[TEST] Skip TensorCore TOPI test when it is not available

[tqchen]

#7146 Introduces a test on tensorcore for batchmatmul

Because tensorcore is only optionally supported in the topi testing stage, it would be great to skip it when it is not available.

[tqchen]

cc @jwfromm @jcf94 @Meteorix

[leandron]

(Posting here as I think it is related)

When running TVM tests on a CPU only machine, I'm getting an error when running /test_topi_batch_matmul_tensorcore.py:test_batch_matmul:

tvm/tests/python/topi/python/test_topi_batch_matmul_tensorcore.py

Lines 66 to 71 in 89e3688

	@tvm.testing.uses_gpu
	def test_batch_matmul():
	verify_batch_matmul(1, 1, 16, 16, 32)
	verify_batch_matmul(5, 5, 16, 16, 32)
	verify_batch_matmul(5, 5, 16, 32, 32)
	verify_batch_matmul(30, 30, 16, 32, 32)

The error I see is:

Check failed: allow_missing == false: Device API gpu is not enabled.

As you can see, the test is marked as uses_gpu.

I wonder whether we should move it to either requires_gpu or even requires_tensorcore? (happy to send a PR, in either case)

[tqchen]

I think requires_tensorcore is the right macro cc @tkonolige

[tkonolige]

Yes, it should be requires_tensorcore.