Quadratic memory usage about enzyme HOT 2 CLOSED

Having not tried out your code, my immediate guess what's happening is that because parts and x_out aren't marked restrict (or noalias) the compiler has to assume that they could overlap and thus it has to cache a bunch.

Try marking the two arguments as restrict. With good Alias Analysis markings I think this code shouldn't require any extra memory in cache.

from enzyme.

Thanks 👍
By replacing the declaration to use restrict with :
void foo( double* __restrict__ parts, double * __restrict__ x_out,int n)

Now it works fine !

from enzyme.