OSDN Git Service

android-x86/external-llvm.git
7 years agoMake SmallPtrSet count and find able to take const PtrType's
Daniel Berlin [Tue, 7 Mar 2017 18:47:48 +0000 (18:47 +0000)]
Make SmallPtrSet count and find able to take const PtrType's

Summary:
For our set/map types, count/find normally take const references.
This works well for non-pointer types, but can suck for pointer
types.

DenseSet<int *> foo;
const int *b = nullptr;
foo.count(b) does not work

but the equivalent reference version does work
(patch to fix DenseSet/DenseMap coming up)
For SmallPtrSet, you have no such option.

The following will not work right now:
SmallPtrSet<int *> foo;
const int *b = nullptr;
foo.count(b);

This makes const correctness hard in some cases.
Example:
SmallPtrSet<Instruction *> InstructionsToErase;

You can't make this SmallPtrSet<const Instruction *> because then you
can't erase the instruction.  If I want to see if something is in the
set, I may only have a const Instruction *.  Given that count and find
are non-mutating, this should just work.

The places in our code base that do this resort to const_cast :(.

This patch makes count and find able to be used with const Instruction
* in the above SmallPtrSet examples.

This is a bit annoying because of where C++ applies the const, so we
have to remove the pointer type from the passed-in-type and rebuild it
with const.

Reviewers: dblaikie

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D30608

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297180 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LV] Consider users that are memory accesses in uniforms expansion step
Matthew Simpson [Tue, 7 Mar 2017 18:47:30 +0000 (18:47 +0000)]
[LV] Consider users that are memory accesses in uniforms expansion step

When expanding the set of uniform instructions beyond the seed instructions
(e.g., consecutive pointers), we mark a new instruction uniform if all its
loop-varying users are uniform. We should also allow users that are consecutive
or interleaved memory accesses. This fixes cases where we have an instruction
that is used as the pointer operand of a consecutive access but also used by a
non-memory instruction that later becomes uniform as part of the expansion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297179 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Add option to specify preferable loop alignment
Sanjoy Das [Tue, 7 Mar 2017 18:47:22 +0000 (18:47 +0000)]
[X86] Add option to specify preferable loop alignment

Summary:
Loop alignment can cause a significant change of
the perfromance for short loops.
To be able to evaluate the impact of loop alignment this change
introduces the new option x86-experimental-pref-loop-alignment.
The alignment will be 2^Value bytes, the default value is 4.

Patch by Serguei Katkov!

Reviewers: craig.topper

Reviewed By: craig.topper

Subscribers: sanjoy, llvm-commits

Differential Revision: https://reviews.llvm.org/D30391

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297178 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[globalisel] Change LLT constructor string into an LLT-based object that knows how...
Daniel Sanders [Tue, 7 Mar 2017 18:32:25 +0000 (18:32 +0000)]
[globalisel] Change LLT constructor string into an LLT-based object that knows how to generate it.

Summary:
This will allow future patches to inspect the details of the LLT. The implementation is now split between
the Support and CodeGen libraries to allow TableGen to use this class without introducing layering concerns.

Thanks to Ahmed Bougacha for finding a reasonable way to avoid the layering issue and providing the version of this patch without that problem.

Reviewers: t.p.northover, qcolombet, rovka, aditya_nandakumar, ab, javed.absar

Subscribers: arsenm, nhaehnle, mgorny, dberris, llvm-commits, kristof.beyls

Differential Revision: https://reviews.llvm.org/D30046

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297177 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFurther reduce testcase
Adrian Prantl [Tue, 7 Mar 2017 18:26:36 +0000 (18:26 +0000)]
Further reduce testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297176 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix test and add missing return for llvm-lto2 error case
Teresa Johnson [Tue, 7 Mar 2017 18:15:13 +0000 (18:15 +0000)]
Fix test and add missing return for llvm-lto2 error case

Summary:
This test was missing the target triple.
Once I fixed that, the case with the invalid character error stopped
returning 1 from llvm-lto2 and the test reported a failure. Fixed by
adding the missing return from llvm-lto2. Apparently we were failing
when we eventually tried to get the target.

Reviewers: pcc

Subscribers: mehdi_amini, llvm-commits

Differential Revision: https://reviews.llvm.org/D30585

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297173 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[GlobalISel] Translate floating-point negation
Volkan Keles [Tue, 7 Mar 2017 18:03:28 +0000 (18:03 +0000)]
[GlobalISel] Translate floating-point negation

Reviewers: qcolombet, javed.absar, aditya_nandakumar, dsanders, t.p.northover, ab

Reviewed By: qcolombet

Subscribers: dberris, rovka, llvm-commits, kristof.beyls

Differential Revision: https://reviews.llvm.org/D30671

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297171 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoUpdate comment in testcase
Adrian Prantl [Tue, 7 Mar 2017 17:55:36 +0000 (17:55 +0000)]
Update comment in testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297170 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRephrase condition for better readability. NFC
Adrian Prantl [Tue, 7 Mar 2017 17:50:51 +0000 (17:50 +0000)]
Rephrase condition for better readability. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297168 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[fs] Make sure to check S_ISLNK() in fillStatus.
Zachary Turner [Tue, 7 Mar 2017 17:48:47 +0000 (17:48 +0000)]
[fs] Make sure to check S_ISLNK() in fillStatus.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297167 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agorevert r297155 because there's a clang test that depends on InstCombine:
Sanjay Patel [Tue, 7 Mar 2017 17:41:45 +0000 (17:41 +0000)]
revert r297155 because there's a clang test that depends on InstCombine:

tools/clang/test/CodeGen/zvector.c

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297166 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "Strip debug info when inlining into a nodebug function."
Adrian Prantl [Tue, 7 Mar 2017 17:28:57 +0000 (17:28 +0000)]
Revert "Strip debug info when inlining into a nodebug function."

