The similar getICmpCode and getPredForICmpCode are already there.
This moves FP for consistency.
I think InstCombine is currently the only user of both.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D120754
This reverts commit 6b3b3ef344.
Jim Ingham informed me that the upper case is a hint to the option
name, like you might see in a menu to show you what the shortcut is.
After merging https://reviews.llvm.org/D120801, Flang no longer builds
with GCC 11:
```
../llvm-project/flang/lib/Semantics/runtime-type-info.cpp:385:22: error: variable ‘lenParam’ set but not used [-Werror=unused-but-set-variable]
385 | for (SymbolRef lenParam : *lenParameters) {
| ^~~~~~~~
```
I'm sending this without a review as a quick fix.
There are two DataExtractors in scope: one from the llvm namespace and
one from the lldb_private namespace. Some Microsoft Visual C++ compilers
(I tested with MSVC 14.23 specifically) cannot handle this situation,
and generate ambiguous symbol errors. This change fixes this compile
error.
Differential revision: https://reviews.llvm.org/D120718
This patch extends first-order recurrence handling to support cases
where we already sunk an instruction for a different recurrence, but
LastPrev comes before Previous.
To handle those cases correctly, we need to find the earliest entry for
the sink-after chain, because this is references the Previous from the
original recurrence. This is needed to ensure we use the correct
instruction as sink point.
Depends on D118558.
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D118642
Two buildbots have started failing recently:
* https://lab.llvm.org/buildbot/#/builders/181/builds/3894
* https://lab.llvm.org/buildbot/#/builders/191/builds/3908
Build error:
```
In file included from /home/tcwg-buildbot/worker/flang-aarch64-rel-assert/llvm-project/flang/examples/FlangOmpReport/FlangOmpReport.cpp:21:
In file included from /home/tcwg-buildbot/worker/flang-aarch64-rel-assert/llvm-project/flang/include/flang/Frontend/FrontendActions.h:15:
In file included from /home/tcwg-buildbot/worker/flang-aarch64-rel-assert/llvm-project/llvm/../mlir/include/mlir/IR/BuiltinOps.h:16:
In file included from /home/tcwg-buildbot/worker/flang-aarch64-rel-assert/llvm-project/llvm/../mlir/include/mlir/IR/FunctionInterfaces.h:17:
In file included from /home/tcwg-buildbot/worker/flang-aarch64-rel-assert/llvm-project/llvm/../mlir/include/mlir/IR/BuiltinTypes.h:12:
/home/tcwg-buildbot/worker/flang-aarch64-rel-assert/llvm-project/llvm/../mlir/include/mlir/IR/BuiltinAttributeInterfaces.h:279:10: fatal error: 'mlir/IR/BuiltinAttributeInterfaces.h.inc' file not found
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1 error generated.
```
I have not been able to reproduce locally, but from this log it is clear
that the rule for `flangFrontend` is missing the
`MLIRBuiltinAttributeInterfacesIncGen` dependency from MLIR. I couldn't
identify a breaking commit. I suspect that until now we have simply been
"lucky" and that dependency just happened to be built before
`flangFrontend`.
I am sending this without a review - the change is rather
straightforward and the only way to verify it is to make the buildbots
test it.
body_start was never used, resulting in the first filtered line to be
skipped.
Fixes the --filter option introduced in D117694.
Differential Revision: https://reviews.llvm.org/D119704
Ensure that `Handler` within the class is interpreted as the as the current template instantiation (instead the class template itself).
Fixes#53447
Reviewed By: rriddle
Differential Revision: https://reviews.llvm.org/D120852
The standard requires[0] member function constraints to be checked when
explicitly instantiating classes. This patch adds this constraints
check.
This issue is tracked as #46029 [1].
