Displaying 2 results from an estimated 2 matches for "a_red_next".
2013 Oct 23
0
[LLVMdev] First attempt at recognizing pointer reduction
...ou have
for ()
*a =
a++;
…
the “*a=“ part being a true use in the loop. This is a problem because at that point you would have to reduce the vectorized reduction.
I.e if we wrote “a" as (pointer reduction, VF=2) as:
%a_red = phi <2 x i32*> [preheader, <%a, 0>], [loop, %a_red_next]
%a_red_next = gep %a_red, <3, 3>
you would have to compute the value of “a” for the loop on every iteration from the vectorized reduction.
%a_red = phi <2 x i32*> [preheader, <%a, 0>], [loop, %a_red_next]
%a = horizontal_add %a_red // This is executed on every iteratio...
2013 Oct 23
2
[LLVMdev] First attempt at recognizing pointer reduction
On 23 October 2013 16:05, Arnold Schwaighofer <aschwaighofer at apple.com>wrote:
> In the examples you gave there are no reduction variables in the loop
> vectorizer’s sense. But, they all have memory accesses that are strided.
>
This is what I don't get. As far as I understood, a reduction variable is
the one that aggregates the computation done by the loop, and is used