**** BEGIN LOGGING AT Sat Mar 19 02:59:58 2016 Mar 19 04:35:04 Hey nerdboy, there? Mar 19 04:35:21 mostly Mar 19 04:38:32 I was a bit confused about the lcm - MAVLink based communicaton.. Mar 19 04:38:33 in our system, we could use the lcm library to communicate between the pru and the sgx ( passing images ) and then between sgx and the main memory ( passing the depth map) Mar 19 04:38:41 is that right ? Mar 19 04:39:34 and mavlink would only have to be used for communication from the ground station to your drone, right ? Mar 19 04:40:49 that was the idea Mar 19 04:41:03 mavlink would do all of that Mar 19 04:41:39 you'd need to connect the camera to the pru and trigger the message when data is ready Mar 19 04:42:52 okay..and then the data is directly sent to the sgx, correct ? Mar 19 04:43:24 the mavconn/mavlink framework would take care of passing data after that Mar 19 04:44:19 you'd need to find/write/modify opencv stuff for processing Mar 19 04:44:42 ohh okay okay Mar 19 04:45:11 pru interface should hopefully be usable outside of mavlink/ros Mar 19 04:46:17 yeahh Mar 19 04:46:18 but the main goal would be using pru as capture interface and optimizing performance Mar 19 04:46:40 and from the sgx, the depth map that is obtained is sent to the main memory, right? Mar 19 04:46:54 again, mavlink/mavconn will take care of that Mar 19 04:46:55 did you see the stuff on rpmsg sample code in the kernel? Mar 19 04:47:40 nope. Mar 19 04:47:47 the virtio module interface Mar 19 04:47:59 post on the mailing list Mar 19 04:48:14 but i have looked at it for another application before, and i have an idea of what it is Mar 19 04:48:35 we wouldnt be using that here, right ? Mar 19 04:49:08 virtio would be the new/current/fast way Mar 19 04:50:21 that would be for the communication between pru and sgx? or for obtaining data into the pru's? Mar 19 04:51:53 the data i/o Mar 19 04:52:26 the framework will move it around once it's available Mar 19 04:53:18 okayy..perfect ! i have an idea..but i will go through the link in the ml now to clearly understand it! Mar 19 04:53:32 do the labs Mar 19 04:53:40 yepp..okay! Mar 19 04:53:41 lab 5 and the last one Mar 19 04:53:43 i think Mar 19 04:54:06 that should give you an idea of the pru interface Mar 19 04:54:25 okay.. will do :) Mar 19 04:54:42 another thing..our entire system has to be integrated with ardupilot, right ? Mar 19 04:54:57 or will that be too much to complete within the timeframe? Mar 19 04:55:04 it can all be built together Mar 19 04:55:27 mavlink/mavconn/arducopter/etc Mar 19 04:55:38 okayy.. Mar 19 04:55:48 ardupilot already works with bbio cape Mar 19 04:56:05 it would need to work with blue sensors/imu Mar 19 04:56:47 this might be weird, but might need to fork/contribute to another project Mar 19 04:57:04 the blue project? Mar 19 04:57:34 Hmmm Mar 19 04:58:05 we haven't confirmed current bmp driver works, although it should... Mar 19 04:58:30 which BMP driver? Mar 19 04:58:32 okayy Mar 19 05:00:44 the 6050 one? Mar 19 05:03:09 hmm? Mar 19 05:03:14 BMP6050? who makes that? Mar 19 05:03:44 inv Mar 19 05:04:06 not confusing it with the MPU6050? Mar 19 05:04:12 what kind of sensor is that? Mar 19 05:04:34 * nerdboy looking at too many sensor drivers Mar 19 05:04:41 mpu6050 Mar 19 05:05:08 ah Mar 19 05:05:21 that should work Mar 19 05:05:28 is it supposed to support the mpu9whatever too? Mar 19 05:07:04 the vendor support pages aren't so clear Mar 19 05:07:23 *for the driver package with funky license Mar 19 05:07:44 unless you need specific features, there is no reason to use that Mar 19 05:08:14 hmm Mar 19 05:21:49 kinda need a blue for testing soon... Mar 19 06:02:24 ds2: it's the mpu9250? same as erlebrain box? Mar 19 06:27:29 <_ZeekHuge> hey nerdboy ! do you think the data from PRU-ICSS could be read at the rate of 3 MBps over the L4 interconnect ? Mar 19 06:41:59 no idea, that's something i'd need to analyze Mar 19 06:44:30 ohk ... Mar 19 06:49:16 ohk so nerdboy, what i am trying to figure