This splits MVE vector stores of a fp_trunc in the same way that we do
for standard trunc's. It extends PerformSplittingToNarrowingStores to
handle fp_round, splitting the store into pieces and adding a VCVTNb to
perform the actual fp_round. The actual store is then converted to an
integer store so that it can truncate bottom lanes of the result.
Differential Revision: https://reviews.llvm.org/D81141
Summary:
Check that the co_await promise.final_suspend() does not potentially throw again after we have resolved dependent types.
This takes care of the cases where promises types are templated.
Added test cases for this scenario and confirmed that the checks happen now.
Also run libcxx tests locally to make sure all tests pass.
Reviewers: Benabik, lewissbaker, junparser, modocache
Reviewed By: modocache
Subscribers: modocache, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D82332
The PREFETCHW instruction was originally part of the 3DNow. But
it was given its own CPUID bit on later CPUs just before 3DNow
was deprecated.
We were setting the -mprfchw flag if -m3dnow was passed or the CPU
supported 3dnow unless -mno-prfchw was passed. But -march=native
on a CPU without the PRFCHW CPUID bit set will pass -mno-prfchw.
So -march=k8 will behave differently than -march=native on a K8
for example.
So remove this implicit setting from the frontend and instead
enable the backend to use PREFETCHW if 3dnow OR prfchw is enabled.
Also enable PRFCHW flag on amdfam10/barcelona which seems to be
where this CPUID bit was introduced. That CPU also supported
3dnow.
If the GEP instruction contanins only constants as its arguments,
then it should be recognized as a constant. For now, there was
also added a flag to turn off this simplification if it causes
any regressions ("disable-gep-const-evaluation") which is off
by default. Once I gather needed data of the effectiveness of
this simplification, the flag will be deleted.
Reviewers: apilipenko, davidxl, mtrofin
Reviewed By: mtrofin
Differential Revision: https://reviews.llvm.org/D81026
Summary:
Standardize the formatting of selector fragments to include the ':',
e.g. for `- (void)foobar:(int)foobar;`, report `{foobar:}` instead of
`{foobar}{:}`. This was normally the case except for a couple of places
where it was split.
This also improves integration with clangd since it relies upon the `:`
to identify ObjC selectors.
NOTE: It is possible to have selector fragments that are just `:` with
no text, we now handle this properly for the first fragment.
Reviewers: sammccall, doug.gregor
Subscribers: ilya-biryukov, dexonsmith, arphaman, kadircet, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D82306
Although this issue is not specific to macOS, Python (in)compatibility
comes up quite often and we've been linking users to this page. This
just adds more details for this particular scenario.
Differential revision: https://reviews.llvm.org/D82507
Summary:
In https://reviews.llvm.org/D82029 we added the conformance check that the expression co_await promise.final_suspend() should not potentially throw.
As part of this expression, in cases when the await_suspend() method of the final suspend awaiter returns a handle, __builtin_coro_resume could be called on the handle to immediately resume that coroutine.
__builtin_coro_resume is not declared with noexcept and it shouldn't. We need to special check this case here.
Reviewers: modocache, lewissbaker, junparser
Reviewed By: lewissbaker
Subscribers: modocache, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D82415
Extend the memop value profile buckets to be more flexible (could accommodate a
mix of individual values and ranges) and to cover more value ranges (from 11 to
22 buckets).
Disabled behind a flag (to be enabled separately) and the existing code to be
removed later.
Differential Revision: https://reviews.llvm.org/D81682
For passes got skipped, this is confusing because the log said it is `running pass`
but it is skipped later.
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D82511
Summary:
A recent Linux kernel commit exposed a performance cliff in Clang. Calls
to SourceManager::getFileIDLocal() when there's a cache miss against
LastFileIDLookup can be relatively expensive, as getFileIDLocal() tries
a few linear probes, then falls back to binary search. The use of
SourceManager::isOffsetInFileID() is also relatively expensive (both
isOffsetInFileID and getFileIDLocal dominated a trace of the performance
cliff case).
As a FIXME notes (and as @kadircet helpfully noted in review of D80681),
there's a few optimizations we can do here since we've already
identified that an offset is local (as opposed to "loaded").
This patch was forked off of D80681, which additionally did this and
modified some caching behavior, as we expect this change to be less
controversial.
In terms of optimizations, we've already determined that the SLocOffset
parameter to SourceManager::getFileIDLocal() is local in the caller
SourceManager::getFileIDSlow(). Also, there's an early continue in the
binary search loop in getFileIDLocal() that are duplicated in
isOffsetInFileID() as pointed out by @kadircet.
Take advantage of these to optimize the binary search patch, and remove
the FIXME.
