Commit Graph

4 Commits

Author SHA1 Message Date
yao_yf 854b7839e5 auto parallel dataset handle in feed mode 2022-08-15 14:41:16 +08:00
wangshengnan12@huawei.com 48065b72a1 adapt_control_flow 2022-05-19 15:50:29 +08:00
zhousiyi d4aba77d3d enable if parallel call transformation by default; fix test case ut/python/pipeline/parse/test_fix_bug.py::test_fix_expanddims_loss_scale
Check only if all branch of middle graph's output are scalar, parallel_if transformation should be ignored

fix test_auto_monad.py::test_while_if geswitch input shape check failed

fix ut/python/pipeline/parse/test_super.py::test_super_cell core dump as varg and kwarg parameter are removed but those flags are not set, it will cause FuncGraph::GetPositionArgsCount be negative

skip test_control_flow in ut/python/paralle as Parallel team is working on contro flow adaption.

fix ut/python/ir/test_row_tensor.py::test_row_tensor_with_control_flow_if

fix for kwonlyargs when eliminate_parameter and add testdases

fix cannot join for st/control/inner/test_012_if_in_for.py

fix pclint errors

fix test/st/scipy_st/sparse/test_linalg.py, if the output cnode of former funcgraph is Depend CNode with input 1 as middle funcgraph call, then keep the output of former funcgraph as Depend other than middle funcgraph tail call.

log as exception

If the output of middle funcgraph contain PyInterpret CNode, then ignore
the transformation as the Value in Abstract will be kAnyValue after join
of the PyInterpret CNode value.

Backend cannot support for RealDiv with Int64, so change one operand to Float
2022-04-29 08:22:56 +00:00
lichenever 3c7cfb7c08 auto_parallel_support_control_flow 2021-07-22 17:31:31 +08:00