Altera Generic Tri-state Controller optimization for ZBT SRAM

[Altera_Forum]

Hello, i'm using EP4CGX150DF31 on custom PCB with dual Cypress NoBL (= Zero Bus Turnaround) SSRAM CY7C1463AV33. 

Theoretical Bandwidth 32 bits x 65 MHZ (memclk speed) = 2080 Mbit/s  QSYS SRAM ios:   .sram_tcm_address_out 21 of 23 hight bits connected to SRAM Address bus  .sram_tcm_byteenable_n_out all 4 bits connected to SRAM BWx pins  .sram_tcm_read_n_out connected to SRAM OE pin  .sram_tcm_write_n_out connected to SRAM WE pin  .sram_tcm_data_out connected to SRAM DATA bus  .sram_tcm_chipselect_n_out connected to CE1, CE2, CE3 according it's polaritys   I used NIOS II EDS Standart memory test program and achieved stable work using the following Generic Tri-state controller parameters:  (see GTC_params PNGs in attachment) Read wait time: 1 Wite wait time: 0 Setup time: 0 Data Hold Time: 1 Maximum pending data transactions: 64 Turnaround time: 0 Timing units: Cycles Read latency: 2 Chipselect through read latency is checked  But when i try to use this memory in Video proccessing (all standart Altera Cores): PAL Video ADC -> CVI -> frame buffer -> CVO -> PAL Video DAC  Clocked Video Output Underflow and my ext monitor lost video synchronization very often PAL video uses total 54 MBit/s maximum bandwith for read and write operation.  I try to analyze behavior of Generic Tri-State Controller with SignalTap LA and got this (stap PNGs in attachment) and saw that reading/writing 1 word use 3-5 clock cycles -> my bandwith is too slow  1) Does this Altera Core support ZBT/NoBL SRAM? 2) Have you any ideas, how to properly setup Generic Tri-State controller for this NoBL SRAM? 3) Does Altera provide any tool for measure external SRAM efficiency/bandwith on Cyclone IV FPGA?   P.S. Sorry for my English, it's not my native language.

[Altera_Forum]

Hello again. 

I've find optimum parameters for Cypress ZBT SRAM.  Maybe it can be usefull for anyone else.  Read wait time: 0 Wite wait time: 0 Setup time: 0 Data Hold Time: 0 Maximum pending data transactions: 64 Turnaround time: 0 Timing units: Cycles Read latency: 3 Chipselect through read latency is unchecked Full preset for Generic Tri-state controller in attachment  And else I had to instatiate PLL and shift 65 MHZ clock which drive SRAM's CLK pin to -3.07 ns. All other logic driven by not shifted clock.   New SignalTap images of Read/Write operation: https://www.alteraforum.com/forum/attachment.php?attachmentid=8439 https://www.alteraforum.com/forum/attachment.php?attachmentid=8440 https://www.alteraforum.com/forum/attachment.php?attachmentid=8441   Good luck

[Altera_Forum]

Hello KWler, 

I found your post very interesting. I have to manage a similar SRAM and I'm struggling with it with a similar QSys system. I would like to ask you some question: 1) Why do you put ".sram_tcm_address_out 21 of 23 hight bits connected to SRAM Address bus" if you have only 21 bits for address? Is there a reason or you could put directly 21? 2) Why do you use only 65 MHz if your memory could work at 133 MHz? (Silly question, but maybe you have a reason to make the SRAM working at half of its maximum speed) 3) What is the type of memory are you using? The 1Mx36 or the 2Mx18 one? I can't understand it from your parameters, you chose Data width 32 but in the module assignments window you put SRAM_Data_width 16; 4) Are Parameters in the "Module Assignment" window important? I don't have a CPU or a NIOS attached (it is a straight system), if I understand well they are parameters only important for software. 5) Have you a reason to put the SRAM clock shifted? I found thin solution also in other projects and I would like to understand why.   Thank you