The nomerge attribute is a statement attribute not a function attribute.

This reverts commit 415f7ee8.

The added tests were failing on the build bots!

The SHL and SRL could be in opposite order so account for that.

Differential Revision: https://reviews.llvm.org/D90586

Based off existing rotate test coverage

This makes us closer to running the test suite on platforms where the
legacy test suite configuration doesn't work.

One notable change after this commit is that the tests will be run with
warnings enabled on GCC too, which wasn't the case before. However,
previous commits should have tweaked the test suite to make sure it
passes with warnings enabled on GCC.

Note that warnings can still be disabled with `--param enable_warnings=False`,
as before.

Differential Revision: https://reviews.llvm.org/D90432

Differential revision: https://reviews.llvm.org/D90500

Update the documentation to fully describe it.

Differential Revision: https://reviews.llvm.org/D90617

As noticed in D90554 ,
the AVX2 costs for 256-bit vectors did not include FMAXNUM entries,
so we fell back to AVX1 which assumes those ops will be split into
128-bit halves or something close to that.

Differential Revision: https://reviews.llvm.org/D90613

PR47997

The code is looking for (sext_inreg (or (shl X, C2), (shr (and Y, C3), C1))).
We need to ensure X and Y are the same.

Differential Revision: https://reviews.llvm.org/D90580

[AggressiveInstCombine] foldGuardedRotateToFunnelShift - generalize rotation to funnel shift matcher.

Replace matchRotate with a more general matchFunnelShift - at the moment this is still just used for rotation patterns.

For PlayStation we offer source code compatibility with
Microsoft's dllimport/export annotations; however, our file
format is based on ELF.

To support this we translate from DLL storage class to ELF
visibility at the end of codegen in Clang.

Other toolchains have used similar strategies (e.g. see the
documentation for this ARM toolchain:

https://developer.arm.com/documentation/dui0530/i/migrating-from-rvct-v3-1-to-rvct-v4-0/changes-to-symbol-visibility-between-rvct-v3-1-and-rvct-v4-0)

This patch adds the ability to perform this translation. Options
are provided to support customizing the mapping behaviour.

Differential Revision: https://reviews.llvm.org/D89970

Add intended and missing prefix check.

The vendor will be prefixed to the "clangd" and can be an arbitrary
string, so account for it in the test.

Reviewed By: sammccall

Differential Revision: https://reviews.llvm.org/D90517

It's only used by Version.cpp, so limit the definition to just that one
file instead of making all of Clang recompile if you change CLANG_VENDOR.

Reviewed By: phosek

Differential Revision: https://reviews.llvm.org/D90516

The issue was unexpected macro expansion when the bot's test output
directory contained a token matching a build system macro (e.g.
"linux"). Switch to using a hardcoded path, which is invalid but is
sufficient for ensuring that the path is passed down to the runtime.

Differential Revision: https://reviews.llvm.org/D90466

We see this fairly often on our Linux bots, which appear to be killed
by GCE from time to time.

This patch calls `lua_close()` on Lua dtor.

This guarantees that the Lua GC finalizers are honored, aside from the
usual internal clean up.

It also guarantees a call to the `__close` metamethod of any active
to-be-closed variable in Lua 5.4.

Since the previous `luaL_openlibs()` was a noop, because the standard
library is cached internally, I've removed it.

Differential Revision: https://reviews.llvm.org/D90557

The `LiveRegUnits` utility (as well as `LivePhysRegs`) considers
callee-saved registers to be alive at the point after the return
instruction in a block. In the ARM backend, the `LR` register is
classified as callee-saved, which is not really correct (from an ARM
eABI or just common sense point of view).  These two conditions cause
the `MachineOutliner` to overestimate the liveness of `LR`, which
results in unnecessary saves/restores of `LR` around calls to outlined
sequences.  It also causes the `MachineVerifer` to crash in some
cases, because the save instruction reads a dead `LR`, for example
when the following program:

int h(int, int);

int f(int a, int b, int c, int d) {
  a = h(a + 1, b - 1);
  b = b + c;
  return 1 + (2 * a + b) * (c - d) / (a - b) * (c + d);
}

int g(int a, int b, int c, int d) {
  a = h(a - 1, b + 1);
  b = b + c;
  return 2 + (2 * a + b) * (c - d) / (a - b) * (c + d);
}

is compiled with `-target arm-eabi -march=armv7-m -Oz`.

This patch computes the liveness of `LR` in return blocks only, while
taking into account the few ARM instructions, which read `LR`, but
nevertheless the register is not mentioned (explicitly or implicitly)
in the instruction operands.

Differential Revision: https://reviews.llvm.org/D89189

[ELF] --emit-relocs: fix st_value of STT_SECTION in the presence of a gap before the first input section

In the presence of a gap, the st_value field of a STT_SECTION symbol is the
address of the first input section (incorrect if there is a gap). Set it to the
output section address instead.

In -r mode, this bug can cause an incorrect non-zero st_value of a STT_SECTION
symbol (while output sections have zero addresses, input sections may have
non-zero outSecOff).  The non-zero st_value can cause the final link to have
incorrect relocation computation (both GNU ld and LLD add st_value of the
STT_SECTION symbol to the output section address).

Reviewed By: grimar

Differential Revision: https://reviews.llvm.org/D90520

The new script keeps the results sorted, which is why the ABI lists appear
to have changed so much. However, this commit makes no actual ABI change.

Instead of having to remember the command-line to use every time, this
commit adds a CMake target to generate the ABI list in the current
configuration, if it is supported.

As a fly-by change, remove scripts that are now unused (sym_match.py
and sym_extract.py).

This reverts commit dbae3d50.
Cannot build with gcc/g++ 7.5.0.

This reverts commit 56299b1e.
Cannot build with gcc/g++ 7.5.0.

Preemptively switch 2 tests to the new PM

Reviewed By: aeubanks

Differential Revision: https://reviews.llvm.org/D90365

ReturnPtrRange checker emits a report if a function returns a pointer which
points out of the buffer. However, end() iterator of containers is always such
a pointer, so this always results a false positive report. This false positive
case is now eliminated.

This patch resolves these tickets:
https://bugs.llvm.org/show_bug.cgi?id=20929
https://bugs.llvm.org/show_bug.cgi?id=25226
https://bugs.llvm.org/show_bug.cgi?id=27701

Patch by Tibor Brunner!

Differential Revision: https://reviews.llvm.org/D83678

This reverts the revert commit 408c4408.

This version of the patch includes a fix for a crash caused by
treating ICmp/FCmp constant expressions as instructions.

Original message:

On some targets, like AArch64, vector selects can be efficiently lowered
if the vector condition is a compare with a supported predicate.

This patch adds a new argument to getCmpSelInstrCost, to indicate the
predicate of the feeding select condition. Note that it is not
sufficient to use the context instruction when querying the cost of a
vector select starting from a scalar one, because the condition of the
vector select could be composed of compares with different predicates.

This change greatly improves modeling the costs of certain
compare/select patterns on AArch64.

I am also planning on putting up patches to make use of the new argument in
SLPVectorizer & LV.

This makes it cleaner to add more instantiations without cluttering the
actual implementation of ios.

This is similar in spirit to what D90313 did for server tests.

This reverts commit 43a38a65,
and follow-up 5a7bc5e2.

The commit breaks check-clang-tools, the test added in the change
does not pass.