The problem is basically fixed in [PR
#5382](https://github.com/apache/incubator-tvm/pull/5382), but I'm still not
very clear of the whole picture.
---
[Visit Topic](https://discuss.tvm.ai/t/warp-memory-in-inferbound/6421/2) to
respond.
You are receiving this because you enabled mailing
Approach 1 does not mean no changes to TVM at all. At least we have to add an
option to turn off the initialize-reduction-buffer-to-zero procedure.
---
[Visit
Topic](https://discuss.tvm.ai/t/rfc-implement-add-to-semantic-in-tvm/3618/19)
to respond.
You are receiving this because you enab
I'm working with a buffer that bound to warp scope. In
`src/te/schedule/message_passing.cc:208`:
```c++
PrimExpr outer = state.at(s->outer);
PrimExpr inner = state.at(s->inner);
PrimExpr factor = dom_map.at(s->inner)->extent;
PrimExpr parent_min = dom_map.at(s->parent)->min;
state[s->parent] =
I think this problem would be trivial if it is able to inline a reduction. For
example, when computing GEMM `C += A * B`, we could simply inline the
multiplication into the addition, so these two stages would become one.
However, inlining a reduction is currently not supported. See [issue
#90
