/*
   TI Davinci driver - C64X+ DSP Library

   (c) Copyright 2008  directfb.org
   (c) Copyright 2007  Telio AG

   Written by Denis Oliver Kropp <dok@directfb.org> and
              Olaf Dreesen <olaf@directfb.org>.

   All rights reserved.

   This library is free software; you can redistribute it and/or
   modify it under the terms of the GNU General Public License
   version 2 as published by the Free Software Foundation.

   This library is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   General Public License for more details.

   You should have received a copy of the GNU General Public
   License along with this library; if not, write to the
   Free Software Foundation, Inc., 59 Temple Place - Suite 330,
   Boston, MA 02111-1307, USA.
*/

//#define DIRECT_ENABLE_DEBUG

#include <config.h>

#include <asm/types.h>

#include <string.h>
#include <stdio.h>
#include <stdlib.h>
#include <fcntl.h>
#include <sys/ioctl.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <unistd.h>

#include <directfb_util.h>

#include <direct/clock.h>
#include <direct/debug.h>
#include <direct/log.h>
#include <direct/messages.h>
#include <direct/util.h>

#include "davinci_c64x.h"


/**********************************************************************************************************************/

#define C64X_DEVICE  "/dev/c64x"
#define C64X_DEVICE0 "/dev/c64x0"
#define C64X_QLEN    direct_page_align( sizeof(c64xTaskControl) )
#define C64X_MLEN    direct_page_align( 0x2000000 )

__attribute__((noinline))
static void
davinci_c64x_queue_error( DavinciC64x *c64x, const char *msg )
{
     c64xTaskControl *ctl       = c64x->ctl;
     uint32_t         dsp       = ctl->QL_dsp;
     uint32_t         arm       = ctl->QL_arm;
     uint32_t         armp      = (arm-1) & C64X_QUEUE_MASK;
     c64xTask        *dsp_task  = &c64x->QueueL[dsp];
     c64xTask        *arm_task  = &c64x->QueueL[arm];
     c64xTask        *armp_task = &c64x->QueueL[armp];

     D_PERROR( "Davinci/C64X+: %s [DSP %d / %d (%s), ARM %d / %d (%s) <- %d / %d (%s)]\n",
               msg,
               dsp,
               (dsp_task->c64x_function >> 2) & 0x3fff,
               state_names[dsp_task->c64x_function & 3],
               arm,
               (arm_task->c64x_function >> 2) & 0x3fff,
               state_names[arm_task->c64x_function & 3],
               armp,
               (armp_task->c64x_function >> 2) & 0x3fff,
               state_names[armp_task->c64x_function & 3] );
}

/*

1. Idle Case

     ARM                                        ARM
     DSP                                        DSP
   |  .  .  .  .  .  .  .  .  |      |  .  .  .  .  .  .  .  .  |          free = length-1


2. Busy Case (ARM after)

        ARM                                     ARM
     DSP                                  DSP
   |  o  .  .  .  .  .  .  .  |      |  .  o  o  .  .  .  .  .  |          free = length-1 - arm + dsp


3. Busy Case (ARM before)

     ARM                                     ARM
                    DSP                                     DSP
   |  .  .  .  .  .  o  o  o  |      |  o  o  .  .  .  .  .  o  |          free = dsp - arm - 1


4. Full Case (ARM after)

                          ARM
     DSP
   |  o  o  o  o  o  o  o  .  |                                            free = 0


5. Full Case (ARM before)

                    ARM                ARM
                       DSP                DSP
   |  o  o  o  o  o  .  o  o  |      |  .  o  o  o  o  o  o  o  |          free = 0

*/

DFBResult
davinci_c64x_emit_tasks( DavinciC64x          *c64x,
                         DavinciC64xTasks     *tasks,
                         DavinciC64xEmitFlags  flags )
{
     c64xTaskControl *ctl     = c64x->ctl;
     uint32_t         arm     = ctl->QL_arm;
     unsigned int     emitted = 0;
     unsigned int     timeout = 23;

     D_MAGIC_ASSERT( tasks, DavinciC64xTasks );

     while (emitted < tasks->num_tasks) {
          uint32_t dsp = ctl->QL_dsp;
          int      free;

          if (arm == dsp)
               free = C64X_QUEUE_LENGTH - 1;
          else if (arm > dsp)
               free = C64X_QUEUE_LENGTH - 1 - arm + dsp;
          else
               free = dsp - arm - 1;

          if (free) {
               int emit = MIN( free, tasks->num_tasks - emitted );
               int copy = MIN( emit, C64X_QUEUE_LENGTH - arm );

               memcpy( (void*) &c64x->QueueL[arm], (void*) &tasks->tasks[emitted], sizeof(c64xTask) * copy );

               if (copy < emit) {
                    memcpy( (void*) &c64x->QueueL[0], (void*) &tasks->tasks[emitted+copy], sizeof(c64xTask) * (emit - copy) );

                    arm = (emit - copy);
               }
               else
                    arm = (arm + copy) & C64X_QUEUE_MASK;

               mb();

               ctl->QL_arm = arm;

               mb();

               emitted += emit;

               timeout = 23;
          }
          else {
               if (!timeout--) {
                    davinci_c64x_queue_error( c64x, "Emit Timeout!" );
                    return DFB_TIMEOUT;
               }
                    
               usleep( 7000 );
          }
     }

     if (flags & C64X_TEF_RESET)
          tasks->num_tasks = 0;

     return DFB_OK;
}

DFBResult
davinci_c64x_tasks_init( DavinciC64xTasks *tasks,
                         unsigned int      size )
{
     tasks->tasks = D_MALLOC( sizeof(c64xTask) * size );
     if (!tasks->tasks)
          return D_OOM();

     tasks->max_tasks = size;
     tasks->num_tasks = 0;

     D_MAGIC_SET( tasks, DavinciC64xTasks );

     return DFB_OK;
}

DFBResult
davinci_c64x_tasks_destroy( DavinciC64xTasks *tasks )
{
     D_MAGIC_ASSERT( tasks, DavinciC64xTasks );
     D_ASSERT( tasks->tasks != NULL );

     D_FREE( (void*) tasks->tasks );

     tasks->tasks = NULL;

     D_MAGIC_CLEAR( tasks );

     return DFB_OK;
}

DFBResult
davinci_c64x_wait_low( DavinciC64x *c64x )
{
     DFBResult        ret;
     c64xTaskControl *ctl = c64x->ctl;

     while (ctl->QL_dsp != ctl->QL_arm) {
          c64xTask *task = c64x_get_task( c64x );

          task->c64x_function = C64X_FLAG_TODO | C64X_FLAG_INTERRUPT;

          c64x_submit_task( c64x, task );

          if (ioctl( c64x->fd, C64X_IOCTL_WAIT_LOW )) {
               c64xTask *dsp_task = &c64x->QueueL[ctl->QL_dsp];

               ret = errno2result( errno );
               D_PERROR( "Davinci/C64X+: C64X_IOCTL_WAIT_LOW failed! [DSP %d / %d (%s), ARM %d / %d (%s)]\n",
                         ctl->QL_dsp,
                         (dsp_task->c64x_function >> 2) & 0x3fff,
                         state_names[dsp_task->c64x_function & 3],
                         ctl->QL_arm,
                         (task->c64x_function >> 2) & 0x3fff,
                         state_names[task->c64x_function & 3] );
               return ret;
          }
     }

     return DFB_OK;
}

/**********************************************************************************************************************/
/*   Benchmarking or Testing                                                                                          */
/**********************************************************************************************************************/

#if 1
#define BRINTF(x...)     do { direct_log_printf( NULL, x ); } while (0)
#else
#define BRINTF(x...)     printf( x )
#endif

static void
bench_mem( const char *name,
           void       *ptr,
           int         length,
           bool        copy,
           bool        from )
{
     int       i, num;
     long long t1, t2, dt, total;
     char      buf[0x100];

     if (length > sizeof(buf))
          length = sizeof(buf);

     num = 0x2000000 / length;

     t1 = direct_clock_get_abs_micros();

     if (copy) {
          if (from)
               for (i=0; i<num; i++)
                    memcpy( buf, ptr, length );
          else
               for (i=0; i<num; i++)
                    memcpy( ptr, buf, length );
     }
     else
          for (i=0; i<num; i++)
               memset( ptr, 0, length );

     t2 = direct_clock_get_abs_micros();

     dt    = t2 - t1;
     total = i * length;

