Looks like these files are generated by the
`generate_formatted_state.py` script as a "status report" of state of
clang-format compliance of files in the LLVM git repo. As such, they do
not belong to the repo itself, so deleting them.
Please see:
https://discourse.llvm.org/t/clang-docs-tools-clang-formatted-files-txt/82803
The "topological" sorting was behaving incorrectly in some cases:
the exit of a loop could have a lower rank than a node in the loop.
This causes issues when structurizing some patterns, and also codegen
issues as we could generate BBs in the incorrect order in regard to the
SPIR-V spec.
Fixing this ordering alone broke other parts of the structurizer, which
by luck worked. Had to fix those.
Added more test cases, especially to test basic patterns.
I also needed to tweak/disable some tests for 2 reasons:
- SPIR-V now required reg2mem/mem2reg to run. Meaning dead stores
are optimized away. Some tests require tweaks to avoid having the
whole function removed.
- Mem2Reg will generate variable & load/stores. This generates
G_BITCAST in several cases. And there is currently something wrong
we do with G_BITCAST which causes MIR verifier to complain.
Until this is resolved, I disabled -verify-machineinstrs flag on
those tests.
---------
Signed-off-by: Nathan Gauër <brioche@google.com>
Many tests that were in test/Analysis/CostModel were actually
loop vectoriser tests. I've moved them as follows:
Analysis/CostModel/X86 -> Transforms/LoopVectorize/X86/CostModel
Analysis/CostModel/AArch64/arith-fp-frem.ll ->
Transforms/LoopVectorize/AArch64/arith-fp-frem-costs.ll
This starts moving `X86Builtins.def` to be a tablegen file. It's quite
large, so I think it'd be good to move things in multiple steps to avoid
a bunch of merge conflicts due to the amount of time this takes to
complete.
SVE2 adds the constructive splice instruction, which takes a tuple.
Even though the register allocator must ensure that the tuple uses
consecutive registers for the tuple, it's likely to be more efficient
than using the destructive splice instruction when the first operand
is reused.
Allow construction of DstOp with VRegAttrs.
Also allow construction with register class or bank and LLT.
Intended to be used in lowering code for reg-bank-select where
new registers need to have both register bank and LLT.
Add support for new type of DstOp in CSEMIRBuilder.
update_mc_test_check script handle the "error case testline" wrong in
three cases:
1. when user select "--llvm-mc-binary" with a path, the script does not
add "not" on top of the "--llvm-mc-binary" and thus getting non-zero
exit code and failed.
2. When "not" is presented in runline while not all testlines are
expected to fail, the script need to check if the "not" is needed when
it execute llvm-mc line by line. Otherwise the script will fail on
testline which is passing.
3. When there are multiple runlines, the error checkline need to use
correct line offset for "[[LINE-X]]"
This patch solve these three issues
When transferring resources, the destination tracker key may not be in
the internal map, invalidating iterators and value references. The added
test creates such situation and would fail before with "Finalized
allocation was not deallocated."
For good measure, fix the same pattern in RTDyldObjectLinkingLayer
which is harder to test because it "only" results in memory managers
being deleted in the wrong order.
This commit adds support for bufferizing external functions that have no
body. Such functions were previously rejected by One-Shot Bufferize if
they returned a tensor value.
This commit is in preparation of removing the deprecated
`func-bufferize` pass. That pass can bufferize external functions.
Also update a few comments.
This commit changes the format of the materialization error message.
Previously: `failed to legalize unresolved materialization from ('f64')
to 'f32' that remained live after conversion`
Now: `failed to legalize unresolved materialization from ('f64') to
('f32') that remained live after conversion`
This commit is in preparation of merging the 1:1 and 1:N dialect
conversions. At that point, target materializations may create more than
one SSA value. I am sending this change as a separate PR to keep the
main PR smaller.
This patch adds methods to `EntryBlockArgs` to access the full list of
entry block argument-related symbols and variables, in their standard
order. This helps centralizing this logic in as few places as possible
to avoid future inconsistencies.
Removes sve-bf16, sve-ebf16, and sve-i8mm since they are obsolete. One
could write target_version("sve+bf16") instead of sve-bf16 for instance.
Approved in ACLE as https://github.com/ARM-software/acle/pull/353
Issue deprecation warning for these directives.
Lowering currently supports parallel master, for all other combined or
composite directives involving master, issue TODO errors.
Note: The first commit changes the formatting and generalizes the
deprecation message emission for reuse in the second commit. I can pull
it out into a separate commit if required.
