ayjay_t has quit [Read error: Connection reset by peer]
ayjay_t has joined ##openfpga
Miyu has joined ##openfpga
genii has joined ##openfpga
Asu has joined ##openfpga
ayjay_t has quit [Read error: Connection reset by peer]
ayjay_t has joined ##openfpga
<whitequark>
azonenberg: i have been informed that PCIe PHY exist that use MDIO
rohitksingh has joined ##openfpga
m4ssi has quit [Remote host closed the connection]
emeb has joined ##openfpga
rohitksingh has quit [Remote host closed the connection]
zng has quit [Ping timeout: 244 seconds]
zng has joined ##openfpga
jevinskie has joined ##openfpga
jevinski_ has quit [Ping timeout: 250 seconds]
jevinskie has quit [Ping timeout: 240 seconds]
jevinski_ has joined ##openfpga
azonenberg_work has joined ##openfpga
<shapr>
whitequark: background - Digium (makers of Asterisk) was planning to develop a hardphone based on the Raspberry Pi, they ordered 10k of the broadcom SoC, order arrived and they started prototyping.
<shapr>
They later discovered their order approval was a clerical error
<whitequark>
what
<shapr>
and they should never have been allowed any
<shapr>
turned into a bunch of drama and frustration, they ended up building their new hardphone on Android instead, and last I heard it's still not done.
<shapr>
I don't work there anymore, haven't kept up with the drama
<shapr>
previous to that, I thought the rpi was an open ecosystem
<Prf_Jakob>
I have my hopes that something rk3399 gets to replace it :)
<shapr>
I'd rather have risc-v, but even that's not as open as it could be
<Prf_Jakob>
And what GPU? :)
<Prf_Jakob>
But to be fair, yeah something RiscV based and some open GPU would be the most ideal.
<shapr>
what's that really cool plotter?
<shapr>
:-D
<Prf_Jakob>
Haha :p
pie_ has joined ##openfpga
pie__ has quit [Ping timeout: 250 seconds]
pie__ has joined ##openfpga
pie_ has quit [Ping timeout: 268 seconds]
pie___ has joined ##openfpga
<catplant>
didn't someone bake a basic gpu in verilog?
pie__ has quit [Ping timeout: 245 seconds]
<tnt>
catplant: define 'basic' ... I meann something that would output a piece of memory onto a screen is trivial.
<tnt>
But something that has accelerated operations and half decent performance for a modern composited desktop ... not so much.
<azonenberg_work>
shapr: i'm honestly amazed the pi even got approval to be made
<azonenberg_work>
broadcom probably figured they were high enough volume potential
<azonenberg_work>
But yeah for the most part broadcom just wont touch anyone who's not making incredibly high volume
<tnt>
catplant: https://github.com/asicguy/gplgpu this might be what you have in mind. I think it was a bit promoted a few years back on various social media.
<sorear>
hoping at least one of vboom, hwacha, sifive’s thing, or ara becomes viable soon
pie_ has joined ##openfpga
<tnt>
catplant: if you look at their demo video, it can render a 2500 triangles teapot at vga resolution at ~ 2 fps. Not exactly usable for anything else than research.
pie___ has quit [Ping timeout: 240 seconds]
<catplant>
ah
<catplant>
lol
<tnt>
some good 2D acceleration would probably be way more useful that a gpgpu architecture for day to day use.
<davidc__>
azonenberg_work: IIRC the pi was founded by a bunch of ex BCM guys
<davidc__>
so they had direct contacts into BCM
pie__ has joined ##openfpga
pie_ has quit [Ping timeout: 240 seconds]
<azonenberg_work>
davidc__: ah, i see
<azonenberg_work>
Because broadcom, marvell, vitesse, and qualcomm are currently all on my blacklist
<azonenberg_work>
of companies that i will not even attempt to do business with, especially for a high volume mass-production design
<azonenberg_work>
They don't want my design wins? Great, they won't get them
<tnt>
Problem is when those are the only ones with chips that do what you want :/
<whitequark>
davidc__: not exactly ex bcm
<whitequark>
it's pretty incestous
<sorear>
I’m not sure 2D acceleration is a thing anymore? Xaa is dead, it’s all GL
<whitequark>
afaik the first pi was basically a way for bcm to get rid of stock of shitty SoC no one wanted
<davidc__>
Vitesse used to be easier to work with :(
<davidc__>
though, they're now microsemi, so who knows
<azonenberg_work>
davidc__: i had problems getting a 1g phy before microsemi bought them
<azonenberg_work>
dont know if its any better now
<davidc__>
azonenberg_work: my experience was ~9y ago
<azonenberg_work>
the chip was on digikey iirc but i couldnt get a datasheet anywhere
<davidc__>
or maybe a bit more
<azonenberg_work>
this was 2-3
pie__ has quit [Ping timeout: 240 seconds]
<tnt>
sorear: I think even if the 2d acceleration is often implemented through GL that you can still provide accelerated 2d ops in your driver in case you don't have gl. I'm definitely no expert though so I might be completely off.
