GlobalISel: Implement narrowing for G_STORE

Legalize stores of types that are too wide by breaking them up into
sequences of smaller stores.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292412 91177308-0d34-0410-b5e6-96231b3b80d8

GlobalISel: Correct copy-pasted comment. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292411 91177308-0d34-0410-b5e6-96231b3b80d8

Don't create a comdat group for a dropped def with initializer

Non-prevailing weak/linkonce odr symbols will be dropped by ThinLTO to
available_externally when possible. If they had an initializer in the
global_ctors list, a comdat group was being created. This code
already had logic to skip available_externally defs, but now the
EliminateAvailableExternally pass will drop these symbols to
declarations earlier. Change the check to skip all declarations for
linker (which includes available_externally along with declarations).

Reviewers: mehdi_amini

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D28737

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292408 91177308-0d34-0410-b5e6-96231b3b80d8

Revert 292404 due to buildbot failures.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292407 91177308-0d34-0410-b5e6-96231b3b80d8

[X86] Minor code cleanup to fix several clang-tidy warnings. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292404 91177308-0d34-0410-b5e6-96231b3b80d8

[ARM] Create SubtargetFeatures from build attrs

An ELFObjectFile can now create SubtargetFeatures from the available
ARM build attributes, in a similar manner to MIPS. I've moved the
MIPS code into its own function and the ARM handler also has a
separate function.

Differential Revision: https://reviews.llvm.org/D28291

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292403 91177308-0d34-0410-b5e6-96231b3b80d8

raw_fd_ostream: Make file handles non-inheritable by default

Summary:
This makes the file descriptors on unix platform non-inheritable (O_CLOEXEC).

There is no change in behavior on windows, as the handles were already
non-inheritable there.

Reviewers: rnk, rafael

Subscribers: llvm-commits, mgorny

Differential Revision: https://reviews.llvm.org/D28854

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292401 91177308-0d34-0410-b5e6-96231b3b80d8

[Assembler] Fix crash when assembling .quad for AArch32.

A 64-bit relocation does not exist in 32-bit ARMELF. Report an error
instead of crashing.

PR23870
Patch by Sanne Wouda (sanwou01).
Differential Revision: https://reviews.llvm.org/D28851

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292373 91177308-0d34-0410-b5e6-96231b3b80d8

[thumb,framelowering] Reset NoVRegs in Thumb1FrameLowering::emitPrologue.

Summary:
In this function, virtual registers can be introduced (for example
through calls to emitThumbRegPlusImmInReg). doScavengeFrameVirtualRegs
will replace those virtual registers with concrete registers later on
in PrologEpilogInserter, which sets NoVRegs again.

This patch fixes the Codegen/Thumb/segmented-stacks.ll test case which
failed with expensive checks.
https://llvm.org/bugs/show_bug.cgi?id=27484

Reviewers: rnk, bkramer, olista01

Reviewed By: olista01

Subscribers: llvm-commits, rengolin

Differential Revision: https://reviews.llvm.org/D28829

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292372 91177308-0d34-0410-b5e6-96231b3b80d8

[InstCombine][AVX2] Add DemandedElts support for VPERMD/VPERMPS shuffles

Simplify a vpermv shuffle mask based on the elements of the mask that are actually demanded.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292371 91177308-0d34-0410-b5e6-96231b3b80d8

Re-revert: [globalisel] Tablegen-erate current Register Bank Information

More missing guards. My build didn't notice it due to a stale file left over
from a Global ISel build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292369 91177308-0d34-0410-b5e6-96231b3b80d8

[InstCombine][AVX2] Tests showing missed opportunities to pass demanded elts through a vpermd/vpermps shuffle

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292368 91177308-0d34-0410-b5e6-96231b3b80d8

Re-commit: [globalisel] Tablegen-erate current Register Bank Information

Summary:
Adds a RegisterBank tablegen class that can be used to declare the register
banks and an associated tablegen pass to generate the necessary code.

Changes since last commit:
The new tablegen pass is now correctly guarded by LLVM_BUILD_GLOBAL_ISEL and
this should fix the buildbots however it may not be the whole fix. The previous
buildbot failures suggest there may be a memory bug lurking that I'm unable to
reproduce (including when using asan) or spot in the source. If they re-occur
on this commit then I'll need assistance from the bot owners to track it down.

