On Tue, Oct 7, 2014 at 11:43 PM, Thomas Preud'homme
<thomas.preudho...@arm.com> wrote:
>> From: Jakub Jelinek [mailto:ja...@redhat.com]
>> Sent: Wednesday, October 08, 2014 2:39 PM
>>
>> Doesn't it turn 16-bit {L,R}ROTATE_EXPR used alone into
>> __builtin_bswap16?
>> For those the question is if the canonical GIMPLE should be the rotation
>> or
>> byteswap, I'd think rotation would be perhaps better. Or depending on
>> if
>> the backend has bswaphi2 or rotate pattern?
>
> Good point. It seems better to keep the status quo.
There is already code which converts __builtin_bswap16 to the rotate
so maybe it does not matter. Though I have not seen code which does
the rotate into a bswaphi2 pattern.
Thanks,
Andrew
>
>>
>> Also, perhaps you could short-circuit this if the rotation isn't by constant
>> or not a multiple of BITS_PER_UNIT. So
>> switch (code)
>> {
>> case BIT_IOR_EXPR:
>> break;
>> case LROTATE_EXPR:
>> case RROTATE_EXPR:
>> if (!tree_fits_uhwi_p (gimple_assign_rhs2 (cur_stmt))
>> || (tree_to_uhwi (gimple_assign_rhs2 (cur_stmt))
>> % BITS_PER_UNIT))
>> continue;
>> break;
>> default:
>> continue;
>> }
>> ?
>
> Right. Thanks for the comments.
>
> Best regards,
>
> Thomas
>
>
>
>
