ARM instructions: invalid operands

[llvmbot]

Bugzilla Link	14145
Resolution	INVALID
Resolved on	Jun 16, 2016 08:09
Version	trunk
OS	MacOS X
Blocks	llvm/llvm-bugzilla-archive#18926
Reporter	LLVM Bugzilla Contributor
CC	@efriedma-quic,@rengolin

Extended Description

Met some build errors when building some assembly code (assembly file, or inline assembly code in C) for iOS, with clang. The errors are:

(1) error: invalid operand for instruction
  vpmin.s16 d24, d24
(2) error: invalid operand in inline asm: 'smlawb ${0:r}, ${1:r}, ${2:r}, ${3:r}'
  asm("smlawb %r0, %r1, %r2, %r3": "=r"(result): "r"(diff),
(3) error: invalid operand for instruction
  __asm __volatile ("ssat %r0, #​16, %r1" : "=r"(out16) : "r"(value32));

Please note that most other Neon or ARM instructions could pass, however.

The version of clang I used is: Apple clang version 4.0 (tags/Apple/clang-421.0.57) (based on LLVM 3.1svn).

[efriedma-quic]

In general, please report bugs in tools distributed by Apple at bugreporter.apple.com .

The error on "vpmin.s16 d24, d24" is fixed on trunk.

For the others, I'm not sure what the "r" operand modifier in "%r0" is supposed to mean.

[rengolin]

Apple Clang 3.1 is old and not upstream. If this problem can be reproduced with a modern upstream toolchain, please include complete reproducible cases (source, command line, etc).

[rengolin]

mentioned in issue llvm/llvm-bugzilla-archive#18926