Note that there's an related open CWG issue (2421[2]) about what to do when
multiple candidates have satisfied constraints. This is particularly an
issue because mangling doesn't contain function constraints, and so the
following code still ICEs with definition with same mangled name
'_ZN1BIiE1fEv' as another definition:
template<class T>
struct B {
int f() requires std::same_as<T, int> {
return 0;
}
int f() requires (std::same_as<T, int> &&
!std::same_as<T, char>) {
return 1;
}
};
template struct B<int>;
Also note that the constraints checking while instantiating *functions*
is still not implemented. I started looking at it but It's a bit more
complicated. I believe in such a case we have to consider the partial
constraints order and potentially choose the best candidate out of the
set of multiple valid ones.
[0]: https://eel.is/c++draft/temp.explicit#10
[1]: https://github.com/llvm/llvm-project/issues/46029
[2]: https://cplusplus.github.io/CWG/issues/2421.html
Differential Revision: https://reviews.llvm.org/D120255
`basic_ios` delays initialization of `__fill_` to `widen(' ')` until `fill()` is called. But, `fill(char_type)` is missing this logic, so the fill character does not get initialized to whitespace if `fill(char_type)` is called first. This patch adds this logic to `fill(char_type)`.
Reviewed By: #libc, ldionne, Quuxplusone
Differential Revision: https://reviews.llvm.org/D120751
https://alive2.llvm.org/ce/z/mJP7XP
This can be viewed as expanding the compare into and/or-of-compares:
https://alive2.llvm.org/ce/z/bkZYWE
followed by reduction of each compare.
This could be extended in several ways:
1. There's a (X & Y) == -1 sibling.
2. We can recurse through more than 1 'or'.
3. The fold could be generalized beyond rotates - any operation that
only changes the order of bits (bswap, bitreverse).
This is a transform noted in D111530.
We will check a bit later that the constant is in fact a function,
so the separate check for a function pointer type is largely
redunant. Also simplify the cast stripping with
stripPointerCasts().
This patch updates the FADDPv* patterns that only use the lower half of
the result register. For those patterns, the second operand does not
matter because its results won't be used.
Instead of introducing new implicit defs for those operands, just use
the first operand. The problem with using new implicit defs is that
register allocation can introduce unnecessary dependencies by using a
different register than the first operand.
For motivating cases, see the changes in the fadd_reduction_*_in_loop
cases. Without this change, the first faddp in the loop has an
unnecessary additional dependency through v0, which is also used for
a cross-iteration reduction.
This can noticeable impact performance. For slightly bigger loops,
this change can improve performance by 15%.
Reviewed By: sdesmalen, t.p.northover
Differential Revision: https://reviews.llvm.org/D120706
Without this patch, when End == Start, we access Actions[Actions.end()]
though we return an empty result.
This fixes an assertion failure in MSVC STL debug build.
This ensures that we generate memref types with matching layout maps. (Especially when using partial bufferization passes.)
Differential Revision: https://reviews.llvm.org/D120893
`ArgInfo` is reduced to only contain a pair of {formal,actual} values.
The specialized function `Fn` and the `Partial` flag are redundant in
this structure. The `Gain` is moved to a new struct `SpecializationInfo`.
The value mappings created by cloneCandidateFunction() are being used
by rewriteCallSites() for matching the formal arguments of recursive
functions.
The list of specializations is passed by reference to calculateGains()
instead of being returned by value.
The `IsPartial` flag is removed from isArgumentInteresting() and
getPossibleConstants() as it's no longer used anywhere in the code.
Differential Revision: https://reviews.llvm.org/D120753
The patch adds very basic cost model for masked memory op on scalable vector.
Reviewed By: frasercrmck
Differential Revision: https://reviews.llvm.org/D117884
Using -isystem marks the headers as system headers, which means that we
don't actually get all the warnings that we'd normally get if we included
the headers as user headers.
The goal of the test suite is normally to mirror as closely as possible
how users would use the library. Technically, this change goes against
that philosophy, since users should be using `-isystem` (if they ever
need to specify the libc++ path explicitly, which should be a rare
occurence). However, I believe fishing out additional warnings from
the headers provides more value, hence this change. Ideally, we'd be
able to still use `-isystem`, but instruct Clang to still emit warnings
from the libc++ headers (e.g. we could tell Clang to emit warnings in
any file inside `<...>/usr/include/c++/v1`).
