#include <pal_master.hch>
#include <stdlib.hch>
#include "weights256.hch"
#include "config.hch"
#include "debug.hch"
#include "xilinxmult.hch"

#define      	PERFORM_FFT_CALCULATION 1
#define		USE_UNSIGNED_AUDIO 0
#define		PRINT_DEBUG 0
/* Define two multi-port RAMs for FFT calculation; one for real and one for imaginary values
 * Extra block RAM settings are defined to make sure read and write actions can be performed
 * within one clock-cycle.
 * Left out extra settings on new board the clock changes TODO !!!!
 */
#if HARDWARE_MULTIPLY
mpram
{
  ram signed 18 rwrite[256];
  rom signed 18 read[256];
} real with {block = "BlockRAM"/*, westart=2.5, welength=1, rclkpos={1.5}, wclkpos={3}, clkpulselen=0.5*/};

mpram 
{
  ram signed 18 rwrite[256];
  rom signed 18 read[256];
} imaginary with {block = "BlockRAM"/*, westart=2.5, welength=1, rclkpos={1.5}, wclkpos={3}, clkpulselen=0.5*/};
#else
mpram
{
  ram signed 24 rwrite[256];
  rom signed 24 read[256];
} real with {block = "BlockRAM"/*, westart=2.5, welength=1, rclkpos={1.5}, wclkpos={3}, clkpulselen=0.5*/};

mpram 
{
  ram signed 24 rwrite[256];
  rom signed 24 read[256];
} imaginary with {block = "BlockRAM"/*, westart=2.5, welength=1, rclkpos={1.5}, wclkpos={3}, clkpulselen=0.5*/};
#endif
// multiplication factors for equalizer function
ram signed 7 eq_settings[16] = {0,2,4,7,10,13,16,19,22,26,30,35,41,48,55,63};


/****************************************************************
* Function:    multiply                                         *
*                                                               *
* Arguments                                                     *
*   x,y        signed variables                                 *
*                                                               *
* Description                                                   *
*   Just a multiplier. But by doing this in a function the      *
*   FPGA space needed is reduced.                               *
*                                                               *
* Return Values                                                 *
*   The result after multiplication                             *
****************************************************************/
macro proc multiply(result, op_a, op_b)
{
#if HARDWARE_MULTIPLY
	xilinxmult(result, op_a, adjs(op_b,18));
#else
	result = (adjs(op_a,38))*(adjs(op_a,38));
#endif	
}




/*******************************************************************
* Function:    calculate_fft                                       *
*                                                                  *
* Arguments                                                        *
*   select_inverse	Boolean that indicates FFT or iFFT calculation *
*                                                                  *
* Description                                                      *
*   This routine performs the Fast Fourier Transform for		   *
*   calculation of the frequency spectrum					       *
*                                                                  *
*******************************************************************/
void calculate_fft(unsigned 1 select_inverse)
{
  unsigned 4 level;
  unsigned 8 point1,point2,j,f,k;
  unsigned 9 e,i;
  signed 16 weight1,weight2;
#if HARDWARE_MULTIPLY
  signed 18 p,q,r,t;
#else
  signed 24 p,q,r,t;	
#endif
  signed a,b;

#if HARDWARE_MULTIPLY
  // Macro to provide rescaling of 36-bit result of fixed point multiply
  // down to an 18-bit result. The range of bits selected depends on the 
  // number that represents the value of "1" in the trig function lookup
  // tables. (Eg. for 16384 == 1, the lowest bit selected should be [14]).
  macro expr rescale (x) = (x[35] @ x[30:14]);
#else
  //Macro to rescale the multiply result down to a 24-bit value.
  macro expr rescale (x) = ((x>>FRACBITS)<-24);
#endif

  for(level=1;level<=NUMBER_OF_COLUMNS;level++) // count all the columns
  {
    	e=1<<(NUMBER_OF_COLUMNS-level+1); // number of points in each block in this column
    	f=(e>>1)<-8; 			  // number of butterflies in each block in this column

    	for(j=1;j<=f;j++) 	// count all the butterflies in each block
    	{
		par
		{
      			// Weight factors for real (the same for FFT and iFFT)
	  		weight1 = weight_re[((j-1)<<(level-1))<-7]; 

			
			// Weight factors for imaginary (opposite for FFT and iFFT)
	  		weight2 = (!select_inverse) ? (weight_im[((j-1)<<(level-1))<-7]) : -(weight_im[((j-1)<<(level-1))<-7]); 

			/* ORIGINAL CODE BELOW, MODIFIED BECAUSE OF MISMATCHING OUTPUT WITH BORLAND TESTAPP
			weight2 = (!select_inverse) ? -(weight_im[((j-1)<<(level-1))<-7]) : weight_im[((j-1)<<(level-1))<-7]; 
			*/
			
