 +0000)]
Reapply "[DWARFv5] Emit file 0 to the line table."

DWARF v5 specifies that the root file (also given in the DW_AT_name
attribute of the compilation unit DIE) should be emitted explicitly to
the line table's list of files.  This makes the line table more
independent of the .debug_info section.
We emit the new syntax only for DWARF v5 and later.

Fixes the bug found by asan. Also XFAIL the new test for Darwin, which
is stuck on DWARF v2, and fix up other tests so they stop failing on
Windows.  Last but not least, don't break "clang -g" of an assembler
file that has .file directives in it.

Differential Revision: https://reviews.llvm.org/D44054

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328805 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PDB] Fix a bug in the explain subcommand.
Zachary Turner [Thu, 29 Mar 2018 17:11:14 +0000 (17:11 +0000)]
[PDB] Fix a bug in the explain subcommand.

We were trying to dig into the super block fields and print a
description of the field at the specified offset, but we were
printing the wrong field due to an off-by-one-field-error.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328804 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ADT] NFC: Fix bogus StringSwitch rule-of-five boilerplate
David Zarzycki [Thu, 29 Mar 2018 16:51:28 +0000 (16:51 +0000)]
[ADT] NFC: Fix bogus StringSwitch rule-of-five boilerplate

Now that 'Str' is constant, the rule-of-file logic needs updating.

Reported by: vit9696@avp.su
Reviewed by: jordan_rose@apple.com

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328803 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRemove unused function.
Zachary Turner [Thu, 29 Mar 2018 16:46:47 +0000 (16:46 +0000)]
Remove unused function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328802 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PDB] Add an explain subcommand.
Zachary Turner [Thu, 29 Mar 2018 16:28:20 +0000 (16:28 +0000)]
[PDB] Add an explain subcommand.

When investigating various things, we often have a file offset
and what to know what's in the PDB at that address.  For example
we may be doing a binary comparison of two LLD-generated PDBs
to look for sources of non-determinism, or we may wish to compare
an LLD-generated PDB with a Microsoft generated PDB for sources
of byte-for-byte incompatibility.  In these cases, we can do a
binary diff of the two files, and once we find a mismatched byte
we can use explain to figure out what that byte is, immediately
honining in on the problem.

This patch implements this by trying to narrow the meaning of
a particular file offset down as much as possible.

Differential Revision: https://reviews.llvm.org/D44959

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328799 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[JumpThreading] Don't select an edge that we know we can't thread
Haicheng Wu [Thu, 29 Mar 2018 16:01:26 +0000 (16:01 +0000)]
[JumpThreading] Don't select an edge that we know we can't thread

In r312664 (D36404), JumpThreading stopped threading edges into
loop headers. Unfortunately, I observed a significant performance
regression as a result of this change. Upon further investigation,
the problematic pattern looked something like this (after
many high level optimizations):

while (true) {
    bool cond = ...;
    if (!cond) {
        <body>
    }
    if (cond)
        break;
}

Now, naturally we want jump threading to essentially eliminate the
second if check and hook up the edges appropriately. However, the
above mentioned change, prevented it from doing this because it would
have to thread an edge into the loop header.

Upon further investigation, what is happening is that since both branches
are threadable, JumpThreading picks one of them at arbitrarily. In my
case, because of the way that the IR ended up, it tended to pick
the one to the loop header, bailing out immediately after. However,
if it had picked the one to the exit block, everything would have
worked out fine (because the only remaining branch would then be folded,
not thraded which is acceptable).

Thus, to fix this problem, we can simply eliminate loop headers from
consideration as possible threading targets earlier, to make sure that
if there are multiple eligible branches, we can still thread one of
the ones that don't target a loop header.

Patch by Keno Fischer!
Differential Revision: https://reviews.llvm.org/D42260

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328798 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago.debug_names: Correctly align the AugmentationStringSize field
Pavel Labath [Thu, 29 Mar 2018 15:12:45 +0000 (15:12 +0000)]
.debug_names: Correctly align the AugmentationStringSize field

We should align the value of the field, not the overall section offset.

This distinction matters if one of the debug_names contributions is not
of size which is a multiple of four. The dwarf producers may choose to
emit rounded contributions, but they are not required to do so. In the
latter case, without this patch we would corrupt the parsing state, as
we would adjust the offset even if subsequent contributions contained
correctly rounded augmentation strings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328796 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-mca] Correctly set the ReadAdvance information for register use operands.
Andrea Di Biagio [Thu, 29 Mar 2018 14:26:56 +0000 (14:26 +0000)]
[llvm-mca] Correctly set the ReadAdvance information for register use operands.

The tool was passing the wrong operand index to method
MCSubtargetInfo::getReadAdvanceCycles(). That method requires a "UseIdx", and
not the operand index. This was found when testing X86 code where instructions
had a memory folded operand.

This patch fixes the issue and adds test read-advance-1.s to ensure that
the ReadAfterLd (a ReadAdvance of 3cy) information is correctly used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328790 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Add support to handle bit-reverse load intrinsics
Krzysztof Parzyszek [Thu, 29 Mar 2018 13:52:46 +0000 (13:52 +0000)]
[Hexagon] Add support to handle bit-reverse load intrinsics

Patch by Sumanth Gundapaneni.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328774 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago.debug_names: Parse DW_IDX_die_offset as a reference
Pavel Labath [Thu, 29 Mar 2018 13:47:57 +0000 (13:47 +0000)]
.debug_names: Parse DW_IDX_die_offset as a reference

Before this patch we were parsing the attributes as section offsets, as
that is what apple_names is doing. However, this is not correct as DWARF
v5 specifies that this attribute should use the Reference form class.

