Optimize (add (mul x, c0), c1) ->
         (add (mul (add x, c1/c0+1), c0), c1%c0-c0),
if c1/c0+1 and c1%c0-c0 are simm12, while c1 is not.

Optimize (add (mul x, c0), c1) ->
         (add (mul (add x, c1/c0-1), c0), c1%c0+c0),
if c1/c0-1 and c1%c0+c0 are simm12, while c1 is not.

Reviewed By: craig.topper, asb

Differential Revision: https://reviews.llvm.org/D111141

Add comments to explain why XXPERMDIs and XXPERMDI have different input register
classes, vsfrc for XXPERMDIs and vsrc for XXPERMDI.

This addresses the comments in abandoned patch D113178, we keep using `f0` instead
of using `vs0` for XXPERMDIs on purpose.

CSKY is a ARCH which supports mixture of 16-bit and 32-bit instructions natively,
and there is not an indivual predictor or feature to enable/disable 16-bit instruction.
So I think it's better to add 16-bit instruction early, and naturally to use 16-bit and 32-bit instructions.

Differential Revision: https://reviews.llvm.org/D112919

Reviewed By: jsji

Differential Revision: https://reviews.llvm.org/D113173

For v8i16 shuffle patterns that are lowered with AND+PACKUS, check to see if the sources are from a 256-bit vector and perform the masking using BLENDW at the 256-bit level.

With the test changes we can see more examples of duplicate XMM/YMM zero vectors (PR26018) :(

This patch add the conversion pattern for fir.extract_value
and fir.insert_value. fir.extract_value is lowered to llvm.extractvalue
anf fir.insert_value is lowered to llvm.insertvalue.
This patch also adds the type conversion for the BoxType and RecordType
needed to have some comprehensive tests.

This patch is part of the upstreaming effort from fir-dev branch.

This patch was landed and reverted once.
TypeBuilderFunc getModel<Fortran::ISO::CFI_index_t>() was clashing
with getModel<long long> on windows since they both are 64 bits
signed interger. On linux CFI_index_t is long. Change CFI_index_t
to getModel<long>.

Reviewed By: awarzynski

Differential Revision: https://reviews.llvm.org/D112961



Co-authored-by: Jean Perier <jperier@nvidia.com>
Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>

For some optimizations on comparisons it's necessary that the
union/intersect is exact and not a superset. Add methods that
return Optional<ConstantRange> only if the result is exact.

For the sake of simplicity this is implemented by comparing
the subset and superset approximations for now, but it should be
possible to do this more directly, as unionWith() and intersectWith()
already distinguish the cases where the result is imprecise for the
preferred range type functionality.

From an API perspective, it does not make a lot of sense that 0
is not a valid argument to this function. Add the exact check needed
to support it.

Somehow every pointer looks like it's tagged on GreenDragon. Removing
the check to unblock the bot until we can get to the bottom of this.

This rewrites the fcvt-fixed.ll test case to be separate functions, not
one large function with volatile global stores. It also adds fp16 and
fptoi.sat testing at the same time.

When accumulating the GEP offset in BasicAA, we should use the
pointer index size rather than the pointer size.

Differential Revision: https://reviews.llvm.org/D112370

The function body proper was removed on Feb 20, 2019 in commit
79b5c384.

D98452 introduced a mismatch between clang expectations for
builtin name for baremetal targets on arm. Fix it by
adding a case for baremetal. This now matches the output of
"clang -target armv7m-none-eabi -print-libgcc-file-name \
-rtlib=compiler-rt"

Reviewed By: mstorsjo

Differential Revision: https://reviews.llvm.org/D113357

Make test_allocator etc. constexpr-friendly so they can be used to test constexpr string and possibly constexpr vector

Reviewed By: Quuxplusone, #libc, ldionne

Differential Revision: https://reviews.llvm.org/D110994

Be more consistent in the naming convention for the various RET instructions to specify in terms of bitwidth.

Helps prevent future scheduler model mismatches like those that were only addressed in D44687.

Differential Revision: https://reviews.llvm.org/D113302

D109434 addressed the v2f64 -> v4f64 case, an internal test has found an equivalent crash for the v4f32 -> v8f32 case.

Noticed on D113212

Currently FoldConstantArithmetic only handles binops, so replacing other uses of FoldConstantVectorArithmetic (in particular for SETCC nodes), still require more work.

The LLVM-C API is relatively small so we've previously added doxygen tags
so it's easier to navigate the LLVM-C web docs. Over the years, more
headers were added without proper doxygen tags, effectively hiding them
from the main LLVM-C doxygen page. This patch adds comments to headers
which did not have them.

Reviewed By: lhames

Differential Revision: https://reviews.llvm.org/D112474

https://sourceware.org/bugzilla/show_bug.cgi?id=22742
uc_mcontext.__reserved probably should not be considered user visible API but
unfortunate it is: it is the only way to access cpu states of some Linux
asm/sigcontext.h extensions. That said, the declaration may be
long double __reserved[256]; (used by musl)
instead of
unsigned char __reserved[4096] __attribute__((__aligned__(16))); (glibc)
to avoid dependency on a GNU variable attribute.

GCC introduced `__attribute__((mode(unwind_word)))` to work around
Cell Broadband Engine SPU (which was removed from GCC in 2019-09),
which is irrelevant to hwasan.
_Unwind_GetGR/_Unwind_GetCFA from llvm-project/libunwind don't use unwind_word.

Using _Unwind_Word can lead to build failures if libunwind's unwind.h is
preferred over unwind.h in the Clang resource directory (e.g. built with GCC).

Nathan Chancellor reported a crash due to commit
3466e007 (Reland "[Attr] support btf_type_tag attribute").

The following test can reproduce the crash:
  $ cat efi.i
  typedef unsigned long efi_query_variable_info_t(int);
  typedef struct {
    struct {
      efi_query_variable_info_t __attribute__((regparm(0))) * query_variable_info;
    };
  } efi_runtime_services_t;
  efi_runtime_services_t efi_0;
  $ clang -m32 -O2 -g -c -o /dev/null efi.i

The reason is that FunctionTypeLoc.getParam(Idx) may return a
nullptr which should be checked before dereferencing the
result pointer. This patch fixed this issue.

This makes it easy to write tests where the irrelevant fields are not needed.

Fix a dangling else that gcc-11 warned about. The EXPECT_EQ macro
expands to an if-else, so the whole construction contains a hidden
dangling else.

Differential Revision: https://reviews.llvm.org/D113346

Add a variant of getEquivalentICmp() that produces an optional
offset. This allows us to create an equivalent icmp for all ranges.

Use this in the with.overflow folding code, which was doing this
adjustment separately -- this clarifies that the fold will indeed
always apply.