Reviewers: t.p.northover, ab, rovka, qcolombet

Reviewed By: qcolombet

Subscribers: aditya_nandakumar, rengolin, kristof.beyls, vkalintiris, mgorny, dberris, llvm-commits, rovka

Differential Revision: https://reviews.llvm.org/D27338

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292367 91177308-0d34-0410-b5e6-96231b3b80d8

[ARM] Create objdump subtarget from build attrs

Enable an ELFObjectFile to read the its arm build attributes to
produce a target triple with a specific ARM architecture.
llvm-objdump now uses this functionality to automatically produce
a more accurate target.

Differential Revision: https://reviews.llvm.org/D28769

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292366 91177308-0d34-0410-b5e6-96231b3b80d8

[InstCombine] Remove unnecessary intrinsics demanded elts handling

As discussed on D28777 - we don't need to handle 'all element' shuffles inside InstCombiner::visitCallInst as InstCombiner::SimplifyDemandedVectorElts will do everything we need.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292365 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed parser error on windows shell evaluation of RUN script line

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292363 91177308-0d34-0410-b5e6-96231b3b80d8

[X86][SSE] Simplify umax knownbits test

combineSRA doesn't detect sign bits splats that it does itself so just use -1 as the demanded input so that its already splatted

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292361 91177308-0d34-0410-b5e6-96231b3b80d8

[X86] Improve mul combine for negative multiplayer (2^c - 1)

This patch improves the mul instruction combine function (combineMul)
by adding new layer of logic.
In this patch, we are adding the ability to fold (mul x, -((1 << c) -1))
or (mul x, -((1 << c) +1)) into (neg(X << c) -x) or (neg((x << c) + x) respective.

Differential Revision: https://reviews.llvm.org/D28232

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292358 91177308-0d34-0410-b5e6-96231b3b80d8

Revert "[XRay][Arm] Repair XRay table emission on Arm32 and add tests to identify such problem earlier"

This reverts commit r292210, as it broke the Thumb buldbot with:

clang-5.0: error: the clang compiler does not support '-fxray-instrument
on thumbv7-unknown-linux-gnueabihf'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292357 91177308-0d34-0410-b5e6-96231b3b80d8

[SystemZ] Proper handling of undef flag while expanding pseudo.

During post-RA pseudo expansion, an 'undef' flag of the source operand should
be propagated by emitGRX32Move().

Review: Ulrich Weigand

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292353 91177308-0d34-0410-b5e6-96231b3b80d8

[X86] Fix for bugzilla 31576 - add support for "data32" instruction prefix

This patch fixes bugzilla 31576 (https://llvm.org/bugs/show_bug.cgi?id=31576).

"data32" instruction prefix was not defined in the llvm.
An exception had to be added to the X86 tablegen and AsmPrinter because both "data16" and "data32" are encoded to 0x66 (but in different modes).

Differential Revision: https://reviews.llvm.org/D28468

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292352 91177308-0d34-0410-b5e6-96231b3b80d8

[LoopDeletion] (cleanup, NFC) Fix one more local variable that didn't
follow LLVM's naming conventions while I'm here.

Again, sorry I didn't spot this earlier to coalesce with other cleanup
changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292333 91177308-0d34-0410-b5e6-96231b3b80d8

[PM] Teach LoopDeletion to correctly update the LPM when loops are
deleted.

I've expanded its test coverage a bit including adding one test that
will crash clearly without this change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292332 91177308-0d34-0410-b5e6-96231b3b80d8

[LoopDeletion] (cleanup, NFC) Make this test actually test what it
claims to test.

LoopSimplify was unifying the multiple exits in this test case, making
it never even test the multiple exit handling of LoopDeletion. Doh.

Now it works (thanks to a great idea from mkuper) and will fail if we
ever change something to make it stop working.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292331 91177308-0d34-0410-b5e6-96231b3b80d8

DAG: Consider nnan in isKnownNeverNaN

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292328 91177308-0d34-0410-b5e6-96231b3b80d8

Revert rL292292 since it causes a SEGV on sanitizer-x86_64-linux-fuzzer build bot.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292327 91177308-0d34-0410-b5e6-96231b3b80d8

[libFuzzer] remove stale code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292325 91177308-0d34-0410-b5e6-96231b3b80d8

[test-release.sh] Add Polly to the list of projects

Reviewers: zinob, hans, grosser

Reviewed By: hans, grosser

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D28712

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292323 91177308-0d34-0410-b5e6-96231b3b80d8

[WebAssembly] Update grow_memory's return type.

