[PATCH] Handle ASHIFTRT with constant shift count >= BITS_PER_WORD in subreg lowering (PR rtl-optimization/50339)

Richard Henderson rth@redhat.com
Thu Feb 21 19:02:00 GMT 2013


On 02/21/2013 10:58 AM, Jakub Jelinek wrote:
> So, do you prefer e.g. the following, or would you instead prefer
> that I just set some rtx to copy_rtx (src_reg) before the
> if (GET_CODE (op) != ZERO_EXTEND)
>   {
>     ...
>   }
> block and do the >> (BITS_PER_WORD - 1) shift only after this (then it can
> be with dest_upper as target and moved only if expand_shift returns
> something different)?

This new patch is fine.

> For the patch below, I've verified no codegen changes from the earlier patch
> on the 3 testcases from the PR.

Excellent.


r~



More information about the Gcc-patches mailing list