7 years agoRemove extra whitespace. NFC.
Junmo Park [Tue, 5 Jan 2016 09:40:03 +0000 (09:40 +0000)]
Remove extra whitespace. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256821 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRemove extra whitespace. NFC.
Junmo Park [Tue, 5 Jan 2016 09:36:47 +0000 (09:36 +0000)]
Remove extra whitespace. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256820 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][SSE] Merge PerformBLENDICombine into PerformShuffleCombine
Simon Pilgrim [Tue, 5 Jan 2016 09:12:17 +0000 (09:12 +0000)]
[X86][SSE] Merge PerformBLENDICombine into PerformShuffleCombine

PBLEND/BLENDPD/BLENDPS are no different to the other target shuffles and this will make future improvements to the target shuffle combines more straightforward.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256819 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Make MOV32ri64 a post-RA pseudo instead of a CodeGenOnly instruction. It was...
Craig Topper [Tue, 5 Jan 2016 07:44:14 +0000 (07:44 +0000)]
[X86] Make MOV32ri64 a post-RA pseudo instead of a CodeGenOnly instruction. It was only needed for rematerialization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256818 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Add OpSize32 to OR32mrLocked instruction to match the normal OR32mr instruction.
Craig Topper [Tue, 5 Jan 2016 07:44:11 +0000 (07:44 +0000)]
[X86] Add OpSize32 to OR32mrLocked instruction to match the normal OR32mr instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256817 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVX512] Add hasSideEffects=0 to kunpck instructions since they lack a pattern in...
Craig Topper [Tue, 5 Jan 2016 07:44:08 +0000 (07:44 +0000)]
[AVX512] Add hasSideEffects=0 to kunpck instructions since they lack a pattern in their instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256816 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SimplifyCFG] Further improve our ability to remove redundant catchpads
David Majnemer [Tue, 5 Jan 2016 07:42:17 +0000 (07:42 +0000)]
[SimplifyCFG] Further improve our ability to remove redundant catchpads

In r256814, we managed to remove catchpads which were trivially redudant
because they were the same SSA value.  We can do better using the same
algorithm but with a smarter datastructure by hashing the SSA values
within the catchpad and comparing them structurally.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256815 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SimplifyCFG] Remove redundant catchpads
David Majnemer [Tue, 5 Jan 2016 06:27:50 +0000 (06:27 +0000)]
[SimplifyCFG] Remove redundant catchpads

Remove duplicate catchpad handlers from a catchswitch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256814 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU: Remove redundant let mayLoad = 1
Matt Arsenault [Tue, 5 Jan 2016 04:50:28 +0000 (04:50 +0000)]
AMDGPU: Remove redundant let mayLoad = 1

This is already set on the SMRD format class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256813 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[RS4GC] Simplify handling of Constants in findBaseDefiningValue(). NFC.
Manuel Jacob [Tue, 5 Jan 2016 04:06:21 +0000 (04:06 +0000)]
[RS4GC] Simplify handling of Constants in findBaseDefiningValue().  NFC.

Previously there were three conditionals, checking for global
variables, undef values and everything constant except these two, all three
returning the same value.  This commit replaces them by one conditional.

Reviewers: reames

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D15818

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256812 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Statepoints] Refactor GCRelocateOperands into an intrinsic wrapper. NFC.
Manuel Jacob [Tue, 5 Jan 2016 04:03:00 +0000 (04:03 +0000)]
[Statepoints] Refactor GCRelocateOperands into an intrinsic wrapper.  NFC.

This commit renames GCRelocateOperands to GCRelocateInst and makes it an
intrinsic wrapper, similar to e.g. MemCpyInst.  Also, all users of
GCRelocateOperands were changed to use the new intrinsic wrapper instead.

Reviewers: sanjoy, reames

Subscribers: reames, sanjoy, llvm-commits

Differential Revision: http://reviews.llvm.org/D15762

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256811 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU/SI: Select non-uniform constant addrspace loads to flat instructions for HSA
Tom Stellard [Tue, 5 Jan 2016 03:40:16 +0000 (03:40 +0000)]
AMDGPU/SI: Select non-uniform constant addrspace loads to flat instructions for HSA

Summary: This fixes a regression caused by r256282.

Reviewers: arsenm, cfang

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D15736

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256810 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[WinEH] Simplify unreachable catchpads
Joseph Tremoulet [Tue, 5 Jan 2016 02:37:41 +0000 (02:37 +0000)]
[WinEH] Simplify unreachable catchpads

At least for CoreCLR, a catchpad which immediately executes an
`unreachable` instruction indicates that the exception can never have a
matching type, and so such catchpads can be removed, and so can their
catchswitches if the catchswitch becomes empty.

