Re: double-buffering with the omap3 parallel interface
Hi Michael, On Monday 17 June 2013 02:44:50 Sakari Ailus wrote: On Wed, Jun 12, 2013 at 06:16:26PM +0200, Michael Jones wrote: Hi Laurent co., I'd like to look at what the maximum possible frame rates are for a sensor connected to the OMAP3 ISP CCDC via the parallel interface, writing frames directly to memory. I understand that there is some minimum amount of time required between frames to pass on the finished frame and set up the address to be written to for the next frame. From the manual it looks like a double buffering scheme would've been available on a different sensor interface, but isn't on the parallel one. Do I see that right? Is it impossible to use double buffering of any sort while using the parallel interface to memory? That's correct. The CCDC has a single destination memory address register, so you can only queue a single buffer to the hardware. I'm still using an older version of the driver, but I've browsed the current state of the code, too. What behavior do you expect if the time between frames is too short for the buffer management? Can it be recovered from? The CCDC is stopped after each frame, reconfigured, and then restarted. If a new frame arrives before the CCDC is restarted the frame will be dropped, but the CCDC might lose synchronization as well (this would need to be verified, the OMAP3 ISP hardware is pretty sensitive to such issues, but I don't remember from the top of my head the details of what synchronization problems affect each block). Has this behavior changed in recent versions? The overall behaviour shouldn't have changed, no. I see from the ISP block diagram that the circular buffer is between the SBL and the MMU. Could this maybe be used to help the situation? It seems to currently not be used at all along this path. Not that I know of. The circular buffer allows allocating less physical memory than what would be required to store a full frame, if the consumer can process image data as it gets written to memory. The memory buffer then essentially acts as a FIFO. The way the hardware is controlled has stayed the same for a very long time. My recollection matches with your findings --- even if double buffering of the buffer pointers would be available in some situations, it isn't used by the driver. You might ask why, and the reason for that is that there are tonds of other things that typically need to be configured (as a result of the configuration given by the user using the private IOCTLs) at that very time. If a block becomes busy while you're configuring it you can say good bye to your frame in any case; whether yousd set up writing it to system memory using DMA or not... Several CCDC registers are latched by the VS sync pulse. It might thus be possible to reconfigure the CCDC right after frame start instead of right after frame end. What comes to the minimum time per frames, I could give you a guesstimate of 1 ms. It depends a lot on how well other drivers in the system behave, but in general that should be enough. Something must be very wrong if you need significantly more than that. -- Regards, Laurent Pinchart -- To unsubscribe from this list: send the line unsubscribe linux-media in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: double-buffering with the omap3 parallel interface
Hi Michael, On Wed, Jun 12, 2013 at 06:16:26PM +0200, Michael Jones wrote: Hi Laurent co., I'd like to look at what the maximum possible frame rates are for a sensor connected to the OMAP3 ISP CCDC via the parallel interface, writing frames directly to memory. I understand that there is some minimum amount of time required between frames to pass on the finished frame and set up the address to be written to for the next frame. From the manual it looks like a double buffering scheme would've been available on a different sensor interface, but isn't on the parallel one. Do I see that right? Is it impossible to use double buffering of any sort while using the parallel interface to memory? I'm still using an older version of the driver, but I've browsed the current state of the code, too. What behavior do you expect if the time between frames is too short for the buffer management? Can it be recovered from? Has this behavior changed in recent versions? I see from the ISP block diagram that the circular buffer is between the SBL and the MMU. Could this maybe be used to help the situation? It seems to currently not be used at all along this path. The way the hardware is controlled has stayed the same for a very long time. My recollection matches with your findings --- even if double buffering of the buffer pointers would be available in some situations, it isn't used by the driver. You might ask why, and the reason for that is that there are tonds of other things that typically need to be configured (as a result of the configuration given by the user using the private IOCTLs) at that very time. If a block becomes busy while you're configuring it you can say good bye to your frame in any case; whether yousd set up writing it to system memory using DMA or not... What comes to the minimum time per frames, I could give you a guesstimate of 1 ms. It depends a lot on how well other drivers in the system behave, but in general that should be enough. Something must be very wrong if you need significantly more than that. -- Kind regards, Sakari Ailus e-mail: sakari.ai...@iki.fi XMPP: sai...@retiisi.org.uk -- To unsubscribe from this list: send the line unsubscribe linux-media in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
double-buffering with the omap3 parallel interface
Hi Laurent co., I'd like to look at what the maximum possible frame rates are for a sensor connected to the OMAP3 ISP CCDC via the parallel interface, writing frames directly to memory. I understand that there is some minimum amount of time required between frames to pass on the finished frame and set up the address to be written to for the next frame. From the manual it looks like a double buffering scheme would've been available on a different sensor interface, but isn't on the parallel one. Do I see that right? Is it impossible to use double buffering of any sort while using the parallel interface to memory? I'm still using an older version of the driver, but I've browsed the current state of the code, too. What behavior do you expect if the time between frames is too short for the buffer management? Can it be recovered from? Has this behavior changed in recent versions? I see from the ISP block diagram that the circular buffer is between the SBL and the MMU. Could this maybe be used to help the situation? It seems to currently not be used at all along this path. thanks, Michael MATRIX VISION GmbH, Talstrasse 16, DE-71570 Oppenweiler Registergericht: Amtsgericht Stuttgart, HRB 271090 Geschaeftsfuehrer: Gerhard Thullner, Werner Armingeon, Uwe Furtner, Erhard Meier -- To unsubscribe from this list: send the line unsubscribe linux-media in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
