Forum Discussion

ggxnd's avatar
ggxnd
Icon for New Contributor rankNew Contributor
2 years ago

Memory Conflicts in onchip ram

I am using IOWR_32DIRECT to write contents in the on-chip ram and using IORD_32DIRECT to read from the same address where I performed write operation. Upto few memory location the write and read data matches. Issue arises when I perform those operation in a loop till I fill the complete on-chip ram addresses (say 2MB), it results in some memory conflicts. Either write data and read data is not matching or the compiler is not processing properly resulting in some weird print character. Does it depends on number of times IOWR_32DIRECT and IORD_32DIRECT execution? On what part of hardware configuration does it depends on? What is the minimum program area needed for such functionality?

Thanks in Advance!

11 Replies

  • ShengN_altera's avatar
    ShengN_altera
    Icon for Super Contributor rankSuper Contributor
    2 years ago

    Hi,

    I had created a simple design that consists of clock, OCR, Nios II, jtag uart and data in/out component. The OCR address is range from 0x40000 (Base) to 0x657ff (End). The corresponding decimal for 0x40000 (Base) and 0x657ff (End) are 262144 and 415743 respectively.

    I had created a code which loops to IOWR_32DIRECT and IORD_32DIRECT the data for each OCR address. The first data is 262144 which being incremented by 1 for each loop until 415743. In the terminal, I'm getting decimal 415743 for address 0x657ff means the data are matching with each address respectively. Attached the project file below for your reference.

    Code:

    # include <stdio.h>

    # include "system.h"

    # include "io.h"

    #define BASE_ADDRESS ONCHIP_MEM_BASE

    #define END_ADDRESS 0x657FF

    int main() {

    int i = 262144;

    unsigned int addr;

    for (addr = BASE_ADDRESS; addr <= END_ADDRESS; addr++) {

    char addr_str[11]; // Assumes 32-bit address

    snprintf(addr_str, sizeof(addr_str), "0x%05x", addr);

    IOWR_32DIRECT(addr_str, 1, i);

    printf("Address: 0x%x, Value: %d\n", addr, IORD_32DIRECT(addr_str, 1));

    i++;

    }

    return 0;

    }

    Terminal screenshot:

    Not sure your case is because of processor speed or not. What will happen if you add small delay to the loop?

    Thanks,

    Best Regards,

    Sheng

    p/s: If any answer from the community or Intel Support are helpful, please feel free to give best answer or rate 4/5 survey.

    • ggxnd's avatar
      ggxnd
      Icon for New Contributor rankNew Contributor
      2 years ago

      Hi,
      Thank you for your reply. I will have a look into the solution that you provided. I have a doubt.
      The following is how I try to perform write and read.

      for (addr = 0; addr <= END_ADDRESS; addr += 4) {

      IOWR_32DIRECT(BASE_ADDRESS, addr, i);

      }

      for (addr = 0; addr <= END_ADDRESS; addr += 4) {

      value = IORD_32DIRECT(BASE_ADDRESS, addr);
      printf("Address: 0x%x, Value: %d\n", addr, value);

      }

      What is the key difference between the your code and the above one. Why is it failing to provide same result here.? Can you provide any insight / idea on this?

  • ShengN_altera's avatar
    ShengN_altera
    Icon for Super Contributor rankSuper Contributor
    2 years ago

    Hi,


    The code provided by me I'm just increasing the address part (addr_str) for each loop. While keeping the address offset at constant 1.

    Compared to yours, you're increasing the address offset (addr) part instead. Why not you just keep the address offset constant and increase the address part for each loop any difference?


    Thanks,

    Best Regards,

    Sheng


    • ggxnd's avatar
      ggxnd
      Icon for New Contributor rankNew Contributor
      2 years ago

      Hi,
      The address part (addr_str) doesn't seem to be increasing.
      How the address translation done here - IOWR_32DIRECT(addr_str, 1, i); and on what address the data 'i' is write into?
      If I want to read the value from a particular address(from the above mentioned range - BASE_ADDR to END_ADDR) outside this loop , how it can be done?
      When I try to read it outside the loop, the same write value is not obtained. so my doubt is how address translation in done in IOWR_32DIRECT ?
      Also IOWR_32DIRECT macro is to perform a memory-mapped I/O write operation , we can directly pass numerical value. What is the need to pass formatted string in that?