     D_INFO( "Davinci/C64X: MEMORY BENCHMARK on %-7s - %-15s   %lld.%03lld MB/sec\n",
             name, copy ? from ? "memcpy() from" : "memcpy() to" : "memset()",
             total / dt, (total * 1000 / dt) % 1000 );
}


/* insert idct code for testing here */


#define DVA_BLOCK_WORD( val, index, EOB )    (((val) << 16) | (((index)&0x3f) << 1) | ((EOB) ? 1 : 0))

static inline void
test_load_block( DavinciC64x *c64x, bool dct_type_interlaced )
{
     int    i;
     int    num = 0;
     short *dst = c64x->mem + 0x01000000;
     int   *src = c64x->mem + 0x01100000;


#if 0
     src[num++] = DVA_BLOCK_WORD( 100, 0, 1 );
     src[num++] = DVA_BLOCK_WORD( 200, 0, 0 );
     src[num++] = DVA_BLOCK_WORD( 210, 1, 0 );
     src[num++] = DVA_BLOCK_WORD( 220, 2, 1 );
     src[num++] = DVA_BLOCK_WORD( 300, 0, 1 );
     src[num++] = DVA_BLOCK_WORD( 400, 0, 0 );
     src[num++] = DVA_BLOCK_WORD( 410, 1, 1 );
     src[num++] = DVA_BLOCK_WORD( 500, 0, 0 );
     src[num++] = DVA_BLOCK_WORD( 510, 63, 1 );
     src[num++] = DVA_BLOCK_WORD( 600, 63, 1 );
#else
     src[num++] = DVA_BLOCK_WORD(136, 0, 0);
     src[num++] = DVA_BLOCK_WORD(-12, 8, 0);
     src[num++] = DVA_BLOCK_WORD(7, 16, 0);
     src[num++] = DVA_BLOCK_WORD(-2, 24, 1);

     src[num++] = DVA_BLOCK_WORD(136, 0, 0);
     src[num++] = DVA_BLOCK_WORD(-12, 8, 0);
     src[num++] = DVA_BLOCK_WORD(7, 16, 0);
     src[num++] = DVA_BLOCK_WORD(-2, 24, 1);


     src[num++] = DVA_BLOCK_WORD(1076, 0, 0);
     src[num++] = DVA_BLOCK_WORD(-204, 8, 0);
     src[num++] = DVA_BLOCK_WORD(-168, 16, 0);
     src[num++] = DVA_BLOCK_WORD(-129, 24, 0);
     src[num++] = DVA_BLOCK_WORD(-100, 32, 0);
     src[num++] = DVA_BLOCK_WORD(-40, 40, 0);
     src[num++] = DVA_BLOCK_WORD(-14, 48, 1);
#if 1
     src[num++] = DVA_BLOCK_WORD(1068, 0, 0);
     src[num++] = DVA_BLOCK_WORD(2, 1, 0);
     src[num++] = DVA_BLOCK_WORD(-202, 8, 0);
     src[num++] = DVA_BLOCK_WORD(-168, 16, 0);
     src[num++] = DVA_BLOCK_WORD(-2, 9, 0);
     src[num++] = DVA_BLOCK_WORD(-129, 24, 0);
     src[num++] = DVA_BLOCK_WORD(-97, 32, 0);
     src[num++] = DVA_BLOCK_WORD(-40, 40, 0);
     src[num++] = DVA_BLOCK_WORD(-13, 48, 1);
#else
     src[num++] = DVA_BLOCK_WORD(1068, 0, 0);
//     src[num++] = DVA_BLOCK_WORD(2, 1, 0);
     src[num++] = DVA_BLOCK_WORD(-202, 8, 0);
     src[num++] = DVA_BLOCK_WORD(-1, 16, 0);
//     src[num++] = DVA_BLOCK_WORD(-2, 9, 0);
     src[num++] = DVA_BLOCK_WORD(-1, 24, 0);
     src[num++] = DVA_BLOCK_WORD(-97, 32, 1);
//     src[num++] = DVA_BLOCK_WORD(-40, 40, 0);
//     src[num++] = DVA_BLOCK_WORD(-13, 48, 1);
#endif

     src[num++] = DVA_BLOCK_WORD(1048, 0, 0);
     src[num++] = DVA_BLOCK_WORD(-26, 8, 0);
     src[num++] = DVA_BLOCK_WORD(4, 16, 0);
     src[num++] = DVA_BLOCK_WORD(5, 24, 0);
     src[num++] = DVA_BLOCK_WORD(-4, 32, 1);

     src[num++] = DVA_BLOCK_WORD(996, 0, 0);
     src[num++] = DVA_BLOCK_WORD(24, 8, 0);
     src[num++] = DVA_BLOCK_WORD(-2, 24, 0);
     src[num++] = DVA_BLOCK_WORD(3, 32, 0);
     src[num++] = DVA_BLOCK_WORD(-4, 48, 1);
#endif

     BRINTF("\n");
     BRINTF("\n\n.======================== Testing load_block (dct_type_interlaced: %s) ========================.\n",
            dct_type_interlaced ? "yes" : "no");
     BRINTF("\n");
     BRINTF( "SOURCE (DVABlockWords)\n" );
     BRINTF("\n");

     for (i=0; i<num; i++)
          BRINTF("0x%08x  (%d, %d, %d)\n", (u32)src[i], src[i] >> 16, (src[i] >> 1) & 0x3f, src[i] & 1);

     BRINTF("\n\n");


     memset( dst, 0x55, 0x100000 );


     // test routine
     davinci_c64x_load_block( c64x, DAVINCI_C64X_MEM+0x01100000, 10, dct_type_interlaced ? 0x7f : 0x3f );

     // copy idct buffer to memory where we can read it
     davinci_c64x_blit_16( c64x, DAVINCI_C64X_MEM+0x01000000, 0, 0xf065c0, 0, 16 * 24, 1 );

     davinci_c64x_write_back_all( c64x );
     davinci_c64x_write_back_all( c64x );
     davinci_c64x_wait_low( c64x );


     BRINTF( "-> IDCT BUFFER (16x16 + [ 8x8 8x8 ] shorts)\n" );
     BRINTF("\n");

     for (i=0; i<16*24; i++) {
          BRINTF("%5d ", dst[i] );
          if ((i&15)==15) {
               BRINTF("\n");
          }
          if ((i&255)==255) {
               BRINTF("\n");
          }
     }

     BRINTF("\n\n");

#if 1
     s16 *blocks = c64x->mem + 0x01200000;
     int offset  = 0;

     memset( blocks, 0, 1024 );

     for (i=0; i<num; i++) {
          blocks[offset + ((src[i] >> 1) & 0x3f)] = src[i] >> 16;

          if (src[i] & 1)
               offset += 64;
     }

	memset( dst, 0x55, 0x100000 );

	// test routine
	for (i=0; i<6; i++)
		davinci_c64x_dva_idct( c64x, DAVINCI_C64X_MEM+0x01000000 + i*128, 16, DAVINCI_C64X_MEM+0x01200000 + i*128 );

	davinci_c64x_write_back_all( c64x );
	davinci_c64x_write_back_all( c64x );
	davinci_c64x_wait_low( c64x );


     BRINTF( "-> SINGLE IDCT (59) BLOCKS (6x 8x8 shorts)\n" );
     BRINTF("\n");

     for (i=0; i<6*64; i++) {
          BRINTF("%5d ", dst[i] );
          if ((i&7)==7) {
               BRINTF("\n");
          }
          if ((i&63)==63) {
               BRINTF("\n");
          }
     }

