Patterns in `LowerContractionToSMMLAPattern` are designed to handle
vector-to-matrix multiplication but not matrix-to-vector. This leads to
the following error when processing `rhs` with rank < 2:
```
iree-compile: /usr/local/google/home/kooljblack/code/iree-build/llvm-project/tools/mlir/include/mlir/IR/BuiltinTypeInterfaces.h.inc:268: int64_t mlir::detail::ShapedTypeTrait<mlir::VectorType>::getDimSize(unsigned int) const [ConcreteType = mlir::VectorType]: Assertion `idx < getRank() && "invalid index for shaped type"' failed.
```
Updates to explicitly check the rhs rank and fail cases that cannot
process.
This PR is a small improvement of parsing of base type name in builtins,
allowing to understand `unsigned ...` types. The test case that fails
without the fix is attached.
We recently optimized the code when the Odd vector was undef to fix a
poison bug.
There are additional optimizations we can do if the even vector is
undef. With Zvbb, we can use a single vwsll. Without Zvbb, we can use a
vzext.vf2 and a vsll.
When writing out a PCM, we compute the set of module maps that did
affect the compilation and we strip the rest to make the output
independent of them. The most common way to read a module map that is
not affecting is with implicit module map search. The other option is to
pass a bunch of unnecessary `-fmodule-map-file=<path>` arguments on the
command-line, in which case the client should probably not give those to
Clang anyway.
This makes serialization of explicit modules faster, mostly due to
reduced file system traffic.
This test just checks for the stdout/stderr of clang, but it
incidentally tries to write to `a.out` in the current directory, which
may be write protected. Typically one would write `clang -o %t.o` for a
writeable dir, but since we only care about stdout/stderr, throw away
the object file and just write to /dev/null instead.
Instead of relying on known-bits for strictly positive, use the
`isKnownPositive` API. This will use `isKnownNonZero` which is more
accurate.
Closes#88170
Adds support for: `{s,u}{add,sub,mul}.with.overflow`
The logic is identical to the the non-overflow binops, we where just
missing the cases.
Closes#87701
This improves a pattern that occurs in 531.deepsjeng_r. Reducing the
dynamic instruction count by 0.5%.
This may be possible to improve in SelectionDAG, but given the special
cases around shXadd formation, it's not obvious it can be done in a
robust way without adding multiple special cases.
I've used a GEP with 2 indices because that mostly closely resembles the
motivating case. Most of the test cases are the simplest GEP case. One
test has a logical right shift on an index which is closer to the
deepsjeng code. This requires special handling in isel to reverse a
DAGCombiner canonicalization that turns a pair of shifts into (srl (and
X, C1), C2).
If the first node is a gather node with extractelement instructions,
still need to put the vector value after all instructions, not after the
very first one.
As pointed out by Fraser, KillSrcReg is always false at this point in
code, and having the inconcistency on whether we check the flag between
the if and else blocks is confusing.
As a followup to my previous commits, this is an implementation of a
single clause, in this case the 'default' clause. This implements all
semantic analysis for it on compute clauses, and continues to leave it
rejected for all others (some as 'doesnt appertain', others as 'not
implemented' as appropriate).
This also implements and tests the TreeTransform as requested in the
previous patch.
This patch moves SYCL-related `Sema` functions into new `SemaSYCL`
class, following the recent example of OpenACC and HLSL. This is a part
of the effort to split `Sema`. Additional context can be found in
#82217, #84184, #87634.
Summary:
The linker wrapper attempts to maintain consistent semantics with
existing host invocations. Static libraries by default only extract if
there are non-weak symbols that remain undefined. However, we have
situations between linkers that put different meanings on ordering. The
ld.bfd linker requires static libraries to be defined after the symbols,
while `ld.lld` relaxes this rule. The linker wrapper went with the
former as it's the easier solution, however this has caused a lot of
issues as I've had to explain this rule to several people, it also make
it difficult to include things like `libc` in the OpenMP runtime because
it would sometimes be linked before or after.
This patch reworks the logic to more or less perform the following logic
for static libraries.
1. Split library / object inputs.
2. Include every object input and record its undefined symbols
3. Repeatedly try to extract static libraries to resolve these
symbols. If a file is extracted we need to check every library
again to resolve any new undefined symbols.
This allows the following to work and will cause fewer issues when
replacing HIP, which does `--whole-archive` so it's very likely the old
logic will regress.
```console
$ clang -lfoo main.c -fopenmp --offload-arch=native
```
Better to walk over SLP nodes rather than single values. Matching
a value to a node is not a 1-to-1 relation, one value may be part of
several nodes and compiler may get wrong node, when trying to map it.