Reviewers: rnk, andrew.w.kaylor, majnemer

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D15846

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256809 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "[X86] Use push-pop for materializing small constants under 'minsize'"
David Majnemer [Tue, 5 Jan 2016 02:32:06 +0000 (02:32 +0000)]
Revert "[X86] Use push-pop for materializing small constants under 'minsize'"

The red zone consists of 128 bytes beyond the stack pointer so that the
allocation of objects in leaf functions doesn't require decrementing
rsp.  In r255656, we introduced an optimization that would cheaply
materialize certain constants via push/pop.  Push decrements the stack
pointer and stores it's result at what is now the top of the stack.
However, this means that using push/pop would encroach on the red zone.
PR26023 gives an example where this corrupts an object in the red zone.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256808 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU/SI: Consolidate FLAT patterns
Tom Stellard [Tue, 5 Jan 2016 02:26:37 +0000 (02:26 +0000)]
AMDGPU/SI: Consolidate FLAT patterns

We had to sets of identical FLAT patterns one inside the
HasFlatAddressSpace predicate and one inside the useFlatForGloabl
predicate.  This patch merges these sets into a single pattern
under the isCIVI predicate.

The reason we can remove the predicates is that when MUBUF instructions
are legal, the instruction selector will prefer selecting those over
FLAT instructions because MUBUF patterns have a higher complexity score.
So, in this case having patterns for FLAT instructions will have no effect.

This change also simplifies the process for forcing global address space
loads to use FLAT instructions, since we no only have to disable the
MUBUF patterns instead of having to disable the MUBUF patterns and
enable the FLAT patterns.

Reviewers: arsenm, cfang

Subscribers: llvm-commits

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256807 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[MDA] Don't be quite as conservative for noalias functions
Philip Reames [Tue, 5 Jan 2016 00:49:14 +0000 (00:49 +0000)]
[MDA] Don't be quite as conservative for noalias functions

If we encounter a noalias call that alias analysis can't analyse, we can fall down into the generic call handling rather than giving up entirely. I noticed this while reading through the code for another purpose.

I can't seem to write a test case which changes; that sorta makes sense given any test case would have to be an inconsistency in AA. Suggestions welcome.

Differential Revision: http://reviews.llvm.org/D15825

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256802 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoX86: Add a testcase for PR25951
Matthias Braun [Tue, 5 Jan 2016 00:48:16 +0000 (00:48 +0000)]
X86: Add a testcase for PR25951

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256801 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoMachineInstrBundle: Fix reversed isSuperRegisterEq() call
Matthias Braun [Tue, 5 Jan 2016 00:45:35 +0000 (00:45 +0000)]
MachineInstrBundle: Fix reversed isSuperRegisterEq() call

Unfortunately this fix had the effect of exposing the
-verify-machineinstrs FIXME of X86InstrInfo.cpp in two testcases for
which I disabled it for now.
Two testcases also have additional pushq/popq where the corrected code
cannot prove that %rax is dead any longer. Looking at the examples, this
could potentially be fixed by improving computeRegisterLiveness() to check
the live-in lists of the successors blocks when reaching the end of a

This fixes http://llvm.org/PR25951.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256799 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix typo in comment
Matthias Braun [Tue, 5 Jan 2016 00:45:31 +0000 (00:45 +0000)]
Fix typo in comment

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256798 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdd explicit string checks in test
Xinliang David Li [Mon, 4 Jan 2016 23:59:14 +0000 (23:59 +0000)]
Add explicit string checks in test

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256796 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU: add +xnack feature
Nicolai Haehnle [Mon, 4 Jan 2016 23:35:53 +0000 (23:35 +0000)]
AMDGPU: add +xnack feature

Enabling this feature will account for the two SGPRs used by the hardware
to store the XNACK_MASK physically.

The hardware only requires this reservation when the XNACK feature is
explicitly enabled. At some point, HSA will probably want to do that, but
it does increase SGPR register pressure, so leave it disabled by default
for now (but do add a small test).

Reviewers: arsenm, tstellarAMD

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D15869

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256794 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstructionCombining] prepareICWorklistFromFunction halts in infinite loop with...
Chen Li [Mon, 4 Jan 2016 23:28:57 +0000 (23:28 +0000)]
[InstructionCombining] prepareICWorklistFromFunction halts in infinite loop with instructions of token type

Summary: This patch fixes a bug in prepareICWorklistFromFunction, where the loop becomes infinite with instructions of token type. The patch checks if the instruction is token type, and if so it updates EndInst with the current instruction.

Reviewers: reames, majnemer

Subscribers: llvm-commits, sanjoy

Differential Revision: http://reviews.llvm.org/D15859

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256792 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoUpdate docs to recommend CMake >= v3.2.
Eric Christopher [Mon, 4 Jan 2016 23:22:43 +0000 (23:22 +0000)]
Update docs to recommend CMake >= v3.2.

CMake v3.2 or newer is necessary to get interactive output when running
Lit via Ninja. Otherwise Ninja will buffer Lit's output, which makes
for a crummy experience -- you can't tell if your tests are hung!

