yongfeng-nv opened a new issue #4890: LoopVectorizer depends on ifthenelse URL: https://github.com/apache/incubator-tvm/issues/4890 My PR (https://github.com/apache/incubator-tvm/pull/4885) is blocked by a test failure (https://ci.tvm.ai/blue/organizations/jenkins/tvm/detail/PR-4885/1/pipeline). The bug can be reproduced without my change. Pick this commit (https://github.com/yongfeng-nv/incubator-tvm/commit/42cef28c3161354cdac66a09e6382b220b7ac383) and run tests/python/relay/test_op_level4.py to reproduce the failure. The original test passes the following IR to LoopVectorizer: ``` produce strided_set { parallel (i0.i1.fused, 0, 12) { vectorized (i2.inner, 0, 16) { if ((i2.inner < 3)) { if ((i2.inner < 3)) { strided_set[((i0.i1.fused*16) + i2.inner)] = tvm_if_then_else(... } } } } } ``` The modified test passes this: ``` produce strided_set { parallel (i0.i1.fused, 0, 12) { vectorized (i2.inner, 0, 16) { strided_set[((i0.i1.fused*16) + i2.inner)] = tvm_if_then_else(... } } } ``` The "if" statements make some difference, but I don't know the exact impact. My PR removes the "if" statements from the original test and exposes this bug.
---------------------------------------------------------------- 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: us...@infra.apache.org With regards, Apache Git Services