Reviewers: kadircet
Reviewed By: kadircet
Subscribers: cfe-commits, kadircet, srhines
Tags: #clang
Differential Revision: https://reviews.llvm.org/D82497
Summary: Dumping contents of `.llvm.call-graph-profile` section of COFF in the same format as ELF.
Reviewers: jhenderson, MaskRay, hans
Reviewed By: jhenderson
Subscribers: grimar, rupprecht, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D81894
Pass EvalCallOptions via runCheckersForEvalCall into defaultEvalCall.
Update the AnalysisOrderChecker to support evalCall for testing.
Differential Revision: https://reviews.llvm.org/D82256
This patch takes the IO redirection logic from ScriptInterpreterPython
and moves it into the interpreter library so that it can be used by
other script interpreters. I've turned it into a RAII object so that we
don't have to worry about cleaning up in the calling code.
Differential revision: https://reviews.llvm.org/D82396
Cleanup MemoryDependenceAnalysis.h as well - GVN.h was also implicitly including AliasAnalysis.h via this.
Fix implicit include dependencies in source files and replace legacy AliasAnalysis typedef with AAResults where necessary.
Summary:
This somewhat matches the --aa-pipeline option, which separates out any
AA analyses to make sure they run before other passes.
Makes check-llvm failures under new PM go from 2356 -> 2303.
AA passes are not handled by PassBuilder::parsePassPipeline() but rather
PassBuilder::parseAAPipeline(), which is why this fixes some failures.
Reviewers: asbirlea, hans, ychen, leonardchan
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D82488
This change enables PowerPC compiler builtins to generate constrained
floating point operations when clang is indicated to do so.
A couple of possibly unexpected backend divergences between constrained
floating point and regular behavior are highlighted under the test tag
FIXME-CHECK. This may be something for those on the PPC backend to look
at.
Patch by: Drew Wock <drew.wock@sas.com>
Differential Revision: https://reviews.llvm.org/D82020
fabs(X) * fabs(Y) --> fabs(X * Y)
fabs(X) / fabs(Y) --> fabs(X / Y)
If both operands of fmul/fdiv are positive, then the result must be positive.
There's a NAN corner-case that prevents removing the more specific fold just
above this one:
fabs(X) * fabs(X) -> X * X
That fold works even with NAN because the sign-bit result of the multiply is
not specified if X is NAN.
We can't remove that and use the more general fold that is proposed here
because once we convert to this:
fabs (X * X)
...it is not legal to simplify the 'fabs' out of that expression when X is NAN.
That's because fabs() guarantees that the sign-bit is always cleared - even
for NAN values.
So this patch has the potential to lose information, but it seems unlikely if
we do the more specific fold ahead of this one.
Differential Revision: https://reviews.llvm.org/D82277
This adds code to lower f32 to f16 fp_trunc's using a pair of MVE VCVT
instructions. Due to v4f16 not being legal, fp_round are often split up
fairly early. So this reconstructs the vcvt's from a buildvector of
fp_rounds from two vector inputs. Something like:
BUILDVECTOR(FP_ROUND(EXTRACT_ELT(X, 0),
FP_ROUND(EXTRACT_ELT(Y, 0),
FP_ROUND(EXTRACT_ELT(X, 1),
FP_ROUND(EXTRACT_ELT(Y, 1), ...)
It adds a VCVTN node to handle this, which like VMOVN or VQMOVN lowers
into the top/bottom lanes of an MVE instruction.
Differential Revision: https://reviews.llvm.org/D81139
Summary:
LDRAA and LDRAB in their writeback variant should softfail when the same
register is used as result and base.
This patch adds a custom decoder that catches such case and emits a
warning when it occurs.
Differential Revision: https://reviews.llvm.org/D82541
Give up folding an expression if the fragment of one of the operands
would require laying out a fragment already being laid out. This
prevents hitting an infinite recursion when a fill size expression
refers to a later fragment since computing the offset of that fragment
would require laying out the fill fragment and thus computing its size
expression.
Reviewed By: echristo
Differential Revision: https://reviews.llvm.org/D79570
This function was modifying and returning pointers to static storage,
which meant that any two accesses to different Scalar objects could
potentially race (depending on which types the objects were storing and
the host endianness).
In the new version the user is responsible for providing a buffer into
which this method will store its binary representation. The main caller
(RegisterValue::GetBytes) already has one such buffer handy, so this did
not require any major rewrites.
To make that work, I've needed to mark the RegisterValue value buffer
mutable -- not an ideal solution, but definitely better than modifying
global storage. This could be further improved by changing
RegisterValue::GetBytes to take a buffer too.
The main interface has been migrated to Align already but a few backends where broadening the type from Align to MaybeAlign.
This patch makes sure all implementations conform to the public API.
Differential Revision: https://reviews.llvm.org/D82465
David Green