mbrookhart commented on a change in pull request #7117:
URL: https://github.com/apache/tvm/pull/7117#discussion_r544452199
##########
File path: tests/python/relay/test_any.py
##########
@@ -1430,6 +1439,21 @@ def test_non_max_suppression():
disable_targets=["nvptx"],
)
+ np_data = np.zeros((1, 0, 6)).astype("float32")
+ np_valid_count = np.array([0]).astype("int32")
+ np_indices = np.zeros((1, 0)).astype("int32")
+ np_max_output_size = -1
+ np_indices_result = np.zeros((1, 0))
+ np_valid_box_count = np.array([[0]]).astype("int32")
+
+ check_result(
+ [np_data, np_valid_count, np_indices, np_max_output_size],
+ mod,
+ [np_indices_result, np_valid_box_count],
+ only_vm=False,
+ disable_targets=["nvptx"],
Review comment:
This tests the empty output VM change :+1:
Why disable nvptx?
##########
File path: tests/python/relay/test_any.py
##########
@@ -199,6 +199,15 @@ def test_any_concat():
ref = np.concatenate([x_np - 3.0, y_np * 5.0], axis=0)
check_result([x_np, y_np], mod, ref)
+ num_inputs = 25
+ x = [relay.var("x", shape=(relay.Any(),), dtype="float32") for _ in
range(num_inputs)]
+ z = relay.op.concatenate(x, axis=0)
Review comment:
this tests the injective schedule :+1:
##########
File path: python/tvm/topi/cuda/nms.py
##########
@@ -754,7 +782,22 @@ def non_max_suppression(
)
score_axis = score_index
score_shape = (batch_size, num_anchors)
- score_tensor = te.compute(score_shape, lambda i, j: data[i, j,
score_axis], tag=tag.ELEMWISE)
+ data_buf = tvm.tir.decl_buffer(data.shape, data.dtype, "data_buf",
data_alignment=8)
Review comment:
This looks fine, but I'm a little surprised it's necessary. Do you have
a test case that breaks the current code, or is this mostly for performance?
##########
File path: python/tvm/topi/cuda/conv2d_transpose_nchw.py
##########
@@ -194,6 +197,8 @@ def _callback(op):
if cfg.is_fallback:
N, F, Y, X = get_const_tuple(conv.shape)
+ if not isinstance(N, int):
+ N = 1
Review comment:
Can we add a test that hits this change?
----------------------------------------------------------------
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]