This also updates all the testcases (except the ones that deliberately
pass a different form) to use the correct form class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328773 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Kaleidoscope] Tiny typo fixes
Sjoerd Meijer [Thu, 29 Mar 2018 12:31:06 +0000 (12:31 +0000)]
[Kaleidoscope] Tiny typo fixes

Fixes for "lets" references which should be "let's" in the Kaleidoscope
tutorial.

Patch by: Robin Dupret

Differential Revision: https://reviews.llvm.org/D44990

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328772 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][SSE] Don't bother re-adding combined target shuffles to the work list
Simon Pilgrim [Thu, 29 Mar 2018 11:18:41 +0000 (11:18 +0000)]
[X86][SSE] Don't bother re-adding combined target shuffles to the work list

We are re-adding all the bitcasts, constant masks and target shuffles to the work list for no apparent gain.

Found while investigating adding SimplifyDemandedVectorElts to target shuffles.

Differential Revision: https://reviews.llvm.org/D44942

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328771 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRename llvm library from libLLVM-X.Y to libLLVM-X
Sylvestre Ledru [Thu, 29 Mar 2018 09:44:09 +0000 (09:44 +0000)]
Rename llvm library from libLLVM-X.Y to libLLVM-X

Summary:
As we are only doing X.0.Z releases (not using the minor version), there is no need to keep -X.Y in the version.

Like patch https://reviews.llvm.org/D41808, I propose that we rename libLLVM-7.0svn.so to libLLVM-7svn.so
This patch will also rename downstream libraries like liblldb-7.0 to liblldb-7

Reviewers: axw, beanz, dim, hans

Reviewed By: dim, hans

Subscribers: mgorny, llvm-commits

Differential Revision: https://reviews.llvm.org/D41869

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328768 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Mips] Remove dead code
Simon Dardis [Thu, 29 Mar 2018 09:21:20 +0000 (09:21 +0000)]
[Mips] Remove dead code

I believe the role of ehDataReg has been replaced by MipsABIInfo::GetEhDataReg, thus removing the dead code.

Patch By: Wei-Ren Chen.

Reviewers: ehostunreach, sdardis

Differential Revision: https://reviews.llvm.org/D44867

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328767 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LoopRotate] Restructuring LoopRotation.cpp to create Loop Rotation Pass with Loop...
David Green [Thu, 29 Mar 2018 08:48:15 +0000 (08:48 +0000)]
[LoopRotate] Restructuring LoopRotation.cpp to create Loop Rotation Pass with Loop Rotation Utility Interface

The existing LoopRotation.cpp is implemented as one of loop passes instead of
being a utility. The user cannot easily perform the loop rotation selectively
(or on demand) under different optimization level. For example, the loop
rotation is needed as part of the logic to convert a loop into a loop with
bottom test for a transformation. If the loop rotation is simply added as a
loop pass before the transformation, the pass is skipped if it is compiled at
–O0 or if it is explicitly disabled by the user, causing the compiler to
generate incorrect code. Furthermore, as a loop pass it will rotate all loops
instead of just the relevant loops.

We provide a utility interface for the loop rotation so that the loop rotation
can be called on demand. The changeset is as follows:

- Create a new file lib/Transforms/Utils/LoopRotationUtils.cpp and move the main
  implementation of class LoopRotate into this file.
- Create a new file llvm/include/Transform/Utils/LoopRotationUtils.h with the
  interface LoopRotation(...).
- Original LoopRotation.cpp is changed to use the utility function LoopRotation
  in LoopRotationUtils.cpp. This is done in the same way community did for
  mem-to-reg implementation.

Patch by Jin Lin!

Differential Revision: https://reviews.llvm.org/D44595

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328766 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Transforms] Make sure to include the c binding header when defining c binding functions
Benjamin Kramer [Thu, 29 Mar 2018 07:56:53 +0000 (07:56 +0000)]
[Transforms] Make sure to include the c binding header when defining c binding functions

Otherwise the definitions can't see the extern C declarations and get
name mangled, making it impossible for users to call them. This breaks
the Go bindings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328765 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[NFC] Fix meaningless assert in SCEV
Max Kazantsev [Thu, 29 Mar 2018 07:54:59 +0000 (07:54 +0000)]
[NFC] Fix meaningless assert in SCEV

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328764 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Don't pass getRegisterName from the InstPrinters into EmitAnyX86InstComments...
Craig Topper [Thu, 29 Mar 2018 04:14:04 +0000 (04:14 +0000)]
[X86] Don't pass getRegisterName from the InstPrinters into EmitAnyX86InstComments. Just always use the function from the ATTPrinter. NFC

The IntelPrinter and the ATTPrinter produce the same strings for the same input. We already use the ATTPrinter explicitly in several other places.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328762 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LLVM-C] Finish exception instruction bindings
Robert Widmann [Thu, 29 Mar 2018 03:43:15 +0000 (03:43 +0000)]
[LLVM-C] Finish exception instruction bindings

Summary:
Add support for cleanupret, catchret, catchpad, cleanuppad and catchswitch and their associated accessors.

Test is modified from SimplifyCFG because it contains many diverse usages of these instructions.

Reviewers: whitequark, deadalnix, echristo

Reviewed By: echristo

Subscribers: llvm-commits, harlanhaskins

Differential Revision: https://reviews.llvm.org/D44496

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328759 91177308-0d34-0410-b5e6-96231b3b80d8