In Zig, we have a tool that updates our CPU model/feature data from
LLVM's. Noticed these typos when running it for LLVM 19.
Note: I don't have commit access.
This reverts commit ef44e4659878f2. The patch was originally reverted
because it was
deemed to introduce a performance regression for small inputs, however
it also fixed
a previous performance regression for larger inputs. So overall, this
patch is desirable.
On ANDNOT capable targets we can always do this profitably, without ANDNOT we only attempt this if we don't introduce an additional NOT
Followup to #112547
Handling is similar to RecordType with following differences:
1. No check for cyclic references
2. No extra processing for lower bounds of array members.
3. No line information as TupleType is a lowering artefact and does not
really represent an entity in the code.
These are primarily meant to test disassembler and that no more than
one variant per instruction is in DisassemblerTables as that can cause
confusion when decoding v0 (vgpr0) whose value when encoded is 0.
Reproducer:
```
//--- a.cppm
export module a;
int func();
static int a = func();
//--- a.cpp
import a;
```
The `func()` should only execute once. However, before this patch we
will somehow import `static int a` from a.cppm incorrectly and
initialize that again.
This is super bad and can introduce serious runtime behaviors.
And also surprisingly, it looks like the root cause of the problem is
simply some oversight choosing APIs.
The trampoline script used on Windows (due to the absence of shebang
support) doesn't properly expand the path to the Python script, as it
leaves out the drive letter.
Functionally equivalent reproducer in action
```
PS C:\Users\mate> gc (gcm git-clang-formatish.bat).Source
@ECHO OFF
echo "%~pn0" %*
PS C:\Users\mate> git-clang-formatish
"\Users\mate\git-clang-formatish"
```
Adding `d` to the variable modifiers [as per the
docs](https://learn.microsoft.com/en-us/windows-server/administration/windows-commands/for)
the drive letter is added. Even in the magical cases when it works.
(I couldn't reproduce, but I suspect it's only tested from some
bash/cygwin variant, where the path becomes `/c/Program Files/...`, but
the drive letter is needed. Without it, I also observed cases when used
via `git clang-format` (without the inital dash) it tries to infer the
drive letter based on the current working directory. In that case it
fails to find `D:\Program Files\LLVM\bin\clang-format.exe`, which
naturally fails, because `Program Files` is on `C:`)
Incomplete types are not considered trivially copyable by clang but we
don't want to warn about invalid argument for memcpy / memset in that
case because we cannot prove they are not Trivially Copyable.
Mark the whole StmtExpr invalid when the last statement in compound
statement is invalid.
Because the last statement need to do copy initialization, it causes
subsequent errors to simply ignore last invalid statement.
Fixed: #113468
This checker has a notion of a guardian variable which is a variable and
keeps the object pointed to by a raw pointer / reference in an inner
scope alive long enough to "guard" it from use-after-free. But such a
guardian variable fails to flawed to keep the object alive if it ever
gets mutated within the scope of a raw pointer / reference.
This PR fixes this bug by introducing a new AST visitor class,
GuardianVisitor, which traverses the compound statements of a guarded
variable (raw pointer / reference) and looks for any operator=, move
constructor, or calls to "swap", "leakRef", or "releaseNonNull"
functions.
This patch converts `SDNodeFlags` into an enumeration as we did for
`FastMathFlags`. It simplifies the implementation and improves
compile-time. This patch is NFC since it doesn't break SDNodeFlags API.
I hope it was missed unintentionally, pushing the same for the review.
Ref: https://github.com/llvm/llvm-project/pull/108745
---------
Co-authored-by: Santhosh Kumar Ellendula <sellendu@hu-sellendu-hyd.qualcomm.com>
Co-authored-by: Santhosh Kumar Ellendula <sellendu@hu-sellendu-lv.qualcomm.com>
This did not consider the IR change to allow a scalar base with a vector
offset part. Reject any users that are not explicitly handled.
In this situation we could handle the vector GEP, but that is a larger
change. This just avoids the IR verifier error by rejecting it.
Add support for generating random hotness in the memprof profile writer,
to be used for testing. The random seed is printed to stderr, and an
additional option enables providing a specific seed in order to
reproduce a particular random profile.
Infers member MayReadFromMemory, MayWriteToMemory, and
MayHaveSideEffects based on intrinsic attributes.
---------
Co-authored-by: Florian Hahn <flo@fhahn.com>