     BRINTF("\n\n");
#endif

#if 0
//     s16 blocks[384];
//     int offset = 0;
     offset = 0;

     memset( blocks, 0, 1024 );

     for (i=0; i<num; i++) {
          blocks[offset + ((src[i] >> 1) & 0x3f)] = src[i] >> 16;

          if (src[i] & 1) {
               int n;

               for (n = 0; n < 8; n++)
                    idct_row (blocks + offset + 8 * n);

               for (n = 0; n < 8; n++)
                    idct_col (blocks + offset + n);

               offset += 64;
          }
     }

     BRINTF( "-> REFERENCE IDCT BLOCKS (6x 8x8 shorts)\n" );
     BRINTF("\n");

     for (i=0; i<6*64; i++) {
          BRINTF("%5d ", blocks[i] );
          if ((i&7)==7) {
               BRINTF("\n");
          }
          if ((i&63)==63) {
               BRINTF("\n");
          }
     }

     BRINTF("\n\n");
#endif
}

static inline void
bench_dezigzag( DavinciC64x *c64x )
{
     int       i, num;
     long long t1, t2, dt, total;
     //int       length = 0x10000;

     num = 0x200000;// / length;

     short *p = c64x->mem + 0x1000000;

     for (i=0; i<64; i++) {
          p[i] = i;
          BRINTF("%3d ", p[i]);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     t1 = direct_clock_get_abs_micros();

     for (i=0; i<num; i++) {
          c64xTask *task = c64x_get_task( c64x );

          task->c64x_function = C64X_DEZIGZAG | C64X_FLAG_TODO;

          task->c64x_arg[0] = (DAVINCI_C64X_MEM+0x01000000)+0x200000;
          task->c64x_arg[1] = (DAVINCI_C64X_MEM+0x01000000)+0x000000;
          //task->c64x_arg[2] = length/4;

          c64x_submit_task( c64x, task );
     }

     davinci_c64x_wait_low( c64x );

     t2 = direct_clock_get_abs_micros();

     p = c64x->mem + 0x1200000;
     for (i=0; i<64; i++) {
          BRINTF("%3d ", p[i]);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     dt    = t2 - t1;
     total = num;// * length;

     D_INFO( "Davinci/C64X: BENCHMARK on DSP - %-15s   %lld Calls/sec\n",
             "de_zigzag()", total * 1000000ULL / dt );
}

#define DUMP_PIXELS 1

static inline void
bench_blend_argb( DavinciC64x *c64x, int sub )
{
     int       i, num;
     long long t1, t2, dt, total;

     num = 1;//0x20000;

     u32 *src = c64x->mem + 0x1000000;
     u32 *dst = c64x->mem + 0x1200000;

     BRINTF( "\nTESTING BLEND_32 SUB %d\n", sub );

     BRINTF( "\nSOURCE               " );

     for (i=0; i<DUMP_PIXELS; i++) {
          src[i] = (i << 26) | ((i & 0x30) << 20) | (i * 0x010204 + 3);

          if (!i)
               src[i] = 0xc0c08001;

          BRINTF("%02x %02x %02x %02x  ", src[i] >> 24, (src[i] >> 16) & 0xff, (src[i] >> 8) & 0xff, src[i] & 0xff);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     BRINTF( "\nDESTINATION          " );

     for (i=0; i<DUMP_PIXELS; i++) {
          dst[i] = i * 0x04040404;

          if (!i)
               dst[i] = 0xe0e0e0e0;

          BRINTF("%02x %02x %02x %02x  ", dst[i] >> 24, (dst[i] >> 16) & 0xff, (dst[i] >> 8) & 0xff, dst[i] & 0xff);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     t1 = direct_clock_get_abs_micros();

     for (i=0; i<num; i++) {
          c64xTask *task = c64x_get_task( c64x );

          task->c64x_function = C64X_BLEND_32 | C64X_FLAG_TODO | (sub << 16);

          task->c64x_arg[0] = (DAVINCI_C64X_MEM+0x01000000)+0x200000;
          task->c64x_arg[1] = 32;
          task->c64x_arg[2] = (DAVINCI_C64X_MEM+0x01000000)+0x000000;
          task->c64x_arg[3] = 32;
          task->c64x_arg[4] = 8;
          task->c64x_arg[5] = 8;
          task->c64x_arg[6] = 0x80;

          c64x_submit_task( c64x, task );
     }

     davinci_c64x_wait_low( c64x );

     t2 = direct_clock_get_abs_micros();

     BRINTF( "\n\nDESTINATION (AFTER)  " );

     for (i=0; i<DUMP_PIXELS; i++) {
          BRINTF("%02x %02x %02x %02x  ", dst[i] >> 24, (dst[i] >> 16) & 0xff, (dst[i] >> 8) & 0xff, dst[i] & 0xff);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     BRINTF("\n\n");

     dt    = t2 - t1;
     total = num;

     D_INFO( "Davinci/C64X: BENCHMARK on DSP - %-15s   %lld Calls/sec\n",
             "blend_32(8x8)", total * 1000000ULL / dt );
}

static inline void
bench_fetch_uyvy( DavinciC64x *c64x, bool interleave, int xoff, int yoff ) {
  int       i, x, y, num=1;
  long long t1, t2, dt, total;
  u8 *yuv = c64x->mem + 0x1000000;
  u8 *src = c64x->mem + 0x1200000;

  BRINTF("\n\n\n.======================== Testing fetch_uyvy (inter %d, xoff %d, yoff %d) ========================.\n\n",
	interleave, xoff, yoff);

  for (y=0; y<20; y++) {
    for (x=0; x<40; x++) {
      int val = (y*40)+x;
      src[y*1440 + x] = val;
      BRINTF("%02x ", val&0xff);
    }
    BRINTF("\n");
  }
  BRINTF("\n");

  memset( yuv, 0xAA, 0x100000 );

  t1 = direct_clock_get_abs_micros();

  for (i=0; i<num; i++) {
    c64xTask *task = c64x_get_task( c64x );

    task->c64x_arg[0] = (DAVINCI_C64X_MEM+0x01000000)+0x000000;
    task->c64x_arg[1] = (DAVINCI_C64X_MEM+0x01000000)+0x200000 + yoff*1440 + xoff * 2;
    task->c64x_arg[2] = 1440;

    task->c64x_function = (21 << 2) | C64X_FLAG_TODO;

    c64x_submit_task( c64x, task );
  }

  davinci_c64x_write_back_all( c64x );
  davinci_c64x_write_back_all( c64x );
  davinci_c64x_wait_low( c64x );

  t2 = direct_clock_get_abs_micros();

  BRINTF( "\n\nDESTINATION\n\nY:\n" );
  for (y=0;y<27;y++) {
    if (y==18) BRINTF("\nUV:\n");
    for (x=0;x<32;x++) {
      BRINTF("%02x ",yuv[y*32+x]);
    }
    BRINTF("\n");
  }

  dt    = t2 - t1;
  total = num;

  D_INFO("\n\nDavinci/C64X: BENCHMARK on DSP - %-15s   %lld Calls/sec\n",
	"blend_fetch_uyvy(16x16)", total * 1000000ULL / dt );
}
#if 0
static inline void
bench_fetch_uyvy( DavinciC64x *c64x, bool interleave, int xoff, int yoff )
{
     int       i, x, y, num;
     long long t1, t2, dt, total;

     num = 1;//0x20000;

     u8 *yuv = c64x->mem + 0x1000000;
     u8 *src = c64x->mem + 0x1200000;

     BRINTF("\n");
     BRINTF("\n\n.======================== Testing fetch_uyvy (inter %d, xoff %d, yoff %d) ========================.\n",
            interleave, xoff, yoff);
     BRINTF("\n");
     BRINTF( "SOURCE (20x20)\n" );

     for (y=0; y<20; y++) {
          for (x=0; x<40; x++) {
               int val = (x & 1) ? (x * 4 + y*0x10) : (x/4 + 0x40 + (x&2) * 0x10 + y*0x08);

               src[y*1440 + x] = val;

