[J-core] new here, project sounds interesting.
BGB
cr88192 at gmail.com
Thu Sep 1 04:51:46 EDT 2016
I recently encountered this project via hearing about it, and watching a
YouTube video talking about it.
alas, I don't have an FPGA nor currently any money to buy one, so will
not be able to test on real HW (my budget it pretty much dead right now).
however, I have briefly looked at the SH-2 ISA, and it seems reasonably
nicely designed. kind of reminds me of MSP430 in some ways. general
instruction coding seems a lot less hairy than Thumb.
granted, I haven't really fully looked into it, so there is always a
potential for hidden "gotchas".
I am not sure off hand if I can do much useful for the project though.
general info (about self / projects):
I have some experience writing compilers and emulators.
most of my compilers have targeted custom bytecode VMs, generally with
interpreter or JIT backends, and some have used C as an input language.
some of my other compilers have used variants of a custom "scripting
language", where C isn't necessarily the best language design to use in
a VM environment. my current script-language sort of resembles a
C/Java/C# hybrid (will refrain from going too much into it here).
but, for anyone curious, here is a code example from a test where I
ported a JPEG decoder over to it (reasonably direct port from C):
https://github.com/cr88192/bgbtech_bsvm2/blob/master/bstest/bs_decjpg.bs
for comparison (a C version):
https://github.com/cr88192/bgbtech_bsvm2/blob/master/bcn_decjpg.c
I have also done a C compiler, where previously I had attempted to
address some of the technical issues which exist with C (in a variant I
had called C-Aux), but found that they weren't so easily defeated (in a
real-world implementation).
though, in a very simple summary: C is a reasonably expensive language
to compile and tends to produce output that is difficult to make
binary-portable across multiple platforms short of pretty much wrapping
everything. I had developed theoretical mechanisms, but realized I would
effectively still need to wrap everything to deploy them.
though, it was still modestly useful as a means to do green-threading
(needed ~ 10us scheduler latency, wasn't going to get this with
preemptive threading, and GCC lacks an option for running code via a
customizable trampoline loop).
for emulators, I have written emulators for x86, x86-64, and MSP430.
have considered an ARM emulator, but haven't gotten to it (for code
intended for ARM, had generally faked it on the PC using x86).
for x86 and x86-64, the emulators had primarily focused on userland
code, which allowed some simplification and trickery (for example, its
MMU was based on page-spans rather than page-tables). my initial tests
focused on 32-bit, using MinGW and custom libraries for compiling
programs, and basically worked.
later, there was a short-lived project to re-implement the Win32 API and
DLLs (for a small subset), but I lost motivation and went onto other
stuff before really getting any Win32 software working. figured
officially I would only support 32-bit x86 with a 486-like ISA.
the MSP430 emulator was fairly straightforward, given how simple it was.
I had noted that some existing emulators existed, but were horribly slow
(vs the real HW) and didn't seem to allow for customizable peripherals
(say, what if one wants to simulate some external electronics the MCU is
hooked up to?...).
I did the emulator more like my other interpreters (via dynamic
translation to call-threaded-code), which ended up considerably faster
than real-time (~ 150 MIPS, or an MSP at ~ 500 MHz or so).
later added logic to slow it down to real-time speeds (as controlled by
CPU registers), as well as to account for the number of clock-cycles
required by various instruction forms (needed to try to match the
behavior of the emulated MSP with the real hardware, and to get more
sensible results with the virtual IO pins).
I did also try to make an electronic simulator to go along with it (in a
vaguely similar sense to something like LTspice mixed with a
board-layout tool), but failed at making the simulator part not suck.
imagination was something like one could drop an MCU onto a board (keyed
to appropriate ROM image), hook components up to it (by routing traces),
and see in real-time how everything would behave. UI basically worked,
but the simulation part was basically unusable (would likely need to be
dropped and tried again, analog signaling was extremely slow).
however, the MCU simulation part did work, and it was still possible to
use it with peripherals implemented using code.
I suspect I could probably do similar for the SuperH ISA if I had a
reason to do so.
if done, can probably get similar speeds to the MSP, or a little faster
if a JIT is used.
any thoughts?...
More information about the J-core
mailing list