Integer specialisation of BitScanForwardNode

Thomas Wuerthinger thomas.wuerthinger at oracle.com
Fri Oct 11 20:22:40 PDT 2013


Yes, absolutely. Could you create a patch for this?

Thanks, thomas

On Oct 11, 2013, at 4:51 PM, D.Sturm <D.Sturm42 at gmail.com> wrote:

> BitScanForwardNode only has a single @NodeIntrinsic method that takes a
> long. As a consequence when computing it for an int we first sign-extend it
> to a long, e.g.:
> 
> cmp    edx,0x0
> je     0x11
> movsxd rax,edx
> bsf    rax,rax
> 
> 
> Could we introduce a new method taking an int to avoid this?



More information about the graal-dev mailing list