This was discovered while working on modules.

Reviewed By: #libc, philnik

Differential Revision: https://reviews.llvm.org/D149351

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D150520

We should not call tryOptimizeLEAtoMOV() in eliminateFrameIndex() when
the base register is a virtual register, because tryOptimizeLEAtoMOV
would assume the base register be physical register. Although we can
also optimize LEA to MOV with virtual register, I'd like to leave the
optimization in another patch.

Differential Revision: https://reviews.llvm.org/D150521

If we know that zero constant operand is already in the right place we do not need
to re-order anything.

Reviewed By: e-kud
Differential Revision: https://reviews.llvm.org/D150249

NFC. Add missing const on affine analysis utils ComputationSliceState
methods.

Differential Revision: https://reviews.llvm.org/D150523

Make isSliceValid const correct. NFC.

There was initially a msan report for use-of-uninitialized value due to
a bug in https://reviews.llvm.org/D138526. find_iterator::Point is
uninitialized for the default constructor of find_iterator, which is
used by IntervalTree::end. This change is not required, but its good
practice to make sure all class members are initialized.

Differential Revision: https://reviews.llvm.org/D149698

This reverts commit ce990b54.

This breaks some build bots - specifically when using GCC to build LLVM and
then -fno-lifetime-dse ends up passed to Clang in some tests like at
https://lab.llvm.org/buildbot/#/builders/139/builds/40594.

Bug: https://github.com/llvm/llvm-project/issues/24952
Bug: https://github.com/llvm/llvm-project/issues/57740

Differential Revision: https://reviews.llvm.org/D150505

Previously, LegalizeVectorOps used the result VT while LegalizeDAG
used the operand VT. This patch makes them both use the operand VT.

This also makes it consistent with how the default cost model works.

I've hacked the AArch64 cost model to maintain old behavior for some
f16 vectors.

Reviewed By: RKSimon

Differential Revision: https://reviews.llvm.org/D149572

Module map generation for the private detail headers is easier done if each private header is by itself in a submodule. Move the __algorithm/pstl_backends into their own submodules.

Reviewed By: philnik, #libc

Differential Revision: https://reviews.llvm.org/D150503

LLVM data structures like llvm::User and llvm::MDNode rely on
the value of object storage persisting beyond the lifetime of the
object (#24952).  This is not standard compliant and causes a runtime
crash if LLVM is built with GCC and LTO enabled (#57740).  Until
these issues are fixed, we need to disable dead store eliminations
eliminations based on object lifetime.

Bug: https://github.com/llvm/llvm-project/issues/24952
Bug: https://github.com/llvm/llvm-project/issues/57740
Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106943

(This was originally committed as 94f7c961 but
I reverted it in b974991f4c4457a2104b648d9797a0ed438ecc9 to fix authorship.)

Reviewed By: MaskRay, thesamesam, nikic

Differential Revision: https://reviews.llvm.org/D150505



Signed-off-by: Sam James <sam@gentoo.org>

This reverts commit 94f7c961.

Differential Revision: https://reviews.llvm.org/D150505

The bit fields have plent of spare bits. Just reserve one for HasName so that we
can access Fragment without bitwise operations. Fragment is commonly accessed.
This change makes my x86-64 release build 5KiB smaller.

Reviewed By: #libc, ldionne, Mordante

Spies: Mordante, libcxx-commits

Differential Revision: https://reviews.llvm.org/D150421

Various ASan interceptors may corrupt memory if passed a
pointer to freed memory (https://github.com/google/sanitizers/issues/321).
This patch fixes the issue for the backtrace interceptor,
by calling REAL(backtrace) with a known-good scratch buffer,
and performing an addressability check on the user-provided
buffer prior to writing to it.

Differential Revision: https://reviews.llvm.org/D150496

This patch changes two instances of an ampersand to a written out and
for more consistency with the rest of the file and brevity. In addition,
the last `cmake --build` reference is removed, again for consistency
with the rest of the file which shows the ninja invocations. This cmake
invocation also passed in the `--parallel` flag which doesn't make sense
with ninja using all threads by default.

This was changed in the previous patch to touch this line
(https://reviews.llvm.org/D88990), but if we want to change this, it
should be done across the entire file.

The current logic is pretty limitted unless the `Op` is a
constant. This at least covers more obvious cases.

Reviewed By: craig.topper, foad

Differential Revision: https://reviews.llvm.org/D149196

Both of these functions recursively call themselves so it makes sense
to limit that upper bound.

Differential Revision: https://reviews.llvm.org/D149195

This is generally handled already in early CSE.

If a specialized pipeline is used, however, its possible for `i1`
operand with known-zero denominator to slip through. Generally the
known-zero denominator is caught and poison is returned, but if it is
indirect enough (known zero through a phi node) we can miss this case
in `InstructionSimplify` and then miss handling `i1`. This is because
`i1` is current handled with the following check:
    `if(Known.countMinLeadingZeros() == Known.getBitWidth() - 1)`

which only works on the assumption we don't know the denominator to be
zero. If we know the denominator to be zero, this check fails:
https://github.com/llvm/llvm-project/issues/62607

This patch simply adds an explicit `if(Known.isZero) return poison;`
which fixes the issue.