out it ... Mar 19 06:49:21 virtio is probably about as fast it gets interface-wise Mar 19 06:49:46 bandwidth calculations left as an exercise for the reader... Mar 19 06:50:17 what we basically do to get data from PRU is ... the device driver reserves memory in the DDR and one of the two PRUs write to that location .. Mar 19 06:50:47 this is done using the OCP master port over the L3 interconnect Mar 19 06:51:29 and that allows for a theoretical speed of 200MBps Mar 19 06:52:23 but if somehow, we are able to achieve the required data transfer speed over the L4 interconnect through the OCP slave port .. Mar 19 06:52:49 that would leave one PRU for us to use Mar 19 06:53:01 what do you think will be better in over all terms ? Mar 19 06:53:45 doesn't remote proc abstract that stuff a bit? Mar 19 06:54:27 as in can't you address each pru individually? Mar 19 06:54:41 yes it does the things in that way. But the communication frameworks use the PRU to communicate .. like the rpmsg Mar 19 06:55:04 address each PRU individually through remoteproc ? Mar 19 15:27:11 hello Mar 19 15:29:11 Hello! Mar 19 15:34:34 Hello all Mar 19 15:35:03 What is the process for applying for gsoc? Mar 19 16:03:29 Could any of the mentors please help me with TI's Code Composer Studio.Whenever I try to download the PRU C compiler,it says the program should have rights to write to the Eclipse folder. Mar 19 16:04:07 However I am installed and am running it as a root user. Mar 19 16:04:21 *installed it Mar 19 16:05:34 Can the same task be not done by Eclipse(as for many other ARM based microcontrollers).Is there support for the PRU C compiler in Eclipse? Mar 19 16:06:54 Any alternate way of writing and compiling code for PRU? Mar 19 16:30:40 hey chanakya_vc ! you can use TI code generation tools through terminal directly on the BBB, no need to install code composer studio. Mar 19 16:31:18 *TI - PRU code generation tools Mar 19 16:55:57 Guest61503: read the wiki page and gsoc site yet? Mar 19 16:57:32 ZeekHuge, So it would compile natively on the BBB? Mar 19 16:58:33 ZeekHuge, Could you please give me any link for this? Mar 19 16:59:34 here it is chanakya_vc : http://software-dl.ti.com/codegen/non-esd/downloads/download.htm#PRU Mar 19 17:01:29 ZeekHuge, Thank you so much!How is your proposal coming? Mar 19 17:02:46 working on it ... The initial idea had a big flaw (was into the idea that didnt see the DMA options ). So had to choose a different one and therefore started late. Mar 19 17:03:03 *was so much into the idea Mar 19 17:22:54 you know you can install that stuff on beaglebone directly Mar 19 17:23:26 plus i would not run that stuff as root Mar 19 17:24:24 directly ? using apt ? Mar 19 17:35:32 how are the proposal coming along? Mar 19 17:35:50 how many projects have been submitted so far? Mar 19 17:36:25 getting ready with that .. still on the hardware side, have to write about the software goals now .. Mar 19 17:37:03 basically, I am still under the heading "What makes me think this can be achieved ?" Mar 19 17:37:49 because it can Mar 19 17:37:52 :D Mar 19 17:37:59 :) Mar 19 17:38:20 nerdboy, Talking about TI's CCS? It requires root privileges for installing drivers and writing to the eclipse folder(wherever that is ) in my opinion. Mar 19 17:39:23 m_w: What do you think about my application draft? Mar 19 17:40:44 can you share a link? Mar 19 17:41:15 https://docs.google.com/document/d/1QGxnjvGWkyppsN7H8Hrc7XFoltBzVDaxKx9nXnQ1fBY/edit?usp=sharing Mar 19 17:41:16 I still haven't been official invited as a mentor. Mar 19 17:42:00 so you are using mcspi in slave mode? Mar 19 17:43:06 yes, exactly Mar 19 17:43:14 that should be interesting Mar 19 17:44:14 would the SPI NOR flash be emulated in kernel or userspace Mar 19 17:44:17 ? Mar 19 17:45:30 in userspace, SPI NOR Flash is simple example Mar 19 17:46:06 so you have to define the API for the SPI slave driver somewhere along the line Mar 19 17:47:02 have you figured out how the driver will signal the userspace when