Example of bsg_mesh_stitch with multiple networks

[dpetrisko]

Can't figure out how to use bsg_mesh_stitch to stitch together multiple networks. 5-D arrays are confusing -- an example would be helpful.

https://github.com/bespoke-silicon-group/basejump_stl/blob/master/bsg_noc/bsg_mesh_stitch.v

[taylor-bsg]

https://github.com/bespoke-silicon-group/bsg_manycore/blob/master/v/bsg_manycore.v

On Sat, Nov 30, 2019 at 6:44 PM Dan Petrisko notifications@github.com wrote:

Can't figure out how to use bsg_mesh_stitch to stitch together multiple networks. 5-D arrays are confusing -- an example would be helpful.

https://github.com/bespoke-silicon-group/basejump_stl/blob/master/bsg_noc/bsg_mesh_stitch.v

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/bespoke-silicon-group/basejump_stl/issues/197?email_source=notifications&email_token=AEFG5AE4T3DHFT66S6EEQYDQWMQH3A5CNFSM4JTJM3LKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4H5BHIPQ, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEFG5AGYWIC22XLP3ROV43TQWMQH3ANCNFSM4JTJM3LA .

[dpetrisko]

https://github.com/bespoke-silicon-group/bsg_manycore/blob/master/v/bsg_manycore.v

That only has nets_p=1. I’d like to use mesh_stitch to combine multiple networks with nets_p=2+, for example mem_cmd/mem_reponse. Unless I’m misunderstanding what nets_p is for.

[taylor-bsg]

In the manycore, it ended up being more convenient to put the plurality of networks in a struct in the innermost array scope rather than the outermost; for example bsg_manycore_link_sif_s in https://github.com/bespoke-silicon-group/bsg_manycore/blob/master/v/bsg_manycore_pkg.v. I think one reason is that the various networks may not necessarily the same width; the other reason is that the natural usage of these networks is that you generally only access the different networks at the leaf nodes, so it does not make sense to put the networks at the first indirection of the array. This can be addressed by https://github.com/bespoke-silicon-group/basejump_stl/blob/master/bsg_misc/bsg_transpose.v but that is very cumbersome.

M

On Sat, Nov 30, 2019 at 11:57 PM Dan Petrisko notifications@github.com wrote:

That only has nets_p=1. I’d like to combine multiple networks with nets_p=2+

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/bespoke-silicon-group/basejump_stl/issues/197?email_source=notifications&email_token=AEFG5ADQ37AJAP7JHUNFGZLQWNU6VA5CNFSM4JTJM3LKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEFRAKSI#issuecomment-560072009, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEFG5AC47LYYGMHSFXWWJGLQWNU6VANCNFSM4JTJM3LA .

[dpetrisko]

Interesting. The same width point is a good one. It seems like nets_p is useless as it currently stands then, because nets_p=1 with a struct more powerful than nets_p=2 with concatenated equal sized nets.

[taylor-bsg]

I think it is potentially useful if you want to specify a parameterizable number of networks, although probably the nets should come last in the order and not first. But agreed that it is a deprecatable feature as it stands!

On Tue, Dec 3, 2019 at 6:19 PM Dan Petrisko notifications@github.com wrote:

Interesting. The same width point is a good one. It seems like nets_p is useless as it currently stands then, because nets_p=1 with a struct more powerful than nets_p=2 with concatenated equal sized nets.

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/bespoke-silicon-group/basejump_stl/issues/197?email_source=notifications&email_token=AEFG5AA765Q5K442QYHGNG3QW4HUZA5CNFSM4JTJM3LKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEF3PTMQ#issuecomment-561445298, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEFG5AH3RVH32XGAVCSPBWLQW4HUZANCNFSM4JTJM3LA .