This reverts commit r296488.

As noted by David Blaikie on llvm-commits, I overlooked the case of a
debug function being inlined into a nodebug function being inlined
into a debug function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297163 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRelax the conflicting function arg verifier to allow for inlined debug
Adrian Prantl [Tue, 7 Mar 2017 17:28:54 +0000 (17:28 +0000)]
Relax the conflicting function arg verifier to allow for inlined debug
info in nodebug functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297161 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoVerfier: Move the reset of DebugFnArgs closer to other similar operations.
Adrian Prantl [Tue, 7 Mar 2017 17:28:49 +0000 (17:28 +0000)]
Verfier: Move the reset of DebugFnArgs closer to other similar operations.
NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297160 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert r297132, it caused PR32171
Nico Weber [Tue, 7 Mar 2017 17:23:52 +0000 (17:23 +0000)]
Revert r297132, it caused PR32171

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297159 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstCombine] shrink truncated splat shuffle
Sanjay Patel [Tue, 7 Mar 2017 16:10:36 +0000 (16:10 +0000)]
[InstCombine] shrink truncated splat shuffle

This is one part of solving a recent bug report:
http://lists.llvm.org/pipermail/llvm-dev/2017-February/110293.html

This keeps with our general approach: changing arbitrary shuffles is off-limts,
but changing splat is ok. The transform is very similar to the existing
shrinkBitwiseLogic() canonicalization.

Differential Revision: https://reviews.llvm.org/D30123

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297155 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Support] Add the option to not follow symlinks on stat.
Zachary Turner [Tue, 7 Mar 2017 16:10:10 +0000 (16:10 +0000)]
[Support] Add the option to not follow symlinks on stat.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297154 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ARM] Correct handling of LSL #0 in an IT block
John Brawn [Tue, 7 Mar 2017 14:42:03 +0000 (14:42 +0000)]
[ARM] Correct handling of LSL #0 in an IT block

The check for LSL #0 in an IT block was checking if operand 4 was zero, but
operand 4 is the condition code operand so it was actually checking for LSLEQ.
Fix this by checking operand 3, which really is the immediate operand, and add
some tests.

Differential Revision: https://reviews.llvm.org/D30692

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297142 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Hexagon] Do not insert instructions before PHI nodes
Krzysztof Parzyszek [Tue, 7 Mar 2017 14:20:19 +0000 (14:20 +0000)]
[Hexagon] Do not insert instructions before PHI nodes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297141 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ARM] Reapply r296865 "[ARM] fpscr read/write intrinsics not aware of each other""
Ranjeet Singh [Tue, 7 Mar 2017 11:17:53 +0000 (11:17 +0000)]
[ARM] Reapply r296865 "[ARM] fpscr read/write intrinsics not aware of each other""

The original patch r296865 was reverted as it broke the chromium builds for
Android https://bugs.llvm.org/show_bug.cgi?id=32134, this patch reapplies
r296865 with a fix to make sure it doesn't cause the build regression.

The problem was that intrinsic selection on int_arm_get_fpscr was failing in
ISel this was because the code to manually select this intrinsic still thought
it was the version with no side-effects (INTRINSIC_WO_CHAIN) which is wrong as
it doesn't semantically match the definition in the tablegen code which says it
does have side-effects, I've fixed this by updating the intrinsic type to
INTRINSIC_W_CHAIN (has side-effects). I've also added a test for this based on
Hans original reproducer.

Differential Revision: https://reviews.llvm.org/D30645

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297137 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SystemZ] Add check VT.isSimple() in canTreateAsByteVector()
Jonas Paulsson [Tue, 7 Mar 2017 09:49:31 +0000 (09:49 +0000)]
[SystemZ] Add check VT.isSimple() in canTreateAsByteVector()

Since BB-vectorizer can produce vectors of for example 3 elements,
this check is needed.

Review: Ulrich Weigand

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297136 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoIn Thumb1, materialize a move between low registers as a `movs`, if CPSR isn't live.
Artyom Skrobov [Tue, 7 Mar 2017 09:38:16 +0000 (09:38 +0000)]
In Thumb1, materialize a move between low registers as a `movs`, if CPSR isn't live.

Summary: Previously, it had always been materialized as a push/pop sequence.

Reviewers: labrinea, jroelofs

Reviewed By: jroelofs

Subscribers: llvm-commits, rengolin

Differential Revision: https://reviews.llvm.org/D30648

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297134 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LoopRotate] Update dbg.value intrinsics
Sam Parker [Tue, 7 Mar 2017 09:34:25 +0000 (09:34 +0000)]
[LoopRotate] Update dbg.value intrinsics

Propagate debug info through the newly inserted PHI nodes.

Differential Revision: https://reviews.llvm.org/D30190

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297132 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Quick fix for build bot failure (r297127) due to unused variable.
Ayman Musa [Tue, 7 Mar 2017 08:56:27 +0000 (08:56 +0000)]
[X86] Quick fix for build bot failure (r297127) due to unused variable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297130 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][AVX512] Adding new LLVM TableGen backend which generates the EVEX2VEX compressi...
Ayman Musa [Tue, 7 Mar 2017 08:11:19 +0000 (08:11 +0000)]
[X86][AVX512] Adding new LLVM TableGen backend which generates the EVEX2VEX compressing tables.

