powderluv

I think the creating a new subgraph should suffice. Maybe it can create something like this:  credit: https://forum.graphviz.org/t/clusters-outline-geometry-attribute/313/2 And then if we can select edges and snap them to...

hit the same issue on my OSX bigsur system. dot -v -Tjpe:cairo:gdk multi.dot > multi_cairo_gdk.jpg is the only one that doesn't get recognized. the others like $ dot -v -Tjpe:cairo:gd...

quartz:quartz crashes with exit code 139. Others all return 0. zsh: segmentation fault dot -v -Tjpe:quartz:quartz multi.dot > multi-quartz-quartz.jpg (pyenv_3.8) powderluv@denali graphs % echo $? 139 (pyenv_3.8) foo@denali graphs %...

Thank you for digging into it. I dont think i care as much about using quartz just that the pydot test fails on a OSX Big Sur install. So if...

This seems like this issue here: https://gitlab.com/graphviz/graphviz/-/issues/1320 I tried to build with the solution recommended at the end but ran into some compile errors. I agree this should be tracked...

please attach a link to the .mlir and iree command line to execute / recreate it

I tested all the way back to https://github.com/google/iree/releases/tag/candidate-20220604.159 and it fails the same way.

In yesterday's meeting we discussed trying to get this in with a big on / off switch. And then make tweaks as necessary. @mariecwhite

This happens on M1 too: https://github.com/nod-ai/SHARK/runs/7636875382?check_suite_focus=true All the following models fail on M1 Vulkan but pass on Linux Vulkan ``` =========================== short test summary info ============================ FAILED tank/convbert-base-turkish-cased_tf/convbert-base-turkish-cased_tf_test.py::ConvBertModuleTest::test_module_static_vulkan FAILED tank/pytorch/MiniLM-L12-H384-uncased/MiniLM-L12-H384-uncased_test.py::MiniLMModuleTest::test_module_dynamic_vulkan...

it is 1:45min when running sequentially -- we can shard cpu / cuda / vulkan and run it at 1/3rd the time.