When widening an IndVar that has LCSSA Phi users outside
the loop, we can safely widen it as usual and then truncate
the result outside the loop without hurting the performance.

Differential Revision: https://reviews.llvm.org/D91593
Reviewed By: skatkov

This was originally a part of D71880 but is separated for simplicity and ease
of reviewing.

Fixes: https://github.com/clangd/clangd/issues/582

Reviewed By: hokein

Differential Revision: https://reviews.llvm.org/D91952

For now, we will hardcode the result as 0.0 if the input is denormal or 0. That will
have the impact the precision. As the fsqrt added belong to the cold path of the
cmp+branch, it won't impact the performance for normal inputs for PowerPC, but improve
the precision if the input is denormal.

Reviewed By: Spatel

Differential Revision: https://reviews.llvm.org/D80974

This patch adds a description about the newly added poison constant to LangRef.

Differential Revision: https://reviews.llvm.org/D92162

We need a real target now, and it was only being created if grpc was
built from source or imported from homebrew.

Differential Revision: https://reviews.llvm.org/D92107

Add a flag that disables caching when computing aliasing results
potentially based on a phi-phi NoAlias assumption. We'll still
insert cache entries temporarily to catch infinite recursion,
but will drop them afterwards, so they won't persist in BatchAA.

Differential Revision: https://reviews.llvm.org/D91936

Also clean it up a bit.

python3-distutils is required to use `import distutils.spawn`, which is
required by the ABI list targets.

Many bots are unhappy, at the very least missed a few codegen tests,
and possibly this has a logic hole inducing a miscompile
(will be really awesome to have ready reproducer..)

Need to investigate.

This reverts commit 2245fb8a.

Without this change mesa fails while looking for llvm components like amdgpu, engine or native:

    Run-time dependency LLVM (modules: amdgpu(missing), bitreader, bitwriter, core, engine(missing), executionengine, instcombine, ipo, mcdisassembler, mcjit, native(missing), scalaropts, transformutils, coroutines)
    Looking for a fallback subproject for the dependency llvm (modules: bitwriter, engine, mcdisassembler, mcjit, core, executionengine, scalaropts, transformutils, instcombine, amdgpu, native, bitreader, ipo)

This change adds component groups (like all-targets, engine, native, amdgpu) to the "all" component.

Differential Revision: https://reviews.llvm.org/D92158

1. It doesn't make sense to enforce that the bonus instruction
   is only used once in it's basic block. What matters is
   whether those user instructions fit within our budget, sure,
   but that is another question.
2. It doesn't make sense to enforce that said bonus instructions
   are only used within their basic block. Perhaps the branch
   condition isn't using the value computed by said bonus instruction,
   and said bonus instruction is simply being calculated
   to be used in successors?

So iff we can clone bonus instructions, to lift these restrictions,
we just need to carefully update their external uses
to use the new cloned instructions.

Notably, this transform (even without this change) appears to be
poison-unsafe as per alive2, but is otherwise (including the patch) legal.

We don't introduce any new PHI nodes, but only "move" the instructions
around, i'm not really seeing much potential for extra cost modelling
for the transform, especially since now we allow at most one such
bonus instruction by default.

This causes the fold to fire +11.4% more (13216 -> 14725)
as of vanilla llvm test-suite + RawSpeed.

The motivational pattern is IEEE-754-2008 Binary16->Binary32
extension code:
https://github.com/darktable-org/rawspeed/blob/ca57d77fb2ba81f21fc712cfac26e54f46406473/src/librawspeed/common/FloatingPoint.h#L115-L120
^ that should be a switch, but it is not now: https://godbolt.org/z/bvja5v
That being said, even thought this seemed like this would fix it: https://godbolt.org/z/xGq3TM
apparently that fold is happening somewhere else afterall,
so something else also has a similar 'artificial' restriction.

[NFC][SimplifyCFG] Add test coverage for FoldBranchToCommonDest xform with live-out bonus instuctions

The uses of the bonus instructions should not be preventing the transformation.

We don't actually update the ABI lists at every release -- it's too much
work, since we'd technically have to do it even for minor releases.
Furthermore, I don't think anybody uses those (I certainly don't rely
on them for anything).

Instead, it is better to rely on the ABI list changelog and the canonical
ABI list that we always keep up to date. If one wants to know what symbols
were shipped in a specific release, that can be discovered easily using
Git, which is a superior tool than keeping textual copies of old versions.

Differential Revision: https://reviews.llvm.org/D92186

Rename prefix from X32 to X86 as we typically use X32 for gnux32 triples

Reviewed By: yonghong-song

Differential Revision: https://reviews.llvm.org/D91990

Currently, we have some confusion in the codebase regarding the
meaning of LocationSize::unknown(): Some parts (including most of
BasicAA) assume that LocationSize::unknown() only allows accesses
after the base pointer. Some parts (various callers of AA) assume
that LocationSize::unknown() allows accesses both before and after
the base pointer (but within the underlying object).

This patch splits up LocationSize::unknown() into
LocationSize::afterPointer() and LocationSize::beforeOrAfterPointer()
to make this completely unambiguous. I tried my best to determine
which one is appropriate for all the existing uses.

The test changes in cs-cs.ll in particular illustrate a previously
clearly incorrect AA result: We were effectively assuming that
argmemonly functions were only allowed to access their arguments
after the passed pointer, but not before it. I'm pretty sure that
this was not intentional, and it's certainly not specified by
LangRef that way.

Differential Revis...

