[Fix] Fix get_valid_count flaky test for cuda (#4901)

* get_valid_count accuracy issue fixed for individual tests but not for all tests running together

* minor fix

* initialize valid_count and PrefixSum buffers

* test updated

* udpate relay test as well

* update document

* fix lint

* address comment

* fix lint

* correct atomicAdd identifier name

tests/python/relay/test_op_level5.py

@@ -221,8 +221,6 @@ def test_get_valid_counts():
         func = relay.Function([x], z.astuple())
         func = run_infer_type(func)
         for target, ctx in ctx_list():
-            if target == 'cuda':
-                return
             intrp = relay.create_executor("debug", ctx=ctx, target=target)
             out = intrp.evaluate(func)(np_data)
             tvm.testing.assert_allclose(out[0].asnumpy(), np_out1, rtol=1e-3, atol=1e-04)

topi/tests/python/test_topi_vision.py

@@ -67,8 +67,8 @@ def verify_get_valid_counts(dshape, score_threshold, id_index, score_index):
         tvm.testing.assert_allclose(tvm_out2.asnumpy(), np_out2, rtol=1e-3)
 
     for device in ['llvm', 'cuda', 'opencl']:
-        # Disable gpu test for now
-        if device != "llvm":
+        # Disable opencl test for now
+        if device != "llvm" and device != "cuda":
             continue
         check_device(device)