[PATCH] Vector mode addresses

Richard Biener richard.guenther@gmail.com
Fri Jan 31 10:39:00 GMT 2014


On Thu, Jan 30, 2014 at 9:47 PM, Jakub Jelinek <jakub@redhat.com> wrote:
> On Thu, Jan 30, 2014 at 08:27:47PM +0000, Paulo Matos wrote:
>> Yes, it looks strange but it was the way we came up with to
>> implement an instruction that loads from a pair of addresses.
>>
>> From what I wrote previously to Richard.
>> "We have an instruction that loads two 32 bit values into a lower
>> and upper parts of a 64bit register using a base register and a 64
>> bit register used as a double index.
>> So,
>> r1 <- [r0, r2]
>> means:
>> low(r1) = [r0 + low(r2)]
>> high(r1) = [r0 + high(r2)]"
>
> That sounds like gather instruction e.g. i?86 AVX2/AVX512F has.
> I don't like using vector mode for the address though, i?86 uses UNSPECs and
> IMNSHO you should too.  Or express it as vec_concat of two MEM loads
> where address of each one will be the base + vec_select of the vector
> register.

Yeah, I don't like it either.  It opens a whole can of worms I think.

Richard.

>         Jakub



More information about the Gcc-patches mailing list