Re: [Fwd: Re: code working on simulator but not on device?]

Started by Jeff Brower May 10, 2006

> I tested with the highest speed possible (156MHz) with a multiplier of
> 13.
> Yes there was increase in the counts of rx
> It was as follows:
> 25
> 0
> 35
> 0
> 34
> 0
> 35
> 0
> and tx was 0 for all. well I cant see the definations of the function
> PEP5416_AIC23_read and write but ya I know what they do. The read16 will
> read a 16-bit signed value from the codec data port and will return false
> if the data port is busy. Same goes with write16. Ya they wont wait, they
> will take the data as it comes.

Your description of the PEP5416_AIC23_read and _write functions sounds Ok,
but why do they take so long? Think about it: if the DSP is truly
running 156 MHz, and the sampling period is truly 22 usec, then each call
to PEP5416_AIC23_read consumes almost 640 nsec -- which is a lot of
instructions (about 100) for one function call that just checks the serial
port status bit.

> Yes the DSP/BIOS is configured but only for memory mapping and global
> settings. I am not using any interrupts or anything in DSP/BIOS.

No tasks? No periodic functions?

> Jeff, in this case the counts are coming 34/35 for rcv but in case when I
> added filter function both of them were 0. Is it becasue filter() is
> taking so long that when the PC comes to read() it reads immediately
> without halting?

Yes. If you make filter() short/fast enough, then rx_wait won't be zero.
For now, you might try a PLL multiplier of 19 -- this will overclock the
5416 to about 225 MHz. That's not recommended for a final solution, but
if it works it might give you a chance to get your filter code running in


> Jeff Brower wrote: Megha- Results with no
> printf were:
> rx_wait was:
> 8
> 0
> 19
> 0
> 19
> 0
> 19
> 0
> and tx_wait were all 0.
> Ok, these counts are really short, but at least they are consistent. The
> Tx values are zero because once Rx is ready, Tx is most likely going to be
> ready at the same time.
> Well as per as clock is concerned. DSP is running at 96MHz (as set in
> DSP/BIOS) and AIC23 is running at 12MHz (as set by PEP config). THe
> CLKMD is set as multiply by 8 (in DSP/BIOS).
> The above counts are just too low for 96 MHz clockrate. First, set your
> multiplier to 15 as I suggested before, and see if the counts increase
> about 1.5x. Second, can you see what PEP5416_AIC23_xxx functions are
> doing? They should just check the serial port RXRDY and TXRDY status
> bits, and do simple read or write if the status bits indicate Ready. I'm
> assuming these functions don't sit there and wait; they should return
> immediately. You have to verify that.
> Also, do you have DSP/BIOS configured and enabled? If so, there could be
> another task getting in the way.
> -Jeff
> Jeff Brower wrote: Megha-
> I did following:int rx_wait[8];
> int tx_wait[8]; for(i=0;i<8;i++)
> {
> rx_wait[i] = 0;
> /* Copy data from microphone inputs to headphone output */
> while (!PEP5416_AIC23_read16(hCodec, &buffer[i]))
> rx_wait[i]++;
> printf("rx wait is %d \n",rx_wait[i]);
> tx_wait[i] = 0;
> while (!PEP5416_AIC23_write16(hCodec, buffer[i]))
> tx_wait[i]++;
> printf("tx wait is %d \n",tx_wait[i]); }
> Don't put printf() statements in the loop -- that's a no-no for real-time
> operation and accurate measurement. Run the experiment again and tell me
> what are your xx_wait[] values. Just use memory watch window to learn the
> array values after the loop is finished -- no need for printf().Regarding
> CLKMD value below, if CLKIN is 12 MHz and multiplier is 8, then the 5416
> should be running 96 MHz. You can try multiplier of 13 also, which would
> be 156 MHz (chip limit is 160).-Jeff
> The values i got was
> rx wait is 13
> tx wait is 0
> rx wait is 0
> tx wait is 0
> rx wait is 0
> tx wait is 0
> rx wait is 0
> tx wait is 0
> rx wait is 0
> tx wait is 0
> rx wait is 0
> tx wait is 0
> rx wait is 0
> tx wait is 0
> rx wait is 0
> tx wait is 0
> I tested with 10, 20 instead of 8. Sometimes rx was 13 twice but rest all
> was same.
> In DSP/BIOS I have set CLKOUT as 96MHz, CLKMD value is 7107 (multiply by
> 8).
> In the master mode, the TLV320AIC23 clock and sampling rates are derived
> from a 12-MHz MCLK signal.
> In the PEP config (PEP5416_AIC23_Config config) I have set CLKOUT as 12MHz
> (and 6MHz) and CLKIN as 12MHz. These values can be either set to 12 or
> 6MHz. In this config, base oversampling rate is set as 250fs and thats
> 48KHz.
> Maximum sampling rate can be 48KHz with filter type as 0 and 96KHz with
> filter type as 3.
> Kindly reply
> Thanks
> Megha