| 06:47 | topi`_ | jnettlet: I was contemplating building Servo on HB... but, does the Vivante drivers support OpenGL 2.0 or just GLES_ |
| 06:48 | topi`_ | I don't know for sure if Servo can also use GLES as backend on linux targets... definitely on android, though |
| 06:51 | jnettlet | topi`_, really just GLES. There is an libGL.so but the GL support it provides is limited. If you grab one of our Debian X11 images you can easily try it |
| 06:55 | topi`_ | ok It seems to use just GLES with Servo: https://github.com/servo/servo/wiki/Building-on-ARM-desktop-Linux |
| 06:56 | topi`_ | jnettlet: what kind of binary blobs does hardware 3D require on Vivante? |
| 06:56 | topi`_ | or is it the open source etnaviv we're talking about? |
| 06:56 | jnettlet | topi`_, you are going to have the best success with Vivante's 3d library for now. |
| 06:57 | topi`_ | so etnaviv is missing some features? |
| 06:57 | jnettlet | GLES makes sense since WebGL is based on it |
| 06:57 | topi`_ | I'm fairly OK with it as long as you guys supply the binary blobs and make sure the kernel won't get tainted ;) |
| 06:58 | jnettlet | unfortunately yes, etnaviv is missing lots of compatibility |
| 06:58 | jnettlet | topi`_, the kernel part of Vivante's drivers is GPL. It is only the userspace that is binaries |
| 06:59 | topi`_ | yeah, so business as usual :) |
| 06:59 | topi`_ | anyhow, I'll just download a recent imx6q image |
| 06:59 | topi`_ | from sr-build.xyz |
| 07:03 | topi`_ | btw does GLES require running on a Dual/Quad SoC or does it also work on the Single? |
| 07:04 | topi`_ | right now I only have two HB's at home... the Gate has only the imx6s but the original HB had a Quad SoC with 2GB ram |
| 07:04 | topi`_ | s/had/has/ |
| 07:05 | topi`_ | but the hardware we shipped to the customer (where I'll try deploying the Servo) is HB Edge with the Dual |
| 07:06 | jnettlet | it works on all of them. The GPU on the iMX6DQ is more powerful |
| 07:06 | topi`_ | OK, so they are compatible |
| 07:07 | topi`_ | how does GC2000 compare to, say, a single-core Mali-400 ? |
| 07:10 | topi`_ | hmm, there is a GPU comparison page on cnx-software |
| 07:11 | jnettlet | topi`_, don't believe it...it is mostly lies :) |
| 07:11 | jnettlet | the only thing that Vivante is really good at is making a small relatively power efficient GPU |
| 07:12 | topi`_ | there is a comparison table between diffrerent Vivante models. it seems GC800 is single-core and GC2000 is quad core |
| 07:13 | topi`_ | GC800 has 6 GFLOPS, GC2000 has 24 GFLOPS |
| 07:13 | jnettlet | well those are the number of shaders. gc800 is single pipeline and gc2k is dual pipeline |
| 07:14 | jnettlet | but really the limiting factor is the iMX6 memory bus limitations |
| 07:15 | topi`_ | ...as always :) |
| 07:16 | topi`_ | I wonder what kind of internal interconnect the i.MX6 has |
| 07:16 | topi`_ | I haven't heard of other SoCs where gbit ethernet is limited to half speed |
| 07:17 | topi`_ | I guess Marvell's chips are much better in that regard :) |
| 08:04 | jnettlet | well a lot of it is performance vs power efficiency |
| 08:06 | jnettlet | the i.MX6 mmdc screw up was a design mistake. A device that wanted to be bus-master on the AXI bus was designated so on every core. So bandwidth was reserved for every busmaster device on ever core whether that device was talking to that core or not. |
| 08:07 | jnettlet | this put an artificial limitation on the overall memory bus throughput |
| 08:07 | jnettlet | this chip was out for 2 years before the community figured out what was going on. |