////////////////////////////////////////////////////////////////////////////////////////
// File   : idct.c  
// Date   : octobre 2015
// author : Alain Greiner
////////////////////////////////////////////////////////////////////////////////////////
// This file define the code of the IDCT (Inverse Discrete Cosinus Transform) thread
// for the MJPEG application.
// It read blocks of 8*8 pixels (one int32_t per pixel) from the <in> MWMR channel.
// It write blocks of 8*8 pixels (one uint8_t per pixel) to the <out> MWMR channel.
////////////////////////////////////////////////////////////////////////////////////////

#include <mwmr_channel.h>
#include <stdio.h>
#include <stdint.h>
#include "mjpeg.h"

#define   INT_MAX   ((int32_t)0x7fffffff)
#define   INT_MIN   ((int32_t)0x80000000)

// macro to use a shared TTY
#define PRINTF(...)    lock_acquire( &tty_lock ); \
                       giet_tty_printf(__VA_ARGS__);  \
                       lock_release( &tty_lock );

// Useful constants 

/* ck = cos(k*pi/16) = s8-k = sin((8-k)*pi/16) times 1 << C_BITS and rounded */
#define c0_1  16384
#define c0_s2 23170
#define c1_1  16069
#define c1_s2 22725
#define c2_1  15137
#define c2_s2 21407
#define c3_1  13623
#define c3_s2 19266
#define c4_1  11585
#define c4_s2 16384
#define c5_1  9102
#define c5_s2 12873
#define c6_1  6270
#define c6_s2 8867
#define c7_1  3196
#define c7_s2 4520
#define c8_1  0
#define c8_s2 0
#define sqrt2 c0_s2

// The number of bits of accuracy in all (signed) integer operations

#define ARITH_BITS      16

// The minimum signed integer value that fits in ARITH_BITS
#define ARITH_MIN       (-1 << (ARITH_BITS-1))
#define ARITH_MAX       (~ARITH_MIN)

// The number of bits coefficients are scaled up before 2-D idct
#define S_BITS           3

// The number of bits in the fractional part of a fixed point constant
#define C_BITS          14

// This version is vital in passing overall mean error test.
#define descale(x, n) (((x) + (1 << ((n) - 1)) - ((x) < 0)) >> (n))

static const int32_t COS[2][8] = 
{
   {c0_1 , c1_1 , c2_1 , c3_1 , c4_1 , c5_1 , c6_1 , c7_1 },
   {c0_s2, c1_s2, c2_s2, c3_s2, c4_s2, c5_s2, c6_s2, c7_s2}
};

////////////////////////////////////
static inline void rot( int32_t   f, 
                        int32_t   k,
                        int32_t   x, 
                        int32_t   y,
                        int32_t*  rx,
                        int32_t*  ry )
{
#define Cos(k)  COS[f][k]
#define Sin(k)  Cos(8-k)
   *rx = (Cos(k) * x - Sin(k) * y) >> C_BITS;
   //    r = (r + (1 << (C_BITS - 1))) >> C_BITS;
   *ry = (Sin(k) * x + Cos(k) * y) >> C_BITS;
   //    r = (r + (1 << (C_BITS - 1))) >> C_BITS;
#undef Cos
#undef Sin
}

/* Butterfly: but(a,b,x,y) = rot(sqrt(2),4,a,b,x,y) */
#define but(a,b,x,y)   do { x = a - b; y = a + b; } while(0)

// Inverse 1-D Discrete Cosine Transform.
// Result Y is scaled up by factor sqrt(8).
// Original Loeffler algorithm.
static inline void idct_1d( int32_t* Y )
{
   int32_t z1[8], z2[8], z3[8];

   /* Stage 1: */
   but(Y[0], Y[4], z1[1], z1[0]);
   rot(1, 6, Y[2], Y[6], &z1[2], &z1[3]);
   but(Y[1], Y[7], z1[4], z1[7]);
   z1[5] = (sqrt2 * Y[3]) >> C_BITS;
   //    r = (r + (1 << (C_BITS - 1))) >> C_BITS;
   z1[6] = (sqrt2 * Y[5]) >> C_BITS;
   //    r = (r + (1 << (C_BITS - 1))) >> C_BITS;

   /* Stage 2: */
   but(z1[0], z1[3], z2[3], z2[0]);
   but(z1[1], z1[2], z2[2], z2[1]);
   but(z1[4], z1[6], z2[6], z2[4]);
   but(z1[7], z1[5], z2[5], z2[7]);

   /* Stage 3: */
   z3[0] = z2[0];
   z3[1] = z2[1];
   z3[2] = z2[2];
   z3[3] = z2[3];
   rot(0, 3, z2[4], z2[7], &z3[4], &z3[7]);
   rot(0, 1, z2[5], z2[6], &z3[5], &z3[6]);

   /* Final stage 4: */
   but(z3[0], z3[7], Y[7], Y[0]);
   but(z3[1], z3[6], Y[6], Y[1]);
   but(z3[2], z3[5], Y[5], Y[2]);
   but(z3[3], z3[4], Y[4], Y[3]);
}

