Lpc2000

At 02:29 PM 11/27/2005, Bertrik Sikken wrote:
>Your were right about the optimisations and the throughput
>limitations.
>
>First of all, I realised thath the results above were obtained
>with code compiled for thumb, so I changed that to arm.
>
>Also I added a dummy write test, which basically just writes
>4 million bytes of value 0xFF (a NOP command for the card).
>The theoretical transfer time for sending it over a 7.5 MHz
>SPI link is just 4.5 sec. However, this took 11 seconds.
>It turns out that the code I was using, calls a subroutine for
>every byte transfered. This subroutine selects the SPI slave
>and transfers just one byte over SPI, then deselects.
>I changed this to do the SPI slave (de)select only once for each
>block and then transfer a block without calling any subroutines.
>
>Finally, I tried a different memory card: an "A-DATA SD 60x 256".
>This card claims 7 MB/s write speed. In my card reader, I got
>3.3 MB/sec.
>
>My results are now (seconds per 4 MB transfered):
>                 A-Data   Kingston
>dummy             6         6
>read              8         8
>write single      9        10
>write multi       7         8

Well, you're up to 680KB/sec for the A-Data dummy mode, whereas your 
computer gets 825KB/sec.  I'm assuming that the 3.3MB/sec is operating in 
SD 4-bit mode, so assuming scaling down to 1-bit mode is 1/4 the speed.  If 
the theoretical max is 4.5 sec and you're getting 6, that's not too 
bad.  If it was heavily assembler optimized you could probably squeeze out 
20-30% more speed to be at the max, but really those speeds aren't bad.


>The dummy write time is still a bit above the theoretical maximum,
>but it is a large improvement over the original 11 seconds.
>
>My conclusions so far:
>* multiple block write is indeed faster than single block write
>* but not by much, because most of the time is spent in SPI0 transfer
>* SPI0 transfer is a bit disappointing because of the 7.5 MHz max speed.

Have you considered trying to use the SSP bus instead of SPI?  I think the 
SSP bus can operate much faster than PCLK/8, no?

-- Sean