When I use Tvm.reduce_axis() to calculate, for example: tvm.reduce_axis(0,end-start), where end and start are both tensor, and an error will be reported:Cannot match type handle64 vs int 32!
But when I enter a fixed value in tvm.reduce_axis(), for example: tvm.reduce_axis(0,2), it works fine.
Since my input must tensor, I would like to ask how to solve this problem or avoid it.
this is unfortunately a limitation of the code generator. To get around it, we could try to add switch clause on the axis dimension and then jump into specialized cases. This is also how the actual code would have been implemented manually