               BRINTF("%02x ", val&0xff);
          }

          BRINTF("\n");
     }

     BRINTF("\n");

     memset( yuv, 0x55, 0x100000 );


     t1 = direct_clock_get_abs_micros();

    for (i=0; i<num; i++) {
	c64xTask *task = c64x_get_task( c64x );

	task->c64x_function = (19 << 2) | C64X_FLAG_TODO;

	task->c64x_arg[0] = (DAVINCI_C64X_MEM+0x01000000)+0x000000;
	task->c64x_arg[1] = (DAVINCI_C64X_MEM+0x01000000)+0x200000 + yoff*1440 + xoff * 2;
	task->c64x_arg[2] = 1440;
	task->c64x_arg[3] = 16;
	task->c64x_arg[4] = interleave ? 1 : 0;

	c64x_submit_task( c64x, task );
    }

    davinci_c64x_write_back_all( c64x );
    davinci_c64x_wait_low( c64x );

    t2 = direct_clock_get_abs_micros();


     BRINTF( "\n\nDESTINATION (17x18 / [9x9 9x9])\n" );

     for (y=0; y<18; y++) {
          for (x=0; x<17; x++) {
               BRINTF("%02x ", yuv[y*32 + x]);
          }

          BRINTF("\n");
     }

     BRINTF("\n");

     for (y=0; y<9; y++) {
          for (x=0; x<9; x++) {
               BRINTF("%02x ", yuv[y*32 + x + 32*18]);
          }

          BRINTF("\n");
     }

     BRINTF("\n");

     for (y=0; y<9; y++) {
          for (x=0; x<9; x++) {
               BRINTF("%02x ", yuv[y*32 + x + 32*18+16]);
          }

          BRINTF("\n");
     }

     BRINTF("\n");

     BRINTF("\n\n");

     dt    = t2 - t1;
     total = num;

     D_INFO( "Davinci/C64X: BENCHMARK on DSP - %-15s   %lld Calls/sec\n",
             "blend_fetch_uyvy(16x16)", total * 1000000ULL / dt );
}
#endif

#if 0
static inline void
bench_put_idct( DavinciC64x *c64x, int dct_type )
{
     int       i, num;
     long long t1, t2, dt, total;
     //int       length = 0x10000;

     num = 0x10000;// / length;

     u8  *dst = c64x->mem + 0x01000000;
     int *src = c64x->mem + 0x01200000;

     src[0] = DVA_BLOCK_WORD( 100, 0, 1 );
     src[1] = DVA_BLOCK_WORD( 200, 0, 0 );
     src[2] = DVA_BLOCK_WORD( 210, 1, 0 );
     src[3] = DVA_BLOCK_WORD( 220, 2, 1 );
     src[4] = DVA_BLOCK_WORD( 300, 0, 1 );
     src[5] = DVA_BLOCK_WORD( 400, 0, 0 );
     src[6] = DVA_BLOCK_WORD( 410, 1, 1 );
     src[7] = DVA_BLOCK_WORD( 500, 0, 0 );
     src[8] = DVA_BLOCK_WORD( 510, 63, 1 );
     src[9] = DVA_BLOCK_WORD( 600, 63, 1 );

     BRINTF("\n");
     BRINTF("\n\n.======================== Testing put_idct (%d) ========================.\n", dct_type);
     BRINTF("\n");

     memset( dst, 0x55, 0x100000 );

     for (i=0; i<10; i++) {
          BRINTF("0x%08x  (%d, %d, %d)\n", (u32)src[i], src[i] >> 16, (src[i] >> 1) & 0x3f, src[i] & 1);
     }

     BRINTF("\n");

     t1 = direct_clock_get_abs_micros();

     {
          c64xTask *task = c64x_get_task( c64x );

          task->c64x_function = C64X_LOAD_BLOCK | C64X_FLAG_TODO;

          task->c64x_arg[0] = DAVINCI_C64X_MEM+0x1200000;
          task->c64x_arg[1] = 10;
          task->c64x_arg[2] = 0x3f;

          c64x_submit_task( c64x, task );
     }

     davinci_c64x_blit_16( c64x, (DAVINCI_C64X_MEM+0x01000000), 0, 0xf06180, 0, 384, 1 );
     davinci_c64x_blit_16( c64x, (DAVINCI_C64X_MEM+0x01100000), 0, 0xf06480, 0, 384/2, 1 );

     davinci_c64x_put_uyvy_16x16( c64x, (DAVINCI_C64X_MEM+0x01300000), 32, 0xf06180, 0 );

     davinci_c64x_write_back_all( c64x );
     davinci_c64x_wait_low( c64x );

     t2 = direct_clock_get_abs_micros();


     for (i=0; i<384; i++) {
          BRINTF("%5d ", dst[i] );
          if (i%8==7) {
               BRINTF("\n");
          }
          if (i%64==63) {
               BRINTF("\n");
          }
     }

     BRINTF("\n\n");


     for (i=0; i<384; i++) {
          BRINTF("%3d ", duv[i] );
          if (i%8==7) {
               BRINTF("\n");
          }
          if (i%64==63) {
               BRINTF("\n");
          }
     }

     BRINTF("\n\n");

     for (i=0; i<16*16*2; i++) {
          BRINTF("%02x ", duy[i]);

          if (i%32==31) {
               BRINTF("\n");
          }
     }

     BRINTF("\n");

     dt    = t2 - t1;
     total = num;// * length;

     D_INFO( "Davinci/C64X: BENCHMARK on DSP - %-15s   %lld Calls/sec\n",
             "block_load()", total * 1000000ULL / dt );
}
#endif

static inline void
bench_put_mc( DavinciC64x *c64x, bool interleave )
{
     int       x, y, i, num;
     long long t1, t2, dt, total;

     num = 1;//720/16*576/16;

     u8 *dst = c64x->mem + 0x1000000;
     u8 *src = c64x->mem + 0x1200000;

     BRINTF("\n");
     BRINTF("\n\n.======================== Testing put_mc (%d) ========================.\n", interleave);
     BRINTF("\n");
     BRINTF("SOURCE (16x16 / [8x8 8x8]\n");

     for (y=0; y<16; y++) {
          for (x=0; x<16; x++) {
               u8 val = (x << 4) + y;
               src[y*16 + x] = val;
               BRINTF("%02x ", val);
          }

          BRINTF("\n");
     }

     BRINTF("\n");

     for (y=0; y<8; y++) {
          for (x=0; x<8; x++) {
               u8 val = (x << 4) + y*2;
               src[y*16 + x + 16*16] = val;
               BRINTF("%02x ", val);
          }

          BRINTF("\n");
     }

     BRINTF("\n");

     for (y=0; y<8; y++) {
          for (x=0; x<8; x++) {
               u8 val = (x << 4) + y*2;
               src[y*16 + x + 16*16 + 8] = val;
               BRINTF("%02x ", val);
          }

          BRINTF("\n");
     }

     BRINTF("\n");

     memset( dst, 0x55, 0x100000 );

     davinci_c64x_blit_32( c64x, C64X_MC_BUFFER_Y, 16, DAVINCI_C64X_MEM+0x1200000, 16, 4, 24 );
     davinci_c64x_write_back_all( c64x );
     davinci_c64x_wait_low( c64x );

     t1 = direct_clock_get_abs_micros();

     for (i=0; i<num; i++) {
          c64xTask *task = c64x_get_task( c64x );

          task->c64x_function = C64X_PUT_MC_UYVY_16x16 | C64X_FLAG_TODO;

          task->c64x_arg[0] = DAVINCI_C64X_MEM+0x1000000;
          task->c64x_arg[1] = 1440;
          task->c64x_arg[2] = interleave ? 1 : 0;

          c64x_submit_task( c64x, task );
     }

     davinci_c64x_write_back_all( c64x );
     davinci_c64x_wait_low( c64x );

     t2 = direct_clock_get_abs_micros();


     BRINTF("\n");
     BRINTF("DESTINATION (16x16 UYVY)\n");

     for (y=0; y<16; y++) {
          for (x=0; x<32; x++)
               BRINTF("%02x ", dst[y*1440 + x]);

          BRINTF("\n");
     }

     BRINTF("\n\n");

     dt    = t2 - t1;
     total = num;

     D_INFO( "Davinci/C64X: BENCHMARK on DSP - %-15s   %lld Calls/sec\n",
             "put_mc_16x16()", total * 1000000ULL / dt );
}

static inline void
bench_put_sum( DavinciC64x *c64x, bool interleave )
{
     int       x, y, i, num;
     long long t1, t2, dt, total;

     num = 1;//720/16*576/16;

     u8  *dst   = c64x->mem + 0x1000000;
     u8  *src   = c64x->mem + 0x1200000;
     u32 *words = c64x->mem + 0x1100000;

