[VPlan] Use inferScalarType in VPInstruction::ResumePhi codegen (NFC).
Use VPlan-based type analysis to retrieve type of phi node. Also adds
missing type inference for ResumePhi and ComputeReductionResult opcodes.
[clang-tidy] use specified type verbatim in modernize-use-using fix (#113837)
Previously, the implementation used the printed type, which contains
expanded
macro arguments, deletes comments, and removes function argument names
from the alias declaration. Instead, this check can be more surgical and
use the
actual written type verbatim.
Fixes #33760
Fixes #37846
Fixes #41685
Fixes #83568
Fixes #95716
Fixes #97009
[libc++][C++03] Use `__cxx03/` headers in C++03 mode (#109002)
This patch implements the forwarding to frozen C++03 headers as
discussed in
https://discourse.llvm.org/t/rfc-freezing-c-03-headers-in-libc. In the
RFC, we initially proposed selecting the right headers from the Clang
driver, however consensus seemed to steer towards handling this in the
library itself. This patch implements that direction.
At a high level, the changes basically amount to making each public
header look like this:
```
// inside <vector>
#ifdef _LIBCPP_CXX03_LANG
# include <__cxx03/vector>
#else
// normal <vector> content
#endif
[4 lines not shown]
Store sysreg names in-line with their descriptors. (#119157)
This wastes some disk space, because we have to size the arrays for
the maximum possible length. However, it eliminates dynamic relocations
from the SysRegsList arrays.
[SandboxVec][DAG] Register move instr callback (#120146)
This patch implements the move instruction notifier for the DAG.
Whenever an instruction moves the notifier will maintain the DAG.
[TableGen][GlobalISel] Add GINodeEquiv for sdivrem/udivrem (#120819)
These nodes are not currently used in DAG patterns. Their GlobalISel
equivalents are primarily useful for testing TableGen backend, as they
produce two results. (There doesn't seem to be other such nodes.)
[mlir python] Port in-tree dialects to nanobind. (#119924)
This is a companion to #118583, although it can be landed independently
because since #117922 dialects do not have to use the same Python
binding framework as the Python core code.
This PR ports all of the in-tree dialect and pass extensions to
nanobind, with the exception of those that remain for testing pybind11
support.
This PR also:
* removes CollectDiagnosticsToStringScope from NanobindAdaptors.h. This
was overlooked in a previous PR and it is duplicated in Diagnostics.h.
---------
Co-authored-by: Jacques Pienaar <jpienaar at google.com>
Revert "Add a pass to collect dropped var stats for MIR (#120780)"
This reverts commit 3bf91ad2a9c75dd045961e45fdd830fd7b7a5455.
(llvmorg-20-init-16123-g3bf91ad2a9c7)
`llvm/CodeGen` should not depend on `llvm/Passes`.
[SelectionDAG] Virtualize isTargetStrictFPOpcode / isTargetMemoryOpcode (#119969)
With this change, targets are no longer required to put memory / strict-fp opcodes after special
`ISD::FIRST_TARGET_MEMORY_OPCODE`/`ISD::FIRST_TARGET_STRICTFP_OPCODE` markers.
This will also allow autogenerating `isTargetMemoryOpcode`/`isTargetStrictFPOpcode (#119709).
Pull Request: https://github.com/llvm/llvm-project/pull/119969
[TableGen][GISel] Learn to import patterns with optional defs (#120470)
The number of skipped patterns reduces for ARM from 4278 to 4257.
This is the only in-tree target that makes use of OptionalDefOperand.
Pull Request: https://github.com/llvm/llvm-project/pull/120470
AMDGPU: Fix assert on physreg MUBUF rsrc operand
The stack case uses a physical register and should not ordinarily
reach here, but strange things happen at -O0. The testcase still
errors because we do not yet attempt to handle arbitrary dynamic
sized allocas yet.