X86EvexToVex machine instruction pass compresses EVEX encoded instructions by replacing them with their identical VEX encoded instructions when possible.
It uses manually supported 2 large tables that map the EVEX instructions to their VEX ideticals.
This TableGen backend replaces the tables by automatically generating them.

Differential Revision: https://reviews.llvm.org/D30451

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297127 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][AVX512] Add missing entries to EVEX2VEX tables
Ayman Musa [Tue, 7 Mar 2017 08:05:53 +0000 (08:05 +0000)]
[X86][AVX512] Add missing entries to EVEX2VEX tables

evex2vex pass defines 2 tables which maps EVEX instructions to their VEX identical when possible. Adding all missing entries.

Differential Revision: https://reviews.llvm.org/D30501

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297126 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "[PowerPC][ELFv2ABI] Allocate parameter area on-demand to reduce stack frame...
Tim Shen [Tue, 7 Mar 2017 07:40:10 +0000 (07:40 +0000)]
Revert "[PowerPC][ELFv2ABI] Allocate parameter area on-demand to reduce stack frame size"

This reverts commit r296771.

We found some wide spread test failures internally. I'm working on a
testcase. Politely revert the patch in the mean time. :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297124 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LoopUnrolling] Fix loop size check for peeling
Sanjoy Das [Tue, 7 Mar 2017 06:03:15 +0000 (06:03 +0000)]
[LoopUnrolling] Fix loop size check for peeling

Summary:
We should check if loop size allows us to peel at least one iteration
before we do so.

Patch by Max Kazantsev!

Reviewers: sanjoy, mkuper, efriedma

Reviewed By: mkuper

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D30632

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297122 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[APInt] Add rvalue reference support to and, or, xor operations to allow their memory...
Craig Topper [Tue, 7 Mar 2017 05:36:19 +0000 (05:36 +0000)]
[APInt] Add rvalue reference support to and, or, xor operations to allow their memory allocation to be reused when possible

This extends an earlier change that did similar for add and sub operations.

With this first patch we lose the fastpath for the single word case as operator&= and friends don't support it. This can be added there if we think that's important.

I had to change some functions in the APInt class since the operator overloads were moved out of the class and can't be used inside the class now. The getBitsSet change collides with another outstanding patch to implement it with setBits. But I didn't want to make this patch dependent on that series.

I've also removed the Or, And, Xor functions which were rarely or never used. I already commited two changes to remove the only uses of Or that existed.

Differential Revision: https://reviews.llvm.org/D30612

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297121 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "AMDGPU: Set MCAsmInfo::PointerSize"
Konstantin Zhuravlyov [Tue, 7 Mar 2017 04:44:33 +0000 (04:44 +0000)]
Revert "AMDGPU: Set MCAsmInfo::PointerSize"

It breaks line tables because the patch is not complete, working on a complete one at the moment

This reverts commit r294031.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297118 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoUse LLVM for all stat-related functionality.
Zachary Turner [Tue, 7 Mar 2017 03:43:17 +0000 (03:43 +0000)]
Use LLVM for all stat-related functionality.

This deletes LLDB's FileType enumeration and replaces all
users, and all calls to functions that check whether a file
exists etc with corresponding calls to LLVM.

Differential Revision: https://reviews.llvm.org/D30624

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297116 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[APInt] Fix test names in unittest to match functions being tested. NFC
Craig Topper [Tue, 7 Mar 2017 03:16:37 +0000 (03:16 +0000)]
[APInt] Fix test names in unittest to match functions being tested. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297115 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[APInt] Add getBitsSetFrom and setBitsFrom to set upper bits starting at a bit
Craig Topper [Tue, 7 Mar 2017 02:58:36 +0000 (02:58 +0000)]
[APInt] Add getBitsSetFrom and setBitsFrom to set upper bits starting at a bit

We currently have methods to set a specified number of low bits, a specified number of high bits, or a range of bits. But looking at some existing code it seems sometimes we want to set the high bits starting from a certain bit. Currently we do this with something like getHighBits(BitWidth, BitWidth - StartBit). Or once we start switching to setHighBits, setHighBits(BitWidth - StartBit) or setHighBits(getBitWidth() - StartBit).

Particularly for the latter case it would be better to have a convenience method like setBitsFrom(StartBit) so we don't need to mention the bit width that's already known to the APInt object.

I considered just making setBits have a default value of UINT_MAX for the hiBit argument and we would internally MIN it with the bit width. So if it wasn't specified it would be treated as bit width. This would require removing the assertion we currently have on the value of hiBit and may not be as readable.

Differential Revision: https://reviews.llvm.org/D30602

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297114 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdd a testcase for r297072.
Adrian Prantl [Tue, 7 Mar 2017 02:49:57 +0000 (02:49 +0000)]
Add a testcase for r297072.

Check that missing debug locations on inlinable calls are a
recoverable error.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297113 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[APInt] Implement getLowBitsSet/getHighBitsSet/getBitsSet using setLowBits/setHighBit...
Craig Topper [Tue, 7 Mar 2017 02:19:45 +0000 (02:19 +0000)]
[APInt] Implement getLowBitsSet/getHighBitsSet/getBitsSet using setLowBits/setHighBits/setBits

This patch implements getLowBitsSet/getHighBitsSet/getBitsSet in terms of the new setLowBits/setHighBits/setBits methods by making an all 0s APInt and then calling the appropriate set method.

This also adds support to setBits to allow loBits/hiBits to be in the other order to match with getBitsSet behavior.

Differential Revision: https://reviews.llvm.org/D30563

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297112 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[APInt] Add setLowBits/setHighBits methods to APInt.
Craig Topper [Tue, 7 Mar 2017 01:56:01 +0000 (01:56 +0000)]
[APInt] Add setLowBits/setHighBits methods to APInt.