      Thanks in Advance!

  • ShengN_altera's avatar
    ShengN_altera
    Icon for Super Contributor rankSuper Contributor
    2 years ago

    Hi,

    There's some mistake in the previous code posted. Done modified as below:

    int main() {

    int i = 262144;

    unsigned int addr;

    unsigned int result;

    for (addr = BASE_ADDRESS; addr <= 0x40019; addr+=4) {

    IOWR_32DIRECT(addr, 0x4, i);

    i++;

    }

    for (result = BASE_ADDRESS; result <= 0x40019; result++ ) {

    printf("Address: 0x%x, Value: %d\n", result, IORD_32DIRECT(result, 0x4));

    }

    printf("Address: 0x40016, Value: %d\n", IORD_32DIRECT(0x40016, 0x4));

    printf("Address: 0x40007, Value: %d\n", IORD_32DIRECT(0x40007, 0x4));

    return 0;

    }

    Terminal screenshot below:

    Address part is increasing correctly and can read value outside the loop correctly. Ya, you're right can pass directly the numerical value.

    You can checkout this link https://fys4220.github.io/part-exercises/exercises_memory_mapped_sw.html on byte-oriented offset for IOWR_32DIRECT and IORD_32DIRECT

    Thanks,

    Best Regards,

    Sheng

    • ggxnd's avatar
      ggxnd
      Icon for New Contributor rankNew Contributor
      2 years ago

      Hi,
      Thanks.
      My question is does the same works for 2MB address location of OCR. What if the end address is increased so that it writes 2MB location?

  • ShengN_altera's avatar
    ShengN_altera
    Icon for Super Contributor rankSuper Contributor
    2 years ago

    Hi,


    I'm not able to complete in one run as well. Because some of the address locations require different data length or format and they're all mixed.

    May be try to read first those address locations which got memory conflicts to know the data format and try to run in separate trial.

    Btw, possible to provide your code for testing out on my platform?


    Thanks,

    Best Regards,

    sheng


    • ggxnd's avatar
      ggxnd
      Icon for New Contributor rankNew Contributor
      2 years ago

      Hi,
      Yes. Is there any other way to address this issue? why the write data is not matching with the read data? Why the read function fails to read the write data? or Is there any memory limitations in onchip ram?

      Code is similar to the one you modified.

      int main() {

      int i = 262144;

      unsigned int addr;

      unsigned int result;

      for (addr =0; addr <=0x1FFFFF ; addr+=4) {

      IOWR_32DIRECT(BASE_ADDRESS, addr, i);

      i++;

      }

      for (addr = 0; addr <= 0x1FFFFF; addr+=4 ) {

      printf("Address: 0x%x, Value: %d\n", BASE_ADDRESS+ addr, IORD_32DIRECT(BASE_ADDRESS, addr));

      }

      return 0;

      }
      Thanks.



  • ShengN_altera's avatar
    ShengN_altera
    Icon for Super Contributor rankSuper Contributor
    2 years ago

    Hi,


    Yes. There're some memory limitations as some of the address locations are not writeable. I would suggest you using this method to verify the matching of the write and read data:

    int main() {

    IOWR_32DIRECT(ONCHIP_MEM_BASE, 0x1FFFFF, "<string>");

    printf("%s\n", IORD_32DIRECT(ONCHIP_MEM_BASE, 0x1FFFFF));


    return 0;

    }


    Thanks,

    Best Regards,

    Sheng




  • ShengN_altera's avatar
    ShengN_altera
    Icon for Super Contributor rankSuper Contributor
    2 years ago

    Another method below can be used as well:

    int main() {

    IOWR(ONCHIP_MEM_BASE, 0x1FFFFF, "<string>");

    alt_printf("%s\n", IORD(ONCHIP_MEM_BASE, 0x1FFFFF));

    return 0;

    }