//////////////////////////////////////////////////////////////
__attribute__ ((constructor)) void idct( unsigned int index )
//////////////////////////////////////////////////////////////
{
    mwmr_channel_t* input  = iqzz_2_idct[index];
    mwmr_channel_t* output = idct_2_libu[index];

    int32_t row;
    int32_t column;
    int32_t block;

    int32_t  bin[64];
    uint8_t  bout[64];
    int32_t  Y[64];

    // get platform parameters
    uint32_t  x_size;
    uint32_t  y_size;
    uint32_t  nprocs;
    giet_procs_number( &x_size , &y_size , &nprocs );

    // get processor coordinates
    unsigned int x , y , p;
    giet_proc_xyp( &x ,&y , &p );

    // private TTY allocation
    // giet_tty_alloc( 0 );

    PRINTF("\n[MJPEG] thread IDCT[%d] starts on P[%d,%d,%d] / trdid = %x\n",
           index , x , y , p, (uint32_t)trdid_idct[index] )


    uint32_t  image = index;
    uint32_t  nblocks = nblocks_h * nblocks_w;

    while( image < MAX_IMAGES )   // one image per iteration
    {
        for ( block = 0 ; block < nblocks ; block++ )
        {
            uint32_t begin;

            // read obe block of coefficients (4 bytes per pixel)
            mwmr_read( input, (uint32_t*)bin , 64 );

            for ( row = 0; row < 8 ; row++ )
            {
                for ( column = 0 ; column < 8 ; column++ )
                {
                    Y[row * 8 + column] = bin[row * 8 + column] << S_BITS;
                }

                idct_1d( &Y[8*row] );

                // Result Y is scaled up by factor sqrt(8)*2^S_BITS.
            }

            for ( column = 0 ; column < 8 ; column++ )
            {
                int32_t Yc[8];

                for ( row = 0 ; row < 8 ; row++ )
                {
                    Yc[row] = Y[8 * row + column];
                }

                idct_1d( Yc );
       
                for ( row = 0 ; row < 8 ; row++ ) 
                {
                    // Result is once more scaled up by a factor sqrt(8). 
                    int32_t r = 128 + descale(Yc[row], 2 * S_BITS);

                    // Clip to 8 bits unsigned
                    r = r > 0 ? (r < 255 ? r : 255) : 0;

                    bout[8*row+column] = r;

                    giet_pthread_assert( ((r & 0xFF) == r ) ,
                    "ERROR in idct() : pixel overflow" ); 
                }
            }

            // write one block to output MWMR channel (one byte per pixel)
            mwmr_write( output, (uint32_t*)bout , 16 );

#if (DEBUG_IDCT > 1)
if ( (index == DEBUG_CLUSTER_INDEX) || (DEBUG_CLUSTER_INDEX == 0XFFFFFFFF) )
 PRINTF("\nIDCT[%d] completes block %d/%d in image %d\n" 
         "  %x  %x  %x  %x  %x  %x  %x  %x\n"
         "  %x  %x  %x  %x  %x  %x  %x  %x\n"
         "  %x  %x  %x  %x  %x  %x  %x  %x\n"
         "  %x  %x  %x  %x  %x  %x  %x  %x\n"
         "  %x  %x  %x  %x  %x  %x  %x  %x\n"
         "  %x  %x  %x  %x  %x  %x  %x  %x\n"
         "  %x  %x  %x  %x  %x  %x  %x  %x\n"
         "  %x  %x  %x  %x  %x  %x  %x  %x\n",
         index , block , nblocks , image ,
         bout[0] , bout[1] , bout[2] , bout[3] , bout[4] , bout[5] , bout[6] , bout[7] ,
         bout[8] , bout[9] , bout[10], bout[11], bout[12], bout[13], bout[14], bout[15],
         bout[16], bout[17], bout[18], bout[19], bout[20], bout[21], bout[22], bout[23],
         bout[24], bout[25], bout[26], bout[27], bout[28], bout[29], bout[30], bout[31],
         bout[32], bout[33], bout[34], bout[35], bout[36], bout[37], bout[38], bout[39],
         bout[40], bout[41], bout[42], bout[43], bout[44], bout[45], bout[46], bout[47],
         bout[48], bout[49], bout[50], bout[51], bout[52], bout[53], bout[54], bout[55],
         bout[56], bout[57], bout[58], bout[59], bout[60], bout[61], bout[62], bout[63]) }
}
#endif
        }  // end for blocks

#if DEBUG_IDCT
if ( (index == DEBUG_CLUSTER_INDEX) || (DEBUG_CLUSTER_INDEX == 0XFFFFFFFF) )
{ PRINTF("\nIDCT[%d] completes image %d at cycle %d\n", index , image , giet_proctime() ) }
#endif

        image = image + x_size*y_size;

    }  // end while (1) on images

    giet_pthread_exit( "IDCT completed" );

}  // end idct()