Summary:
There are quite a few places in the code base that do something like the following to set the high or low bits in an APInt.

KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - 1);

For BitWidths larger than 64 this creates a short lived APInt with malloced storage. I think it might even call malloc twice. Its better to just provide methods that can set the necessary bits without the temporary APInt.

I'll update usages that benefit in a separate patch.

Reviewers: majnemer, MatzeB, davide, RKSimon, hans

Reviewed By: hans

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D30525

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297111 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoUn-XFAIL Fuzzer/test/dump_coverage.test. This now passes on Darwin (after r297094).
Kuba Mracek [Tue, 7 Mar 2017 01:01:38 +0000 (01:01 +0000)]
Un-XFAIL Fuzzer/test/dump_coverage.test.  This now passes on Darwin (after r297094).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297110 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoremove Cmake option for LLVM_DISABLE_ABI_BREAKING_CHECKS_ENFORCING
Bob Wilson [Tue, 7 Mar 2017 00:51:07 +0000 (00:51 +0000)]
remove Cmake option for LLVM_DISABLE_ABI_BREAKING_CHECKS_ENFORCING

This is a follow-up to my change in r295090, which added support for
disabling these checks selectively based on setting the preprocessor
macro without relying on the Cmake setting. Swift has moved over to use
that approach, so we can clean up here and remove the Cmake setting.

https://reviews.llvm.org/D30578

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297109 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SLP] Revert r296863 due to miscompiles.
Michael Kuperstein [Mon, 6 Mar 2017 23:54:51 +0000 (23:54 +0000)]
[SLP] Revert r296863 due to miscompiles.

Details and reproducer are on the email thread for r296863.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297103 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoGlobalISel: restrict G_EXTRACT instruction to just one operand.
Tim Northover [Mon, 6 Mar 2017 23:50:28 +0000 (23:50 +0000)]
GlobalISel: restrict G_EXTRACT instruction to just one operand.

A bit more painful than G_INSERT because it was more widely used, but this
should simplify the handling of extract operations in most locations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297100 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[libfuzzer] do not use xargs for shell expansion
Mike Aizatsky [Mon, 6 Mar 2017 23:42:14 +0000 (23:42 +0000)]
[libfuzzer] do not use xargs for shell expansion

Summary:
After https://reviews.llvm.org/rL296904 xargs shouldn't be necesssary.

Differential Revision: https://reviews.llvm.org/D30371

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297094 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstCombine] use dyn_cast instead of isa+cast; NFCI
Sanjay Patel [Mon, 6 Mar 2017 23:25:28 +0000 (23:25 +0000)]
[InstCombine] use dyn_cast instead of isa+cast; NFCI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297092 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ObjectYAML] Support for DW_FORM_implicit_const DWARF5 form
Chris Bieneman [Mon, 6 Mar 2017 23:22:49 +0000 (23:22 +0000)]
[ObjectYAML] Support for DW_FORM_implicit_const DWARF5 form

This patch adds support to the DWARF YAML reader and writer for the new DWARF5 abbreviation form, DW_FORM_implicit_const.

The attribute was added in r291599.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297091 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[GlobalISel] Fix G_FPEXT’s description. NFC.
Volkan Keles [Mon, 6 Mar 2017 22:47:19 +0000 (22:47 +0000)]
[GlobalISel] Fix G_FPEXT’s description. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297088 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DAG] refactor related div/rem folds; NFCI
Sanjay Patel [Mon, 6 Mar 2017 22:32:40 +0000 (22:32 +0000)]
[DAG] refactor related div/rem folds; NFCI

This is known incomplete and not called in the right order relative to
other folds, but that's the current behavior. I'm just trying to clean
this up before making actual functional changes to make the patch smaller.

The logic here should mimic the IR equivalents that are in InstSimplify's
simplifyDivRem().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297086 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DWARFv5] Update definitions to match published spec.
Paul Robinson [Mon, 6 Mar 2017 22:20:03 +0000 (22:20 +0000)]
[DWARFv5] Update definitions to match published spec.

Some late additions to DWARF v5 were not in Dwarf.def; also one form
was redefined.  Add the new cases to relevant switches in different
parts of LLVM.  Replace DW_FORM_ref_sup with DW_FORM_ref_sup[4,8].

I did not add support for DW_FORM_strx3/addrx3 other that defining the
constants. We don't have any infrastructure to support these.

Differential Revision: http://reviews.llvm.org/D30664

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297085 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Outliner] Fixed Asan bot failure in r296418
Jessica Paquette [Mon, 6 Mar 2017 21:31:18 +0000 (21:31 +0000)]
[Outliner] Fixed Asan bot failure in r296418

Fixed the asan bot failure which led to the last commit of the outliner being reverted.
The change is in lib/CodeGen/MachineOutliner.cpp in the SuffixTree's constructor. LeafVector
is no longer initialized using reserve but just a standard constructor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297081 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Ensure proper ordering of subtarget feature names
Krzysztof Parzyszek [Mon, 6 Mar 2017 21:26:49 +0000 (21:26 +0000)]
[TableGen] Ensure proper ordering of subtarget feature names

Recommit r297039 without the testcase. The MIR testcase did not work
well with MC code emitter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297080 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AArch64][Redundant Copy Elim] Add support for CMN and shifted imm.
Chad Rosier [Mon, 6 Mar 2017 21:20:00 +0000 (21:20 +0000)]
[AArch64][Redundant Copy Elim] Add support for CMN and shifted imm.

