openhwgroup / cva6

The CORE-V CVA6 is an Application class 6-stage RISC-V CPU capable of booting Linux
https://docs.openhwgroup.org/projects/cva6-user-manual/
Other
2.21k stars 673 forks source link

schema : realigner at the wrong place #674

Closed ld3804 closed 1 year ago

ld3804 commented 3 years ago

I think the schema ariane_overview.png (cva6/docs/_static/ariane_overview.png) is wrong : the realigner has to be in frontend stage before instruction queue, and not in decode stage after the instruction queue.

ariane_overview

[EDIT] : This change is between the 4.1.2 and 4.2.0 version. Changes are in this commit : https://github.com/openhwgroup/cva6/commit/830540b7574f57c2bf155cf4d75cd1849549eb0b

Bill94l commented 3 years ago

According to the CVA6 paper : https://arxiv.org/pdf/1904.05442.pdf

Instruction Decode: Re-aligns potentially unaligned instructions, de-compresses them and decodes them. Decoded instructions are then put into an issue queue in the issue stage. [source : CVA6 Paper]

ld3804 commented 3 years ago

According to the CVA6 paper : https://arxiv.org/pdf/1904.05442.pdf

Instruction Decode: Re-aligns potentially unaligned instructions, de-compresses them and decodes them. Decoded instructions are then put into an issue queue in the issue stage. [source : CVA6 Paper]

This doc was edited before the 4.2.0 release, so i think this part is out of date because it's a change between Ariane 4.1.2 and Ariane 4.2.0

I checked the Ariane 4.1.2, and a "soft realigner" was on the frontend stage and the re-aligner module was after the instruction queue, in the id_stage (like on the schema). With Ariane 4.2.0, the re-aligner is now placed before the instruction queue, and no re-aligner was on the id_stage.

The associated commit : https://github.com/openhwgroup/cva6/commit/830540b7574f57c2bf155cf4d75cd1849549eb0b

For Ariane 2.0.0, here the source code :

Doc is also out of date (no realignment in id_stage)

I will edit the issue to add the link to the commit.

MikeOpenHWGroup commented 1 year ago

Hi @Bill94l and @ld3804, there have been no further comments for ~1.5 years and the last responses are "thumbsup". Given that, I believe this issue is resolved, and can be closed. If that is not the case, please feel free to reopen this issue or create a new one.