2 Replies Latest reply on May 4, 2015 5:32 PM by jake.wegman

    "Artificial" SPI when using GPIF-II 32-bit mode




      Please bear with my multi-threaded ignorance...


      Also, I originally posted this on 'Known Problems and Solutions' when it should've gone here...


      For our current application, we're using the 32-bit GPIF-II mode, and thus the SPI peripheral is not available.  We have another device off to the side that we've connected to three GPIO on the FX3 to act as chip select, clock, and data (we did basically the same thing with a project we did that used the FX2 chip).  We want this interface to work at close to its maximum speed, which is 20MHz according to the interfacing component's data sheet.  In our previous iteration, we'd drive chip select and the clock low, drive the data to its desired value, run a couple of 'nop' calls for setup time, bring the clock high, run a couple of 'nop' calls for hold time, and then bring the clock signal low again, all in a loop until the message was finished.  I tried to implement something similar on the FX3, but it runs quite a bit slower than I anticipated.  I'm thinking it's due to the multi-threaded nature of the underlying OS, but I'm not an expert on this, so I've come to the forums.  Here's my code:

      void SwXTimesFiveNsDelay( uint8_t x ) {     int i;     for (i = 0; i < x; i++)     {         __nop();     } }  void SwSpiWriteToDac( uint16_t val ) {     signed char b;     /* Don't let anything get in the way of this routine; disable all      * interrupts, but make sure we know which interrupts were enabled before we      * disabled them. */     uint32_t    currentIRQs = CyU3PVicDisableAllInterrupts();      /* Turn on the chip select */     CyU3PGpioSetValue( GPIO_HVCTRLCS_L, CyFalse );     /* Force clock low if it isn't already */     CyU3PGpioSetValue( GPIO_HVCTRLSCK, CyFalse );      for(b = 15; b >= 0; b--)     {         if (val & (1 << b))             CyU3PGpioSetValue( GPIO_HVCTRLSDI, CyTrue );         else             CyU3PGpioSetValue( GPIO_HVCTRLSDI, CyFalse );         SwXTimesFiveNsDelay(6);  // 50% duty cycle clock, ~16.8MHz         CyU3PGpioSetValue( GPIO_HVCTRLSCK, CyTrue );  // Bring clock high         SwXTimesFiveNsDelay(6);  // Hold high for the rest of the duty cycle         CyU3PGpioSetValue( GPIO_HVCTRLSCK, CyFalse );  // Bring clock low     }     /* Disable DAC */     CyU3PGpioSetValue( GPIO_HVCTRLCS_L, CyTrue );      /* Re-enable previously disabled interrupts */     CyU3PVicEnableInterrupts( currentIRQs ); }

      The actual logic of the code is fine; it does exactly what I expect it to logically on the scope.  However, the calls to CyU3PGpioSetValue appear to take ~1.2us, and the __nop's take ~63ns.  I didn't really now how long the CyU3PGpioSetValue call was going to take, and my calculations were that a __nop would take 4.96ns when using a system clock of 403.2MHz (CPU clock is half the system clock).


      Thread priority is '8', and it's one of two threads running, the other being a low-priority debug thread.


      Thanks in advance.