dace ignoring auto_optimize flag when using to_sdfg? #1379
-
I am not sure if this is a bug or a misunderstanding from my part, I have tested this on v0.14.4 of dace. @dace.program(auto_optimize=True) I define example numpy matrices as follows: m = 200 When I try to generate the SDFG using the following commands: madd_dace_sdfg = madd.to_sdfg(vA, vB, m, n) I see that the auto_optimize flag for the function gets ignored i.e. it produces the same graph regardless of whether the auto_optimize flag is set to True or False. On the other hand, if I generate the graph using the following command (without going through the intermediate to_sdfg(): madd_dace_compiled = madd.compile(vA, vB, m, n) I see that the graph changes based on the auto_optimize flag. Is there something I am missing? Thanks in advance! |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 3 replies
-
Hi, thank you for noting this! This behavior is by design, but |
Beta Was this translation helpful? Give feedback.
Hi, thank you for noting this! This behavior is by design, but
to_sdfg
should raise a warning and it does not. Could you please report an issue? This should be a two-line fix in https://github.com/spcl/dace/blob/master/dace/frontend/python/parser.py#L235