Reviewed By: #libc, ldionne, #libc_abi
Spies: Mordante, EricWF, mstorsjo, mgorny, aheejin, arichardson, philnik, jloser, libcxx-commits
Differential Revision: https://reviews.llvm.org/D118616
Based off Agner and AMD SoG tables, the XOP VPHADD/VPHSUB unary horizontal ops are as fast as basic arithmetic ops, not the slower SSSE3 binary horizontal add/sub ops. This also matches what the bdver2 model already lists.
Noticed while investigating reduction add optimizations.
This takes care of normalizing newlines back to single LF instead
of CRLF.
Fix up a couple tests that accidentally pass binary data to stdout.
Differential Revision: https://reviews.llvm.org/D120623
Note, reducing ios.width() in put_long_double instead of using variable
padding, when using a variable width symbol. Some of those tests didn't
actually trigger any padding in the existing form, with a longer
currency symbol; reduce the width so there's no actual padding with the
slightly shorter currency symbol either.
The tests for the international currency symbol use the same amount of
padding on all platforms, so they still exercise the padding properly.
Differential Revision: https://reviews.llvm.org/D120317
This commit deletes the old dialect conversion-based bufferization patterns, which are now obsolete.
Differential Revision: https://reviews.llvm.org/D120883
The costs of vector shifts was 2 as opposed to 1, as the nodes are
marked custom. Fix this like the others and mark the nodes as cheap.
Differential Revision: https://reviews.llvm.org/D120773
VectorBuilder wraps around an IRBuilder and
VectorBuilder::createVectorInstructions emits VP intrinsics as if they
were regular instructions.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D105283
The problem with leak bug reports is that the most interesting event in the code
is likely the one that did not happen -- lack of ownership change and lack of
deallocation, which is often present within the same function that the analyzer
inlined anyway, but not on the path of execution on which the bug occured. We
struggle to understand that a function was responsible for freeing the memory,
but failed.
D105819 added a new visitor to improve memory leak bug reports. In addition to
inspecting the ExplodedNodes of the bug pat, the visitor tries to guess whether
the function was supposed to free memory, but failed to. Initially (in D108753),
this was done by checking whether a CXXDeleteExpr is present in the function. If
so, we assume that the function was at least party responsible, and prevent the
analyzer from pruning bug report notes in it. This patch improves this heuristic
by recognizing all deallocator functions that MallocChecker itself recognizes,
by reusing MallocChecker::isFreeingCall.
Differential Revision: https://reviews.llvm.org/D118880
To make this actually trigger, we also need to check whether the
function types differ, which is a hidden cast under opaque pointers.
The transform is somewhat less relevant there because it is
primarily about pointer bitcasts, but it can also happen with other
bit- or pointer-castable types.
Byval handling is easier with opaque pointers because there is no
need to adjust the byval type, we only need to make sure that it's
still a pointer.
The logic for handling this was fixed in
8d7f118ab2, but the check for byval
on the callee was retained. This resulted in a weird situation
where the transform would work depending on whether the byval
was only on the call or on both the call and the function.
The compiler currently crashes for scalable types when compiling with
+sme, e.g.
define <vscale x 4 x i32> @foo(<vscale x 4 x i32> %a) {
ret <vscale x 4 x i32> %a
}
since it doesn't know how to legalize the types. SME implies a subset of
SVE (+streaming-sve), the hasSVE predication in the backend needs
extending to consider types/operations that are legal in Streaming SVE.
This is the first patch adding legal types <-> register classes. Before
making the change +sve(2) was temporarily replaced with +sme in all the
intrinsics tests to see what failed, and again after making the change.
For all the tests that passed after adding the legal types another RUN
line has been added for +streaming-sve. More patches to follow.
Reviewed By: sdesmalen
Differential Revision: https://reviews.llvm.org/D118561
Previously, we didin't build a DeclRefExpr which refers to an invalid declaration.
In this patch, we handle this case by building an empty RecoveryExpr,
which will preserve more broken code (AST parent nodes that contain the
RecoveryExpr is preserved in the AST).
Differential Revision: https://reviews.llvm.org/D120812
Craig Topper