[Libre-soc-dev] daily kan-ban update 16nov2020
Luke Kenneth Casson Leighton
lkcl at lkcl.net
Sat Nov 21 20:17:01 GMT 2020
On 11/21/20, Cole Poirier <colepoirier at gmail.com> wrote:
> . I am explicitly
> requesting the list of tasks and their priorities from your perspective.
excellent. that's clear.
for you, there's just the one high-priority task: the JTAG checklist.
nothing else. it's dead simple. includes things like "make sure
power is disconnected" and "verify table X TCK pin is correctly
with both Jean-Paul focussing on coriolis2 and Staf on the Cell
Library and both of them being hyper-busy *and* under NDA *and* the
core being on code-freeze there is nothing that can happen on the core
or the ASIC unless i get word from JP or Staf that something is
at which point everything goes into overdrive for me to get it sorted.
therefore, so as not to interfere with that, we plan ahead for *after*
the tape-out, without actually doing any HDL work (yet).
this therefore involves getting SimpleV designed on top of PowerISA
and approved as well as the Compressed instructions designed, SV
designed, and so on.
all this basically getting us back to where things were with RISC-V
spike simulator and the SV Spec. sigh.
then we need binutils to support the SV augmentations, followed
closely by modifications to the ISACaller python simulator in order to
let us confirm that the two work as expected.
then we need ideally a "full" simulator (probably gem5-power)
after that, being able to write assembly code, Lauri can finally begin
writing Video and Audio accelerated algorithms and we get onto the
feedback loop that allows us to assess the impact of what we are doing
and if it is towards or away from the goal.
this ultimately makes the design of Compressed, SV Prefix, SV
predication and so on our highest priority tasks for the immediate
forseeable future, placing SV on the critical path that blocks several
other critical tasks (the entire NLnet Video Grant is held up for
example) for several months.
tasks that happen not to be (mostly) on this critical path are Kazan
and the 3D MESA driver.
they will be later but are not right now.
also the integration of the MMU, and the SIMD HDL backend, as well as
the scoreboard system, these too are blocked until January (after the
tapeout), as is the FPU ops.
this leaves SV as the "thing that needs critical focus" where that
"thing" has huge tentacles stretching in about 5 simultaneous
inter-related directions: predication, vectorisation, prefixing,
swizzles and 16 bit Compressed. at least 5 more once we include
Video, Audio and 3D opcodes but those come later.
More information about the Libre-soc-dev