This patch extends the current functionality of the AArch64 redundant copy
elimination pass to handle CMN instructions as well as a shifted
immediates.

Differential Revision: https://reviews.llvm.org/D30576.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297078 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoDisable gvn-hoist (PR32153)
Hans Wennborg [Mon, 6 Mar 2017 21:10:40 +0000 (21:10 +0000)]
Disable gvn-hoist (PR32153)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297075 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoVerifier: Change Assert to AssertDI.
Adrian Prantl [Mon, 6 Mar 2017 21:05:14 +0000 (21:05 +0000)]
Verifier: Change Assert to AssertDI.
This error can be recovered from by stripping debug info.
This is NFC for +asserts builds.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297072 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ObjectYAML] NFC. Refactor DWARFYAML CompileUnit dump code
Chris Bieneman [Mon, 6 Mar 2017 20:52:12 +0000 (20:52 +0000)]
[ObjectYAML] NFC. Refactor DWARFYAML CompileUnit dump code

Summary: This patch refactors the DWARFYAML code for dumping compile units to use a visitor pattern. Using this design will, in the future, enable the DWARF YAML code to perform analysis and mutations of the DWARF DIEs. An example of such mutations would be calculating the length of a compile unit and updating the CU's Length field before writing the DIE. This support will make it easier to craft or modify DWARF tests by hand.

Reviewers: lhames

Subscribers: mgorny, fhahn, jgosnell, aprantl, llvm-commits

Differential Revision: https://reviews.llvm.org/D30357

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297067 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoTest commit. Added a blank line.
Jessica Paquette [Mon, 6 Mar 2017 20:45:33 +0000 (20:45 +0000)]
Test commit. Added a blank line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297064 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert r297039, it's causing some mysterious buildbot failures
Krzysztof Parzyszek [Mon, 6 Mar 2017 20:24:21 +0000 (20:24 +0000)]
Revert r297039, it's causing some mysterious buildbot failures

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297062 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU/R600: Fix ALU clause markers use detection
Jan Vesely [Mon, 6 Mar 2017 20:10:05 +0000 (20:10 +0000)]
AMDGPU/R600: Fix ALU clause markers use detection

also exit early on kill instead of redefinition.

Differential Revision: https://reviews.llvm.org/D30230

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297060 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoNewGVN: We were not really failing this testcase, because the instructions it was...
Daniel Berlin [Mon, 6 Mar 2017 20:01:31 +0000 (20:01 +0000)]
NewGVN: We were not really failing this testcase, because the instructions it was looking for are unused. GVN value numbers unused instructions, NewGVN does not.  Fix the instructions to be used, so we eliminate the redundancies it's checking for, and un-XFAIL it

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297058 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DWARF] NFC. A few bits of minor code cleanup.
Chris Bieneman [Mon, 6 Mar 2017 19:25:07 +0000 (19:25 +0000)]
[DWARF] NFC. A few bits of minor code cleanup.

David Blaikie pointed out that the `setForceChildren` API is no longer needed and should be removed from the DWARF Generator APIs.

Also the DWARFDebugInfoTest file had some copy pasted comments that are not relevant. I've removed them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297056 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[opt-diff] Fix the case when the script is invoked with directories
Adam Nemet [Mon, 6 Mar 2017 19:15:22 +0000 (19:15 +0000)]
[opt-diff] Fix the case when the script is invoked with directories

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297055 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[IfConversion] Only renormalize probabilities if branches are analyzable
Krzysztof Parzyszek [Mon, 6 Mar 2017 19:12:42 +0000 (19:12 +0000)]
[IfConversion] Only renormalize probabilities if branches are analyzable

If a block has non-analyzable branches, the listed successors don't need
to add up to one. For example, if a block has a conditional tail call,
that tail call will not have a corresponding successor in the successor
list, but will still be a possible branch.

Differential Revision: https://reviews.llvm.org/D30556

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297054 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstSimplify] refactor related div/rem folds; NFCI
Sanjay Patel [Mon, 6 Mar 2017 19:08:35 +0000 (19:08 +0000)]
[InstSimplify] refactor related div/rem folds; NFCI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297052 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoGlobalISel: don't emit degenerate G_INSERT instructions.
Tim Northover [Mon, 6 Mar 2017 19:04:17 +0000 (19:04 +0000)]
GlobalISel: don't emit degenerate G_INSERT instructions.

Before, we were producing G_INSERT instructions that were actually closer to a
cast or even a COPY when both input and output sizes are the same. This doesn't
really make sense and means that everything interpreting a G_INSERT also has to
handle all these kinds of casts.

So now we detect these degenerate cases and emit real casts instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297051 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstSimplify] add tests for vector div/rem with UB potential; NFC
Sanjay Patel [Mon, 6 Mar 2017 18:45:39 +0000 (18:45 +0000)]
[InstSimplify] add tests for vector div/rem with UB potential; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297048 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoNewGVN: Remove DebugUnknownExprs, just mark the instructions as unused
Daniel Berlin [Mon, 6 Mar 2017 18:42:39 +0000 (18:42 +0000)]
NewGVN: Remove DebugUnknownExprs, just mark the instructions as unused

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297047 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoNewGVN: Only call isInstructionTrivially dead once per instruction.
Daniel Berlin [Mon, 6 Mar 2017 18:42:27 +0000 (18:42 +0000)]
NewGVN: Only call isInstructionTrivially dead once per instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297046 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Fix arg copy elision for illegal types
Reid Kleckner [Mon, 6 Mar 2017 18:39:39 +0000 (18:39 +0000)]
[X86] Fix arg copy elision for illegal types

Use the store size of the argument type, which will be a byte-sized
quantity, rather than dividing the size in bits by 8.