     BRINTF("\n");
     BRINTF("\n\n.======================== Testing put_sum (%d) ========================.\n", interleave);
     BRINTF("\n");
     BRINTF("WORDS (6x IDCT with one value)\n");

     words[0] = DVA_BLOCK_WORD(   0, 0, 1 );
     words[1] = DVA_BLOCK_WORD(  50, 0, 1 );
     words[2] = DVA_BLOCK_WORD( 100, 0, 1 );
     words[3] = DVA_BLOCK_WORD( 150, 0, 1 );
     words[4] = DVA_BLOCK_WORD( 200, 0, 1 );
     words[5] = DVA_BLOCK_WORD( 250, 0, 1 );

     BRINTF("\n");
     BRINTF("\n");

     memset( dst, 0x55, 0x100000 );

     for (i=0; i<6; i++) {
          BRINTF("0x%08x  (%d, %d, %d)\n", (u32)words[i], words[i] >> 16, (words[i] >> 1) & 0x3f, words[i] & 1);
     }

     {
          c64xTask *task = c64x_get_task( c64x );

          task->c64x_function = C64X_LOAD_BLOCK | C64X_FLAG_TODO;

          task->c64x_arg[0] = DAVINCI_C64X_MEM+0x1100000;
          task->c64x_arg[1] = 6;
          task->c64x_arg[2] = 0x3f;

          c64x_submit_task( c64x, task );
     }

     BRINTF("\n");
     BRINTF("SOURCE (16x16 / [8x8 8x8]\n");

     for (y=0; y<16; y++) {
          for (x=0; x<16; x++) {
               u8 val = (x << 4) + y;
               src[y*16 + x] = val;
               BRINTF("%02x ", val);
          }

          BRINTF("\n");
     }

     BRINTF("\n");

     for (y=0; y<8; y++) {
          for (x=0; x<8; x++) {
               u8 val = (x << 4) + y*2;
               src[y*16 + x + 16*16] = val;
               BRINTF("%02x ", val);
          }

          BRINTF("\n");
     }

     BRINTF("\n");

     for (y=0; y<8; y++) {
          for (x=0; x<8; x++) {
               u8 val = (x << 4) + y*2;
               src[y*16 + x + 16*16 + 8] = val;
               BRINTF("%02x ", val);
          }

          BRINTF("\n");
     }

     BRINTF("\n");

     memset( dst, 0x55, 0x100000 );

     davinci_c64x_blit_32( c64x, C64X_MC_BUFFER_Y, 16, DAVINCI_C64X_MEM+0x1200000, 16, 4, 24 );
     davinci_c64x_write_back_all( c64x );
     davinci_c64x_wait_low( c64x );

     t1 = direct_clock_get_abs_micros();

     for (i=0; i<num; i++) {
          c64xTask *task = c64x_get_task( c64x );

          task->c64x_function = C64X_PUT_SUM_UYVY_16x16 | C64X_FLAG_TODO;

          task->c64x_arg[0] = DAVINCI_C64X_MEM+0x1000000;
          task->c64x_arg[1] = 1440;
          task->c64x_arg[2] = interleave ? 1 : 0;

          c64x_submit_task( c64x, task );
     }

     davinci_c64x_write_back_all( c64x );
     davinci_c64x_wait_low( c64x );

     t2 = direct_clock_get_abs_micros();


     BRINTF("\n");
     BRINTF("DESTINATION (16x16 UYVY)\n");

     for (y=0; y<16; y++) {
          for (x=0; x<32; x++)
               BRINTF("%02x ", dst[y*1440 + x]);

          BRINTF("\n");
     }

     BRINTF("\n\n");

     dt    = t2 - t1;
     total = num;

     D_INFO( "Davinci/C64X: BENCHMARK on DSP - %-15s   %lld Calls/sec\n",
             "put_sum_16x16()", total * 1000000ULL / dt );
}

static inline void
bench_sat_mc( DavinciC64x *c64x )
{
     int       x, y, i, num;
     long long t1, t2, dt, total;

     num = 1;//720/16*576/16;

     u8 *dst = c64x->mem + 0x1000000;
     u8 *src = c64x->mem + 0x1200000;

     BRINTF("\n\n.======================== Testing sat_mc ========================.\n");
     BRINTF("\n");
     BRINTF("SOURCE (16x16 / [8x8 8x8]\n");

     for (y=0; y<16; y++) {
          for (x=0; x<16; x++) {
               u8 val = (x << 4) + y;
               src[y*16 + x] = val;
               BRINTF("%02x ", val);
          }

          BRINTF("\n");
     }

     BRINTF("\n");

     for (y=0; y<8; y++) {
          for (x=0; x<8; x++) {
               u8 val = (x << 4) + y*2;
               src[y*16 + x + 16*16] = val;
               BRINTF("%02x ", val);
          }

          BRINTF("\n");
     }

     BRINTF("\n");

     for (y=0; y<8; y++) {
          for (x=0; x<8; x++) {
               u8 val = (x << 4) + y*2;
               src[y*16 + x + 16*16 + 8] = val;
               BRINTF("%02x ", val);
          }

          BRINTF("\n");
     }

     BRINTF("\n");

     memset( dst, 0x55, 0x100000 );

     t1 = direct_clock_get_abs_micros();

     for (i=0; i<num; i++) {
          c64xTask *task = c64x_get_task( c64x );

          task->c64x_function = (57 << 2) | C64X_FLAG_TODO;

          task->c64x_arg[0] = DAVINCI_C64X_MEM+0x1000000;
          task->c64x_arg[1] = DAVINCI_C64X_MEM+0x1200000;
          task->c64x_arg[2] = 16;

          c64x_submit_task( c64x, task );
     }

     davinci_c64x_write_back_all( c64x );
     davinci_c64x_wait_low( c64x );

     t2 = direct_clock_get_abs_micros();


     BRINTF("\n");
     BRINTF("DESTINATION (16x16 / [8x8 8x8]\n");

     for (y=0; y<16; y++) {
          for (x=0; x<16; x++)
               BRINTF("%02x ", dst[y*16 + x]);

          BRINTF("\n");
     }

     BRINTF("\n");

     for (y=0; y<8; y++) {
          for (x=0; x<8; x++)
               BRINTF("%02x ", dst[y*16 + x + 16*16]);

          BRINTF("\n");
     }

     BRINTF("\n");

     for (y=0; y<8; y++) {
          for (x=0; x<8; x++)
               BRINTF("%02x ", dst[y*16 + x + 16*16 + 8]);

          BRINTF("\n");
     }

     BRINTF("\n\n");

     dt    = t2 - t1;
     total = num;

     D_INFO( "Davinci/C64X: BENCHMARK on DSP - %-15s   %lld Calls/sec\n",
             "sat_mc_16x16()", total * 1000000ULL / dt );
}

static inline void
bench_uyvy_1( DavinciC64x *c64x, bool progressive )
{
     c64xTask *task;
     int       i, num;
     long long t1, t2, dt, total;

     num = 720/16*576/16;

     u8 *u = c64x->mem + 0x1200000;
     u8 *p = c64x->mem + 0x1000000;

