[libre-riscv-dev] New Person

[lkcl]

On Sat, Oct 6, 2018 at 5:49 AM Daniel Benusovich
<flyingmonkeys1996 at gmail.com> wrote:

> I am Daniel Benusovich and a newcomer to the RISC-V and Open
> Source world in general and was hoping you could use some help in one way or another.

 much appreciated. yes!

> I saw your post asking for help with development for the Vulkan
> 3D GPU project and while I do not have any experience in this
> area I would love to help out and get my foot in the door
> contributing to RISC-V development. I have experience working
> in C, C++, Java, VHDL and some assembly though I am happy
> to learn any languages and tools that are used for this project specifically.

 that's fantastic.

> Let me know what needs to be done and I will get right on it.

 thanks daniel.

 ok, so the preliminary roadmap is here:
 https://libre-riscv.org/3d_gpu/roadmap/

 and there are currently two parallel tracks: i'm working at the
beginning of a chain of dependencies, implementing SV (a
hardware-level parallelism API for RISCV, which is best described as a
"hardware-level macro-expansion system" that effectively applies a
for-loop to *scalar* assembly instructions) and starting with the
simulator, spike:

https://git.libre-riscv.org/?p=riscv-isa-sim.git;a=shortlog;h=refs/heads/sv
https://git.libre-riscv.org/?p=riscv-tests.git;a=shortlog;h=refs/heads/sv

 the second track is jacob's domain, he's focussing on the vulkan3d
software-renderer, where, once that's done, we will be able to begin
doing benchmarks, and, based on those benchmarks (following the
techniques pioneered by jeff bush from the nyuzi project), actually
design custom RISC-V instructions and a microarchitecture that will
bring the speed up and power consumption down to match an entry-level
embedded GPU such as MALI400, Vivante GC800 and so on.

 in terms of "numbers" that's around 5 GFLOPS, 100 MPixels/sec and i
believe something like 30 MTriangles/sec which are extremely modest
and perfectly reasonably achievable: the tricky bit is doing so in
well under 1 Watt @ 28nm.

anyway that is a long way ahead: the most immediately useful thing
would be to replicate and document the build environment for spike-sv
and the unit tests.  this is slightly tricky as it's based on
riscv-tools which is not being properly maintained.  i set up a build
environment over a year ago and haven't dared update it since, just in
case it breaks.

i started a page here:
https://libre-riscv.org/3d_gpu/spike_sv/

the unit tests themselves also need a bit of work, as i have made some
mistakes, plus more tests are also needed.

lots of things: even just asking questions helps.

thanks daniel,

l.