• State
  • Locked Locked
  • Replies 3 replies
  • Subscribers 81 subscribers
  • Views 600 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

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.

CLA MPU memory read conflict and incorrect read

evs
Genius 5910 points

I got an incorrect memory read by the CLA.

With the following code:

    encoder_dir=1;
    Cla1ForceTask8 ();

__interrupt void Cla1Task8 ( void )
{
    dir=encoder_dir;
}

Somehow this gives a conflict. The result of this code is that when dir = 0.

If I change it to this:


    Cla1ForceTask8wait ();

    encoder_dir=1;

It works correctly.

So somehow when the 28035 and CLA try to write/read the same memory location the CLA got an incorrect value. So how do I solve this?

Thanks!

edit:

Can the memory alignment form 16 -> 32 bit be a problem?

This is the variable definition:

volatile int dir;
#pragma DATA_SECTION(encoder_dir,"CpuToCla1MsgRAM")
volatile  int  encoder_dir;

  • 0
    TI__Mastermind 20850 points

    Hi,

    Don't use int for shared variables. int is 16-bits on the C28x and 32-bits on the CLA. It could be causing the code on the CLA side to look up the wrong address, especially if the linker has placed the 16-bit variable at an odd address.

    I would declare it as volatile short or volatile int16_t (include stdint.h)

  • 0 in reply to Vishal_Coelho
    Genius 5910 points

    Hi Vishal,

    Thanks for the reply. That seems to be the problem. I changed it to long.

    Can you please explain the difference between int and short or int16_t are they not all typedef int?

    Is not the only way to solve it  something like:

     #pragma align memory even

    And why  it isn't done automatically by the compiler with this command:

    #pragma DATA_SECTION(encoder_dir,"CpuToCla1MsgRAM")

  • 0 in reply to evs
    TI__Mastermind 20850 points

    Hi evs,

    evs said:
    Can you please explain the difference between int and short or int16_t are they not all typedef int?

    int and short are primitive C type. The C language sets short to be 16-bits and int to match the native bit-format of the machine. In the case of the c28 its 16-bits but the CLA is more geared toward 32-bit operations so it made sense to set int to 32-bits on the CLA. int16_t is typedef'd to 16-bits on both cores so it is always safer to use the explicit types defined in stdint.h

    evs said:

    Is not the only way to solve it  something like:

     #pragma align memory even

    And why  it isn't done automatically by the compiler with this command:

    #pragma DATA_SECTION(encoder_dir,"CpuToCla1MsgRAM")

    I guess you could have got away with even aligning the int, but consider if you had an array of ints - the c28 allocates 16 bits per entry, and when you access it on the CLA, say a[0], you end up reading a[0]:a[1] as the CLA will do a 32-bit read.