     BRINTF("\n\n\n.======================== Testing put_uyvy (%s) ========================.\n\n",
            progressive ? "progressive" : "interlaced");

     for (i=0; i<256; i++) {
          p[i] = i - 128;
          BRINTF("Y%-3d ", p[i]);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     for (i=0; i<64; i++) {
          p[256+i] = i-32;
          BRINTF("U%-3d ", p[256+i]);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     for (i=0; i<64; i++) {
          p[320+i] = i-32;
          BRINTF("V%-3d ", p[320+i]);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     BRINTF("\n");

     for (i=0; i<384; i++) {
          BRINTF("%4d ", p[i]);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     BRINTF("\n");

     memset( u, 0x55, 720*576*2 );

     t1 = direct_clock_get_abs_micros();

     for (i=0; i<num; i++) {
          davinci_c64x_dva_begin_frame( c64x, 720 * 2, (DAVINCI_C64X_MEM+0x01000000)+0x200000+i*16*16*2, 0, 0, progressive ? 0x100 : 0 );

          task = c64x_get_task( c64x );

          task->c64x_function = C64X_PUT_UYVY_16x16 | C64X_FLAG_TODO;

          task->c64x_arg[0] = (DAVINCI_C64X_MEM+0x01000000)+0x200000+i*16*16*2;
          task->c64x_arg[1] = 720 * 2;
          task->c64x_arg[2] = (DAVINCI_C64X_MEM+0x01000000);
          task->c64x_arg[3] = 0;

          c64x_submit_task( c64x, task );
     }

     BRINTF("\n");

     davinci_c64x_wait_low( c64x );

     t2 = direct_clock_get_abs_micros();

     for (i=0; i<16*16*2; i++) {
          BRINTF("%02x ", u[i/32*720*2 + i%32]);

          if (i%32==31) {
               BRINTF("\n");
          }
     }

     BRINTF("\n");

     dt    = t2 - t1;
     total = num;

     D_INFO( "Davinci/C64X: BENCHMARK on DSP - %-15s   %lld Calls/sec\n",
             "put_uyvy_16x16()", total * 1000000ULL / dt );
}

static inline void
bench_uyvy_2( DavinciC64x *c64x, bool progressive )
{
     c64xTask *task;
     int       i, num;
     long long t1, t2, dt, total;

     num = 1;//720/16*576/16;

     u8 *u = c64x->mem + 0x0200000;
     u8 *p = c64x->mem + 0x0000000;

     BRINTF("\n\n\n.======================== Testing put_uyvy (%s) ========================.\n\n",
            progressive ? "progressive" : "interlaced");

     for (i=0; i<256; i++) {
          p[i] = i/8;
          BRINTF("Y%-3d ", p[i]);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     for (i=0; i<64; i++) {
          p[256+i] = i/8 + 128;
          BRINTF("U%-3d ", p[256+i]);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     for (i=0; i<64; i++) {
          p[320+i] = i/8 + 240;
          BRINTF("V%-3d ", p[320+i]);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     BRINTF("\n");

     for (i=0; i<384; i++) {
          BRINTF("%4d ", p[i]);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     BRINTF("\n");

     memset( u, 0x55, 720*576*2 );

     t1 = direct_clock_get_abs_micros();

     for (i=0; i<num; i++) {
          davinci_c64x_dva_begin_frame( c64x, 720 * 2, (DAVINCI_C64X_MEM+0x01000000)+0x200000+i*16*16*2, 0, 0, progressive ? 0x100 : 0 );

          task = c64x_get_task( c64x );

          task->c64x_function = C64X_PUT_UYVY_16x16 | C64X_FLAG_TODO;

          task->c64x_arg[0] = DAVINCI_C64X_MEM+0x200000+i*16*16*2;
          task->c64x_arg[1] = 720 * 2;
          task->c64x_arg[2] = DAVINCI_C64X_MEM;
          task->c64x_arg[3] = 0;

          c64x_submit_task( c64x, task );
     }

     BRINTF("\n");

     davinci_c64x_write_back_all( c64x );
     davinci_c64x_wait_low( c64x );

     t2 = direct_clock_get_abs_micros();

     for (i=0; i<16*16*2; i++) {
          BRINTF("%02x ", u[i/32*720*2 + i%32]);

          if (i%32==31) {
               BRINTF("\n");
          }
     }

     BRINTF("\n");

     dt    = t2 - t1;
     total = num;

     D_INFO( "Davinci/C64X: BENCHMARK on DSP - %-15s   %lld Calls/sec\n",
             "put_uyvy_16x16()", total * 1000000ULL / dt );
}

static inline void
bench_uyvy_3( DavinciC64x *c64x, bool progressive )
{
     c64xTask *task;
     int       i, num;
     long long t1, t2, dt, total;

     num = 1;//720/16*576/16;

     u8 *u = c64x->mem + 0x1200000;
     u8 *p = c64x->mem + 0x1000000;

     BRINTF("\n\n\n.======================== Testing put_uyvy (%s) ========================.\n\n",
            progressive ? "progressive" : "interlaced");

     for (i=0; i<256; i++) {
          p[i] = i%8;
          BRINTF("Y%-3d ", p[i]);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     for (i=0; i<64; i++) {
          p[256+i] = i%8 + 128;
          BRINTF("U%-3d ", p[256+i]);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     for (i=0; i<64; i++) {
          p[320+i] = i%8 + 240;
          BRINTF("V%-3d ", p[320+i]);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     BRINTF("\n");

     for (i=0; i<384; i++) {
          BRINTF("%4d ", p[i]);
          if (i%8==7) {
               BRINTF("\n");
          }
     }

     BRINTF("\n");

     memset( u, 0x55, 720*576*2 );

     t1 = direct_clock_get_abs_micros();

     for (i=0; i<num; i++) {
          davinci_c64x_dva_begin_frame( c64x, 720 * 2, (DAVINCI_C64X_MEM+0x01000000)+0x200000+i*16*16*2, 0, 0, progressive ? 0x100 : 0 );

          task = c64x_get_task( c64x );

          task->c64x_function = C64X_PUT_UYVY_16x16 | C64X_FLAG_TODO;

          task->c64x_arg[0] = (DAVINCI_C64X_MEM+0x01000000)+0x200000+i*16*16*2;
          task->c64x_arg[1] = 720 * 2;
          task->c64x_arg[2] = (DAVINCI_C64X_MEM+0x01000000);
          task->c64x_arg[3] = 0;

          c64x_submit_task( c64x, task );
     }

     BRINTF("\n");

     davinci_c64x_wait_low( c64x );

     t2 = direct_clock_get_abs_micros();

     for (i=0; i<16*16*2; i++) {
          BRINTF("%02x ", u[i/32*720*2 + i%32]);

          if (i%32==31) {
               BRINTF("\n");
          }
     }

     BRINTF("\n");

     dt    = t2 - t1;
     total = num;

     D_INFO( "Davinci/C64X: BENCHMARK on DSP - %-15s   %lld Calls/sec\n",
             "put_uyvy_16x16()", total * 1000000ULL / dt );
}

static inline void
bench_mc( DavinciC64x *c64x, int func, int w, int h, bool avg, const char *name )
{
     int       i, x, y, num;
     long long t1, t2, dt, total;

     num = 0x1;//0000;

     u8 *dst = c64x->mem + 0x1200000;
     u8 *dsr = c64x->mem + 0x1100000;
     u8 *src = c64x->mem + 0x1000000;

     BRINTF("\n\n.============ Testing %s ============.\n", name);
     BRINTF("\n");
     BRINTF("SRC REF\n");

     for (y=0; y<h+1; y++) {
          for (x=0; x<w+1; x++) {
               src[x+y*32] = x*y;
               BRINTF("%-3d ", src[x+y*32]);
          }
          BRINTF("\n");
     }

     BRINTF("\n");

     BRINTF("DST REF\n");

     for (y=0; y<h; y++) {
          for (x=0; x<w; x++) {
               dsr[x+y*32] = w*h-1-x*y;
               BRINTF("%-3d ", dsr[x+y*32]);
          }
          BRINTF("\n");
     }

     BRINTF("\n");


     for (i=0; i<0x100000; i++) {
          dst[i] = i;
     }


     t1 = direct_clock_get_abs_micros();

     for (i=0; i<num; i++) {
          c64xTask *task = c64x_get_task( c64x );

          task->c64x_function = (func << 2) | C64X_FLAG_TODO;

          task->c64x_arg[0] = DAVINCI_C64X_MEM + 0x01200000;
          task->c64x_arg[1] = 32;
          task->c64x_arg[2] = DAVINCI_C64X_MEM + 0x01000000;
          task->c64x_arg[3] = DAVINCI_C64X_MEM + 0x01100000;
          task->c64x_arg[4] = 32;
          task->c64x_arg[5] = h;

          c64x_submit_task( c64x, task );
     }

     davinci_c64x_wait_low( c64x );

     t2 = direct_clock_get_abs_micros();