The grow_memory instruction now returns the previous memory size. Add the
return type to the LLVM intrinsic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292322 91177308-0d34-0410-b5e6-96231b3b80d8

MIRParser: Allow regclass specification on operand

You can now define the register class of a virtual register on the
operand itself avoiding the need to use a "registers:" block.

Example: "%0:gr64 = COPY %rax"

Differential Revision: https://reviews.llvm.org/D22398

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292321 91177308-0d34-0410-b5e6-96231b3b80d8

[Target, Transforms] Fix some Clang-tidy modernize and Include What You Use warnings; other minor fixes (NFC).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292320 91177308-0d34-0410-b5e6-96231b3b80d8

[libFuzzer] exit(1) on failed merge

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292319 91177308-0d34-0410-b5e6-96231b3b80d8

[libFuzzer] better link for trophies

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292318 91177308-0d34-0410-b5e6-96231b3b80d8

[NVPTX] Support global variables of integer type larger than i64.

Reviewers: tra, majnemer

Subscribers: llvm-commits, jholewinski

Differential Revision: https://reviews.llvm.org/D28825

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292316 91177308-0d34-0410-b5e6-96231b3b80d8

2 returns next to each other =). NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292315 91177308-0d34-0410-b5e6-96231b3b80d8

Skip loop header while we can when computing loop safety info

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292310 91177308-0d34-0410-b5e6-96231b3b80d8

[LIT] Make util.executeCommand python3 friendly

Summary: The parameter `input` to `subprocess.Popen.communicate(...)` must be an object of type `bytes` . This is strictly enforced in python3. This patch (1) allows `to_bytes` to be safely called redundantly. (2) Explicitly convert `input` within `executeCommand`. This allows for usages like `executeCommand(['clang++', '-'], input='int main() {}\n')`.

Reviewers: ddunbar, BinaryKhaos, modocache, dim, EricWF

Reviewed By: EricWF

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D28736

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292308 91177308-0d34-0410-b5e6-96231b3b80d8

[NVPTX] Standardize asm printer on "foo \tbar".

Some instructions were printed as "foo\tbar", but most are printed as
"foo \bar".  Standardize on the latter form.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292306 91177308-0d34-0410-b5e6-96231b3b80d8

[NVPTX] Clean up nested !strconcat calls.

!strconcat is a variadic function; it will concatenate an arbitrary
number of strings.  There's no need to nest it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292305 91177308-0d34-0410-b5e6-96231b3b80d8

[NVPTX] Implement min/max in tablegen, rather than with custom DAGComine logic.

Summary:
This change also lets us use max.{s,u}16.  There's a vague warning in a
test about this maybe being less efficient, but I could not come up with
a case where the resulting SASS (sm_35 or sm_60) was different with or
without max.{s,u}16.  It's true that nvcc seems to emit only
max.{s,u}32, but even ptxas 7.0 seems to have no problem generating
efficient SASS from max.{s,u}16 (the casts up to i32 and back down to
i16 seem to be implicit and nops, happening via register aliasing).

In the absence of evidence, better to have fewer special cases, emit
more straightforward code, etc.  In particular, if a new GPU has 16-bit
min/max instructions, we want to be able to use them.

Reviewers: tra

Subscribers: jholewinski, llvm-commits

Differential Revision: https://reviews.llvm.org/D28732

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292304 91177308-0d34-0410-b5e6-96231b3b80d8

[NVPTX] Lower integer absolute value idiom to abs instruction.

Summary: Previously we lowered it literally, to shifts and xors.

Reviewers: tra

Subscribers: jholewinski, llvm-commits

Differential Revision: https://reviews.llvm.org/D28722

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292303 91177308-0d34-0410-b5e6-96231b3b80d8

[NVPTX] Improve lowering of llvm.ctpop.

Summary:
Avoid an unnecessary conversion operation when using the result of
ctpop.i32 or ctpop.i16 as an i32, as in both cases the ptx instruction
we run returns an i32.

(Previously if we used the value as an i32, we'd do an unnecessary
zext+trunc.)

