FrozenGene commented on a change in pull request #7635:
URL: https://github.com/apache/tvm/pull/7635#discussion_r593110478
##########
File path: python/tvm/relay/op/strategy/arm_cpu.py
##########
@@ -198,7 +231,9 @@ def conv2d_strategy_arm_cpu(attrs, inputs, out_type,
target):
elif layout == "NHWC":
assert kernel_layout == "HWOI"
strategy.add_implementation(
-
wrap_compute_conv2d(topi.arm_cpu.compute_depthwise_conv2d_nhwc),
+ wrap_compute_conv2d(
+ topi.arm_cpu.compute_depthwise_conv2d_nhwc,
need_auto_scheduler_layout=True
Review comment:
> If AutoSchedule is enabled, use
`wrap_topi_schedule(topi.arm_cpu.schedule_depthwise_conv2d_nhwc)` is the same
as `naive_schedule`. So the then path and the else path are the same.
>
> This will only take effect if auto_schedule is not enabled.
My concern is `topi.arm_cpu.compute_depthwise_conv2d_nhwc` maybe changed for
auto tvm template (like spatial pack). However, auto scheduler's compute
template should be generic computation. The way I mentioned could guard it.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
