[llvm-commits] We are selecting a CMP64mi32 instead of a CMP64mi8

[Chris Lattner]

Fixed in r116053/116054, thanks for the great testcase!

-Chris

On Oct 7, 2010, at 1:15 PM, Rafael Espindola wrote:

> I am debugging a case where llc + llvm-mc produces different results
> from using "llc -filetype=obj". The issues is that for the attached
> test llc selects CMP64mi32. This prints as
> 
> cmpq	$0, -8(%rsp)
> 
> Which gets parsed by llvm-mc as the 8 bit immediate version. Why are
> we selecting CMP64mi32? The attached patch fixes this particular
> issue, but it would be better if we could this in a less brittle way.
> 
> Cheers,
> -- 
> Rafael Ávila de Espíndola
> <patch><SparseBitVectorTest.ll>