Patch by Justin Lebar!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256791 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoClarify that the bypassSlowDivision optimization operates on a single BB [v2]
Eric Christopher [Mon, 4 Jan 2016 23:18:58 +0000 (23:18 +0000)]
Clarify that the bypassSlowDivision optimization operates on a single BB [v2]

Update some comments to be more explicit.

Change bypassSlowDivision and the functions it calls so that they take
BasicBlock*s and Instruction*s, rather than Function::iterator&s and

Change the APIs so that the caller is responsible for updating the
iterator, rather than the callee. This makes control flow much easier
to follow.

Patch by Justin Lebar!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256789 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LICM] Fix a small oversight introduced in r256763
David Majnemer [Mon, 4 Jan 2016 23:16:22 +0000 (23:16 +0000)]
[LICM] Fix a small oversight introduced in r256763

r256763 had promoteLoopAccessesToScalars check for the existence of a
catchswitch when the exit blocks were populated but
promoteLoopAccessesToScalars may be called with a prepopulated set of
exit blocks which would also need to be checked.

This fixes PR26019.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256788 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[MemoryBuiltins] Remove isOperatorNewLike by consolidating non-null inference handling
Philip Reames [Mon, 4 Jan 2016 22:49:23 +0000 (22:49 +0000)]
[MemoryBuiltins] Remove isOperatorNewLike by consolidating non-null inference handling

This patch removes the isOperatorNewLike predicate since it was only being used to establish a non-null return value and we have attributes specifically for that purpose with generic handling. To keep approximate the same behaviour for existing frontends, I added the various operator new like (i.e. instances of operator new) to InferFunctionAttrs. It's not really clear to me why this isn't handled in Clang, but I didn't want to break existing code and any subtle assumptions it might have.

Once this patch is in, I'm going to start separating the isAllocLike family of predicates. These appear to be being used for a mixture of things which should be more clearly separated and documented. Today, they're being used to indicate (at least) aliasing facts, CSE-ability, and default values from an allocation site.

Differential Revision: http://reviews.llvm.org/D15820

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256787 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PGO] Simplify string parsing
Xinliang David Li [Mon, 4 Jan 2016 22:09:26 +0000 (22:09 +0000)]
[PGO] Simplify string parsing

Patch Suggested by Vedant.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256785 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PGO] Refactor string writer code
Xinliang David Li [Mon, 4 Jan 2016 22:01:02 +0000 (22:01 +0000)]
[PGO] Refactor string writer code

For readability and code sharing.
(Adapted from Suggestions by Vedant).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256784 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LIR] General refactoring to simplify code and the ease future code review
Haicheng Wu [Mon, 4 Jan 2016 21:43:14 +0000 (21:43 +0000)]
[LIR] General refactoring to simplify code and the ease future code review

This is a resubmission of r256336 which was reverted in r256361. The issue was the lack of the invariant check of the memset value in processLooMemSet().

The original message:

Move several checks into isLegalStores. Also, delineate between those stores that are memset-able and those that are memcpy-able.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256783 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][SSE] Ensure BLENDPD/BLENDPS/PBLEND inputs are both of the correct input type
Simon Pilgrim [Mon, 4 Jan 2016 21:41:11 +0000 (21:41 +0000)]
[X86][SSE] Ensure BLENDPD/BLENDPS/PBLEND inputs are both of the correct input type

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256782 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PGO]: Use efficient 'join' API for uncompressed string
Xinliang David Li [Mon, 4 Jan 2016 21:31:09 +0000 (21:31 +0000)]
[PGO]: Use efficient 'join' API for uncompressed string

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256781 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PGO]: reserve space for string to avoid excessive memory realloc/copy (non linear)
Xinliang David Li [Mon, 4 Jan 2016 20:26:05 +0000 (20:26 +0000)]
[PGO]: reserve space for string to avoid excessive memory realloc/copy (non linear)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256776 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU/SI: Move VI SMEM pattern back into VIInstructions.td
Tom Stellard [Mon, 4 Jan 2016 20:23:10 +0000 (20:23 +0000)]
AMDGPU/SI: Move VI SMEM pattern back into VIInstructions.td

Summary: This was accidently moved to CIInstructions.td in r256282

Reviewers: cfang, arsenm

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D15763

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256775 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoUpdate documantation
Xinliang David Li [Mon, 4 Jan 2016 20:00:47 +0000 (20:00 +0000)]
Update documantation

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256774 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRemove dead instructions before Redoing
Aditya Nandakumar [Mon, 4 Jan 2016 19:48:14 +0000 (19:48 +0000)]
Remove dead instructions before Redoing

Before reevaluating instructions, iterate over all instructions
to be reevaluated and remove trivially dead instructions and if
any of it's operands become trivially dead, mark it for deletion
until all trivially dead instructions have been removed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256773 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert previous test commit.
Haicheng Wu [Mon, 4 Jan 2016 19:13:29 +0000 (19:13 +0000)]
Revert previous test commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256768 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoThis is a test commit to check my commit access works.
Haicheng Wu [Mon, 4 Jan 2016 19:12:36 +0000 (19:12 +0000)]
This is a test commit to check my commit access works.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256767 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AArch64] Optimize some simple TBZ/TBNZ cases.
Geoff Berry [Mon, 4 Jan 2016 18:55:47 +0000 (18:55 +0000)]
[AArch64] Optimize some simple TBZ/TBNZ cases.

Add some AArch64 dag combines to optimize some simple TBZ/TBNZ cases:

 (tbz (and x, m), b) -> (tbz x, b)
 (tbz (shl x, c), b) -> (tbz x, b-c)
 (tbz (shr x, c), b) -> (tbz x, b+c)
 (tbz (xor x, -1), b) -> (tbnz x, b)

Reviewers: jmolloy, mcrosier, t.p.northover

Subscribers: aemerson, rengolin, llvm-commits

Differential Revision: http://reviews.llvm.org/D15702

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256765 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoClang-format my previous change (r256313)
Paul Robinson [Mon, 4 Jan 2016 18:49:15 +0000 (18:49 +0000)]
Clang-format my previous change (r256313)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256764 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LICM] Don't insert instructions after a catchswitch when performing loop promotion
David Majnemer [Mon, 4 Jan 2016 17:42:19 +0000 (17:42 +0000)]
[LICM] Don't insert instructions after a catchswitch when performing loop promotion

Inserting after a catchswitch results in verifier errors, bail out on
promotion if a catchswitch is a loop exit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256763 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix comment in typo. NFC
Nick Lewycky [Mon, 4 Jan 2016 16:44:44 +0000 (16:44 +0000)]
Fix comment in typo. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256761 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[WinEH] Update CoreCLR EH state numbering
Joseph Tremoulet [Mon, 4 Jan 2016 16:16:01 +0000 (16:16 +0000)]
[WinEH] Update CoreCLR EH state numbering

Fix the CLR state numbering to generate correct tables, and update the lit
test to verify them.

The CLR numbering assigns one state number to each catchpad and

It also computes two tree-like relations over states:
 1) Each state has a "HandlerParentState", which is the state of the next
    outer handler enclosing this state's handler (same as nearest ancestor
    per the ParentPad linkage on EH pads, but skipping over catchswitches).
 2) Each state has a "TryParentState", which:
    a) for a catchpad that's not the last handler on its catchswitch, is
       the state of the next catchpad on that catchswitch.
    b) for all other pads, is the state of the pad whose try region is the
       next outer try region enclosing this state's try region.  The "try
       regions are not present as such in the IR, but will be inferred
       based on the placement of invokes and pads which reach each other
       by exceptional exits.

Catchswitches do not get their own states, but each gets mapped to the
state of its first catchpad.

Table generation requires each state's "unwind dest" state to have a lower
state number than the given state.

Since HandlerParentState can be computed as a function of a pad's
ParentPad, and TryParentState can be computed as a function of its unwind
dest and the TryParentStates of its children, the CLR state numbering
algorithm first computes HandlerParentState in a top-down pass, then
computes TryParentState in a bottom-up pass.

Also reword some comments/names in the CLR EH table generation to make the
distinction between the different kinds of "parent" clear.

Reviewers: rnk, andrew.w.kaylor, majnemer

Subscribers: AndyAyers, llvm-commits

Differential Revision: http://reviews.llvm.org/D15325

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256760 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU: Avoid assertions after SGPR spilling failed
Nicolai Haehnle [Mon, 4 Jan 2016 15:50:01 +0000 (15:50 +0000)]
AMDGPU: Avoid assertions after SGPR spilling failed

The comment explains it: emitError does not necessarily exit the compilation
process, and then using NoRegister leads to assertions later on.
This generates incorrect code, of course, but the user should know to not use
the result when an error has been emitted.

It would be nice to have a test-case for this inside the LLVM repository,
but llc exits on error. shader-db tests trigger the underlying issue at least
on Tonga.

Reviewers: arsenm, tstellarAMD, mareko

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D15826

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256757 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVX512] add PSRAD and PSRAQ Intrinsic
Michael Zuckerman [Mon, 4 Jan 2016 13:45:45 +0000 (13:45 +0000)]
[AVX512] add PSRAD and PSRAQ Intrinsic

Differential Revision: http://reviews.llvm.org/D15851

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256754 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVX512] add PSRAW Intrinsic
Michael Zuckerman [Mon, 4 Jan 2016 12:50:36 +0000 (12:50 +0000)]
[AVX512] add PSRAW Intrinsic

Differential Revision: http://reviews.llvm.org/D15850

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256751 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[MC] Fix file name in file header
Jeroen Ketema [Mon, 4 Jan 2016 12:22:34 +0000 (12:22 +0000)]
[MC] Fix file name in file header

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256749 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVX512] add PSRLV Intrinsic
Michael Zuckerman [Mon, 4 Jan 2016 11:39:06 +0000 (11:39 +0000)]
[AVX512] add PSRLV Intrinsic

Differential Revision: http://reviews.llvm.org/D15838

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256747 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix a horrible infloop in value tracking in the face of dead code.
Chandler Carruth [Mon, 4 Jan 2016 07:23:12 +0000 (07:23 +0000)]
Fix a horrible infloop in value tracking in the face of dead code.

