prevent over unrolling in optimzer

This commit is contained in:
George Hotz 2023-10-31 11:45:18 -07:00
commit c59ea32f90

View file

@ -368,7 +368,7 @@ class OptimizedKernel(Kernel):
break
# if last dim is small(ish) and it's a reduce dim, upcast the reduce (loop unrolling). no simplify needed since it's just an upcast. NOTE: careful, this has broken VALIDHACKS
if self.first_reduce < (self.shape_len-self.upcasted) and (len(list(self.shape_offsets(self.full_buf_index))) <= 4 or not any(r for _,_,r in self.upcasted_axis(self.full_buf_index))):
if self.first_reduce < (self.shape_len-self.upcasted) and (len(list(self.shape_offsets(self.full_buf_index))) <= 4 or not any(r for _,_,r in self.upcasted_axis(self.full_buf_index))) and (self.upcasted == 0 or prod(self.full_shape[-self.upcasted:]) < 64):
if (s:=self.full_unupcasted_shape[-1]) <= 32 and isinstance(s, int): # NOTE: cannot loop unroll symbolic axis
self.apply_opt(Opt(OptOps.UNROLL, len(self.full_unupcasted_shape)-1-self.first_reduce, 0))
# if it's small, upcast a second reduce dimension too