Currently there are no such issues detected, but they may appear in
future.
BinomialCoefficient computes the value of W-bit IV at iteration It of a loop. When W is 1, we can call multiplicative inverse on 0 which triggers an assert since 1b76120.
Since the arithmetic is supposed to wrap if It or K does not fit in W bits, do the truncation into W bits after we do the shift.
Fixes#87798
This ports https://github.com/openxla/xla/pull/10503 by @pearu. The new
implementation matches mpmath's results for most inputs, see caveats in
the linked pull request. In addition to the filecheck test here, the
accuracy was tested with XLA's complex_unary_op_test and its MLIR
emitters.
Avoid replacing replacing a chunk with one from a different type. It's
mostly a concern for ARM64X, where we don't want to merge aarch64 and
arm64ec chunks, but it may also in theory happen between arm64ec and
x86_64 chunks.
Although the identifier-naming.cpp lit test expected macro arguments not
to be renamed, the code seemed to already allow it. The code was simply
not being exercised because a SourceManager argument wasn't being
provided. With this change, renaming of macro arguments that expand to
renamable decls is permitted.
There has been an optimization for `SizeOfPackExprs` since c5452ed9, in
which
we overlooked a case where the template arguments were not yet
formed into a `PackExpansionType` at the token annotation stage. This
led to a problem in that a template involving such expressions may
lose its nature of being dependent, causing some false-positive
diagnostics.
Fixes https://github.com/llvm/llvm-project/issues/84220
Fixes https://github.com/llvm/llvm-project/issues/63818 for control flow
out of an expressions.
#### Background
A control flow could happen in the middle of an expression due to
stmt-expr and coroutine suspensions.
Due to branch-in-expr, we missed running cleanups for the temporaries
constructed in the expression before the branch.
Previously, these cleanups were only added as `EHCleanup` during the
expression and as normal expression after the full expression.
Examples of such deferred cleanups include:
`ParenList/InitList`: Cleanups for fields are performed by the
destructor of the object being constructed.
`Array init`: Cleanup for elements of an array is included in the array
cleanup.
`Lifetime-extended temporaries`: reference-binding temporaries in
braced-init are lifetime extended to the parent scope.
`Lambda capture init`: init in the lambda capture list is destroyed by
the lambda object.
---
#### In this PR
In this PR, we change some of the `EHCleanups` cleanups to
`NormalAndEHCleanups` to make sure these are emitted when we see a
branch inside an expression (through statement expressions or coroutine
suspensions).
These are supposed to be deactivated after full expression and destroyed
later as part of the destructor of the aggregate or array being
constructed. To simplify deactivating cleanups, we add two utilities as
well:
* `DeferredDeactivationCleanupStack`: A stack to remember cleanups with
deferred deactivation.
* `CleanupDeactivationScope`: RAII for deactivating cleanups added to
the above stack.
---
#### Deactivating normal cleanups
These were previously `EHCleanups` and not `Normal` and **deactivation**
of **required** `Normal` cleanups had some bugs. These specifically
include deactivating `Normal` cleanups which are not the top of
`EHStack`
[source1](92b56011e6/clang/lib/CodeGen/CGCleanup.cpp (L1319)),
[2](92b56011e6/clang/lib/CodeGen/CGCleanup.cpp (L722-L746)).
This has not been part of our test suite (maybe it was never required
before statement expressions). In this PR, we also fix the emission of
required-deactivated-normal cleanups.
This turns the current `Pointer` class into a discriminated union of
`BlockPointer` and `IntPointer`. The former is what `Pointer` currently
is while the latter is just an integer value and an optional
`Descriptor*`.
The `Pointer` then has type check functions like
`isBlockPointer()`/`isIntegralPointer()`/`asBlockPointer()`/`asIntPointer()`,
which can be used to access its data.
Right now, the `IntPointer` and `BlockPointer` structs do not have any
methods of their own and everything is instead implemented in Pointer
(like it was before) and the functions now just either assert for the
right type or decide what to do based on it.
This also implements bitcasts by decaying the pointer to an integral
pointer.
`test/AST/Interp/const-eval.c` is a new test testing all kinds of stuff
related to this. It still has a few tests `#ifdef`-ed out but that
mostly depends on other unimplemented things like
`__builtin_constant_p`.
After commit 18c5f3c3 ("[RegisterScavenger][RISCV] Don't search for
FrameSetup instrs if we were searching from Non-FrameSetup instrs"), we
can revert the `sp` adjustment 4e2364a2 ("[LoongArch] Add emergency
spill slot for GPR for large frames") to generate better code, as the
issue with `RegScavenger` has been resolved.
Fixes#88109