Amazingly, we just never triggered this without:
1) Moving code around for MetadataTracking so that a certain *different*
   amount of inlining occurs in the per-TU compile step.
2) Then you LTO opt or clang with a bootstrap, and get inlining, loop
   opts, and GVN line up everything *just* right.

I don't really know how we didn't hit this before. We really need to be
fuzz testing stuff, it shouldn't be hard to trigger. I'm working on
crafting a reduced nice test case, and will submit that when I have it,
but I want to get LTO build bots going again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256735 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Fix a typo in r256733.
Craig Topper [Mon, 4 Jan 2016 06:35:08 +0000 (06:35 +0000)]
[TableGen] Fix a typo in r256733.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256734 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Use some free space in Init to store the opcode for UnOpInit/BinOpInit...
Craig Topper [Mon, 4 Jan 2016 06:28:49 +0000 (06:28 +0000)]
[TableGen] Use some free space in Init to store the opcode for UnOpInit/BinOpInit/TernOpInit allowing those types to be a little smaller. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256733 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Call llvm_shutdown on exit so that all the ManagedStatic objects in the...
Craig Topper [Mon, 4 Jan 2016 04:51:51 +0000 (04:51 +0000)]
[TableGen] Call llvm_shutdown on exit so that all the ManagedStatic objects in the support library will be deleted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256732 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Make hasFP constant time
David Majnemer [Mon, 4 Jan 2016 04:49:41 +0000 (04:49 +0000)]
[X86] Make hasFP constant time

We need a frame pointer if there is a push/pop sequence after the
prologue in order to unwind the stack.  Scanning the instructions to
figure out if this happened made hasFP not constant-time which is a
violation of expectations.  Let's compute this up-front and reuse that
computation when we need it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256730 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LICM] Make instruction sinking funclet-aware
David Majnemer [Mon, 4 Jan 2016 03:37:39 +0000 (03:37 +0000)]
[LICM] Make instruction sinking funclet-aware

We had two bugs here:
- We might try to sink into a catchswitch, causing verifier failures.
- We will succeed in sinking into a cleanuppad but we didn't update the
  funclet operand bundle.

This fixes PR26000.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256728 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Change TGParser::SetValue to take an ArrayRef instead of std::vector refer...
Craig Topper [Mon, 4 Jan 2016 03:15:08 +0000 (03:15 +0000)]
[TableGen] Change TGParser::SetValue to take an ArrayRef instead of std::vector reference. Use None in many places where a default constructed vector was being passed. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256726 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Fix a bug that caused the wrong name for a record built from a multiclass...
Craig Topper [Mon, 4 Jan 2016 03:05:14 +0000 (03:05 +0000)]
[TableGen] Fix a bug that caused the wrong name for a record built from a multiclass containing a defm called NAME that references another multiclass that contains a defm that uses NAME concatenated with other strings.

It would end up doing the concatenations from the second multiclass twice. This occured because SetValue detected a self assignment when trying to set the value of NAME to a VarInit called NAME. NAME is special here and it will get cleaned up later. So add a flag to suppress the self assignment check for this case.

Strangely the self-assignment error was returning false indicating it wasn't an error, but it wasn't doing the right thing. So this also changes it to report an error.

This fixes the names of some AVX512 FMA instructions that showed this double expansion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256725 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agollvm/include/llvm/ProfileData/InstrProfData.inc: Add c++-mode in the header.
NAKAMURA Takumi [Mon, 4 Jan 2016 00:47:33 +0000 (00:47 +0000)]
llvm/include/llvm/ProfileData/InstrProfData.inc: Add c++-mode in the header.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256721 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix one file that I didn't convert properly in r256707.
Dimitry Andric [Sun, 3 Jan 2016 22:33:32 +0000 (22:33 +0000)]
Fix one file that I didn't convert properly in r256707.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256720 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoUse std::is_sorted and std::none_of instead of manual loops. NFC
Craig Topper [Sun, 3 Jan 2016 19:43:40 +0000 (19:43 +0000)]
Use std::is_sorted and std::none_of instead of manual loops. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256719 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PGO] Cleanup: remove reduncant calls in lowering
Xinliang David Li [Sun, 3 Jan 2016 19:38:51 +0000 (19:38 +0000)]
[PGO] Cleanup: remove reduncant calls in lowering

