The scope of DT updates are very limited when unrolling loops: the DT
should only need updating for
* new blocks added
* exiting blocks we simplified branches

This can be done manually without too much extra work.
MergeBlockIntoPredecessor also needs to be updated to support direct
DT updates.

This fixes excessive time spent in DTU for same cases. In an internal
example, time spent in LoopUnroll with this patch goes from ~200s to 2s.

It also is slightly positive for CTMark:
* NewPM-O3: -0.13%
* NewPM-ReleaseThinLTO: -0.11%
* NewPM-ReleaseLTO-g: -0.13%

Notable improvements are mafft (~ -0.50%) and lencod (~ -0.30%), with no
workload regressed.

https://llvm-compile-time-tracker.com/compare.php?from=78a9ee7834331fb4360457cc565fa36f5452f7e0&to=687e08d011b0dc6d3edd223612761e44225c7537&stat=instructions:u

Reviewed By: kuhar

Differential Revision: https://reviews.llvm.org/D141487

Upper bound and step size should be symbols instead of dims.

Differential Revision: https://reviews.llvm.org/D142136

Reviewers: dvyukov

Reviewed-By: dvyukov

Differental Revision: https://reviews.llvm.org/D140688

Implement InferIntRangeInterface for all operations in the Index dialect. The
inference implementation, unlike the one for Arith, accounts for the
fact that Index can be either 64 or 32 bits long by evaluating both
cases. Bounds are stored as if index were i64, but when inferring new
bounds, we compute both f(...) and f(trunc(...)). We then compare
trunc(f(...)) to f(trunc(...)). If they are equal in the relevant
range components, we use the 64-bit range computation, otherwise we
give the range ext(f(trunc(...))) union f(...).

Note that this can cause surprising behavior as seen in the tests,
where, for example, the order of min and max operations impacts the
behavior of the inference. The inference could perhaps be made more
precise in the future (ex. by tracking 32 and 64-bit results
separately and having them influence each other somehow) butt, since
my project targets an index=i32 platform and doesn't see index-valued
values > uint32_max, I'm not too concerned about it.

Depends on https://reviews.llvm.org/D141299

Depends on https://reviews.llvm.org/D141296

Reviewed By: Mogball

Differential Revision: https://reviews.llvm.org/D140899

Summary:
This patch adds OpenMP runtime to the linker command line if -fopenmp is specifed for AIX.

Reviewed by: daltenty

Differential Revision: https://reviews.llvm.org/D141862

Current version there is a fix-it for

template <class> constexpr int x = 0;
template <> constexpr int x<int>; // fix-it here
but it will cause

template <> constexpr int x = 0<int>;

Differential Revision: https://reviews.llvm.org/D139705

This reverts commit a0f8bdbb.

Several bots are failing in shtest-format.py, likely because of this.

Reviewed By: jeanPerier, PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D142070

The std::optional implementation in MSVC causes this code to produce a
sign comparison warning. This ensures the types are the same sign.

This patch adds the %f/F/e/E/g/G/a/A conversions for scanf, as well as
accompanying tests. This implementation matches the definition set forth
in the standard, which may conflict with some other implementations.

Reviewed By: sivachandra

Differential Revision: https://reviews.llvm.org/D141091

The scanf implementation needs a dynamically resizing string class. This
patch adds a minimal version of that class along with tests to check the
current functionality.

Reviewed By: sivachandra

Differential Revision: https://reviews.llvm.org/D141162

-> Use file pathname from the Flang frontend. It is the frontend
that is in-charge of finding the files and is hence the canonical
source for paths.
-> Convert pathname to absolute pathname while creating the moduleOp.

Co-authored-by: Peter Klausler <pklausler@nvidia.com>

Reviewed By: PeteSteinfeld, vzakhari, jeanPerier, awarzynski

Differential Revision: https://reviews.llvm.org/D141674

Implements parts of
- P2286R8 Formatting Ranges

Depends on D140653

Reviewed By: ldionne, #libc

Differential Revision: https://reviews.llvm.org/D141761

shuffle_vector instructions are serialized targeting SVE fixed vectors, see
https://reviews.llvm.org/D139111. This patch disables
optimizeExtendOrTruncateConversion peepholes that generates shuffle_vector.

Differential Revision: https://reviews.llvm.org/D141439

Implements parts of
- P2286R8 Formatting Ranges

Depends on D140653

Reviewed By: ldionne, #libc

Differential Revision: https://reviews.llvm.org/D141290

Only allow replacements of nodes that have a single user. This is better as
simple instructions (e.g. XGRK) are one cycle faster, and it helps in cases
where both inputs share a common node.

Review: Ulrich Weigand

This corresponds to the cmake change in 81ca5aa4

AFAICT all in-tree lit tests have been converted to use `target=...`
and so there is no longer any need for triples being special.
Some project config files still define their own features based on
the triple, but those are normal feature words (although now are
redundant with target= checks).

