Jerry Zhao comments

Results 391 comments of


                                            Jerry Zhao

Is there any possibility that ```RegNext(rob.io.flush.valid)``` and ```brupdate.b2.mispredict``` are valid in the same cycle, which makes the mispredict FTQ entry's information writen into the redirect FTQ entry?

I believe this should be impossible. The flush signal would have killed the in-flight branch in the previous cycle, so it would be impossible for the branch mispredict signal to...

Is there any possibility that ```RegNext(rob.io.flush.valid)``` and ```brupdate.b2.mispredict``` are valid in the same cycle, which makes the mispredict FTQ entry's information writen into the redirect FTQ entry?

The valid signal is gated off by the flush, which suppresses the mispredict signal the next cycle.

Is there any possibility that ```RegNext(rob.io.flush.valid)``` and ```brupdate.b2.mispredict``` are valid in the same cycle, which makes the mispredict FTQ entry's information writen into the redirect FTQ entry?

I see. This is just a performance bug, then, not a functionality bug? The incorrectly updated entry should only cause some incorrect branch predictor updates.

Dose Boom V3 core support B expansion？

Not currently, possibly in the future.

Is BOOM affected by Spectre and Meltdown?

The secureboom branch is currently outdated. We implemented a prototype "Spectre-resistant" data cache, and its sitting in a private branch currently. We'll try to move it to the public secureboom...

Is BOOM affected by Spectre and Meltdown?

Our implementation is comparable to InvisiSpec

Is BOOM affected by Spectre and Meltdown?

The secureboom branch is stale, and it will be challenging to make it work again (you'd have to track down old versions of several dependencies). I paused work on secureboom...

Add logging printf DPI widgets that direct the log to a file

The point of this is to avoid post-processing with grep or sed. You can pass a string into a blackbox module through the IO, as a fixed-length array of bytes....

What if bpd_end_idx has exceeded bpd_end_idx when starting the check of repair end in Line 259-266?

> what if bpd_end_idx has exceeded bpd_end_idx What do you mean here? > When handling mispredict repair (which takes some cycle to finish), what if the redirect generated by rob.flush...

What if bpd_end_idx has exceeded bpd_end_idx when starting the check of repair end in Line 259-266?

Hmm, that's a good observation, and it would certainly be a bug. I believe in practice this is unlikely, as it is unlikely for there to many invalid FTQ entries...