jtuyls opened a new pull request #8989: URL: https://github.com/apache/tvm/pull/8989
While experimenting with `dyn.strided_slice` I noticed that some of the DynamicToStatic pass tests are not testing dynamic expressions. For example, the created strided_slice function here https://github.com/apache/tvm/blob/02f885aacde5739c060507afa0b534019d5439cc/tests/python/relay/test_pass_dynamic_to_static.py#L435 looks like this: ``` fn (%x: Tensor[(1, 3, 10, 10), float32]) { strided_slice(%x, begin=[0, 0, 0, 0], end=[1, 3, 10, 10], strides=[1, 1, 1, 1], axes=None) } ``` which I think doesn't test the DynamicToStatic pass being executed on this expression. With the small changes in this PR, the above function looks like this: ``` fn (%x: Tensor[(1, 3, 10, 10), float32], %begin: Tensor[(4), int64], %end: Tensor[(4), int64], %strides: Tensor[(4), int64]) { %0 = cast_like(0, %begin); %1 = shape_of(%x, dtype="int32"); %2 = cast_like(%1, %begin); %3 = slice_like(%2, %begin, axes=None); %4 = less(%begin, %0); %5 = add(%begin, %3); %6 = where(%4, %5, %begin); %7 = greater_equal(%6, %3); %8 = where(%7, %3, %6); dyn.strided_slice(%x, %8, %end, %strides, begin=None, end=None, strides=None, axes=None) } ``` which I think is a better test for the DynamicToStatic pass. Similarly, tests for topk, resize2d, upsampling, upsampling3d and pad have been adjusted. @mbrookhart Could you help with reviewing this PR? -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