			

			for(i=0@j;i<=NUMBER_OF_POINTS;i+=e)   // count all the blocks in this column
			{	// Butterfly calculation
        			par
				{
					point1 = ((i<-8)-1);
	      				point2 = (((i<-8)+f)-1);
	    			}
				
				par
				{
					p = (real.read[point1] >> 1) + (real.rwrite[point2] >> 1);
		  			q = (imaginary.read[point1] >> 1) + (imaginary.rwrite[point2] >> 1);
				}
				
				par
				{
		      			r = (real.read[point1] >> 1) - (real.rwrite[point2] >> 1);
					t = (imaginary.read[point1] >> 1) - (imaginary.rwrite[point2] >> 1);
				}		

				multiply(a,r,weight1);
				multiply(b,t,weight2);

       		        	par
 				{
        	  			real.rwrite[point2] = (rescale(a-b));
          				imaginary.rwrite[point1] = q;
    				}

				multiply(a,t,weight1);
				multiply(b,r,weight2);

				par
				{	
					real.rwrite[point1] = p;
		  			imaginary.rwrite[point2] = (rescale(a+b));
		  		}

		  	}
		}
	}
  }

  j=1;
  for(i=1;i<NUMBER_OF_POINTS;i++)
  {
  	if(i<(0@j))
   	{
   		par
		{
			point1=j-1;
   			point2=(i-1)<-8;
		}
		/*
		  COPYING ARRAY VALUES FROM ONE PLACE TO ANOTHER IN THE ARRAT MUST BE DONE IN 
		  2 STEPS. FIRSTLY THE VALUES ARE COPIED TO SEPARATE VARIABLES AFTER THAT THEY
		  ARE COPIED BACK TO THEIR NEW POSITION IN THE ARRAY. THIS MUST BE DONE TO 
		  PREVENT TIMING ISSUES FROM OCCURING.
		*/
   		par
		{
			p = real.read[point1];
   			q = imaginary.read[point1];
		}
		par
		{
			r = real.read[point2];
   			t = imaginary.read[point2];
		}
		par
		{
			real.rwrite[point1] = r;   	
   			imaginary.rwrite[point1] = t;
		}
  		par
		{
			real.rwrite[point2] = p;
   			imaginary.rwrite[point2] = q;
		}
   	}

  	k = NUMBER_OF_POINTS>>1;


  	while(k<j)
   	{
   		j = j-k;
   		k = k>>1;
   	}

  	j+=k;
  }

}

/********************************************************************/
#if HARDWARE_MULTIPLY
void perform_fft(signed 18 *pcm_audio)
#else
void perform_fft(signed 16 *pcm_audio)
#endif
{
  	unsigned 8 k;
	
	
#if HARDWARE_MULTIPLY
	signed 18 sample;
	k=0;
	sample = adjs(pcm_audio[k],18);
#else
	signed 24 sample;
	k=0;
	sample = adjs(pcm_audio[k],24);
#endif
    	
	//initialize variables for the copying pipeline

	
	// copy audio data to real-array before starting FFT calculation
	// and set imaginary values to zero
	do
	{
		//Copying the array values has been pipelined to prevent parallel access to the
		//pcm_audio array. This copying procedure must be finished before another 
		//sample is read from the audio input. The time available for this loop is 
		//determined by the sampling rate of 44,1 Khz
		par
		{
			//COPYING NEEDS TO BE DONE IN 2 STEPS, BECAUSE THE VALUE THAT NEEDS TO WRITTEN
			//TO THE REAL-RAM NEEDS TO BE AVAILABLE ON THE START OFF THE CLOCKCYCLE.
#if HARDWARE_MULTIPLY
			sample = adjs(pcm_audio[k+1],18);
#else
			sample = adjs(pcm_audio[k+1],24);
#endif
			real.rwrite[k] = sample;
      			imaginary.rwrite[k] = 0;
			k++;
		}		
    	}  while (k);

	

#if PERFORM_FFT_CALCULATION
	calculate_fft(0);
#endif


}

/********************************************************************/
#if HARDWARE_MULTIPLY
void perform_ifft(signed 18 *modified_audio, unsigned 6 *ifft_info)
#else
void perform_ifft(signed 16 *modified_audio, unsigned 6 *ifft_info)
#endif
{
  	unsigned 6 k;
#if HARDWARE_MULTIPLY 
  	signed 18 p;
#else
	signed 24 p;
#endif
#if PERFORM_FFT_CALCULATION	
	calculate_fft(1);
#endif

    	k=0;
//initialize variables for the copying pipeline
#if PERFORM_FFT_CALCULATION	
	#if HARDWARE_MULTIPLY 
		p = (real.read[(0@k)+95] << NUMBER_OF_COLUMNS);
	#else
		p = (real.read[(0@k)+95] >> NUMBER_OF_COLUMNS);
	#endif
#else
		p = (real.read[(0@k)+95]);
#endif

