Hello, I am also struggling the tvm.transform.Sequential API usage，could you
please help me to understand the following code piece ?
# driver/build_module.py下面的_build_for_device函数
opt_mixed += [
tvm.tir.transform.ThreadSync("shared"),
tvm.tir.transform.ThreadSync("warp"),
tvm.tir.transform.InferFragment(),
tvm.tir.transform.LowerThreadAllreduce(),
tvm.tir.transform.MakePackedAPI(),
tvm.tir.transform.SplitHostDevice(),
]
mod_mixed = tvm.transform.Sequential(opt_mixed)(mod_mixed)
Basically my questions are:
1. Where is this 2nd paramter mod_mixed passed to in the opt_mixed listed
functions? The functions in the list actually does not need this IRModule param
as an input.
2. When I dump the final output of mod_mixed, it is a IRModule that contains
Relay Expressions. But according to the doc, the tvm.tir API is used to convert
Relay IR to lower IR right ? So why mod_mixed still contains the Relay
Expression.
I guess that I mixed up some understandings, please help, thanks.
---
[Visit
Topic](https://discuss.tvm.apache.org/t/tir-relay-why-tvm-tir-api-generates-relay-ir/9132/1)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
here](https://discuss.tvm.apache.org/email/unsubscribe/de6a415c47693d736a48cf2212ad8ab960bf232fef1eefec6b7d136652a9125c).
