Not sure whether the decision has already been done but I'll put my two cents
here.
You have covered all the tradeoffs. Since in most cases, the divisor is
positive, I believe the important move was to add one of floordiv or
euclideandiv (which you are doing now). The difference between floordi
Is there a way to enable the new infra (and disable the old one) to be able to
evaluate it, for example, for unit testcases, bugs we encounter in customized
forks of TVM, etc.?
---
[Visit
Topic](https://discuss.tvm.ai/t/integer-constant-folding-for-division-and-mod/2008/9)
to respond.
Y
You're right. I just checked those specific rules in the new infra and they
were fixed. I had mistakenly assumed that the rules of the new infra were
copied from the old infra without further changes.
On an orthogonal but related topic: I believe it's still important to fix bugs
in old relea
The investigation I'm doing will affect both the old and new infra. So far I
have found out that we currently have simplification rules that are not
correct. For example, we simplify (x*4 + y) / 2 to x*2 + y/2 which is not
correct in general but is true if we prove that x*4+y and y are both
n
I tried the fix but a few TVM testcases started to fail. I think those
testcases were assuming that division was Euclidean and that's why they were
passing before. I need more time to investigate and fix them.
---
[Visit
Topic](https://discuss.tvm.ai/t/integer-constant-folding-for-divisio
I think `div_imp` should simply return `a/b`
---
[Visit
Topic](https://discuss.tvm.ai/t/integer-constant-folding-for-division-and-mod/2008/2)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
here](https://discuss.tvm.ai/e
@sgrechanik-h @tqchen @yzhliu
My understanding is that TVM uses truncated division/mod, unlike Halide that
uses Euclidean division/mod. Nevertheless, TVM uses `div_imp` (imlementation
below) to do constant folding for division. Seems like this implementation is
coming from Halide source code.
