fbcon.c

arm平台上的uclinux系统全部源代码

static __inline__ void fast_memmove(char *dst, char *src, size_t size)
{
  if (!size)
    return;
  if (dst < src)
    __asm__ __volatile__
      ("1:"
       "  moveml %0@+,%/d0/%/d1/%/a0/%/a1\n"
       "  moveml %/d0/%/d1/%/a0/%/a1,%1@\n"
       "  addql #8,%1; addql #8,%1\n"
       "  dbra %2,1b\n"
       "  clrw %2; subql #1,%2\n"
       "  jcc 1b"
       : "=a" (src), "=a" (dst), "=d" (size)
       : "0" (src), "1" (dst), "2" (size / 16 - 1)
       : "d0", "d1", "a0", "a1", "memory");
  else
    __asm__ __volatile__
      ("1:"
       "  subql #8,%0; subql #8,%0\n"
       "  moveml %0@,%/d0/%/d1/%/a0/%/a1\n"
       "  moveml %/d0/%/d1/%/a0/%/a1,%1@-\n"
       "  dbra %2,1b\n"
       "  clrw %2; subql #1,%2\n"
       "  jcc 1b"
       : "=a" (src), "=a" (dst), "=d" (size)
       : "0" (src + size), "1" (dst + size), "2" (size / 16 - 1)
       : "d0", "d1", "a0", "a1", "memory");
}


/* Sets the bytes in the visible column at d, height h, to the value
 * val for a 4 plane screen. The the bis of the color in 'color' are
 * moved (8 times) to the respective bytes. This means:
 *
 * for(h times; d += bpr)
 *   *d     = (color & 1) ? 0xff : 0;
 *   *(d+2) = (color & 2) ? 0xff : 0;
 *   *(d+4) = (color & 4) ? 0xff : 0;
 *   *(d+6) = (color & 8) ? 0xff : 0;
 */

static __inline__ void memclear_4p_col(void *d, size_t h, u_long val, int bpr)
{
	__asm__ __volatile__
		("1: movepl %4,%0@(0)\n\t"
		  "addal  %5,%0\n\t"
		  "dbra	  %1,1b"
		  : "=a" (d), "=d" (h)
		  : "0" (d), "1" (h - 1), "d" (val), "r" (bpr)
		);
}

/* Sets a 4 plane region from 'd', length 'count' bytes, to the color
 * in val1/val2. 'd' has to be an even address and count must be divisible
 * by 8, because only whole words and all planes are accessed. I.e.:
 *
 * for(count/8 times)
 *   *d     = *(d+1) = (color & 1) ? 0xff : 0;
 *   *(d+2) = *(d+3) = (color & 2) ? 0xff : 0;
 *   *(d+4) = *(d+5) = (color & 4) ? 0xff : 0;
 *   *(d+6) = *(d+7) = (color & 8) ? 0xff : 0;
 */

static __inline__ void memset_even_4p(void *d, size_t count, u_long val1,
                                      u_long val2)
{
  u_long *dd = d;

  count /= 8;
  while (count--)
    {
      *dd++ = val1;
      *dd++ = val2;
    }
}

/* Copies a 4 plane column from 's', height 'h', to 'd'. */

static __inline__ void memmove_4p_col (void *d, void *s, int h, int bpr)
{
  u_char *dd = d, *ss = s;

  while (h--)
    {
      dd[0] = ss[0];
      dd[2] = ss[2];
      dd[4] = ss[4];
      dd[6] = ss[6];
      dd += bpr;
      ss += bpr;
    }
}


/* This expands a 4 bit color into a long for movepl (4 plane) operations. */

static __inline__ u_long expand4l(u_char c)
{
	u_long	rv;

	__asm__ __volatile__
		("lsrb	 #1,%2\n\t"
		  "scs	 %0\n\t"
		  "lsll	 #8,%0\n\t"
		  "lsrb	 #1,%2\n\t"
		  "scs	 %0\n\t"
		  "lsll	 #8,%0\n\t"
		  "lsrb	 #1,%2\n\t"
		  "scs	 %0\n\t"
		  "lsll	 #8,%0\n\t"
		  "lsrb	 #1,%2\n\t"
		  "scs	 %0\n\t"
		  : "=&d" (rv), "=d" (c)
		  : "1" (c)
		);
	return(rv);
}

/* This expands a 4 bit color into two longs for two movel operations
 * (4 planes).
 */

static __inline__ void expand4dl(u_char c, u_long *ret1, u_long *ret2)
{
	u_long	rv1, rv2;