     BRINTF("-> DST\n");

     for (y=0; y<h; y++) {
          for (x=0; x<w; x++) {
               BRINTF("%-3d ", dst[x+y*32]);
          }
          BRINTF("\n");
     }

     BRINTF("\n");

     dt    = t2 - t1;
     total = num;

     BRINTF( "BENCHMARK on DSP - %-15s   %lld Calls/sec\n",
             name, total * 1000000ULL / dt );
}

static inline void
bench_div( DavinciC64x *c64x, u32 nom, u32 den )
{
     c64xTask *task = c64x_get_task( c64x );

     BRINTF("\n\n.============ Testing div ============.\n");
     BRINTF("\n");

     task->c64x_function = (63 << 2) | C64X_FLAG_TODO;

     task->c64x_arg[0] = nom;
     task->c64x_arg[1] = den;

     c64x_submit_task( c64x, task );

     davinci_c64x_wait_low( c64x );

     BRINTF("%x / %x = %x\n\n\n", nom, den, task->c64x_return);
}

static inline void
bench_dither_argb( DavinciC64x *c64x )
{
     int       i, x, y, num, w = 8, h = 17;
     long long t1, t2, dt, total;

     num = 0x10000;

     u16 *dr  = c64x->mem + 0x1200000;
     u8  *da  = c64x->mem + 0x1100000;
     u32 *src = c64x->mem + 0x1000000;

     BRINTF("\n\n.======================== Testing dither_argb ========================.\n");
     BRINTF("\n");
     BRINTF("SOURCE ARGB\n");

     for (y=0; y<h-1; y++) {
          for (x=0; x<w; x++) {
               src[x+y*32] = 0x10101010 * y + 0x888888 * x;
               BRINTF("%08x ", src[x+y*32]);
          }
          BRINTF("\n");
     }
     for (x=0; x<w; x++) {
          src[x+(h-1)*32] = 0xffffffff;
          BRINTF("%08x ", src[x+y*32]);
     }

     BRINTF("\n");
     BRINTF("\n");

     memset( dr, 0x55, 0x100000 );
     memset( da, 0x55, 0x100000 );


     t1 = direct_clock_get_abs_micros();

     for (i=0; i<num; i++) {
          c64xTask *task = c64x_get_task( c64x );

          task->c64x_function = C64X_DITHER_ARGB | C64X_FLAG_TODO;

          task->c64x_arg[0] = DAVINCI_C64X_MEM + 0x01200000;
          task->c64x_arg[1] = DAVINCI_C64X_MEM + 0x01100000;
          task->c64x_arg[2] = 64;
          task->c64x_arg[3] = DAVINCI_C64X_MEM + 0x01000000;
          task->c64x_arg[4] = 128;
          task->c64x_arg[5] = w;
          task->c64x_arg[6] = h;

          c64x_submit_task( c64x, task );
     }

     davinci_c64x_wait_low( c64x );

     t2 = direct_clock_get_abs_micros();

     BRINTF("-> DST RGB\n");

     for (y=0; y<h; y++) {
          for (x=0; x<w; x++) {
               BRINTF("    %04x ", dr[x+y*32]);
          }
          BRINTF("\n");
     }

     BRINTF("\n");

     BRINTF("-> DST ALPHA\n");

     for (y=0; y<h; y++) {
          for (x=0; x<w; x++) {
               if (x&1)
                    BRINTF(" %x       ", da[x/2+y*64] & 0xF);
               else
                    BRINTF(" %x       ", da[x/2+y*64] >> 4);
          }
          BRINTF("\n");
     }

     BRINTF("\n");

     dt    = t2 - t1;
     total = num;

     BRINTF( "BENCHMARK on DSP - %-15s   %lld Calls/sec\n",
             "dither_argb", total * 1000000ULL / dt );
}




/**********************************************************************************************************************/
/*** 32 bit scaler ****************************************************************************************************/
/**********************************************************************************************************************/

typedef struct {
     DFBRegion   clip;
     const void *colors;
     ulong       protect;
     ulong       key;
} StretchCtx;

typedef void (*StretchHVx)( void             *dst,
                            int               dpitch,
                            const void       *src,
                            int               spitch,
                            int               width,
                            int               height,
                            int               dst_width,
                            int               dst_height,
                            const StretchCtx *ctx );

#define STRETCH_NONE           0
#define STRETCH_SRCKEY         1
#define STRETCH_PROTECT        2
#define STRETCH_SRCKEY_PROTECT 3
#define STRETCH_NUM            4

typedef struct {
     struct {
          StretchHVx     up[STRETCH_NUM];
          StretchHVx     down[STRETCH_NUM];
     } f[DFB_NUM_PIXELFORMATS];
} StretchFunctionTable;


#define DST_FORMAT              DSPF_ARGB
#define TABLE_NAME              stretch_32
#define FUNC_NAME(UPDOWN,K,P,F) stretch_32_ ## UPDOWN ## _ ## K ## P ## _ ## F
#define SHIFT_R8                8
#define SHIFT_L8                8
#define X_00FF00FF              0x00ff00ff
#define X_FF00FF00              0xff00ff00
#define MASK_RGB                0x00ffffff
#define HAS_ALPHA

#include <gfx/generic/stretch_up_down_32.h>

#undef DST_FORMAT
#undef TABLE_NAME
#undef FUNC_NAME
#undef SHIFT_R8
#undef SHIFT_L8
#undef X_00FF00FF
#undef X_FF00FF00
#undef MASK_RGB
#undef HAS_ALPHA


static inline void
bench_stretch_32( DavinciC64x *c64x, int sw, int sh, int dw, int dh )
{
     int       i, x, y, num;
     long long t1, t2, dt, total;
     bool      down = (dw < sw) && (dh < sh);

#if 0
     int SW = (sw + 5) & ~3;
     int SH = (sh + 5) & ~3;
     int DW = (dw + 5) & ~3;
     int DH = (dh + 5) & ~3;
#else
     int SW = sw;
     int SH = sh;
     int DW = dw;
     int DH = dh;
#endif

     num = 1;//0x10000;

     u32  cpu[DW * DH];
     u32 *dst = c64x->mem + 0x1200000;
     u32 *src = c64x->mem + 0x1000000;

     memset( src, 0x55, 0x100000 );

     for (y=0; y<sh; y++) {
          for (x=0; x<sw; x++) {
               src[x + y*SW] = 0xffffffff * x;//  0x10010203 * x + 0x04202020 * (y + 1);
          }
     }


     BRINTF("\n\n.======================== Testing stretch_32( %dx%d -> %dx%d ) ========================.\n", sw, sh, dw, dh);
     BRINTF("\n");
     BRINTF("SOURCE IMAGE (%dx%d) [%dx%d]\n", sw, sh, SW, SH);

     for (y=0; y<SH; y++) {
          for (x=0; x<SW; x++) {
               BRINTF("%08x ", src[x + y*SW]);
          }
          BRINTF("\n");
     }

     BRINTF("\n");
     BRINTF("\n");

     memset( dst, 0x55, 0x100000 );
     memset( cpu, 0x55, sizeof(cpu) );


     t1 = direct_clock_get_abs_micros();

     for (i=0; i<num; i++) {
          c64xTask *task = c64x_get_task( c64x );

          task->c64x_function = (down ?
                                 C64X_STRETCH_32_down :
                                 C64X_STRETCH_32_up ) | C64X_FLAG_TODO;

          task->c64x_arg[0] = DAVINCI_C64X_MEM + 0x1200000;
          task->c64x_arg[1] = DAVINCI_C64X_MEM + 0x1000000;
          task->c64x_arg[2] = (DW * 4) | ((SW * 4) << 16);
          task->c64x_arg[3] = dh       | (dw       << 16);
          task->c64x_arg[4] = sh       | (sw       << 16);
          task->c64x_arg[5] = (dw - 1) | ((dh - 1) << 16);
          task->c64x_arg[6] = 0        | (0        << 16);

          c64x_submit_task( c64x, task );
     }

     davinci_c64x_write_back_all( c64x );

     davinci_c64x_wait_low( c64x );

     t2 = direct_clock_get_abs_micros();