Fixes PR32136 and re-enables copy elision from i64 arguments.

Reverts the workaround in from r296950.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297045 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoGlobalISel: add buildUndef method to MachineIRBuilder. NFC.
Tim Northover [Mon, 6 Mar 2017 18:36:40 +0000 (18:36 +0000)]
GlobalISel: add buildUndef method to MachineIRBuilder. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297044 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoGlobalISel: refactor legalization of G_INSERT.
Tim Northover [Mon, 6 Mar 2017 18:23:04 +0000 (18:23 +0000)]
GlobalISel: refactor legalization of G_INSERT.

Now that G_INSERT instructions can only insert one register, this code was
overly general. In another direction it didn't handle registers that crossed
split boundaries properly, which needed to be fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297042 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstSimplify] regenerate checks; NFC
Sanjay Patel [Mon, 6 Mar 2017 18:13:01 +0000 (18:13 +0000)]
[InstSimplify] regenerate checks; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297040 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Ensure proper ordering of subtarget feature names
Krzysztof Parzyszek [Mon, 6 Mar 2017 18:08:37 +0000 (18:08 +0000)]
[TableGen] Ensure proper ordering of subtarget feature names

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297039 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRemove the sample pgo annotation heuristic that uses call count to annotate basic...
Dehao Chen [Mon, 6 Mar 2017 17:49:59 +0000 (17:49 +0000)]
Remove the sample pgo annotation heuristic that uses call count to annotate basic block count.

Summary: We do not need that special handling because the debug info is more accurate now. Performance testing shows no regression on google internal benchmarks.

Reviewers: davidxl, aprantl

Reviewed By: aprantl

Subscribers: llvm-commits, aprantl

Differential Revision: https://reviews.llvm.org/D30658

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297038 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Hexagon] Early-if-convert branches that may exit the loop
Krzysztof Parzyszek [Mon, 6 Mar 2017 17:24:04 +0000 (17:24 +0000)]
[Hexagon] Early-if-convert branches that may exit the loop

Merge the tail block into the loop in cases where the main loop body
exits early, subject to profitability constraints. This will coalesce
the loop body into fewer blocks.

For example:
  loop:                           loop:
    // loop body                      // loop body
    if (...) jump exit      -->       // more body
  more:                               if (...) jump exit
    // more body                      jump loop
    jump loop

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297033 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Hexagon] Mark dead defs as <dead> in expand-condsets
Krzysztof Parzyszek [Mon, 6 Mar 2017 17:09:06 +0000 (17:09 +0000)]
[Hexagon] Mark dead defs as <dead> in expand-condsets

The code in updateDeadFlags removed unnecessary <dead> flags, but there
can be cases where such a flag is not set, and yet a register has become
dead. For example, if a mux with identical inputs is replaced with a COPY,
the predicate register may no longer be used after that.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297032 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Hexagon] Pick a dot-old instruction that matches the architecture
Krzysztof Parzyszek [Mon, 6 Mar 2017 17:03:16 +0000 (17:03 +0000)]
[Hexagon] Pick a dot-old instruction that matches the architecture

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297031 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstSimplify] remove misleading comments; NFC
Sanjay Patel [Mon, 6 Mar 2017 16:49:35 +0000 (16:49 +0000)]
[InstSimplify] remove misleading comments; NFC

Div/rem-of-0 does not cause faults/undef (not the same as div/rem-by-0).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297029 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DAGCombiner] simplify div/rem-by-0
Sanjay Patel [Mon, 6 Mar 2017 16:36:42 +0000 (16:36 +0000)]
[DAGCombiner] simplify div/rem-by-0

Refactoring of duplicated code and more fixes to follow.

This is motivated by the post-commit comments for r296699:
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20170306/435182.html

Ie, we can crash if we're missing obvious simplifications like this that
exist in the IR simplifier or if these occur later than expected.

The x86 change for non-splat division shows a potential opportunity to improve
vector codegen: we assumed that since only one lane had meaningful results, we
should do the math in scalar. But that means moving back and forth from vector
registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297026 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SLP] A test for vectorization of users of extractelement instructions,
Alexey Bataev [Mon, 6 Mar 2017 16:26:00 +0000 (16:26 +0000)]
[SLP] A test for vectorization of users of extractelement instructions,
NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297024 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix minor typo introduce in r297014
Tobias Grosser [Mon, 6 Mar 2017 16:03:26 +0000 (16:03 +0000)]
Fix minor typo introduce in r297014

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297020 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoSilence a warning "hiding virtual function".
Vassil Vassilev [Mon, 6 Mar 2017 15:50:59 +0000 (15:50 +0000)]
Silence a warning "hiding virtual function".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297018 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[x86] add tests to show missing div/rem simplifications; NFC
Sanjay Patel [Mon, 6 Mar 2017 15:50:07 +0000 (15:50 +0000)]
[x86] add tests to show missing div/rem simplifications; NFC

These are not x86-specific, but the problem is not visible for all targets
because it is masked by other transforms. These can lead to compiler crashes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297017 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[BasicBlockUtils] Check for nullptr before updating LoopInfo.
Michael Kruse [Mon, 6 Mar 2017 15:33:05 +0000 (15:33 +0000)]
[BasicBlockUtils] Check for nullptr before updating LoopInfo.

LoopInfo::getLoopFor returns nullptr if a BB is not in a loop and only
then can the loop be updated to contain the newly created BBs. Add the
missing nullptr check to SplitBlockAndInsertIfThen.

