Ah whoops, yep I really though I picked a different email to reply to.... thanks for the heads up, I can ask the right lot now!
Andrew -----Original Message----- From: Laurent Pinchart [mailto:laurent.pinch...@ideasonboard.com] Sent: Monday, 13 September 2010 8:52 AM To: linux-uvc-devel@lists.berlios.de Cc: Andrew Leech Subject: Re: [Linux-uvc-devel] Maximising iram throughput? Hi Andrew, On Friday 10 September 2010 08:14:30 Andrew Leech wrote: > Hi all, > > Just wondering if anyone's found some tricks to speed up iram data > throughput, specifically with dma transfers. Are you sure you've sent this mail to the right mailing list ? > My application is basically sending data (real time video) from the > external static ram interface over USB. The catch is the data buffer > sizes of the external data and the usb data need to be quite different > so I've had to set up a double buffering arrangement, where I'm using > the DMA to copy from external port to a large rolling buffer space in > iram, and then on a separate dma channel copying the (smaller) chunks > of data from the large buffer to my usb buffer/structs. The usb > packets needs a header on each one, so I can't just give the usb a > pointer to a location in the main buffer because it obviously won't have the header. > > The system is technically working, but I'm running very low on memory > bus bandwidth. The main issue is coming from latency in the external > interface, I currently have it running with a MPMCStaticWaitRd0 of 1 > and have a couple of errors in transmission. The errors are completely > gone if I up the WaitRd to 2 but then the dma from buffer to usb is not keeping up. > > The external ram interface is in 16 bit mode, and the write buffers > (in > MPMCStaticConfig) are disabled, I'm using interface strictly read only > and enabling them slowed it down more. > > Both dma channels are running in burst mode (16 byte chunks). > > Clocks are all at csp default settings again, so mpmc, dma, ebi all at > 90Mhz, core at 180Mhz. > > I've tried SYS_REGS->eshctrl_sup4 = 0; to enable high speed > performance mode on EBI, but that doesn't really seem to help much, > not that I expected it to really. > > I've also tried enabling instruction and data caches at start of main: > > // Set virtual address of MMU table > cp15_set_vmmu_addr((void *) ISROM_MMU_TTB_BASE); > cp15_set_mmu(1); > cp15_set_icache(1); > cp15_set_dcache(1); > > But I didn't notice any difference in application speed, can anyone > confirm whether this is the correct way to enable the buffers? > Although I guess these are buffers for the cpu, so wouldn't affect the dma anyway. > > Basically I'd really like to speed up the iram - iram dma transfer to > allow more bus bandwidth / time to devote to external ram - iram dma > channel so I can slow the mpmc WaitRd down a touch and still kepp up > with the video stream. -- Regards, Laurent Pinchart _______________________________________________ Linux-uvc-devel mailing list Linux-uvc-devel@lists.berlios.de https://lists.berlios.de/mailman/listinfo/linux-uvc-devel
