Made more progress over the weekend. As the hashing data is generated it sets stores in one of 4096 buckets (just separated areas of memory) but to know where the next slot in the bucket is, there needs to be a list of buckets and how many items they have in it. I already had some look up RAM for this, but now made this a 3 way cycling RAM bugger. So one ram is being erased, when one gets updated to be written to and the 3rd is one we just read from. It seemed simple in my mind beginning with but had to MUX the control signals/ addr and data.
Synthesized it this morning and it will happy run at 277Mhz, and only uses 4.5block RAM's and around 102 LUT's. So pretty happy with that result. Now just need to stick this block in. This is going to cause me to modify proven code, but it will make the interface for the sorting block much easier to share RAM. Especially when the sorting does the 9 levels, I just get it to cycle the internal 3 RAM's. So the RAM that was one being written to can now be read, and the RAM we no longer need can be wiped. We could have done this with 2 RAM blocks, but then we would need to wait for the RAM to erase.This method we save 4096 cycles of 275Mhz which is around 15uS. Saving us around 150uS though the entire code. When the entire code should be around 125,000uS, its a bit of saving, only around 0.1%, but every little helps.
Synthesized it this morning and it will happy run at 277Mhz, and only uses 4.5block RAM's and around 102 LUT's. So pretty happy with that result. Now just need to stick this block in. This is going to cause me to modify proven code, but it will make the interface for the sorting block much easier to share RAM. Especially when the sorting does the 9 levels, I just get it to cycle the internal 3 RAM's. So the RAM that was one being written to can now be read, and the RAM we no longer need can be wiped. We could have done this with 2 RAM blocks, but then we would need to wait for the RAM to erase.This method we save 4096 cycles of 275Mhz which is around 15uS. Saving us around 150uS though the entire code. When the entire code should be around 125,000uS, its a bit of saving, only around 0.1%, but every little helps.