Alive2 Link for tests:
    https://alive2.llvm.org/ce/z/VTw54n

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D150142

We can detect the case under the following circumstances:
Take `(Pow2_Ceil(C1) - (1 << C0))` as `C2`.
    1) `C2` is NOT a power of 2.
    2) `C2 + LeastSignificantBit(C2)` is a nonzero power of 2.
    3) `C2 u>= C1`

The motivation is the middle end transforms:
    `(-x << C0) & C1`
to
    `(x * (Pow2_Ceil(C1) - (1 << C2))) & C1`

As it saves IR instructions. On X86 the two instruction, `sub` and
`shl`, and better than the `mul` so we want to undo the transform.

This comes up when shifting a bit-mask by a byte-misalignment i.e:
    `y << ((-(uintptr)x * 8) & 63)`

Alive2 Proofs (including all cases with undefs in the vector):
https://alive2.llvm.org/ce/z/f-65b6

Reviewed By: RKSimon, pengfei

Differential Revision: https://reviews.llvm.org/D150294

Differential Revision: https://reviews.llvm.org/D150293

LLVM data structures like llvm::User and llvm::MDNode rely on
the value of object storage persisting beyond the lifetime of the
object (#24952).  This is not standard compliant and causes a runtime
crash if LLVM is built with GCC and LTO enabled (#57740).  Until
these issues are fixed, we need to disable dead store eliminations
eliminations based on object lifetime.

Bug: https://github.com/llvm/llvm-project/issues/24952
Bug: https://github.com/llvm/llvm-project/issues/57740
Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106943

Reviewed By: MaskRay, thesamesam, nikic

Differential Revision: https://reviews.llvm.org/D150505

Leverage perf-training flow for BOLT profile collection, enabling reproducible
BOLT optimization. Remove the use of bootstrapped build for profile collection.

Test Plan:
- Regular (single-stage) build
```
$ cmake ... -C .../clang/cmake/caches/BOLT.cmake
$ ninja clang-bolt
...
[21/24] Instrumenting clang binary with BOLT
[21/24] Generating BOLT profile for Clang
[23/24] Merging BOLT fdata
Profile from 2 files merged.
[24/24] Optimizing Clang with BOLT
...
          1291202496 : executed instructions (-1.1%)
            27005133 : taken branches (-71.5%)
...
```
- Two stage build (ThinLTO+InstPGO)
```
$ cmake ... -C .../clang/cmake/caches/BOLT.cmake -C .../clang/cmake/caches/BOLT-PGO.cmake
$ ninja clang-bolt
$ ninja stage2-clang-bolt
...
[2756/2759] Instrumenting clang binary with BOLT
[2756/2759] Generating BOLT profile for Clang
[2758/2759] Merging BOLT fdata
[2759/2759] Optimizing Clang with BOLT
...
BOLT-INFO: 7092 out of 184104 functions in the binary (3.9%) have non-empty execution profile
           756531927 : executed instructions (-0.5%)
            15399400 : taken branches (-40.3%)
...
```

Reviewed By: beanz

Differential Revision: https://reviews.llvm.org/D143553

Split off from D143938. This moves the planning logic to select the
vectorization factor to LoopVectorizationPlanner as a step towards only
computing costs for individual VFs in LoopVectorizationCostModel and do
planning in LVP.

Reviewed By: Ayal

Differential Revision: https://reviews.llvm.org/D150197

NFC. Pass affine copy options by const ref.

Differential Revision: https://reviews.llvm.org/D150507

The owner of the last two failing buildbots updated CMake.

This reverts commit e8e8707b.

The reason for this patch is to allow the MCDisassembler used in tests
to disassemble instructions that are only available when a specific
feature is enabled.

For example, on RISC-V it's currently not possible to use
decode_operand() on a compressed instruction. This patch fixes this.

Reviewed By: lhames

Differential Revision: https://reviews.llvm.org/D149523

Like done D149543 this validates the other formatter specializations.

Reviewed By: #libc, ldionne

Differential Revision: https://reviews.llvm.org/D150041

Split off refactoring from D150197 to reduce diff.

Currently, there is no documentation on what platforms and architectures
llvm-exegesis is supported on. This patch adds in user-facing
documentation in the CommandGuide about what architectures are supported
as well as developer facing documentation detailing the technical
reasons for why certain platforms are supported and some aren't.

This is a follow-up after discussion in
https://discourse.llvm.org/t/clarification-on-platform-support-for-llvm-exegesis/70206.

Reviewed By: kpdev42

Differential Revision: https://reviews.llvm.org/D149378

The Assembler tests have been disabled for years in tree and at this
point don't test anything other than common MC infrastructure that is
already tested in other parts of the tree. This patch removes them due
to the mentioned reasons.

Reviewed By: courbet

Differential Revision: https://reviews.llvm.org/D149819

This patch fixes the man page build. It currently doesn't work as
SOURCE_DIR isn't set correctly (just undefined) within the
add_sphinx_target function. This patch also moves around the creation of
targets for autogenerated rst files so that both the man page and html
build can depend upon them as before only the html build depended on
them.

Fixes #62540

Reviewed By: tstellar

Differential Revision: https://reviews.llvm.org/D149809

At the moment, lower-matrix-intrinsics accepts mis-matches between
shapes for operations. See shape-verification.ll for an example where
@llvm.matrix.column.major.load specifies 6x1 and then the use
(@llvm.matrix.multiply) specifies the operand to have 1x6.

This patch adds verification for shapes to check if shapes match.

Reviewed By: thegameg

Differential Revision: https://reviews.llvm.org/D147438