[MetaSchedule] Fix tensorcore winograd task extraction

[masahi]

When both AutoTVM and AS/MS winograd are applicable, the latter is never picked by the op strategy because the AutoTVM one has a higher plevel (20 vs 15)

tvm/python/tvm/relay/op/strategy/cuda.py

Lines 271 to 276 in b16a64d

	strategy.add_implementation(
	wrap_compute_conv2d(topi.cuda.conv2d_nhwc_tensorcore),
	wrap_topi_schedule(topi.cuda.schedule_conv2d_nhwc_tensorcore),
	name="conv2d_nhwc_tensorcore.cuda",
	plevel=20,
	)

This has been fixed, and now we can properly extract FP16 winograd tasks for MS tuning.

@vinx13 @junrushao @zxybazh

[tvm-bot]

Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment.

• cc @Hzfengsy, @elvin-n, @junrushao, @quic-sanirudh _{See #10317 for details}

_{Generated by tvm-bot}

[junrushao]

Interesting! I was completely unaware of the issue, and thanks for pointing it out!