     BRINTF("-> DSP RESULT (%dx%d) [%dx%d]\n", dw, dh, DW, DH);

     for (y=0; y<DH; y++) {
          for (x=0; x<DW; x++) {
               BRINTF("%08x ", dst[x + y*DW]);
          }
          BRINTF("\n");
     }

     BRINTF("\n");


     {
          StretchHVx func = (down ?
                              stretch_32.f[DFB_PIXELFORMAT_INDEX(DSPF_ARGB)].down[STRETCH_NONE] :
                              stretch_32.f[DFB_PIXELFORMAT_INDEX(DSPF_ARGB)].up[STRETCH_NONE]);
          StretchCtx ctx  = { .clip = DFB_REGION_INIT_FROM_RECTANGLE_VALS( 0, 0, dw, dh ) };
     
          func( cpu, DW * 4, src, SW * 4, sw, sh, dw, dh, &ctx );
     
          BRINTF("-> CPU RESULT (%dx%d) [%dx%d]\n", dw, dh, DW, DH);
     
          for (y=0; y<DH; y++) {
               for (x=0; x<DW; x++) {
                    BRINTF("%08x ", cpu[x + y*DW]);
               }
               BRINTF("\n");
          }
     
          BRINTF("\n");
     }

     dt    = t2 - t1;
     total = num;

     BRINTF( "BENCHMARK on DSP - stretch_32_up   %lld Calls/sec\n", total * 1000000ULL / dt );
}

static inline void
run_benchmarks( const char *name,
                void       *ptr,
                int         length )
{
     bench_mem( name, ptr, length, false, false );
     bench_mem( name, ptr, length, true,  false );
     bench_mem( name, ptr, length, true,  true  );
}

/**********************************************************************************************************************/
/*   Public Functions                                                                                                 */
/**********************************************************************************************************************/

DFBResult
davinci_c64x_open( DavinciC64x *c64x )
{
     DFBResult  ret;
     int        fd;
     void      *map_m;
     void      *map_q = NULL;

     mknod( C64X_DEVICE, 0666 | S_IFCHR, makedev( 400, 0 ) );

     fd = direct_try_open( C64X_DEVICE, C64X_DEVICE0, O_RDWR, true );
     if (fd < 0)
          return DFB_IO;

     map_q = mmap( NULL, C64X_QLEN, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0 );
     if (map_q == MAP_FAILED) {
          ret = errno2result( errno );
          D_PERROR( "Davinci/C64X: Mapping %lu bytes at %lu via '%s' failed!\n", C64X_QLEN, 0UL, C64X_DEVICE );
          goto error;
     }

//     run_benchmarks( "Queue", map_q, C64X_QLEN );

     map_m = mmap( NULL, C64X_MLEN, PROT_READ | PROT_WRITE, MAP_SHARED, fd, C64X_QLEN );
     if (map_m == MAP_FAILED) {
          ret = errno2result( errno );
          D_PERROR( "Davinci/C64X: Mapping %lu bytes at %lu via '%s' failed!\n", C64X_MLEN, C64X_QLEN, C64X_DEVICE );
          goto error;
     }

//     run_benchmarks( "Memory", map_m, C64X_MLEN );

     c64x->fd  = fd;
     c64x->ctl = map_q;
     c64x->mem = map_m;
     c64x->QueueL = map_m + 0x01e00000;

     D_INFO( "Davinci/C64X: Low ARM %d / DSP %d, High ARM %d / DSP %d\n",
             c64x->ctl->QL_arm, c64x->ctl->QL_dsp, c64x->ctl->QH_arm, c64x->ctl->QH_dsp );

     D_MAGIC_SET( c64x, DavinciC64x );

if (getenv("C64X_TEST")) {
//     test_load_block( c64x, false );

//     test_load_block( c64x, true );

//     bench_dither_argb( c64x );

#if 0
     bench_uyvy_1( c64x, true );
     bench_uyvy_1( c64x, false );
     bench_uyvy_2( c64x, true );
     bench_uyvy_2( c64x, false );
     bench_uyvy_3( c64x, true );
     bench_uyvy_3( c64x, false );
#endif

#if 0
     bench_blend_argb( c64x, 0 );
     bench_blend_argb( c64x, 1 );
     bench_blend_argb( c64x, 2 );
     bench_blend_argb( c64x, 3 );
#endif

#if 0
     bench_stretch_32( c64x, 2, 1, 16, 1 );
     bench_stretch_32( c64x, 2, 2, 16, 2 );

     bench_stretch_32( c64x, 2, 1,  3, 1 );
     bench_stretch_32( c64x, 4, 1,  6, 1 );

     bench_stretch_32( c64x, 3, 1,  2, 1 );
     bench_stretch_32( c64x, 6, 1,  4, 1 );
#endif

#if 1
     bench_fetch_uyvy( c64x, false, 0, 0 );
     bench_fetch_uyvy( c64x, false, 1, 0 );
     bench_fetch_uyvy( c64x, false, 0, 1 );
     bench_fetch_uyvy( c64x, false, 1, 1 );
     bench_fetch_uyvy( c64x, true, 0, 0 );
     bench_fetch_uyvy( c64x, true, 1, 0 );
     bench_fetch_uyvy( c64x, true, 0, 1 );
     bench_fetch_uyvy( c64x, true, 1, 1 );
#endif

#if 0
     bench_put_mc( c64x, false );
     bench_put_mc( c64x, true );

     bench_put_sum( c64x, false );
     bench_put_sum( c64x, true );

     bench_sat_mc( c64x );
#endif

#if 0
     bench_mc( c64x, 32, 8, 8, false, "mc_put_o_8" );
     bench_mc( c64x, 33, 8, 8, false, "mc_put_x_8" );
     bench_mc( c64x, 34, 8, 8, false, "mc_put_y_8" );
     bench_mc( c64x, 35, 8, 8, false, "mc_put_xy_8" );
     bench_mc( c64x, 36, 16, 16, false, "mc_put_o_16" );
     bench_mc( c64x, 37, 16, 16, false, "mc_put_x_16" );
     bench_mc( c64x, 38, 16, 16, false, "mc_put_y_16" );
     bench_mc( c64x, 39, 16, 16, false, "mc_put_xy_16" );
#endif

#if 0
     bench_mc( c64x, 40, 8, 8, true, "mc_avg_o_8" );
     bench_mc( c64x, 41, 8, 8, true, "mc_avg_x_8" );
     bench_mc( c64x, 42, 8, 8, true, "mc_avg_y_8" );
     bench_mc( c64x, 43, 8, 8, true, "mc_avg_xy_8" );
     bench_mc( c64x, 44, 16, 16, true, "mc_avg_o_16" );
     bench_mc( c64x, 45, 16, 16, true, "mc_avg_x_16" );
     bench_mc( c64x, 46, 16, 16, true, "mc_avg_y_16" );
     bench_mc( c64x, 47, 16, 16, true, "mc_avg_xy_16" );
#endif

#if 0
     bench_div( c64x, 1, 3 );
     bench_div( c64x, 1000, 333 );
     bench_div( c64x, 1000, 334 );
     bench_div( c64x, 6666, 2222 );
     bench_div( c64x, 1234, 1234 );
     bench_div( c64x, 4000, 0 );
     bench_div( c64x, 5000, 0 );
     bench_div( c64x, 10000, 3 );
     bench_div( c64x, 14, 3 );
     bench_div( c64x, 0x10000, 0x1000 );
     bench_div( c64x, 0x1000, 0x100 );
     bench_div( c64x, 0x100000, 2 );
#endif
}

     return DFB_OK;


error:
     if (map_q)
          munmap( map_q, C64X_QLEN );

     close( fd );

     return ret;
}

DFBResult
davinci_c64x_close( DavinciC64x *c64x )
{
     D_MAGIC_ASSERT( c64x, DavinciC64x );

     munmap( (void*) c64x->mem, C64X_MLEN );
     munmap( (void*) c64x->ctl, C64X_QLEN );

     close( c64x->fd );

     D_MAGIC_CLEAR( c64x );

     return DFB_OK;
}