Update the arch_spec example with memory placement#282
Conversation
|
@tancheng Is this ok? Or we still need to add some other things? |
There was a problem hiding this comment.
Pull request overview
Updates the arch_spec_example.yaml test/example architecture spec to illustrate off-chip memory placement intent (memory_side) alongside an explicit concrete realization via tile_overrides.
Changes:
- Add
memory_side: "West"as an architectural-intent annotation incgra_defaults. - Remove memory ops from
tile_defaults.fu_typesand instead explicitly mark memory-capable tiles viatile_overrides. - Expand the example
tile_overridesto show multiple west-edge tiles (CGRA (0,0)) havingmem/mem_indexed.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
You can also share your feedback on Copilot code review. Take the survey.
tancheng
left a comment
tancheng
left a comment
There was a problem hiding this comment.
Why there is no test failing? It should fail a lot due to the mapping changes..
So indeed many tests failed? |
yes. I am working on it. |
|
I guess though many kernels are affected, the mappedII didn't change a lot? @ShangkunLi @n0thingNoob |
Not changed too much, but the mapping time increased a lot orz.... |
For the one with longest compilation time, can you try also make the tiles on the first row to be able to access memory (not only the first column), to see whether the compilation time can be improved? |
The configuration of this pr has already set that both the first col & row can access the memory. I think the mapping time introduced by this pr is acceptable (<5min in total). The mapping time introduced in #278 is much larger (~10 minutes for all tests). |
#148
To determine the memory placement