	do
	{
		//Copying the array values has been pipelined to prevent parallel access to the
		//pcm_audio array. This copying procedure must be finished before another 
		//sample is read from the audio input. The time available for this loop is 
		//determined by the sampling rate of 44,1 Khz
	   	par
		{
#if PERFORM_FFT_CALCULATION	
	#if HARDWARE_MULTIPLY 
		p = (real.read[(0@k)+95] << NUMBER_OF_COLUMNS);
	#else
		p = (real.read[(0@k)+95] >> NUMBER_OF_COLUMNS);
	#endif
#else
		p = (real.read[(0@k)+95]);
#endif
#if HARDWARE_MULTIPLY
      			modified_audio[k] = p ;
#else
			modified_audio[k] = (p<-16);
#endif
			ifft_info[k] = (p[17:12]); 		
			k++;
		}
    	} while(k);
}

/********************************************************************/
void equalize_audio(unsigned 4 *eq_level, unsigned 7 *fft_info)
{
#if HARDWARE_MULTIPLY
  signed 18 p,q;
#else
  signed 24 p,q;
#endif
  signed 18 a;
  unsigned 8 i, mirror_i, bit, m, n;
  unsigned 7 old_value;
  unsigned 9 tmp;
  
  //macro expr equalize_bar = multiply(q,a)[29:6];
  
  macro proc equalize_bar(retval)
  {
  	 signed result;
 	 multiply(result, q,a);
#if HARDWARE_MULTIPLY
	 retval = result[23:6]; //drop last 6 bit to compensate the maximum multiplication with 64 from the eq_settings array
#else
	 retval = result[29:6]; //drop last 6 bit to compensate the maximum multiplication with 64 from the eq_settings array
#endif
  } 

  p = real.read[0] - DC_COMPONENT; // remove DC component for calculations
  real.rwrite[0] = p;
//  imaginary.rwrite[0] = 0; 				   		// remove DC component 

  
  for(i=0;i!=NUMBER_OF_FREQUENCIES;i++)	  
  {  
	// set multiplication factor (0..64) for current frequency bar, The first frequency band must be equalized at 100% (63) since there is no DC-component taken into account.
	a = (i==0) ? 63 : adjs(eq_settings[eq_level[i<-7]],18);


	// multiply frequency with this factor and divide by 64 (drop 6 LSB's)
	q = real.read[i];
	equalize_bar(p);
	real.rwrite[i] = p;

    	q = imaginary.read[i];
	equalize_bar(p);
	imaginary.rwrite[i] = p;

	// the upper part(128..255) of the spectrum is mirrored to the lower part; 
	// these values need to be adjusted too
	if ((i<-7)!=0) // if not in DC component bar
	{
		mirror_i = (NUMBER_OF_POINTS-1)-i+1;
		q = real.read[mirror_i];
		equalize_bar(p);
		real.rwrite[mirror_i] = p;

		q = imaginary.read[mirror_i];
		equalize_bar(p);
		imaginary.rwrite[mirror_i] = p;
	}
  }
  
  //write data to fft_info for display purposes
  for(i=0;i<NUMBER_OF_FREQUENCIES;i++)
  {
   	p = real.read[i];
    	q = imaginary.read[i];
#if HARDWARE_MULTIPLY
    	if (p[17] == 1) p = -p; else delay;
	if (q[17] == 1) q = -q; else delay;
#else
    	if (p[23] == 1) p = -p; else delay;
	if (q[23] == 1) q = -q; else delay;
#endif
	p = (p<q) ? q : p; // This is done to get the best visual frequency result
	 
	if (display_log)
	{

		bit=126;
#if HARDWARE_MULTIPLY
		while ((p[15] == 0) && (bit != 0))
#else
		while ((p[21] == 0) && (bit != 0))
#endif
			par
			{
				p = p<<1;
				bit = bit - 18;
			}
		old_value = fft_info[i<-7];
		tmp = ((0@old_value) + (0@bit))>>1;
		fft_info[i<-7] = (old_value <= (tmp<-7)) ? (tmp<-7) : old_value-1;
	} 
	else 
	{
		old_value = fft_info[i<-7];
#if HARDWARE_MULTIPLY
		fft_info[i<-7] = (old_value<=(unsigned)(p[15:9])) ? (unsigned)(p[15:9]) : old_value-1;
#else
		fft_info[i<-7] = (old_value<=(unsigned)(p[21:15])) ? (unsigned)(p[21:15]) : old_value-1;
#endif
	}
  }

  // add DC component again before inverse FFT calculation is performed

  p = real.read[0] + DC_COMPONENT; 
  real.rwrite[0] = p;
}