<shapr>
I much prefer the beaglebone black over the raspberry pi
<shapr>
also, jason kridner is a good guy :-)
<shapr>
I know the BeagleBone came from having a bunch of contacts inside TI, otherwise wouldn't have happened.
<shapr>
funny how having a bunch of contacts makes life easier
<tnt>
"How to generate custom LED pattern in Trident3" KB0027683 ... wow the broadcom KB has such useful articles :p
pie_ has joined ##openfpga
<whitequark>
cc rqou
<rqou>
lol
<rqou>
yeah, broadcom has a custom microcontroller just to do LEDs
<TD-Linux>
at least on linux, 2d accelerated ops are still exposed on X, but less and less useful
<TD-Linux>
much of the problem simply being that it's not any faster than just using the cpu
<tnt>
TD-Linux: I guess it depends of the cpu :)
<TD-Linux>
also there are less ops to accelerate now. I think the only ones that still get used are blits
<Prf_Jakob>
The problem with accelerating 2d is that doing small things like rendering text is faster on the CPU, but doing something "simple" as filling the screen or blitting large surfaces is much faster on the gpu.
<TD-Linux>
gdi style line drawing doesn't have antialiasing for example so modern guis don't use it
<tnt>
right, I was mostly thinking things like 2d blit with alpha or fill, that kind of stuff.
<whitequark>
Prf_Jakob: there are really fast gpu based text renderers now
<Prf_Jakob>
Problem two, switcing between say rendering text and blitting, is not as free on a GPU as a CPU. And every switch costs.
<TD-Linux>
note that this very much requires a "real" gpu
<TD-Linux>
fwiw 2d accelerators have never rendered text
<TD-Linux>
they just blit characters
<Prf_Jakob>
Yeah, but thinking here about 3D GPUs doing text rendering.
<TD-Linux>
oh okay. well pathfinder works
<TD-Linux>
see the benchmarks
<whitequark>
we don't really have 3D GPUs anymore
<TD-Linux>
webrender, a similar concept is already in firefox nightly
<whitequark>
they happen to do 3d well
<TD-Linux>
(but for css objects and not fonts)
<Prf_Jakob>
Hehe true that
<whitequark>
but the architecture doesn't inherently favor 3d
<TD-Linux>
there are still some 3d biases. most notably the texture sampler and early z
<TD-Linux>
and fixed function blender on desktop
<whitequark>
texture sampler is 3d bias?
<whitequark>
early z, doesn't this benefit anything with z structure? like webrender?
<TD-Linux>
I mean it's useful 2d too
<TD-Linux>
but some texture sampler features like aniso are kinda 3d specific and also only useful for 2d/3d not general purpose compute
<TD-Linux>
but yeah I thought you meant vs general compute. basically all the 3d features are useful for 2d oto
<Prf_Jakob>
I would say the derivative HW is pretty 3D biased. Along with anisotropic filtering.
<whitequark>
point
<Prf_Jakob>
But if you look at a chip layout of say a AMD GPU it's like 60% cumpute, 20%, 9% IO, 1% 3D. (Numbers taken out of my ass, but to make a point).
<Prf_Jakob>
30% Cache.
azonenberg_work has quit [Ping timeout: 240 seconds]
azonenberg_work has joined ##openfpga
<sorear>
rasterization is weirdly shaped for vector units, I'm not sure exactly how it works
<sorear>
and tiling is a thing too
<sorear>
been too long since I read relevant documentation
<sorear>
certainly if you just have a nice vector unit and ALUs with good support for operations that are common in vertex/fragment shaders you'll get a long way
pie__ has joined ##openfpga
pie_ has quit [Ping timeout: 240 seconds]
<sorear>
"fossi AV1 en/decoder" is somewhere on my TOOMANYPROJECTIDEAS list
<Prf_Jakob>
sorear: Oh you have one of those as well :p
<TD-Linux>
sorear, you mean an fpga impl?
<sorear>
ideally usable for either
<sorear>
really it's a big scope learning exercise in codec tech
<TD-Linux>
I'm around to help obvs :)
<sorear>
what's your relevant knowledge/experience