• State Accepted Answer
  • +1 person also asked this people also asked this
  • Locked Locked
  • Replies 3 replies
  • Subscribers 347 subscribers
  • Views 3326 views
  • Users 0 members are here
Options
Related
How was your experience today?
This discussion has been locked.
You can no longer post new replies to this discussion. If you have a question you can start a new discussion

TCM and ARM1136

hongkongrr hongkongrr
Offline
Note: This was originally posted on 13th January 2009 at http://forums.arm.com

Hi,

I am trying to use DTCM to see if I can improve performance of my program on ARM1136.

I wrote a test program that just read/write TCM and read/write to RAM.  However anlayzing the timing of the test program there seemed to be no performance increase at all.

Therefore, I am suspecting that I am not enabling TCM correctly.  Could someone please help me confirm, I am really stuck here :)?
-> I arbitrarily just picked 0x80000000 to use, since I know that nothing else is addressed there.  Can I do this? Or do I need to have some specific HW knowledge here?
-> Next I enabled it to use 4K TCM by ORing my base address it with 0xd.
-> If I used the following code, will any load/store that I subsequently do to 0x80000000 be going to TCM or am I missing something? 
MCR p15, 0, r0, c9, c1, 0;r0 was filled with (0x80000000|0xd)
  • 0 Offline
    Note: This was originally posted on 13th January 2009 at http://forums.arm.com

    The device provides 4K TCM according to the reading of the register.

    Also, is my assumption about arbitrarily picking 0x80000000 to use correct? Or do I need to have some specific HW knowledge here to pick the right address?

    Thanks a lot for your help... I am completely stumped
  • 0 Offline
    Note: This was originally posted on 13th January 2009 at http://forums.arm.com

    The size field of this CP15 register is ignored on write; the available TCM on the I and on the D interface is defined at synthesis time. How much TCM does your device actually provide (reading the register will tell you how much TCM is available)?

    The second point would be to make sure that you put data in D TCM, not I TCM. Loads and stores to I-TCM are a cycle slower than loads and stores to D TCM.
  • +1 Offline
    Note: This was originally posted on 13th January 2009 at http://forums.arm.com

    Choosing 0x80000000 is fine, provided that it does not overlay anything critical in your physical memory map.

    It is worth noting that 0x80000000 is a Physical Address; assuming that you have the MMU enabled you will need to provide a sensible VA mapping of this address in the translation tables to make it available to the code that is running.

    If you don't have MMU enabled, you will want to look at enabling it - this enables the D-cache to function on ARM11 which gives very large speed increases!