CoverageMapping data's section and alignment is
already set during creation. No need to call it again
during lowering.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256716 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PGO] Cleanup: Use covmap header definition in the template file
Xinliang David Li [Sun, 3 Jan 2016 19:26:07 +0000 (19:26 +0000)]
[PGO] Cleanup: Use covmap header definition in the template file

This is one last remaining instrumentatation related structure
that needs to be migrate to use the centralized template
definition.  With this change, instrumentation code
related to coverage module header will be kept in sync
with the coverage mapping reader. The remaining code
which makes implicit assumption about covmap control
structure layout in the the lowering pass will cleaned
up in a different patch. This patch is not intended to
have no functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256715 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][MMX] Regenerated vector insertion test.
Simon Pilgrim [Sun, 3 Jan 2016 19:17:37 +0000 (19:17 +0000)]
[X86][MMX] Regenerated vector insertion test.

Shows the true horror of what is going on....

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256713 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PGO] Code refactoring to use header struct def /NFC
Xinliang David Li [Sun, 3 Jan 2016 18:57:40 +0000 (18:57 +0000)]
[PGO] Code refactoring to use header struct def /NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256712 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PGO] Introduce coverage map header structure in PGO data template file /NFC
Xinliang David Li [Sun, 3 Jan 2016 18:35:56 +0000 (18:35 +0000)]
[PGO] Introduce coverage map header structure in PGO data template file /NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256710 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SelectionDAG] Pulled out common code for CONCAT_VECTORS node creation
Simon Pilgrim [Sun, 3 Jan 2016 18:24:19 +0000 (18:24 +0000)]
[SelectionDAG] Pulled out common code for CONCAT_VECTORS node creation

Pulled out the similar CONCAT_VECTORS creation code from the 2/3 operand getNode() calls (to handle all UNDEF and all BUILD_VECTOR cases). Added a similar handler to the general getNode() call as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256709 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][SSE] Added tests for insertion of zero elements into vectors
Simon Pilgrim [Sun, 3 Jan 2016 17:33:32 +0000 (17:33 +0000)]
[X86][SSE] Added tests for insertion of zero elements into vectors

Many of these could be much better if we just lowered them all as shuffles - especially for the 256-bit vectors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256708 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix several accidental DOS line endings in source files
Dimitry Andric [Sun, 3 Jan 2016 17:22:03 +0000 (17:22 +0000)]
Fix several accidental DOS line endings in source files

There are a number of files in the tree which have been accidentally checked in with DOS line endings.  Convert these to native line endings.

There are also a few files which have DOS line endings on purpose, and I have set the svn:eol-style property to 'CRLF' on those.

Reviewers: joerg, aaron.ballman

Subscribers: aaron.ballman, sanjoy, dsanders, llvm-commits

Differential Revision: http://reviews.llvm.org/D15848

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256707 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][SSE41] Added test cases for improving insertps shuffles
Simon Pilgrim [Sun, 3 Jan 2016 17:14:15 +0000 (17:14 +0000)]
[X86][SSE41] Added test cases for improving insertps shuffles

As mentioned on D14261, an upcoming patch will improve combines of insertps instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256706 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][SSE] Added v4f32 shuffle with zero tests
Simon Pilgrim [Sun, 3 Jan 2016 17:02:56 +0000 (17:02 +0000)]
[X86][SSE] Added v4f32 shuffle with zero tests

This is mainly test cases for improvements to insertps matching, but pre-SSE41 shuffles could be improved as well

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256705 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Simplify some code slightly. No need to check if the arrays are empty...
Craig Topper [Sun, 3 Jan 2016 08:57:41 +0000 (08:57 +0000)]
[TableGen] Simplify some code slightly. No need to check if the arrays are empty before printing. The loop can be made to print the same thing if the loop is empty. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256703 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoUse an ArrayRef to simplify repeated calculation of the array end. NFC
Craig Topper [Sun, 3 Jan 2016 08:45:36 +0000 (08:45 +0000)]
Use an ArrayRef to simplify repeated calculation of the array end. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256702 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoUse std::is_sorted instead of manual loops. NFC
Craig Topper [Sun, 3 Jan 2016 07:33:45 +0000 (07:33 +0000)]
Use std::is_sorted instead of manual loops. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256701 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoUse range-based for loop. NFC
Craig Topper [Sun, 3 Jan 2016 07:33:42 +0000 (07:33 +0000)]
Use range-based for loop. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256700 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Replace a logically negated xor of bools with just an equality comparison...
Craig Topper [Sun, 3 Jan 2016 07:33:39 +0000 (07:33 +0000)]
[TableGen] Replace a logically negated xor of bools with just an equality comparison for readability. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256699 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Use std::find_if and a lambda instead of manual loops.
Craig Topper [Sun, 3 Jan 2016 07:33:36 +0000 (07:33 +0000)]
[TableGen] Use std::find_if and a lambda instead of manual loops.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256698 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Fix a bug introduced in r256627. If the switch was not emitted we still...
Craig Topper [Sun, 3 Jan 2016 07:33:34 +0000 (07:33 +0000)]
[TableGen] Fix a bug introduced in r256627. If the switch was not emitted we still emitted a closing curly brace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256697 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Use range-based for loops. NFC
Craig Topper [Sun, 3 Jan 2016 07:33:30 +0000 (07:33 +0000)]
[TableGen] Use range-based for loops. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256696 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PGO] simple refactoring (NFC)
Xinliang David Li [Sun, 3 Jan 2016 04:38:13 +0000 (04:38 +0000)]
[PGO] simple refactoring (NFC)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256695 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoWinEHPrepare.cpp: Suppress a warning for -Asserts. [-Wunused-variable]
NAKAMURA Takumi [Sun, 3 Jan 2016 01:41:00 +0000 (01:41 +0000)]
WinEHPrepare.cpp: Suppress a warning for -Asserts. [-Wunused-variable]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256694 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Verifier] Add braces to satisfy buildbots. NFC
Joseph Tremoulet [Sat, 2 Jan 2016 15:50:34 +0000 (15:50 +0000)]
[Verifier] Add braces to satisfy buildbots. NFC

