Fix a few things in one swoop.
# Add some negative tests.
# Fix some formatting issues.
# Add some missing IsThumb / ARMv8
# Fix some outs / ins mistakes.
llvm-svn: 189490
This is just enough to get "llvm-ranlib foo.a" working and tested. Making
llvm-ranlib a symbolic link to llvm-ar doesn't work so well with llvm's option
parsing, but ar's option parsing is mostly custom anyway.
This patch also removes the -X32_64 option. Looks like it was just added in
r10297 as part of implementing the current command line parsing. I can add it
back (with a test) if someone really has AIX portability problems without it.
llvm-svn: 189489
The usual default of "dmb ish" (inner-shareable) isn't even a valid instruction
on v6M or v7M (well, it does the same thing but software is strongly
discouraged from using it) so we should emit a full-system barrier there.
llvm-svn: 189483
The vqdmlal and vqdmlls instructions are really just a fused pair consisting of
a vqdmull.sN and a vqadd.sN. This adds patterns to LLVM so that we can switch
Clang's CodeGen over to generating these instead of the special vqdmlal
intrinsics.
llvm-svn: 189480
These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes,
are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as
a branch/mov sequence to evaluate to 0 or 1.
Note: The resulting code is sub-optimal since it doesnt seem to be possible
to feed the result of an intrinsic directly into a brcond. At the moment
it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily
evaluates the boolean twice.
llvm-svn: 189478
For now just handles simple comparisons of an ANDed value with zero.
The CC value provides enough information to do any comparison for a
2-bit mask, and some nonzero comparisons with more populated masks,
but that's all future work.
llvm-svn: 189469
The MSA control registers have been added as reserved registers,
and are only used via ISD::Copy(To|From)Reg. The intrinsics are lowered
into these nodes.
llvm-svn: 189468
Two changes:
* Don't add an extra penalty on breaking the same token multiple times.
Generally, we should prefer not to break, but once we break, the
normal line breaking penalties apply.
* Slightly increase the penalty for breaking comments. In general, the
author has put some thought into how to break the comment and we
should not overwrite this unnecessarily.
With a 40-column column limit, formatting
aaaaaa("aaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaa");
Leads to:
Before:
aaaaaa(
"aaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaa "
"aaaaaaaaaaaaaaaa");
After:
aaaaaa("aaaaaaaaaaaaaaaa "
"aaaaaaaaaaaaaaaa "
"aaaaaaaaaaaaaaaa");
llvm-svn: 189466
These operations "vector add high-half narrow" actually correspond to the
sequence:
%sum = add <4 x i32> %lhs, %rhs
%high = lshr <4 x i32> %sum, <i32 16, i32 16, i32 16, i32 16>
%res = trunc <4 x i32> %high to <4 x i16>
Now that LLVM can spot this, Clang should emit the corresponding LLVM IR.
llvm-svn: 189463
The NEON intrinsics vqdmlal and vqdmlsl are really just combinations of a
saturating-doubling-multiply (vqdmull) and a saturating add/sub, so now that
LLVM can spot those patterns Clang should emit them instead of specialised
intrinsics.
Feature already tested by existing ARM NEON intrinsics tests.
llvm-svn: 189462
The code leading to a segfault was:
#pragma omp threadprivate(y)), // long comment leading to a line break
This fixes llvm.org/PR16513.
llvm-svn: 189460
If escaped newlines are aligned right
(FormatStyle.AlignEscapedNewlinesLeft == false), and a line contained
too many characters to fit into the column limit, this would result in
a (virtually) endless loop creating a negative number of spaces.
Instead, allow the escaped newlines to be pushed past the column limit
in this case.
This fixes llvm.org/PR16515.
llvm-svn: 189459
In
@implementation ObjcClass
- (void)method;
{
}
@end
the ObjC compiler seems to accept the superfluous comma after "method",
but clang-format used to assert on the subsequent "{".
This fixes llvm.org/PR16604.
llvm-svn: 189453
Summary:
-fno-exceptions does not implicitly attach a nothrow specifier to every operator
new. Even in this mode, non-nothrow new must not return a null pointer. Failure
to allocate memory can be signalled by other means, or just by killing the
program. This behaviour is consistent with the compiler - even with
-fno-exceptions, the generated code never tests for null (and would segfault if
the opeator actually happened to return null).
Reviewers: jordan_rose
CC: cfe-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D1528
llvm-svn: 189452
The previous default was almost, but not quite enough digits to
represent a floating-point value in a manner which preserves the
representation when it's read back in. The larger default is much
less confusing.
I spent some time looking into printing exactly the right number of
digits if a precision isn't specified, but it's kind of complicated,
and I'm not really sure I understand what APFloat::toString is supposed
to output for FormatPrecision != 0 (or maybe the current API specification
is just silly, not sure which). I have a WIP patch if anyone is interested.
llvm-svn: 189442
As Chandler pointed out, we should not be using -backend-option because this
will cause crashes for users of the tooling interface, etc. A better way to fix
this will be to provide the unrolling pass-manager flag to the loop vectorizer
directly.
Original commit message:
Disable loop vectorizer unrolling when no unrolling requested
In addition to the regular loop unrolling transformation, the loop vectorizer
can also unroll loops. If no unrolling has specifically been requested (by
-fno-unroll-loops), and the loop vectorizer will be used, then add the backend
option to (also) prevent the loop vectorizer from unrolling loops.
I confirmed with Nadav (off list) that disabling vectorizer loop unrolling when
-fno-unroll-loops is provided is the desired behavior.
llvm-svn: 189441
Hans Wennborg