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

[Jacob Lifshay]

On Mon, Feb 24, 2020, 13:32 Staf Verhaegen <staf at fibraservi.eu> wrote:

> All,
>
> I may have missed it, but up to now I haven't seen anything on how
> branches will be handled coupled with the Dependency Matrix etc.
> Will branch prediction be included and if so how will a branch
> misprediction be handled ?


We will have branch prediction in some form (not specifically decided yet),
mispredictions are handled by cancelling all instructions afterwards then
updating the branch prediction tables (which must not be updated till after
the mispredicted branch is non-speculative, since otherwise it would be a
spectre-style data leak through the branch predictor)

Jacob