Fix build break introduced by r256691.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256692 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[WinEH] Verify catchswitch handlers
Joseph Tremoulet [Sat, 2 Jan 2016 15:25:25 +0000 (15:25 +0000)]
[WinEH] Verify catchswitch handlers

The handler list must be nonempty and consist solely of CatchPads.

Reviewers: rnk, andrew.w.kaylor, majnemer

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D15842

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256691 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[WinEH] Tighten parentPad verifier checks
Joseph Tremoulet [Sat, 2 Jan 2016 15:24:24 +0000 (15:24 +0000)]
[WinEH] Tighten parentPad verifier checks

Summary: A catchswitch cannot be a parent of a cleanuppad or another catchswitch.

Reviewers: rnk, andrew.w.kaylor, majnemer

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D15841

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256690 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[WinEH] Update catchrets with cloned successors
Joseph Tremoulet [Sat, 2 Jan 2016 15:22:36 +0000 (15:22 +0000)]
[WinEH] Update catchrets with cloned successors

Add a pass to update catchrets when their successors get cloned; the
existing pass doesn't catch these because it walks the funclet whose
blocks are being cloned but the catchret is in a child funclet.

Also update the test for removing incoming PHI values; when the
predecessor is a catchret, the relevant color is the catchret's parentPad,
not its block's color.

Reviewers: andrew.w.kaylor, rnk, majnemer

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D15840

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256689 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoCorrect misleading formatting of several ifs followed by two statements without braces.
Yaron Keren [Sat, 2 Jan 2016 13:40:36 +0000 (13:40 +0000)]
Correct misleading formatting of several ifs followed by two statements without braces.
While the original code would work with or without braces, it makes sense to
set HaveSemi to true only if (!HaveSemi), otherwise it's already true, so I
put the assignment inside the if block. This addresses PR25998.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256688 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[WinEH] Add additional verification
David Majnemer [Sat, 2 Jan 2016 09:26:36 +0000 (09:26 +0000)]
[WinEH] Add additional verification

Recolor the IR to make sure our computed colors are not hiding any bugs.
Also, verifyFunction if we are running some post-preparation operations;
some of these operations can hide latent bugs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256687 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Add intrinsics for reading and writing to the flags register
David Majnemer [Fri, 1 Jan 2016 06:50:01 +0000 (06:50 +0000)]
[X86] Add intrinsics for reading and writing to the flags register

LLVM's targets need to know if stack pointer adjustments occur after the
prologue.  This is needed to correctly determine if the red-zone is
appropriate to use or if a frame pointer is required.

Normally, LLVM can figure this out very precisely by reasoning about the
contents of the MachineFunction.  There is an interesting corner case:
inline assembly.

The vast majority of inline assembly which will perform a push or pop is
done so to pair up with pushf or popf as appropriate.  Unfortunately,
this inline assembly doesn't mark the stack pointer as clobbered
because, well, it isn't.  The stack pointer is decremented and then
immediately incremented.  Because of this, LLVM was changed in r256456
to conservatively assume that inline assembly contain a sequence of
stack operations.  This is unfortunate because the vast majority of
inline assembly will not end up manipulating the stack pointer in any
way at all.

