place_application_graph vs FixedLocation

[Christian-B]

The place_application_graph does a single pass through all vertices

Therefor if it hits a fixed_location these may already have been taken by another vertex/ same_chip_group

Discussing with @rowleya we decided to leave this until after #436

To replicate edit/ clone
test_application_placer in PACMAN/unittests/operations_tests/placer_algorithms_tests/test_application_placer.py

vertex = make_vertices(app_graph, 10, 1, 1, f"app_vertex{i}")
if i ==55:
for mv in vertex.machine_vertices:
mv.add_constraint(ChipAndCoreConstraint(1, 1)

[Christian-B]

Will require either a sort of the vertices to process or a two pass approach

[Christian-B]

One simplification would be to only allow users to add ChipAndCoreConstraint to the WHOLE Application vertex.

this includes machine_vertices such as GFE ones that that are the ONLy one of their Application_vertex

[Christian-B]

The coverage branch unittests/operations_tests/placer_algorithms_tests/test_application_placer.py
Has a Test that raise SkipTest due to this.

[Christian-B]

APPEARS TESTED AND WORKING

PACMAN/unittests/operations_tests/placer_algorithms_tests/test_application_placer.py

Line 113 in ecade56

def test_application_placer_late_fixed():

[Christian-B]

PACMAN/unittests/operations_tests/placer_algorithms_tests/test_application_placer.py

Line 113 in f689c54

def test_application_placer_late_fixed():

test still skips