Reviewers: tra

Subscribers: jholewinski, llvm-commits

Differential Revision: https://reviews.llvm.org/D28721

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292302 91177308-0d34-0410-b5e6-96231b3b80d8

[NVPTX] Add lowering for llvm.bitreverse.

Reviewers: tra

Subscribers: llvm-commits, jholewinski

Differential Revision: https://reviews.llvm.org/D28720

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292301 91177308-0d34-0410-b5e6-96231b3b80d8

[NVPTX] Fix function names in ctlz.ll test.  Test-only change.

Looks like a copy/paste mistake, all the functions in ctlz.ll were named
"ctpop".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292300 91177308-0d34-0410-b5e6-96231b3b80d8

[NVPTX] Improve lowering of llvm.ctlz.

Summary:
* Disable "ctlz speculation", which inserts a branch on every ctlz(x) which
  has defined behavior on x == 0 to check whether x is, in fact zero.

* Add DAG patterns that avoid re-truncating or re-expanding the result
  of the 16- and 64-bit ctz instructions.

Reviewers: tra

Subscribers: llvm-commits, jholewinski

Differential Revision: https://reviews.llvm.org/D28719

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292299 91177308-0d34-0410-b5e6-96231b3b80d8

[IR] Grammar police: "intact" is one word.  NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292298 91177308-0d34-0410-b5e6-96231b3b80d8

[InstCombine] add tests to show missed shrinkage; NFC

A patch to partially solve this:
https://reviews.llvm.org/D28625

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292296 91177308-0d34-0410-b5e6-96231b3b80d8

[libFuzzer] add ATTRIBUTE_NO_SANITIZE_MEMORY to sanitizer hooks

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292295 91177308-0d34-0410-b5e6-96231b3b80d8

Introduce -unroll-partial-threshold to separate PartialThreshold from Threshold in loop unorller.

Summary: Partial unrolling should have separate threshold with full unrolling.

Reviewers: efriedma, mzolotukhin

Reviewed By: efriedma, mzolotukhin

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D28831

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292293 91177308-0d34-0410-b5e6-96231b3b80d8

[RegisterCoalescing] Remove partial redundent copy.

The patch is to solve the performance problem described in PR27827.
Register coalescing sometimes cannot remove a copy because of interference.
But if we can find a reverse copy in one of the predecessor block of the copy,
the copy is partially redundent and we may remove the copy partially by moving
it to the predecessor block without the reverse copy.

Differential Revision: https://reviews.llvm.org/D28585

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292292 91177308-0d34-0410-b5e6-96231b3b80d8

Fix GettingStarted doc so that the example build command for cmake LLVM_ENABLE_PROJECTS works on linux

I tested the previous one on macOS, however building libc++ on Linux
requires libcxxabi as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292290 91177308-0d34-0410-b5e6-96231b3b80d8

[libfuzzer] fixing collected pc addresses for coverage

Summary: The causes google/ossfuzz#84

Reviewers: kcc

Subscribers: mgorny

Differential Revision: https://reviews.llvm.org/D28827

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292289 91177308-0d34-0410-b5e6-96231b3b80d8

[ADT] Add SparseBitVector::find_last().

Differential Revision: https://reviews.llvm.org/D28817

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292288 91177308-0d34-0410-b5e6-96231b3b80d8

[libFuzzer] use table of recent compares for memcmp/strcmp (to unify the code between cmp and memcmp handling)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292287 91177308-0d34-0410-b5e6-96231b3b80d8

[libFuzzer] copy the options inside MutationDispatcher to avoid use-after-scope in mutator tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292286 91177308-0d34-0410-b5e6-96231b3b80d8

GlobalISel: fix comparison order for G_FCMP

As with G_ICMP we'd written the CSET instructions backwards.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292285 91177308-0d34-0410-b5e6-96231b3b80d8

GlobalISel: add callseq instructions to record stack usage

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292284 91177308-0d34-0410-b5e6-96231b3b80d8

GlobalISel: correctly handle varargs

Some platforms (notably iOS) use a different calling convention for unnamed vs
named parameters in varargs functions, so we need to keep track of this
information when translating calls.