Instead, let's provide a more principled solution: an intrinsic.
FWIW, other compilers (MSVC and GCC among them) also provide this
functionality as an intrinsic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256685 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LibCallSimplifier] propagate FMF when shrinking binary calls
Sanjay Patel [Thu, 31 Dec 2015 23:40:59 +0000 (23:40 +0000)]
[LibCallSimplifier] propagate FMF when shrinking binary calls

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256682 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Remove a return after llvm_unreachable.
Craig Topper [Thu, 31 Dec 2015 22:40:48 +0000 (22:40 +0000)]
[X86] Remove a return after llvm_unreachable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256681 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Move shuffle decoding for constant pool into the X86CodeGen library to remove...
Craig Topper [Thu, 31 Dec 2015 22:40:45 +0000 (22:40 +0000)]
[X86] Move shuffle decoding for constant pool into the X86CodeGen library to remove a layering violation in the Util library.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256680 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LibCallSimplifier] propagate FMF when shrinking unary calls
Sanjay Patel [Thu, 31 Dec 2015 21:52:31 +0000 (21:52 +0000)]
[LibCallSimplifier] propagate FMF when shrinking unary calls

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256679 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agochange function names to avoid accidentally matching the substring
Sanjay Patel [Thu, 31 Dec 2015 21:25:25 +0000 (21:25 +0000)]
change function names to avoid accidentally matching the substring

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256678 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoadd 'fast' attribute to calls to show that the flag isn't being propagated
Sanjay Patel [Thu, 31 Dec 2015 21:12:19 +0000 (21:12 +0000)]
add 'fast' attribute to calls to show that the flag isn't being propagated

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256677 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoVariable names start with an upper case letter; NFC
Sanjay Patel [Thu, 31 Dec 2015 16:16:58 +0000 (16:16 +0000)]
Variable names start with an upper case letter; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256676 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agofix formatting; NFC
Sanjay Patel [Thu, 31 Dec 2015 16:10:49 +0000 (16:10 +0000)]
fix formatting; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256675 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoadd FMF for CreateCall variant
Sanjay Patel [Thu, 31 Dec 2015 15:39:34 +0000 (15:39 +0000)]
add FMF for CreateCall variant

The version with OpBundles was missed in:

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256674 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVX512] add PSRLQ and PSRLD Intrinsic
Michael Zuckerman [Thu, 31 Dec 2015 15:22:04 +0000 (15:22 +0000)]
[AVX512] add PSRLQ and PSRLD Intrinsic

Differential Revision: http://reviews.llvm.org/D15770

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256673 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Avoid folding scalar loads into unary sse intrinsics
Michael Kuperstein [Thu, 31 Dec 2015 09:45:16 +0000 (09:45 +0000)]
[X86] Avoid folding scalar loads into unary sse intrinsics

Not folding these cases tends to avoid partial register updates:
sqrtss (%eax), %xmm0
Has a partial update of %xmm0, while
movss (%eax), %xmm0
sqrtss %xmm0, %xmm0
Has a clobber of the high lanes immediately before the partial update,
avoiding a potential stall.

Given this, we only want to fold when optimizing for size.
This is consistent with the patterns we already have for some of
the fp/int converts, and in X86InstrInfo::foldMemoryOperandImpl()

Differential Revision: http://reviews.llvm.org/D15741

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256671 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][PKU] Add {RD,WR}PKRU intrinsics
Asaf Badouh [Thu, 31 Dec 2015 08:31:13 +0000 (08:31 +0000)]
[X86][PKU] Add {RD,WR}PKRU intrinsics

Differential Revision: http://reviews.llvm.org/D15808

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256670 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Modify the AsmMatcherEmitter to only apply the table growth from r252440...
Craig Topper [Thu, 31 Dec 2015 08:18:23 +0000 (08:18 +0000)]
[TableGen] Modify the AsmMatcherEmitter to only apply the table growth from r252440 to the Hexagon target.

This restores the previous behavior of not including the mnemonic in the classes table for every target that starts instruction lines with the mnemonic. Not only did the table size increase by 1 entry, but the class enum increased in size which caused every class in the array to increase in size. It also grew the size of the function that parsers tokens into classes by a substantial amount.

This adds a new HasMnemonicFirst flag to all AsmParsers. It's set to 1 by default and Hexagon target overrides it to 0.

For the X86 target alone this recovers 324KB of size on the llvm-mc executable.

I believe the current state is still a bad design choice for the Hexagon target as it causes most of the parsing to do a linear search through the entire match table to comparing operands against every instruction until it finds one that works. At least for the other targets we do a binary search based on mnemonic over which to do the linear scan.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256669 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen] Use range-based for loops. NFC
Craig Topper [Thu, 31 Dec 2015 08:18:20 +0000 (08:18 +0000)]
[TableGen] Use range-based for loops. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256668 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PGO]: Implement Func PGO name string compression
Xinliang David Li [Thu, 31 Dec 2015 07:57:16 +0000 (07:57 +0000)]
[PGO]: Implement Func PGO name string compression

This is part of the effort/prepration to reduce the size
instr-pgo (object, binary, memory footprint, and raw data).

The functionality is currently off by default and not yet
used by any clients.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256667 91177308-0d34-0410-b5e6-96231b3b80d8