Within LLVM, the only user of this function that also passes a LoopInfo
to be updated is InnerLoopVectorizer::predicateInstructions().
As the method's name implies, the BB operataten on will always be within
a loop, but out-of-tree users may also use it differently (here: Polly).

All other uses of LoopInfo::getLoopFor in the file properly check its
return value for nullptr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297016 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DAG] fix formatting; NFC
Sanjay Patel [Mon, 6 Mar 2017 15:27:57 +0000 (15:27 +0000)]
[DAG] fix formatting; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297015 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoNew Test-Case for Region Analysis
Tobias Grosser [Mon, 6 Mar 2017 15:20:38 +0000 (15:20 +0000)]
New Test-Case for Region Analysis

While working on improvements to region info analysis, this test case caused an
incorrect region bb2 => bb3 to be detected.

Reviewers: grosser

Contributed-by: Nandini Singhal <cs15mtech01004@iith.ac.in>
Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D30652

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297014 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DAG] fix typo in comment; NFC
Sanjay Patel [Mon, 6 Mar 2017 15:07:43 +0000 (15:07 +0000)]
[DAG] fix typo in comment; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297011 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoCMake: Add a build target for generating a source RPM
Tom Stellard [Mon, 6 Mar 2017 14:26:50 +0000 (14:26 +0000)]
CMake: Add a build target for generating a source RPM

Summary:
'make srpm' or 'ninja srpm' will tar up the current source code and then
build a source RPM package.

By default it will use the llvm.spec file to generate the source RPM,
but you can specify your own custom spec file with the
LLVM_SRPM_USER_BINARY_SPECFILE option.  CMake will perform variable
substitution on your custom specfile, so you can reference CMake
variables in it.  For example:

Version:        @LLVM_RPM_SPEC_VERSION@

Note that everything in the source directory will be included in the
tarball so if you have a clang check out in tools/clang, then all
the clang source will end up in the tarball to.  It is recommended
to only use this build target with a clean source tree.

Reviewers: beanz

Reviewed By: beanz

Subscribers: mgorny, llvm-commits

Differential Revision: https://reviews.llvm.org/D30093

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297007 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PowerPC] Fix failure with STBRX when store is narrower than the bswap
Nemanja Ivanovic [Mon, 6 Mar 2017 07:32:13 +0000 (07:32 +0000)]
[PowerPC] Fix failure with STBRX when store is narrower than the bswap

Fixes a crash caused by r296811 by truncating the input of the STBRX node
when the bswap is wider than i32.

Fixes https://bugs.llvm.org/show_bug.cgi?id=32140

Differential Revision: https://reviews.llvm.org/D30615

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297001 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[XRay] Allow logging the first argument of a function call.
Dean Michael Berris [Mon, 6 Mar 2017 06:48:56 +0000 (06:48 +0000)]
[XRay] Allow logging the first argument of a function call.

Summary:
Functions with the "xray-log-args" attribute will have a special XRay sled kind
emitted, for compiler-rt to copy any call arguments to your logging handler.

For practical and performance reasons, only the first argument is supported, and
only up to 64 bits.

Reviewers: dberris

Reviewed By: dberris

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D29702

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296998 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[APInt] Move operator~ out of line to make it better able to reused memory allocation...
Craig Topper [Mon, 6 Mar 2017 06:30:47 +0000 (06:30 +0000)]
[APInt] Move operator~ out of line to make it better able to reused memory allocation from temporary objects

Summary:
This makes operator~ take the APInt by value so if it came from a temporary APInt the move constructor will get invoked and it will be able to reuse the memory allocation from the temporary.

This is similar to what was already done for 2s complement negation.

Reviewers: hans, davide, RKSimon

Reviewed By: davide

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D30614

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296997 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[APInt] Remove add and sub functions from APIntOps namespace.
Craig Topper [Mon, 6 Mar 2017 04:35:01 +0000 (04:35 +0000)]
[APInt] Remove add and sub functions from APIntOps namespace.

They aren't used in tree and using the overloaded operators has more optimization opportunities.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296995 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[APInt] Remove unused And/Or/Xor methods. They just forward to the corresponding...
Craig Topper [Mon, 6 Mar 2017 00:24:53 +0000 (00:24 +0000)]
[APInt] Remove unused And/Or/Xor methods. They just forward to the corresponding operator overload.

I plan to enhance the operator overloads to handle rvalues and these methods would not longer be optimal to use.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296993 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SCEV] Decrease the recursion threshold for CompareValueComplexity
Sanjoy Das [Sun, 5 Mar 2017 23:49:17 +0000 (23:49 +0000)]
[SCEV] Decrease the recursion threshold for CompareValueComplexity

Fixes PR32142.

r287232 accidentally increased the recursion threshold for
CompareValueComplexity from 2 to 32.  This change reverses that change
by introducing a separate flag for CompareValueComplexity's threshold.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296992 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[APInt] Remove the And/Or/Xor/Not functions from the APIntOps namespace.
Craig Topper [Sun, 5 Mar 2017 16:41:11 +0000 (16:41 +0000)]
[APInt] Remove the And/Or/Xor/Not functions from the APIntOps namespace.

Summary:
They aren't used anywhere in tree and its preferable to use the &, |, ^, or ~ operators.

With my patch to add rvalue reference support to &, |, ^ operators it also becomes less performant to use these functions.