Since not many platforms are involved, the guts of the special handling is in
the ValueHandler class (with a generic implementation that should work for most
targets).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292283 91177308-0d34-0410-b5e6-96231b3b80d8

[LoopDeletion] (cleanup, NFC) Use the dedicated helper to get a single
unique exit block if available rather than rolling it ourselves.

This is a little disappointing because that helper doesn't do anything
clever to short-circuit the (surprisingly expensive) computation of all
exit blocks. What's worse is that the way we compute this is hopelessly,
hilariously inefficient. We're literally computing the same information
two different ways and multiple times each way:
- hasDedicatedExits computes the exit block set and then looks at the
  predecessors of each
- getExitingBlocks computes the set of loop blocks which have exiting
  successors
- getUniqueExitBlock(s) computes the set of non-loop blocks reached from
  loop blocks (sound familiar?)

Anyways, at some point we should clean all of this up in the LoopInfo
API, but for now just simplifying the user I'm about to touch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292282 91177308-0d34-0410-b5e6-96231b3b80d8

[LV] Add requires asserts to test case

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292280 91177308-0d34-0410-b5e6-96231b3b80d8

[LoopDeletion] (cleanup, NFC) Fix another variable name to match LLVM
conventions, missed this one in a previous cleanup patch (sorry).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292279 91177308-0d34-0410-b5e6-96231b3b80d8

[GlobalISel] track predecessor mapping during switch lowering.

Correctly populating Machine PHIs relies on knowing exactly how the IR level
CFG was lowered to MachineIR. This needs to be tracked by any translation
phases that meddle (currently only SwitchInst handling).

This reapplies r291973 which was reverted because of testing failures. Fixes:

 + Don't return an ArrayRef to a local temporary.
 + Incorporate Kristof's suggested comment improvements.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292278 91177308-0d34-0410-b5e6-96231b3b80d8

[X86][SSE] Split UMIN and UMAX known bits tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292277 91177308-0d34-0410-b5e6-96231b3b80d8

[LoopDeletion] (cleanup, NFC) Remove a pointless comment.

I hope that for any code, it is changed only with good reason and only
when the author knows what they are doing...

There is of course good reason to comment here about the subtlety of the
process, and I've left that comment in tact.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292275 91177308-0d34-0410-b5e6-96231b3b80d8

[LoopDeletion] (cleanup, NFC) Make simple helper functions static
instead of members.

No state was being provided by the object so this seems strictly
simpler.

I've also tried to improve the name and comments for the functions to
more thoroughly document what they are doing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292274 91177308-0d34-0410-b5e6-96231b3b80d8

[LoopDeletion] (cleanup, NFC) Stop passing around reference to a vector
that we know has exactly one element when all we are going to do is get
that one element out of it.

Instead, pass around that one element.

There are more simplifications to come in this code...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292273 91177308-0d34-0410-b5e6-96231b3b80d8

[PM] Clean up variable and parameter names to match modern LLVM naming
conventions more conistently before hacking on this code to integrate
nicely with new PM's loop pass infrastructure. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292272 91177308-0d34-0410-b5e6-96231b3b80d8

Silence some Sphinx diagnostics in an attempt to get the documentation builder back to green (lab.llvm.org:8011/builders/llvm-sphinx-docs/builds/1895).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292271 91177308-0d34-0410-b5e6-96231b3b80d8

Add a test case for LICM when promoting locals that may be read after the throw within the loop. NFCI.

Summary: Add a test case for LICM when promoting locals that may be read after the throw within the loop.

Reviewers: eli.friedman, hfinkel, sanjoy

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D28822

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292261 91177308-0d34-0410-b5e6-96231b3b80d8

[InstCombine] refactor foldICmpShlConstant(); NFCI

This reduces the size of and increases the symmetry with the planned functional change in:
https://reviews.llvm.org/D28406

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292260 91177308-0d34-0410-b5e6-96231b3b80d8

[bpf] fix stack-use-after-scope

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292258 91177308-0d34-0410-b5e6-96231b3b80d8

Enabled -fsanitize-address-use-after-scope for -DLLVM_USE_SANITIZER=Address

Subscribers: mgorny, llvm-commits

Differential Revision: https://reviews.llvm.org/D28823

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292256 91177308-0d34-0410-b5e6-96231b3b80d8

[cmake] Update SOVERSION for the new versioning scheme

Update SOVERSION to use just the major version number rather than
major+minor, to match the new versioning scheme where only major is used
to indicate API/ABI version.