I transaction comes in? Mar 19 17:47:36 do you think that the response latency will be good enough handle the flash protocol in userspace? Mar 19 17:48:53 yes, I am collecting information on how to do exactly Mar 19 17:50:24 I think it will be tricky timing wise Mar 19 17:51:25 i'm writing communication between LKM and application. Mar 19 17:52:18 it works, API sends char array Mar 19 17:52:37 sure it works Mar 19 17:53:02 but it time for the driver to respond? Mar 19 17:53:43 SPI transactions happen pretty fast Mar 19 17:54:32 I have to look like it is done in the SPI master Mar 19 17:54:34 a command comes in on the MOSI and the master expects a response immediately on the MISO Mar 19 17:54:51 same transaction Mar 19 17:56:43 if the command comes in, the driver says "hey userspace what is the response", the driver waits an indeterminate time or the userspace answer, meanwhile the SPI host keep clocking when result may not be in yet Mar 19 17:57:06 you will need to handle to protocol in realtime Mar 19 17:57:50 and linux userspace is not likely going to keep up Mar 19 17:58:20 the SPI driver really needs to know the answer before the request comes in Mar 19 17:58:27 yes, i know, but I wanted to check the status of a loop Mar 19 18:00:16 operation on files is very fast in linux Mar 19 18:01:07 how fast are you allowing the SCLK to go? Mar 19 18:01:47 up to 24Mhz Mar 19 18:01:55 i think think you will have time to go to a file between clock pulses even with slow rates Mar 19 18:02:34 think about it Mar 19 18:02:57 Is there a better way? Mar 19 18:03:42 you dont know the command until the last bit comes in and you are expect to provide results in the next clock cycle after that Mar 19 18:04:18 I would imagine the only way to handle it so quickly is with an atomic memory read Mar 19 18:05:11 the driver would have to know the answer before the question came in Mar 19 18:05:44 userspace allows it? Mar 19 18:06:32 they userspace can point the driver to the awnsers but not provide them quickly probably on the fly Mar 19 18:08:23 something has a similar problem in the kernel? Mar 19 18:08:40 hmmmm Mar 19 18:12:11 nothing is coming to mind Mar 19 18:13:55 I will search Mar 19 18:15:51 hey Zeekhuge,The file that I downloaded from the link that you gave me requires some version on ccs on my desktop.I would have to install it on my desktop or directly to bbb? Mar 19 18:17:07 http://www.ti.com/lit/ug/tidua38a/tidua38a.pdf this is interesting but probably unrelated Mar 19 18:19:23 ok thanks but the application draft is probably ok? Mar 19 18:21:59 yeah Mar 19 18:27:55 you may want to mention the challenges I have listed Mar 19 18:29:45 or think about come up with plans to address them Mar 19 18:32:43 ZeekHuge do you have a link to the latest version of your proposal Mar 19 18:32:44 ? Mar 19 18:33:02 I think about it I still have a few days Mar 19 18:33:19 here it is https://docs.google.com/document/d/1GfnoWusWUeH2VsLbt1qJaRi20uNxCC8O_Xw1CHoGY9A/edit Mar 19 18:33:56 chanakya_vc, download the ARM tagged file, install on BBB Mar 19 18:34:02 hello all mentors. i have finished writing my application. eagerly waiting for feedback from u guys Mar 19 18:34:03 https://docs.google.com/document/d/1syUlWQkTPHuA0m6owekqa2zO-G0uMyQ7i66Z5aMkltY/edit?usp=sharing Mar 19 18:39:16 cool Mar 19 18:39:16 ZeekHuge, Okay and this would allow me to code in C for the PRU Mar 19 18:39:41 ZeekHuge, ? Mar 19 18:53:11 chanakya_vc : yes Mar 19 18:54:22 can we use clocks from a different system like the UART to clock the ADC ? Mar 19 18:54:29 m_w ? Mar 19 18:56:55 not sure on that one Mar 19 20:10:32 * ZeekHuge realizes that there is lot much more to learn than he thought ..... Mar 19 21:56:08 here i have written my gsoc application on implementation of real time stereo vision on PRU using BBB as ardupilot platform Mar 19 21:56:11 https://docs.google.com/document/d/1syUlWQkTPHuA0m6owekqa2zO-G0uMyQ7i66Z5aMkltY/edit?usp=sharing Mar 19 21:56:32 i will request mentors to give their valueable feedback Mar 19 21:57:53 nerdboy Mar 19 21:58:31 thank you for your papers. that really helped initially to get on the right track Mar 19 21:58:53 you mean the meier references? Mar 19 21:59:03 yes Mar 19 22:00:09 sorry what is meier Mar 19 22:06:54 the drone guy at ETH? Mar 19 22:07:22 see ml or mavlink/pixhawk pages for references Mar 19 22:07:46 the draft looks okay, timeline might need a little adjustment Mar 19 22:08:25 i'd like to see a little more detail on processing pipeline Mar 19 22:08:41 performance estimates even Mar 19 22:12:31 sir im not clear what exactly u mean by processing pipeline ? Mar 19 22:13:12 is it about the PRU code? Mar 19 22:13:16 processing the image data Mar 19 22:14:17 from the camera to the command generation Mar 19 22:15:41 take an image resolution, chunk it up for processing, same thing for output Mar 19 22:15:51 what are the sizes/times? Mar 19 22:16:11 can you do a little math maybe? Mar 19 22:17:38 there's also previous work to get some performance data from Mar 19 22:17:42 i will update the doc by assigning an input resplution and the number of parts in which stereo images will be divided to give to shared memory for processing Mar 19 22:17:58 *resolution Mar 19 22:20:25 is it that in timeline you want performance comparison between opencv and using opencv with pru to be added and shown in documentation? Mar 19 22:21:08 i think there's at least one reference with some data Mar 19 22:22:31 more than one i think Mar 19 22:23:20 sorry to bother you. but it will be very helpful if you can comment down in google doc. So i can focus very carefully on each line u have commented? Mar 19 22:24:13 i dont understand by what u mean 'at least one reference with some data' Mar 19 22:26:33 data on vision processing performance Mar 19 22:27:23 can you cite previous work and make some target estimates for your "design"? Mar 19 22:28:04 add a data flow diagram with some size/timing data Mar 19 22:28:43 this has (mostly) been done before on same/similar hardware Mar 19 22:28:56 just not using the prus Mar 19 22:29:32 you should really discuss said previous work and show how yours is "better" Mar 19 22:30:03 *estimated performance using some math/assumptions is okay Mar 19 22:32:43 ok ill add th data flow diagram and some performance estimations too that will be based on some assumptions.. Mar 19 22:35:00 google beaglebone computer vision Mar 19 22:35:00 also we electronics student did not use gitHub like the comps studetsmuch. it is from these 2-3 months that we have started it. so it will take me some time to make repo ofall my projects i have doneand upload the codes i have prepared before Mar 19 22:35:50 you need to add stuff to gihub unless you need it for something or it's project-related Mar 19 22:35:55 *don't even Mar 19 22:38:03 like i think having a good github profile will prove me that i will be able to do this project? coz having uploaded just two of my previous project codes makes me feel embarrasing.. Mar 19 22:39:47 https://pixhawk.ethz.ch/overview/publications Mar 19 22:40:05 there are others you can find with data Mar 19 22:49:57 thanks for the links. so in all i need to make some performance estimation and compare it with the previously implemented drones to show how my idea is better Mar 19 22:50:21 also data flow diagram has to be added. Mar 19 22:50:55 is this all or iam missing something? Mar 19 22:52:37 sounds good Mar 19 22:55:08 thanks. Mar 19 22:55:33 remember it's part proposal Mar 19 22:57:21 also can i write a proposal for this project " Need a sonic anemometer for open source weather station (use PRUs to calculate sonic velocity and wind components). Needs ultrasonic ping sensors and mounting framework". it is also quite appaling to me. any suggestions? Mar 19 22:58:10 *appealing Mar 19 22:58:35 if you really want to do two... Mar 19 22:58:37 i didnt get u on part proposal thing? Mar 19 22:59:13 the gsoc application is like a proposal **** ENDING LOGGING AT Sun Mar 20 02:59:58 2016