[libre-riscv-dev] branches, prediction and backtrack

[Cole Poirier]

> On Feb 24, 2020, at 1:37 PM, Luke Kenneth Casson Leighton <lkcl at lkcl.net> wrote:
> 
> if you have mitch alsup's book chapters (let me know if i have sent
> them to you already) it is described, there, and correctly implemented
> in the soc git repo score6600.py.

Can you send me Mitch Alsup’s book chapters? I promise to respect his wishes that they not be shared widely nor published online.

> 
> i literally just described this mechanism 3 minutes ago and it is here:
> http://bugs.libre-riscv.org/show_bug.cgi?id=186#c18
> 
> shadows are used for exceptions, branch speculative cancellation,
> predication, LD/ST page faults - everything.
> 
> hypothetically we could execute both halves of the branch however i
> could not get it to work, and instead it only takes one path (that can
> be cancelled).
> 
> l.

In light of this, is the branch prediction information on this page (https://libre-riscv.org/3d_gpu/microarchitecture/) still accurate? Before you responded Luke, this is what I was going to reply to Staf with, so I’m curious if that would have actually been unhelpful.

Cole