06:43 <jduchniewicz> tacokoneko what are you doing? I had some struggles with working with the newest Imagination Debian image but got 32-bit headless rendering working

08:27 <iobus> hello, what the diff bt: init-eMMC-flasher-v3-no-eeprom.sh and init-eMMC-flasher-v3.sh

10:17 <iobus> oh no, is BB-AI+TIDL configured by default to use DSP in BLAS and other?!

11:57 <iobus> how I can make BLAS use TIDL acceleration?

13:09 <zmatt> I don't know a ton about tidl, but afaik it's a high-level framework for executing neural network inference, it sounds to me like you're more looking for a lower-level mechanism like opencl

13:12 <jduchniewicz> TIDL is not very open for lower-level operations

13:12 <jduchniewicz> wanted to deploy YOLO with it but faced several problems

13:17 <zmatt> annoying

13:17 <iobus> I can find how to make lib BLAS or LAPACK use it, the TI website mention that by using something called LINAGL, but the link is broken :( https://processors.wiki.ti.com/index.php/Processor_SDK_Linear_Algebra_Library

13:18 <iobus> s/can/can not

13:18 <iobus> I saw this too: https://git.ti.com/gitweb?p=dense-linear-algebra-libraries/linalg.git;a=tree

13:18 <zmatt> https://www.ti.com/microcontrollers-mcus-processors/processors/digital-signal-processors/libraries/linear-algebra.html

13:19 <iobus> but I'm having bad time to understand it's compiling, also weather or not it's included by default in the distro

13:19 <zmatt> well this doesn't have anything to do with TIDL

13:19 <iobus> @zmatt exactly, the " LINALG User’s Guide." is dead link

13:20 <zmatt> yeah it's really annoying TI deprecated the wiki, especially given how there are links to that wiki literally on their site

13:21 <zmatt> you can find archived versions though: https://web.archive.org/web/20191107184128/http://processors.wiki.ti.com/index.php/Processor_SDK_Linear_Algebra_Library

13:22 <iobus> wow, i will read read

13:23 <zmatt> the documentation seems heavily oriented around TI's processor SDKs... I'm sure they can be made to work on the beagleboard.org debian images too but instructions may require adaptation

13:24 <zmatt> and my guess would be it's incompatible with tidl

13:24 <zmatt> kind of a shame there's no EVE equivalent... the BBAI has four EVEs, and those are basically bigass linear algebra engines

13:25 <iobus> my end goal is just to make "Dlib" ML framework used BLAS that use the DSP/hardware acc

13:26 <iobus> one frame is taking 1600ms while in 1.6GHz i5 4th gen 30ms

13:27 <iobus> and I can't even find a way to convert Dlib models to tensorflow or cafee, ...

13:27 <zmatt> I can't you with that, I have zero experience with machine learning

13:29 <iobus> no problem, many thanks

13:34 <iobus> @jduchniewicz was u able to make BLAS or LAPACK use hardware acceleration?

13:38 <zmatt> now I kinda want to try getting the EVEs working using UIO hehe

13:54 <iobus> @rcn-ee hi, are you https://github.com/rcn-ee ?

13:55 <iobus> I'm trying to find a package or a way to make BLAS or LAPACK use hardware acceleration in BB-AI, if you have an information regarding that. (for Dlib)

14:00 <rcn-ee> iobus, the BBAI does have an OpenCL layer on the DSP, not sure if it'll speed up anything..

14:04 <rcn-ee> iobus, here are TI examples: https://git.ti.com/gitweb?p=opencl/ti-opencl.git;a=tree;f=examples;h=fa6cec9f4486037e3b025e9c362ae90f3ac6a6fe;hb=b1dfed9af21a1236380b9d643c432e4f0c619fce

14:08 <iobus> yep it's installed by default in /usr/share/ti/opencl

14:08 <rcn-ee> correct, figured the git version is easier to search...

14:09 <rcn-ee> if you do some google fu, there are lapack/opencl things..

15:23 <Guest757> anyone know where to get the environmenal parameters for the BeagleBone Black boards? I found the operating temp for the CPU, but what about any vibration testing performed? Is there an ISO or MIL specification that anyone has tested to?