|
Hello friends: I left this one out on my previous post. I am curious to know whether or not the EDMA mechanism changed from C62x to the C64X processors. By reading SPRU190 (Peripherals Reference Guide), it seems it did not. The EDMA wait times on C64x for the exact same design I had on C62x are annoyingly longer (in terms of clock-cycles and percent of total time) than those on the C62x. My question is, do I necessarily have to do more processing than on C62X to reduce the EDMA wait times ? I mean, real processing, not just increasing the amount of data transferred/processed. Looking forward to your comments and ideas. Thanks to Mike and Indrajit for sharing their views on optimization. Regards ka |
|
|
EDMA with c64x
Started by ●September 3, 2003
Reply by ●September 4, 20032003-09-04
|
ka,
The 64x EDMA does appear to be an extension of the 621x/671x EDMA.
There does seem to be some functions relating to cache and EDMA that are longer
[in clock cycles] - but overall I like the performance of the c64.
When you say 'EDMA wait times', I am not sure of the
meaning. Are you referring to latency?? or setup time??
mikedunn
Anand K <a...@yahoo.com> wrote:
|
|
|
Reply by ●September 4, 20032003-09-04
|
Mike, when I say "EDMA wait time" - I mean the idle time, when the CPU sits doing nothing waiting for an EDMA transfer to complete. For e.g. in a ping-pong buffering scheme, CPU has processed ping buffer and waiting for pong buffer to be transferred into the Internal SRAM. (Apologize for the ambiguity, is there a more tangible term for this ?) I am also wondering if the extra priority level on the C64x would help mitigate the long wait times, havent experimented with that yet. ka --- In , Mike Dunn <mike-dunn@s...> wrote: > ka, > > The 64x EDMA does appear to be an extension of the 621x/671x EDMA. There does seem to be some functions relating to cache and EDMA that are longer [in clock cycles] - but overall I like the performance of the c64. > > When you say 'EDMA wait times', I am not sure of the meaning. Are you referring to latency?? or setup time?? > > mikedunn > > Anand K <akalya@y...> wrote: > Hello friends: > > I left this one out on my previous post. I am curious to know whether > or not the EDMA mechanism changed from C62x to the C64X processors. > By reading SPRU190 (Peripherals Reference Guide), it seems it did > not. The EDMA wait times on C64x for the exact same design I had on > C62x are annoyingly longer (in terms of clock-cycles and percent of > total time) than those on the C62x. > > My question is, do I necessarily have to do more processing than on > C62X to reduce the EDMA wait times ? I mean, real processing, not > just increasing the amount of data transferred/processed. > > Looking forward to your comments and ideas. Thanks to Mike and > Indrajit for sharing their views on optimization. > > Regards > ka > |
|
|
Reply by ●September 5, 20032003-09-05
|
ka,
I was wondering if that was what you meant - most of us would love to have
that problem [:-)
It seems to me that the wait times are longer because you have more
horsepower.
I do not know your architecture - are you just EDMAing from SDRAM to IRAM
or is this data coming from IO?? If you are just coping RAM and your DSP
has nothing to do, you could try to operate on SDRAM directly and see what that
does for your overall processing time.
You might review/experiment with you EDMA setup. Also, if your algo
permits it, you might play with the transfer count.
mikedunn
Anand K <a...@yahoo.com> wrote:
|
