Zhiru Zhang

Is there a description of the rules we use to do the casting? Why do we have to add these implicit casts on the intermediate variables? HLS tool will do...

Again, Is there a description of the rules we use to do the casting? What's the solution to fix this bug?

I did not know that the Xilinx user manual describes how the HLS tool casts the intermediate variables? Can you send me the link? It's not clear to me what...

> The solution is to remove the casting to make LLVM happy. Under what circumstances do we not cast? Only for array indices? We have to clearly describe our solution....

> In our case, since both i and j are int32, the output width would be int33 according to the manual. I understood this part well. The question is when...

> int32 o = (int32) (int33(i) + int33(j)) The cast to int33 seems totally unnecessary to me. HLS tool does this automatically. >> int32 = int3 + int5 This is...

Right, we should do max_width+1 for the intermediate result. But again, no need to generate cast in the HLS code.

Sounds good. It's fine that we generate redundant HLS code for now. But we should clean it up. In addition, we don't need to generate warning if ap_int is used...

@hanchenye can you provide the complete list of transforms that the pattern class currently supports? Also is there a difference between transform and rewrite? I would prefer we shorten the...

@seanlatias we should first support multi-dimensional array in our internal IR.