Similar to D92113. Currently `clang -fstack-size-section -fno-unique-section-names`
sets the linked-to symbol to the first `.text`, which is:

* incorrect for COMDAT sections
* inferior for non-COMDAT sections in -ffunction-sections mode (poor --gc-sections: .stack_sizes cannot be separately discarded)

Note, if the section symbol can be referenced in more places (if the
function begin symbol does not apply), we probably should consider
defining a different BeginSymbol for sections with ",unique" linkage.

Reviewed By: grimar, jhenderson

Differential Revision: https://reviews.llvm.org/D92151

This patch enables passing non variadic vector type parameters on the caller and callee side and vector return on AIX that are passed in vector registers only.

So far, support is enabled for only the AIX extended Altivec ABI Calling convention.

Reviewed By: sfertile, DiggerLin

Differential Revision: https://reviews.llvm.org/D86476

The test case isn't using the AST matchers for all checks as there doesn't seem to be support for
matching TemplateTypeParmDecl default arguments. Otherwise this is simply importing the
default arguments.

Also updates several LLDB tests that now as intended omit the default template
arguments of several std templates.

Reviewed By: martong

Differential Revision: https://reviews.llvm.org/D92103

Using sysctl requires including headers that are considered internal on
Linux, like <sys/sysctl.h> & friends. Instead, sysconf is defined by POSIX
(and we have a fallback for Windows), so all the systems we support should
be happy with just sysconf.

Differential Revision: https://reviews.llvm.org/D92135

I'm not 100% sure what the issue actually is since I can't reproduce it
locally, however what I explain in the comment is my best attempt to
explain what's going on.

Differential Revision: https://reviews.llvm.org/D92131

The rewrite logic has an optimization to drop a cast operation after
rewriting block arguments if the cast operation has no users. This is
unsafe as there might be a pending rewrite that replaced the cast operation
itself and hence would trigger a second free.

Instead, do not remove the casts and leave it up to a later canonicalization
to do so.

Differential Revision: https://reviews.llvm.org/D92184

FreeBSD's locale data uses the same U+2027 separator as Glibc 2.27 and newer.

Reviewed By: #libc, emaste, ldionne

Differential Revision: https://reviews.llvm.org/D91165

This strips out a lot of the code that should no longer be needed from
the MVETailPredictionPass, leaving the important part - find active lane
mask instructions and convert them to VCTP operations.

Differential Revision: https://reviews.llvm.org/D91866

Revert rG12d59b69 "[DAG] Legalize umin(x,y) -> sub(x,usubsat(x,y)) and umax(x,y) -> add(x,usubsat(y,x)) iff usubsat is legal"

This reverts commit 12d59b69.

Prematurely pushed this to trunk

[DAG] Legalize umin(x,y) -> sub(x,usubsat(x,y)) and umax(x,y) -> add(x,usubsat(y,x)) iff usubsat is legal

If usubsat() is legal, this is likely to result in smaller codegen expansion than the default cmp+select codegen expansion.

Allows us to move the x86-specific lowering to the generic expansion code.

Depends on D83697

Original patch by Daniel Grumberg.

Differential Revision: https://reviews.llvm.org/D83698

A crash/assertion failure in the greedy register allocator was tracked
down to a debug instr being passed to LiveIntervals::getInstructionIndex.
Normally this should not occur as debug instructions are collected and
removed by LiveDebugVariables before RA, and reinserted afterwards.
However, when a function has no debug info, LiveDebugVariables simply
strips any debug values that are present as they're not needed (this
situation will occur when a function with debug info is inlined into a
nodebug function). The problem is, it only removes DBG_VALUE instructions,
leaving DBG_LABELs (the cause of the crash).

This patch updates the LiveDebugVariables nodebug path to remove all debug
instructions. The test case verifies that DBG_VALUE/DBG_LABEL instructions
are present, and that they are stripped.

When -experimental-debug-variable-locations is enabled, certain variable
locations are represented by DBG_INSTR_REF instead of DBG_VALUE. The test
case verifies that a DBG_INSTR_REF is emitted by the option, and that it
is also stripped.

Differential Revision: https://reviews.llvm.org/D92127

Same idea as in D92103 and D92106, but I realised after creating those reviews that there are
also TemplateTemplateParmDecls that can have default arguments, so here's hopefully the
last patch for default template arguments.

Reviewed By: martong

Differential Revision: https://reviews.llvm.org/D92119

Update VPReplicateRecipe to inherit from VPValue. This still does not
update scalarizeInstruction to set the result for the VPValue of
VPReplicateRecipe, because this first requires tracking scalar values in
VPTransformState.

Reviewed By: reames

Differential Revision: https://reviews.llvm.org/D91500

This change is required so that bufferization can properly identify
the linalg.yield as a terminator with an associated parent op.

Differential Revision: https://reviews.llvm.org/D92173

When bailing out in rewriteLoopExitValues() you could be left with PHI
nodes in the DeadInsts vector. Those would be not handled by the use of
RecursivelyDeleteTriviallyDeadInstructions() in IndVarSimplify. This
resulted in the IndVarSimplify pass returning an incorrect modified
status. This was caught by the expensive check introduced in D86589.

This patches changes IndVarSimplify so that it deletes those PHI nodes,
using RecursivelyDeleteDeadPHINode().

This fixes PR47486.

Reviewed By: mkazantsev

Differential Revision: https://reviews.llvm.org/D91153

The cost-model is not getting the cost right for a mul with <2 x i64>
operands, i.e. we don't have a MUL.2d, and this is precommitting some
tests before adjusting this.