I am getting a lot less floating point horsepower from the 300 Mhz 6747 than I did from the 200 Mhz 6713. All specs seem to point to this not being the case.
This thread has been locked.
If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.
I am getting a lot less floating point horsepower from the 300 Mhz 6747 than I did from the 200 Mhz 6713. All specs seem to point to this not being the case.
Something fundamentally is wrong. Ensure L1P and L1D cache are enabled. Ensure L2 Cache in enabled. Ensure MAR bits are set for both L3 (0x8...) and external memory (0xc...)
Gagan
Would this not be something the DSP BIOS configuration would set up? I do have L1P and L1D enabled, I tried with L2 disabled and enabled (128K), that part made no difference.
Thank you, I manually set up MAR196 for SDRAM and enabled 128K of L2 and that did the trick. But I am having cache coherence issues with EDMA. The 0x8 and 0xC are not making sense to me, as the MARxx bits are only one bit for enable. Which register are you referring too?
> The 0x8 and 0xC are not making sense to me, as the MARxx bits are only one bit for enable. Which register are you referring too?
As you already know, each MAR register represent a range of memory that it controls for enabling caching. See here:
http://tiexpressdsp.com/index.php/Enabling_64x%2B_Cache
On C6747, the L3 ram is at address 0x80000000 - 0x8001ffff and the external memory at: 0xC0000000 - 0xDFFFFFFF. You need to ensure that the MAR registers are set for the memory that you need to enable caching for. by 0x8 and 0xC I meant the L3 RAM and external memory and not really MAR bits. Sorry for confusion.
For cache coherence with EDMA, see the cache UG for details http://focus.ti.com/lit/ug/sprug82a/sprug82a.pdf. There are DSP-BIOS BCACHE module APIs that you can use for cache coherence
Gagan
Got it all running great, seeing almost double the performance of 6713 (as it should be). Cache coherence is not an issue, I am pretty familiar with that.
The main issue is that the DSP BIOS config for 674x and all of the examples in the EVM have the MAR bits set incorrectly (unlike all previous chips I have played with). So I didn't even think to look at it.
Probably should be corrected in the EVM material.
Edmund,
I will give feedback about that. For now I created the page:
http://tiexpressdsp.com/index.php/C6747_Audio_Example_cache_configuration