[llvm-commits] [PATCH] Fix broken decoding for VCVT (between floating-point and fixed-point, Floating-point)

[Jim Grosbach]

Hi Richard,

This is great. Thanks for doing this.

Two things, if you don't mind. First, please add tests for the other instructions that are changing, not just vcvt. For obvious reasons, not using a register numbered zero is a good idea, and one I wish more of our other tests followed… Second, much more nitty, please format the tests grouped and with the CHECK lines following the instructions being tested (like the tests towards the end of the file). That particular file hasn't yet been fully converted to that style yet, so I certainly understand the confusion.

Thanks again, and great bug catch.

-Jim

On Mar 9, 2012, at 10:20 AM, Richard Barton wrote:

> Hello
> 
> The attached patch fixes VCVT (between floating-point and fixed-point,
> Floating-point) instruction decoding in MC.
> 
> The previous support omitted the register operand from the encoding. My patch
> adds a new class to the hierarchy to handle the two different encodings for
> double-precision and single-precision registers.
> 
> I have added a regression test for single and double precision flavours of this
> instruction.
> 
> Thanks,
> Richard Barton
> 
> <vcvt_floating_to_fixed.patch>_______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits