**** BEGIN LOGGING AT Sat May 24 02:59:58 2014
May 24 03:36:52 <Abhishek_> jkridner: I had ccd you.
May 24 03:37:06 <jkridner> yeah, I saw that....
May 24 03:37:10 <jkridner> haven't read the code yet
May 24 03:37:42 <Abhishek_> k
May 24 03:48:05 <Abhishek_> panto: Do you have any hints on removing data fast enough, or some alternative approach? It's absolutely no guarantee that buffers allotted with malloc would be contiguous chunks of mem, right?
May 24 06:06:29 <panto> Abhishek_, I am sleeping you know :)
May 24 06:06:50 <panto> Abhishek_, cool
May 24 06:07:10 <panto> Abhishek_, that's where vrings go into effect
May 24 06:07:22 <panto> you can do scatter gather which is required for transfer to/from user-space
May 24 06:07:39 <panto> Abhishek_, send me a link to the sources
May 24 06:07:42 <panto> I'll take a look Monday :)
May 24 06:07:46 <Abhishek_> Already :)
May 24 06:08:30 <Abhishek_> the sources are not too large,
May 24 06:54:42 <Abhishek_> panto: Your counter-delta approach worked so well, I even put it into the PRU0 firmware to check for stalls and underruns in transfer [as an extra register pushed out via XOUT for free with the data]
May 24 09:01:37 <Abhishek_> mranostay: did you see the subsequent mail I sent you all yesterday?
May 24 12:50:40 * Abhishek_ is still looking for a fast memcpy
May 24 13:03:10 <_av500_> Linaro has one
May 24 13:03:26 <_av500_> it makes Android twice as fast
May 24 13:03:30 <Abhishek_> link?
May 24 13:03:41 <Abhishek_> I think I am running linaro currently
May 24 13:04:17 <Abhishek_> but the speed I got was 281ms for 4 MB :/
May 24 13:04:45 <_av500_> https://www.youtube.com/watch?v=mrQRYmYip6Q
May 24 13:08:47 <Abhishek_> _av500_: Is this toolchain okay? https://gist.github.com/abhishek-kakkar/cb7417fe9fcfa41baa38
May 24 13:15:41 <_av500_> should be fine
May 24 13:15:51 <_av500_> difference between toolchains are not huge
May 24 13:16:23 <Abhishek_> _av500_: my requirement is mostly to move 4 MB of data in less than 20ms
May 24 13:20:33 <_av500_> thats 200MB/s
May 24 13:20:46 <_av500_> move between where and where?
May 24 13:21:00 <Abhishek_> from DDR to DDR
May 24 13:22:04 <Abhishek_> the PRU fw writes into two buffers of 4 MB alternately and I need to remove data fast before the next chunk comes in
May 24 13:23:36 <_av500_> ah
May 24 13:23:43 <_av500_> https://groups.google.com/forum/#!topic/beagleboard/Ese9QV-8CZM
May 24 13:24:36 <_av500_> see there for some speeds achieved on omap3
May 24 13:24:45 <_av500_> so on the ARM side your goal seems reachable
May 24 13:25:01 <_av500_> hmm, the PRU also writes into SDRAM, no?
May 24 13:25:02 <Abhishek_> _av500_: I don't think it's possible to pass a pointer to physical memory directly to the PRU, does it?
May 24 13:25:11 <Abhishek_> _av500_: yes
May 24 13:25:19 <_av500_> so its SDRAM to SDRAM copy
May 24 13:25:58 <Abhishek_> I tried using PRU Shared RAM -> RAM copy but data comes in way too fast for ARM to keep up. Interrupts fire at 61us
May 24 13:26:09 <_av500_> right
May 24 13:26:51 <_av500_> do you have a testcase?
May 24 13:27:00 <_av500_> something simple on the ARM side to benchmark
May 24 13:27:32 <Abhishek_> _av500_: Does this qualify? https://gist.github.com/abhishek-kakkar/b8460278f2d86d5e71ec
May 24 13:27:41 <Abhishek_> but it doesn't list the times yet
May 24 13:27:51 <Abhishek_> lemme add a few lines of code.
May 24 13:33:48 <_av500_> Abhishek_: fast memcpy on cortex A8 is a much discussed topic
May 24 13:34:01 <_av500_> many posts with benchamrks to compare against
May 24 13:34:52 <Abhishek_> is PLE available on the AM335x?
May 24 13:56:57 <_av500_> dont know
May 24 13:57:04 <_av500_> probably
May 24 13:57:19 <_av500_> but even non-ple should be fast enough
May 24 13:57:35 <_av500_> maybe try some of Mans' code just to see if you are in the same ballpark
May 24 15:15:14 <karki> what other ways are there of creating a "/dev/<char-driver>" entry other than explicitly using mknod?
May 24 15:36:13 <Abhishek_> are VFP and neon mutually exclusive?
May 24 16:47:54 <Abhishek_> _av500_: Do I need to add -mfpu=neon to my C application while using his assembly code?
May 24 17:48:30 <Abhishek_> _av500_: I get a segfault with that
May 24 17:48:33 <Abhishek_> any ideas?
May 24 17:56:16 <vvu> praveendath92: ping
May 24 18:09:51 <Abhishek_> panto: Could the slow memcpy speeds be due to the L3 bus being saturated with write bursts from the PRU and congestion?
May 24 18:35:20 <praveendath92> vvu: pong
May 24 18:36:24 <vvu> i've seen u pushed some endpoint code to the repo
May 24 18:36:30 <vvu> does it output something?
May 24 18:36:35 <praveendath92> No no.
May 24 18:36:41 <praveendath92> It's not a working version.
May 24 18:36:52 <praveendath92> I was working with the simplectrl.c
May 24 18:37:31 <praveendath92> I couldn't get the Android device to ADK mode.
May 24 18:38:31 <vvu> simplectrl.c ?
May 24 18:39:01 <praveendath92> https://gist.github.com/praveendath92/9641145
May 24 18:39:31 <vvu> ah i remmeber this
May 24 18:39:38 <praveendath92> Yeah
May 24 18:40:13 <praveendath92> Wait I will show you my modifications.
May 24 18:42:41 <praveendath92> I updated the code in the same gist
May 24 18:44:49 <vvu> where does it break ?
May 24 18:45:04 <praveendath92> line 192
May 24 18:45:45 <praveendath92> I think it is not able to find a device the vid and accessory_pid
May 24 18:46:08 <vvu> yes, do a search again
May 24 18:46:13 <vvu> or do dmesg
May 24 18:46:13 <praveendath92> When I replace accessory_pid with pid (actual pid of my device).
May 24 18:46:29 <vvu> to see if a change of vid is being done
May 24 18:46:47 <praveendath92> I get a handle. Which I guess indicates that the device hasn't gone to accessory mode.
May 24 18:47:07 <praveendath92> Kernel log doesn't show any change.
May 24 18:47:38 <vvu> so the ctrl transfers are not working
May 24 18:48:39 <praveendath92> I guess so but I don't get any errors for them.
May 24 18:51:07 <praveendath92> Also I get a message on my device to open an app for that accessory.
May 24 19:12:28 <vvu> so it should switch at that point as i remmeber
May 24 19:12:42 <vvu> after monday i will have time to look exactly how it works, still 2 more exams pending
May 24 19:14:44 <praveendath92> vvu: It's fine.
May 24 19:15:29 <praveendath92> I'm working on writing the module to send commands.
May 24 19:16:04 <praveendath92> Once I have a structure this can go in I suppose.
May 24 22:51:33 * mranostay wonders why the compiler never uses SBCO now
May 25 02:12:28 <Abhishek_> Morning
May 25 02:31:00 <vvu> morning Abhishek_
May 25 02:50:48 <Abhishek_> mranostay: did you see the code?
**** ENDING LOGGING AT Sun May 25 02:59:58 2014