Since two-digit SOVERSIONs were introduced post 3.9 branching, this
change does not risk any SOVERSION collisions. In the past,
two-component X.Y SOVERSIONs were shortly used but those will not
interfere with the new ones since the new versions start at 4.

Differential Revision: https://reviews.llvm.org/D28730

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292255 91177308-0d34-0410-b5e6-96231b3b80d8

[LV] Mark non-consecutive-like pointers non-uniform

If a memory instruction will be vectorized, but it's pointer operand is
non-consecutive-like, the instruction is a gather or scatter operation. Its
pointer operand will be non-uniform. This should fix PR31671.

Reference: https://llvm.org/bugs/show_bug.cgi?id=31671
Differential Revision: https://reviews.llvm.org/D28819

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292254 91177308-0d34-0410-b5e6-96231b3b80d8

[WebAssembly] Add triple support for the new wasm object format

Differential Revision: https://reviews.llvm.org/D26701

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292252 91177308-0d34-0410-b5e6-96231b3b80d8

Rename scalar_promote.ll to scalar-promote.ll and scalar_promote-unwind.ll to scalar-promote-unwind.ll. NFCI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292251 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor out LoopInfo computation so that it can be used by
other test cases.

Summary: Refactor out LoopInfo computation so that it can be
used by other test cases.

So i am changing this test proactively for later commit, which will use
this function.

Reviewers: sanjoy, hfinkel

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D28778

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292250 91177308-0d34-0410-b5e6-96231b3b80d8

[EarlyCSE] Don't DSE across readnone functions that may throw

Summary: Depends on D28740

Reviewers: dberlin, chandlerc, hfinkel, majnemer

Subscribers: mcrosier, llvm-commits

Differential Revision: https://reviews.llvm.org/D28741

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292249 91177308-0d34-0410-b5e6-96231b3b80d8

[InstCombine] add tests for shl nsw + icmp sle; NFC

We want to handle these cases similarly to icmp sgt, so add the tests for it.
See: https://reviews.llvm.org/D28406

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292248 91177308-0d34-0410-b5e6-96231b3b80d8

[TLI] Appease spurious MSVC warning using llvm_unreachable. NFC.

r292188 confused MSVC because of the combined lack of a default
case and return statement.

Move the unreachable outside of the NumLibFuncs case, to make it
obvious that all cases should be handled.

llvm_unreachable is __declspec(noreturn), so I'm assuming this
does appease MSVC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292246 91177308-0d34-0410-b5e6-96231b3b80d8

Remove an overeager assert from r288844.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292244 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r291640 change to fold X86 comparison with atomic_load_add.

Even with the fix from r291630, this still causes problems. I get
widespread assertion failures in the Swift runtime's WeakRefCount::increment()
function. I sent a reduced testcase in reply to the commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292242 91177308-0d34-0410-b5e6-96231b3b80d8

[PM] Teach the LoopPassManager to automatically canonicalize loops by
runnig LCSSA over them prior to running the loop pipeline.

This also teaches the loop PM to verify that LCSSA form is preserved
throughout the pipeline's run across the loop nest.

Most of the test updates just leverage this new functionality. One has to be
relaxed with the new PM as IVUsers is less powerful when it sees LCSSA input.

Differential Revision: https://reviews.llvm.org/D28743

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292241 91177308-0d34-0410-b5e6-96231b3b80d8

[ValueTracking] recognize a 'not' of an assumed condition as false

Also, add the corresponding match to the AssumptionCache's 'Affected Values' list.

Differential Revision: https://reviews.llvm.org/D28485

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292239 91177308-0d34-0410-b5e6-96231b3b80d8

[InstCombine] Fold ((C1 OP zext(X)) & C2) -> zext((C1 OP X) & C2)

This further extends r292179 to support additional binary operators
beyond subtraction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292238 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r292231.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292237 91177308-0d34-0410-b5e6-96231b3b80d8

[X86][AVX512] Add all_of/any_of avx512vl tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292235 91177308-0d34-0410-b5e6-96231b3b80d8

[ValueTracking] Extend known bits to understand @llvm.bitreverse.

