You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository was archived by the owner on Apr 28, 2023. It is now read-only.
halide2isl: do not require init statement for reduction detection
Since the previous commit, the init statement is no longer used
by the TC mapper, so there is no longer any need to keep track of it.
This in turn means that there is also no longer any need to require
the presence of an init statement.
The new code does require that there is at least one reduction
dimension in order to prune out trivial reductions.
For example, in the TC below, both statements would
otherwise be treated as reduction updates.
def fun(float(N,K) A, float(K,M) B, float(N,M) C) -> (O) {
O(i,j) +=! A(i,k) * B(k,j)
O(i,j) = O(i,j) + C(i,j)
}
Note that the reduction is still identified by the tensor name,
meaning that multiple reductions on the same tensor are still
not allowed. Identifying the reduction by the update statement
is left for future work.
0 commit comments