PROCESSOR-SDK-AM64X: Xin and Xout instructions with 0 bytes

Hello Madhurya,

What exactly are you trying to accomplish by doing an XIN / XOUT with 0 bytes? For example, are you trying to get the PRU core to pause for one clock cycle? If you are just trying to get the PRU core to pause for a clock cycle, you would want to use NOP (no operation) instead of an actual assembly command.

Regards,

Nick

Hi Nick!

Nop(e), I'm not trying to pause the PRU core. .

I have an algorithm that increments the Xout bytes from 0,4,8,16,32. Since I wrote the program in this way, I need to xout 0 bytes in the first instant.

I found a fix by clearing the data from the respective memory location to nullify the xout of 2 bytes that was caused by xout 0 bytes instruction.

I still suggest this should be documented as a default for xout instructions.

Best Regards,

Madhurya

Hello Madhurya,

Nice pun :P

To make sure I understand what you are observing:
xout & xin were not designed to move 0 bytes of data. Because of this, that behavior is undefined. It looks like when you passed in length = 0, the assembler defaulted to 32 bits or 4 bytes. I am not sure if that behavior is replicable across all compiler / assembler settings, and all compiler / assembler releases.

I am unclear on your workaround. Did you observe xout with length = 0 moving 2 bytes at one point in time, and moving 4 bytes at a different point in time? If I understand correctly, you just wrote zeros into the memory addresses that would get moved with your xout with length = 0 command?

Regards,

Nick

Hi Nick!

No, I did not observe xout with length = 0 moving 2 bytes at one point in time and moving 4 bytes at a different point in time. It always xout's 4 bytes if I try to xout 0 bytes.

Yes, the workaround is to clear the content in the memory.

clear:
    zero &r2, 32
    ldi R10.w0, 0x0001;
    xout 0x1E, &R10, 2
    xout 0x1E, &R2 , 32 ; empty contents of the previous load with Xout 0 bytes.

Best Regards,

Madhurya

Hello Madhurya,

Apologies for the delayed response. I am sending your thread to a member of our compiler team to double-check if there is a way to xout 0 bytes, or if your workaround is the best approach.

Regards,

Nick

I assembled similar code that uses XOUT.  Here is the assembler listing ...

       1 00000000 0000002F0F3F8A          xout    0x1e, &r10, 0
       2 00000004 0000002F0F018A          xout    0x1e, &r10, 4

Notice how the encoding of an XOUT of 0 bytes is different from the encoding of the same XOUT of 4 bytes.  So, as far as the assembler is concerned, it is valid to XOUT zero bytes.  What the PRU does with that particular instruction encoding is beyond my expertise.

Thanks and regards,

-George