[libre-riscv-dev] GPU design

[lkcl]

On Thu, Dec 6, 2018 at 2:05 PM Jacob Lifshay <programmerjake at gmail.com> wrote:
>
> the Alpha 21164 has a scoreboard and two load units (integer pipe 0 & 1),
> also, the d-cache has two read ports, so I would expect that it can execute
> multiple loads simultaneously.
>
> http://www.eecg.toronto.edu/~moshovos/ACA07/lecturenotes/21164%2520(micro).pdf

http://users.utcluj.ro/~sebestyen/_Word_docs/Cursuri/SSC_course_5_Scoreboard_ex.pdf

examples always appear to be the 6600 (first time SB appeared),
definitely 2+ LOADs stall in that design.  ARM Cortex A8 apparently is
dual-issue superscalar, however it seems to be a hybrid where the
compiler is expected to allocate instructions based on an alternating
cycle to "help" avoid conflicts.

l.