Hello,
I'm optimizing the operation of my application running on a C6747 @ 375 MHz. Also, I have configured L1P full cache, L1D full cache and L2 64 kB cache / 192 kB RAM.
I placed the .const section in the internal shared RAM (128 kB space) as opposed to the external SDRAM on EMIFB @ 125 MHz. I was thinking that it was quicker for the DSP core to access the shared RAM and cache the contents instead of accessing the external SDRAM and cache the contents.The system interconnect block in section 3.2 of SPRUFK4D shows that the path to EMIFB has an additional Switched Central Resource (SRC) and then a 32-bit bus instead of 64 bits for shared RAM.
After testing, it seems to be the opposite, the overall performance of my application is faster when .const data is in SDRAM. Would this be because shared RAM contents are not cached in L2 and/or L1. I went through the datasheets and the cache user guide but I cannot find information on shared RAM vs. cache. How can this behaviour be explained?
Thanks,
SC