Fixes: SWDEV-503538
[libc] Remove old RPC host call extension for GPU (#120800)
Summary:
This was originally a hacked together function that served to just
implement some features for OpenMP. That has been moved into OpenMP
itself now that we have exported RPC properly. This can now be deleted.
[TableGen][GISel] Delete unused `Src` arguments (NFC) (#120445)
The last uses were removed in #120332 and #120426.
When emitting renderers, we shouldn't look at the source DAG at all. The
required information is provided by the destination DAG and by the
instructions referenced in that DAG. Sometimes, we do want to know if a
result was referenced in the source DAG; this can be checked by calling
`RuleMatcher::hasOperand`. Any other use of the source DAG when emitting
renderers is likely an error.
Pull Request: https://github.com/llvm/llvm-project/pull/120445
[libc] Make a couple of math smoke tests more robust. (#120808)
Make sure to clear out all FE_ALL_EXCEPT bits both before and after
invoking function under test -- otherwise the very first check for the
exception bits in the unit test may fail due to bits set prior to test
invocation.
[llvm-gsymutil] Ensure stable ordering of merged functions (#120796)
Previously, the test
`llvm/test/tools/llvm-gsymutil/ARM_AArch64/macho-gsym-merged-callsites-dsym.yaml`
[was disabled](https://github.com/llvm/llvm-project/pull/119957/files)
due to failing on Linux.
The issue is that the merged functions appear in a different order in
the final produced `gSYM`.
To ensure deterministic ordering of merged functions, we change the
sorting of functions to use the `stable_sort` algorithm. Before merged
functions, this was not an issue as the address is used as a comparator
in the sorting algorithm so there was no chance of two functions testing
as identical according to the comparator.
Confirmed that test now passes locally with
`-DLLVM_ENABLE_EXPENSIVE_CHECKS=ON`.
[memprof] Undrift MemProf profile even when some frames are missing (#120500)
This patch makes the MemProf undrifting process a little more lenient.
Consider an inlined call hierarchy:
foo -> bar -> ::new
If bar tail-calls ::new, the profile appears to indicate that foo
directly calls ::new. This is a problem because the perceived call
hierarchy in the profile looks different from what we can obtain from
the inline stack in the IR.
Recall that undrifting works by constructing and comparing a list of
direct calls from the profile and that from the IR. This patch
modifies the construction of the latter. Specifically, if foo calls
bar in the IR, but bar is missing the profile, we pretend that foo
directly calls some heap allocation function. We apply this
transformation only in the inline stack leading to some heap
allocation function.
Thread Safety Analysis: Support passing scoped locks between functions with appropriate annotations (#110523)
This is helpful when multiple functions operate on the same
capabilities, but we still want to use scoped lockable types for
readability and exception safety.
- Introduce support for thread safety annotations on function parameters
marked with the 'scoped_lockable' attribute.
- Add semantic checks for annotated function parameters, ensuring
correct usage.
- Enhance the analysis to recognize and handle parameters annotated for
thread safety, extending the scope of analysis to track these across
function boundries.
- Verify that the underlying mutexes of function arguments match the
expectations set by the annotations.
Limitation: This does not work when the attribute arguments are class
members, because attributes on function parameters are parsed
differently from attributes on functions.
[Clang-CL][DXC] Expose -fdiagnostic-color= to clang-cl and clang-dxc (#120644)
Exposing `-fdiagnostic-color=` to clang-cl and clang-dxc.
`-fcolor-diagnostics` and `-fno-color-diagnostics` are already allowed
in both of these and `-fdiagnostics-color=` allows one additional value,
`auto`.
I've added the tests for clang-cl to `cl-options.c` as per the comments
in the issue linked below. I couldn't finding a suitable existing file
to add the clang-dxc tests to so I've created a new one.
Resolves #119184
[flang][cuda] Correctly allocate descriptor in managed memory when reboxing (#120795)
Reboxing might create a new in memory descriptor. If this one was
allocate with managed memory, allocate the new one in managed memory as
well.