Downstream tests that use triple substrings will need to convert.
For example:
    UNSUPPORTED: -aix
    XFAIL: arm
becomes
    UNSUPPORTED: target={{.*}}-aix{{.*}}
    XFAIL: target=arm{{.*}}

You can do
    git log --grep "special handling for triples"
to find many examples of updates to the upstream tests.

https://discourse.llvm.org/t/rfc-lits-requires-and-triples/66041

Differential Revision: https://reviews.llvm.org/D141007

Pointer association to unlimited polymorphic target is allowed for
unlimited polymorphic pointer and non-extensible derived-type.
This is checked by the semantic and this patch allows it in the
fir.rebox operation.

Reviewed By: jeanPerier, PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D142104

Result must carry the polymorphic type information
from the source.

Reviewed By: jeanPerier, PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D142095

CLASS DEFAULT needs to be the last attribute when fir.select_type op is created.
It needs to be at its actual position in the Fortran code when the TypeGuardStmt
are processed. The current lowering was crashing when CLASS DEFAULT was not at
the last position.
This patch fixes the issue by tracking the actual position of the CLASS DEFAULT
type guard and set it at the correct position after the fir.select_type op
is created.

Reviewed By: jeanPerier, PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D142091

Implements parts of
- P2286R8 Formatting Ranges
- P2585R0 Improving default container formatting

Depends on D140651

Reviewed By: ldionne, #libc

Differential Revision: https://reviews.llvm.org/D140653

By switching to --check-globals. Also make sure that the
!tbaa.struct metadata mapping is preserved.

Expanding upon https://reviews.llvm.org/D138203, allow null indices in
InsertElts to be matched with any value and be duplicated if the fixed
vector the scalar values are inserted into is poison, and the scalable vector
the subvector being inserted into is poison.

Differential Revision: https://reviews.llvm.org/D141846

Currently, the code assumes that all boolean-typed values are an instance of
`BoolValue` (or its subclasses). Yet, lvalues violate this assumption. This
patch drops the assumption and strengthens the check to confirm the shape of
both values being joined.

The patch also notes as FIXMES a number of problems discovered fixing this bug.

Differential Revision: https://reviews.llvm.org/D141709

Compare to other component ref lowering, the hlfir.designate result type
computation is different, and the allocatable/pointer/contiguous must
be set on the hlfir.designate so that the component attributes are
kept in the IR.

Differential Revision: https://reviews.llvm.org/D142111

If we're only changing the type of the load, preserve the noundef
metadata.

This patch is to diagnose the case when a type bound procedure is passed as an actual procedure argument.

   call sub0(t%t3%t2%t%info1)

Fix: https://github.com/llvm/llvm-project/issues/55826

Committed on behalf of DanielCChen

Differential Revision: https://reviews.llvm.org/D141506

The !noundef metadata is currently dropped.

As Armv9-a implies SVE2 it implies SVE (added in D141411) and so it
should also imply FP16, which this patch adds. This helps get the target
features correct when using `target("arch=armv9-a")` attributes.

There is also an adjustment to AssertSameExtensionFlags in this patch to
make it print cpu names, useful when the TargetParser unit tests are run
through lit to distinguish which cpu is failing.

Differential Revision: https://reviews.llvm.org/D142087

The entries inside a "target data end" is processed in three steps:

  1. Query internal data maps for the entries and dispatch any necessary
     device-side operations (i.e., data retrieval);
  2. Synchronize the such operations;
  3. Update the host-side pointers and remove any entry which reference
     counter reached zero.

Such steps may be executed by multiple threads which may even operate on
the same entries. The current implementation (D121058) tries to
synchronize these threads by tracking the "owner" for the deletion of
each entry using their thread ID. Unfortunately it may failed to do so
because of the following reasons:

  1. The owner is always assigned at the first step only if the
     reference count is 0 when the map is queried. This does not work
     when such owner thread is faster than a previous one that is also
     processing the same entry on another "target data end", leading to
     user-after-free problems.
  2. The entry is only added for post-processing (step 3) if its
     reference count was 0 at query time (step 1). This does not allow
     for threads to exchange responsibility for the deletion, leading
     again to user-after-free problems.
  3. An entry may appear multiple times in the arguments array of a
     "target data end", which may lead to deleting the entry
     prematurely, leading, again, to user-after-free problems.

This patch addresses these problems by tracking all the threads that are
using an entry at "target data end" region through a counter, ensuring
only the last one deletes it when needed. It also ensures that all
entries that are successfully found inside the data maps in step 1 are
also processed in step 3, regardless if their reference count was zeroed
or not at query time. This ensures the deletion ownership may be passed
to any thread that is using such entry.

Reviewed By: ye-luo

Differential Revision: https://reviews.llvm.org/D132676

DexExpectStepOrder uses the line to expect a debugger step from the actual line
of the command in the Dexter source file. Now Dexter scripts have mainly moved
to thier own script files instead of the actual source, there should be a
option to override this behaviour to choose your own debugger step location.

Reviewed By: Orlando

Differential Revision: https://reviews.llvm.org/D142099

And regenerate test checks to pick up new names.

I made a typo here, this was supposed to be !align rather than
!aligned. But then !align can only be applied to loads, not calls
(where one would use the return attribute instead). And freeze
can't be pushed through loads anyway, so there's no way to test
this case (same as !nonnull).

Also adds uses of the new printing in analysis inner loop.

Differential Revision: https://reviews.llvm.org/D141716