Reviewers: RKSimon, davide, hans

Reviewed By: RKSimon

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D30613

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296990 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SelectionDAG] Fix vector splitting for *_EXTEND_VECTOR_INREG instructions
Simon Pilgrim [Sun, 5 Mar 2017 15:52:18 +0000 (15:52 +0000)]
[SelectionDAG] Fix vector splitting for *_EXTEND_VECTOR_INREG instructions

Found by fuzz testing after rL296985 landed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296989 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoNew Test-Case for Region Analysis
Tobias Grosser [Sun, 5 Mar 2017 14:08:28 +0000 (14:08 +0000)]
New Test-Case for Region Analysis

While working on improvements to the region info analysis, this test case caused
an incorrect region 1 => 2 to be detected.  It is incorrect because entry has an
outgoing edge to 3.  This is interesting because 1 dom 2 and 2 pdom 1, which
should have been enough to prevent incoming forward edges into the region and
outgoing forward edges from the region.

Reviewers: grosser

Subscribers: llvm-commits

Contributed-by: Nandini Singhal <cs15mtech01004@iith.ac.in>
Differential Revision: https://reviews.llvm.org/D30603

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296988 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Silence GCC enum compare warning.
Benjamin Kramer [Sun, 5 Mar 2017 12:53:20 +0000 (12:53 +0000)]
[X86] Silence GCC enum compare warning.

X86ISelLowering.cpp:26506:36: error: enumeral mismatch in conditional
expression: 'llvm::X86ISD::NodeType' vs 'llvm::ISD::NodeType'
[-Werror=enum-compare]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296986 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][SSE] Lower 128-bit vectors to SIGN/ZERO_EXTEND_VECTOR_IN_REG ops
Simon Pilgrim [Sun, 5 Mar 2017 09:57:20 +0000 (09:57 +0000)]
[X86][SSE] Lower 128-bit vectors to SIGN/ZERO_EXTEND_VECTOR_IN_REG ops

As described on PR31712, we miss a variety of legalization combines because we lower these to X86ISD::VSEXT/VZEXT despite them having the same functionality. This patch makes 128-bit (SSE41) SIGN/ZERO_EXTEND_VECTOR_IN_REG ops legal, adds the necessary tablegen plumbing and uses a helper 'getExtendInVec' to decide when to use SIGN/ZERO_EXTEND_VECTOR_IN_REG or VSEXT/VZEXT.

We're missing a couple of shuffle combines that will be added in a future patch for review.

Later patches can then support the AVX2 cases as a mixture of SIGN/ZERO_EXTEND and SIGN/ZERO_EXTEND_VECTOR_IN_REG, and then finally deal with the AVX512 cases.

Differential Revision: https://reviews.llvm.org/D30549

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296985 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert commit r296967, no typo
Sylvestre Ledru [Sun, 5 Mar 2017 07:46:24 +0000 (07:46 +0000)]
Revert commit r296967, no typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296984 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SimplifyCFG] Use APInt::operator| instead of APInt::Or. NFC
Craig Topper [Sun, 5 Mar 2017 01:08:19 +0000 (01:08 +0000)]
[SimplifyCFG] Use APInt::operator| instead of APInt::Or. NFC

I'm looking to improve operator| to support rvalue references and may remove APInt::Or.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296982 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DAGCombine] Use APInt::operator|(uint64_t) instead of creating a temporary APInt...
Craig Topper [Sun, 5 Mar 2017 01:08:16 +0000 (01:08 +0000)]
[DAGCombine] Use APInt::operator|(uint64_t) instead of creating a temporary APInt and calling APInt::Or. NFC

This is more efficient by itself. But this is prep for a future patch that may remove APInt::Or while making operator| support rvalue references similar to add/sub.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296981 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[x86] don't require a zext when forming ADC/SBB
Sanjay Patel [Sat, 4 Mar 2017 20:35:19 +0000 (20:35 +0000)]
[x86] don't require a zext when forming ADC/SBB

The larger goal is to move the ADC/SBB transforms currently in
combineX86SetCC() to combineAddOrSubToADCOrSBB() because we're
creating ADC/SBB in lots of places where we shouldn't.

This was intended to be an NFC change, but avx-512 has something
strange going on. It doesn't seem like any of the affected tests
should really be using SET+TEST or ADC; a simple ADD could replace
several instructions. But that's another bug...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296978 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DAGCombiner] allow transforming (select Cond, C +/- 1, C) to (add(ext Cond), C)
Sanjay Patel [Sat, 4 Mar 2017 19:18:09 +0000 (19:18 +0000)]
[DAGCombiner] allow transforming (select Cond, C +/- 1, C) to (add(ext Cond), C)

select Cond, C +/- 1, C --> add(ext Cond), C -- with a target hook.

This is part of the ongoing process to obsolete D24480.  The motivation is to
canonicalize to select IR in InstCombine whenever possible, so we need to have a way to
undo that easily in codegen.

PowerPC is an obvious winner for this kind of transform because it has fast and complete
bit-twiddling abilities but generally lousy conditional execution perf (although this might
have changed in recent implementations).

x86 also sees some wins, but the effect is limited because these transforms already mostly
exist in its target-specific combineSelectOfTwoConstants(). The fact that we see any x86
changes just shows that that code is a mess of special-case holes. We may be able to remove
some of that logic now.

My guess is that other targets will want to enable this hook for most cases. The likely
follow-ups would be to add value type and/or the constants themselves as parameters for the
hook. As the tests in select_const.ll show, we can transform any select-of-constants to
math/logic, but the general transform for any 2 constants needs one more instruction
(multiply or 'and').

ARM is one target that I think may not want this for most cases. I see infinite loops there
because it wants to use selects to enable conditionally executed instructions.

Differential Revision: https://reviews.llvm.org/D30537

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296977 91177308-0d34-0410-b5e6-96231b3b80d8