[P1] Adapt existing MRRG mapping to support spatial-only accelerators

[tancheng]

MRRG mapping is prototyped in #58 and got refactored/improved by #73 with backtracking, which can perform heuristic mapping like https://github.com/tancheng/CGRA-Mapper now.

We should be able to easily (or somehow) change certain parameters to make the mapper support spatial-only CGRA. Basically,

• I view the difference is just to not allow the placement as long as a tile is occupied by anything at anytime: https://github.com/coredac/dataflow/blob/cc7f4a6dbe3e878d7b89a511a004dfdbe9ac3f5f/lib/NeuraDialect/Mapping/mapping_util.cpp#L455C26-L455C47
• The mapping can still be performed in MRRG way, i.e., expand resources across time

Filed on @ShangkunLi for now as the hybrid accelerator design is one of our target platforms.

[tancheng]

CCing @syifan @n0thingNoob @Jackcuii for visibility, as the simulator would also model such spatial-only (sub-category of spatial-temporal) designs.

[syifan]

CCing @syifan @n0thingNoob @Jackcuii for visibility, as the simulator would also model such spatial-only (sub-category of spatial-temporal) designs.

So, this means we will not model temporal mapping, right? So, one PE only executes one task for one program.

[tancheng]

CCing @syifan @n0thingNoob @Jackcuii for visibility, as the simulator would also model such spatial-only (sub-category of spatial-temporal) designs.

So, this means we will not model temporal mapping, right? So, one PE only executes one task for one program.

No. I mean we will support both and hybrid heterogeneous, if possible.