Using GCNDownwardRPTracker or GCNUpwardRPTracker the pass collects register pressure values for a function and prints these values next to instructions. Output can be used to generate Filecheck rules in mir tests.

Opaque ptr cleanup effort.

This makes it easier to write .sh.cpp tests that perform multiple clang-verify checks in the same test.

- Document legacy ops lowered by `convert-func-to-llvm` and link it to a
GH issue
- Document the need for bufferization before `convert-linalg-to-loops`
- Document the ops lowered by `expand-strided-metadata`.

Needed for -riscv-experimental-rv64-legal-i32 and probably GISel.

Opaque ptr cleanup effort. NFC.

Co-authored-by: Sylvestre Ledru <sylvestre@debian.org>

Do not issue a hard error when the function in acc routine directive is
not present in the current translation unit. Only issue a warning.

When SI_PC_ADD_REL_OFFSET is expanded to S_GETPC/S_ADD/S_ADDC, the
GlobalAddress operands have to be adjusted by 4 or 12 bytes to account
for the offset from the end of the S_GETPC instruction to the literal
operands. Do this all in SIInstrInfo::expandPostRAPseudo instead of
duplicating the adjustment code in both AMDGPULegalizerInfo and
SITargetLowering. NFCI.

This is essentially a revert of
1ed71083. It is safe to use the
pull_request_target event for pr-subscriber, because it does not
checkout any code from the pull request branch.

We started going over our API limits and suspect this was the cause.
For now, we'll run it once a day while we sort this out.

Constant materialization can use bset x0, 11 to create 2048.

This is essentially a revert of
91fdb209. It is safe to use the
pull_request_target event for new-prs, because it does not checkout any
code from the pull request branch.

The last use was removed by:

  commit 18d9f7eb
  Author: Johannes Doerfert <johannes@jdoerfert.de>
  Date:   Wed Jul 5 15:35:24 2023 -0700

There is no PIC support for -mcmodel=large

(https://github.com/ARM-software/abi-aa/blob/main/sysvabi64/sysvabi64.rst)
and Clang recently rejects -mcmodel= with PIC (#70262).

The current backend code assumes that the large code model is non-PIC.
This patch adds `!getTargetMachine().isPositionIndependent()` conditions
to clarify that the support is non-PIC only. In addition, add some tests
as change detectors in case PIC large code model is supported in the
future.

If other front-ends/JITs use the large code model with PIC, they will
get small code model code sequence, instead of potentially-incorrect
MOVZ/MOVK sequence, which is only suitable for non-PIC. The sequence
will cause text relocations using ELF linkers.

(The small code model code sequence is usually sufficient as ADRP+ADD or
ADRP+LDR targets [-2**32,2**32), which has a doubled range of x86-64
R_X86_64_REX_GOTPCRELX/R_X86_64_PC32 [-2**32,2**32).)

LoopFlatten uses IndVars mechanics to widen induction variables, but then
fails to cleanup the dead instructions returned via the API.  This isn't
a functional problem, but means that this test is highly sensative to
indvars internals in an unhelpful way.  Use DCE to "normalize" the test
output to avoid false positives in a later change.

[clang][modules] Don't prevent translation of FW_Private includes when explicitly building FW (#70714)

We prevent translating `#include <FW/PrivateHeader.h>` into an import of
FW_Private when compiling the implementation of FW or FW_Private. This
is specified via `-fmodule-name=` on the TU command line (used to be
`-fmodule-implementation-of`).

This logic is supposed to only kick in when imported directly from a TU,
but it currently also kicks in when compiling the public FW module
explicitly (since it also has `-fmodule-name=` on the command line).

This patch makes sure this logic only kicks in for the case that used to
be `-fmodule-implementation-of` (for the TU), and not for all
`-fmodule-name=` cases (especially for the explicit compile of a
module).

rdar://101051277; related: rdar://37500098&38434694

The compute and data constructs implement getNumDataOperands and
getDataOperand. The acc.loop operation similarly has multiple data
operands - thus it makes sense to expose them the same way.

For loop, only private and reduction operands are exposed this way.
Technically, acc.loop also holds cache operands - but these are hints
not a data attribute.

Some test improvements, but also some regressions that need to be
fixed.

This reverts commit c0a18579.

A shared_ptr assertion always triggers causes all bots to fail.

Rather than write a bunch of logic to shepherd between enums with the
same sets of values, add the ability for EnumArgument to refer to an
external enum in the first place.

We've gotten ~750 commits in the last 7 days. Upping the fetch depth to
2000 will make it more likely that PRs up to 2 weeks old will have
enough fetch history to find a common parent. This _might_ address some
of the failures we're seeing in the clang-format action where it cannot
find a common base commit.

Constructors and destructors on the device do not take any arguments,
also not the implicit dyn_ptr argument other kernels automatically take.

Refactor some code from https://github.com/llvm/llvm-project/pull/69493.

#70712 was reverted due to linking failures. So, `-debug-info-correlate` remains unchanged and no new flag added.

Co-authored-by: Konstantin Zhuravlyov <kzhuravl@amd.com>

We used to perform team reduction on global memory allocated in the
runtime and by clang. This was racy as multiple instances of a kernel,
or different kernels with team reductions, would use the same locations.
Since we now have the kernel launch environment, we can allocate dynamic
memory per-launch, allowing us to move all the state into a non-racy
place.

Fixes: https://github.com/llvm/llvm-project/issues/70249

Closes #69736

This upstreams more of the Clang API Notes functionality that is
currently implemented in the Apple fork:
https://github.com/apple/llvm-project/tree/next/clang/lib/APINotes

This adds the first compiler options related to API Notes to the
upstream Clang: `-iapinotes-modules` and `-fapinotes-swift-version=`.
However, this does not add the `-fapinotes` flag that enables API Notes,
since the feature is not fully functional yet.

Currently tryToGatherExtractElements function analyzes the whole vector,
regrdless number of actual registers, used in this vector. It may
prevent some optimizations, because per-register analysis may allow to
simplify the final code by reusing more already emitted vectors and
better shuffles.

Differential Revision: https://reviews.llvm.org/D148855

Add an option to target SPIR-V to the clang-dxc driver, which sets the
target triple's architecture to logical SPIR-V. This does not yet
address the need for a Vulkan target environment in the triple, tracked
in Issue #70051.

Depends on #70330

This patch fixes:

  mlir/include/mlir/Bytecode/BytecodeImplementation.h:415:4: warning:
  extra ‘;’ [-Wpedantic]

  mlir/include/mlir/Bytecode/BytecodeWriter.h:110:4: warning: extra
  ‘;’ [-Wpedantic]

Reducing spurious diff in upcoming changes.

CloneModule is not currently designed to handle un-materialized Modules,
for example one created via a lazy initializer like
getLazyBitcodeModule(). In this case we get a somewhat cryptic
segmentation fault without a clear path forward.

In this patch, we add a comment to inform CloneModule users of this
shortcoming, and an assert to test for empty function bodies before the
segmentation fault is triggered.

Fixup for #70722.

This patch adds the feature flag FDOT2/FDOT4 and the
assembly/disassembly
for the following instructions of NEON and SVE2:
 * NEON:
   - FDOTlane
   - FDOT
 * SVE2:
   - FDOT_ZZZI_BtoH
   - FDOT_ZZZ_BtoH
   - FDOT_ZZZI_BtoS
   - FDOT_ZZZ_BtoS

That is according to this documentation:
https://developer.arm.com/documentation/ddi0602/2023-09