Differential Revision: https://reviews.llvm.org/D28780

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292233 91177308-0d34-0410-b5e6-96231b3b80d8

[lit] Limit parallelism of sanitizer tests on Darwin [llvm part]

Running lit tests and unit tests of ASan and TSan on macOS has very bad performance when running with a high number of threads. This is caused by xnu (the macOS kernel), which currently doesn't handle mapping and unmapping of sanitizer shadow regions (reserved VM which are several terabytes large) very well. The situation is so bad that increasing the number of threads actually makes the total testing time larger. The macOS buildbots are affected by this. Note that we can't easily limit the number of sanitizer testing threads without affecting the rest of the tests.

This patch adds a special "group" into lit, and limits the number of concurrently running tests in this group. This helps solve the contention problem, while still allowing other tests to run in full, that means running lit with -j8 will still with 8 threads, and parallelism is only limited in sanitizer tests.

Differential Revision: https://reviews.llvm.org/D28420

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292231 91177308-0d34-0410-b5e6-96231b3b80d8

[InstCombine] reduce indent; NFCI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292230 91177308-0d34-0410-b5e6-96231b3b80d8

Recommit r292214 "[Support/Compression] - Change zlib API to return Error instead of custom status"

No any changes, will follow up with D28807 commit containing APLi change for clang
to fix build issues happened.

Original commit message:
[Support/Compression] - Change zlib API to return Error instead of custom status.

Previously API returned custom enum values.
Patch changes it to return Error with string description.
That should help users to report errors in universal way.

Differential revision: https://reviews.llvm.org/D28684

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292226 91177308-0d34-0410-b5e6-96231b3b80d8

[AMDGPU] Assembler: fix v_mac_f16 immediates

Reviewers: vpykhtin, artem.tamazov, tstellarAMD

Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, tony-tye

Differential Revision: https://reviews.llvm.org/D28802

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292224 91177308-0d34-0410-b5e6-96231b3b80d8

[X86][SSE] Tests showing horizontal all_of/any_of of vector comparison results

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292223 91177308-0d34-0410-b5e6-96231b3b80d8

[llvm-objdump tests] Copy the inputs of tests closer to tests.

Summary:
Tests under tools/llvm-objdump should not use inputs from Object. Copied the
required inputs and aligned the new tests to be more consistent with the existing
tests in this respect.

Reviewers: ioeric

Reviewed By: ioeric

Subscribers: davide, djasper, cfe-commits

Differential Revision: https://reviews.llvm.org/D28799

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292222 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r292214 "[Support/Compression] - Change zlib API to return Error instead of custom status."

It broked clang:
http://lab.llvm.org:8080/green//job/clang-stage1-cmake-RA-incremental_build/34218/consoleFull#46141505449ba4694-19c4-4d7e-bec5-911270d8a58c

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292217 91177308-0d34-0410-b5e6-96231b3b80d8

BrainF example: fixing output buffering issue
Differential Revision: https://reviews.llvm.org/D27824

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292216 91177308-0d34-0410-b5e6-96231b3b80d8

[Support/Compression] - Change zlib API to return Error instead of custom status.

Previously API returned custom enum values.
Patch changes it to return Error with string description.
That should help users to report errors in universal way.

Differential revision: https://reviews.llvm.org/D28684

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292214 91177308-0d34-0410-b5e6-96231b3b80d8

[XRay][Arm] Repair XRay table emission on Arm32 and add tests to identify such problem earlier

Summary:
Emission of XRay table was occasionally disabled for Arm32, but this bug was not then detected because earlier (also by mistake) testing of XRay was occasionally disabled on 32-bit Arm targets. This patch should fix that problem and detect such problems in the future.
This patch is one of a series, see also
- https://reviews.llvm.org/D28623

Reviewers: rengolin, dberris

Reviewed By: dberris

Subscribers: llvm-commits, aemerson, rengolin, dberris, iid_iunknown

Differential Revision: https://reviews.llvm.org/D28624

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292210 91177308-0d34-0410-b5e6-96231b3b80d8

[InstCombine][X86][AVX] Add DemandedElts support for VPERMILPD/VPERMILPS instructions

Simplify a vpermilvar shuffle mask based on the elements of the mask that are actually demanded.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292209 91177308-0d34-0410-b5e6-96231b3b80d8

Update the release tester for MIPS. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292208 91177308-0d34-0410-b5e6-96231b3b80d8