	__asm__ __volatile__
		("lsrb	 #1,%3\n\t"
		  "scs	 %0\n\t"
		  "extw	 %0\n\t"
		  "swap	 %0\n\t"
		  "lsrb	 #1,%3\n\t"
		  "scs	 %0\n\t"
		  "extw	 %0\n\t"
		  "lsrb	 #1,%3\n\t"
		  "scs	 %1\n\t"
		  "extw	 %1\n\t"
		  "swap	 %1\n\t"
		  "lsrb	 #1,%3\n\t"
		  "scs   %1\n\t"
		  "extw	 %1"
		  : "=&d" (rv1), "=&d" (rv2), "=d" (c)
		  : "2" (c)
		);
	*ret1 = rv1;
	*ret2 = rv2;
}


/* This duplicates a byte 4 times into a long. */

static __inline__ u_long dup4l(u_char c)
{
	ushort	tmp;
	ulong	rv;

	__asm__ __volatile__
		("moveb  %2,%0\n\t"
		  "lslw   #8,%0\n\t"
		  "moveb  %2,%0\n\t"
		  "movew  %0,%1\n\t"
		  "swap   %0\n\t"
		  "movew  %1,%0"
		  : "=&d" (rv), "=d" (tmp)
		  : "d" (c)
		);

	return(rv);
}


/* Sets the bytes in the visible column at d, height h, to the value
 * val1,val2 for a 8 plane screen. The the bis of the color in 'color' are
 * moved (8 times) to the respective bytes. This means:
 *
 * for(h times; d += bpr)
 *   *d      = (color & 1) ? 0xff : 0;
 *   *(d+2)  = (color & 2) ? 0xff : 0;
 *   *(d+4)  = (color & 4) ? 0xff : 0;
 *   *(d+6)  = (color & 8) ? 0xff : 0;
 *   *(d+8)  = (color & 16) ? 0xff : 0;
 *   *(d+10) = (color & 32) ? 0xff : 0;
 *   *(d+12) = (color & 64) ? 0xff : 0;
 *   *(d+14) = (color & 128) ? 0xff : 0;
 */

static __inline__ void memclear_8p_col(void *d, size_t h, u_long val1,
                                       u_long val2, int bpr)
{
	__asm__ __volatile__
		("1: movepl %4,%0@(0)\n\t"
	      "movepl %5,%0@(8)\n\t"
		  "addal  %6,%0\n\t"
		  "dbra	  %1,1b"
		  : "=a" (d), "=d" (h)
		  : "0" (d), "1" (h - 1), "d" (val1), "d" (val2), "r" (bpr)
		);
}

/* Sets a 8 plane region from 'd', length 'count' bytes, to the color
 * val1..val4. 'd' has to be an even address and count must be divisible
 * by 16, because only whole words and all planes are accessed. I.e.:
 *
 * for(count/16 times)
 *   *d      = *(d+1)  = (color & 1) ? 0xff : 0;
 *   *(d+2)  = *(d+3)  = (color & 2) ? 0xff : 0;
 *   *(d+4)  = *(d+5)  = (color & 4) ? 0xff : 0;
 *   *(d+6)  = *(d+7)  = (color & 8) ? 0xff : 0;
 *   *(d+8)  = *(d+9)  = (color & 16) ? 0xff : 0;
 *   *(d+10) = *(d+11) = (color & 32) ? 0xff : 0;
 *   *(d+12) = *(d+13) = (color & 64) ? 0xff : 0;
 *   *(d+14) = *(d+15) = (color & 128) ? 0xff : 0;
 */

static __inline__ void memset_even_8p(void *d, size_t count, u_long val1,
                                      u_long val2, u_long val3, u_long val4)
{
  u_long *dd = d;

  count /= 16;
  while (count--)
    {
      *dd++ = val1;
      *dd++ = val2;
      *dd++ = val3;
      *dd++ = val4;
    }
}

/* Copies a 8 plane column from 's', height 'h', to 'd'. */

static __inline__ void memmove_8p_col (void *d, void *s, int h, int bpr)
{
  u_char *dd = d, *ss = s;

  while (h--)
    {
      dd[0] = ss[0];
      dd[2] = ss[2];
      dd[4] = ss[4];
      dd[6] = ss[6];
      dd[8] = ss[8];
      dd[10] = ss[10];
      dd[12] = ss[12];
      dd[14] = ss[14];
      dd += bpr;
      ss += bpr;
    }
}


/* This expands a 8 bit color into two longs for two movepl (8 plane)
 * operations.
 */

static __inline__ void expand8dl(u_char c, u_long *ret1, u_long *ret2)
{
	u_long	rv1, rv2;

	__asm__ __volatile__
		("lsrb	 #1,%3\n\t"
		  "scs	 %0\n\t"
		  "lsll	 #8,%0\n\t"
		  "lsrb	 #1,%3\n\t"
		  "scs	 %0\n\t"
		  "lsll	 #8,%0\n\t"
		  "lsrb	 #1,%3\n\t"
		  "scs	 %0\n\t"
		  "lsll	 #8,%0\n\t"
		  "lsrb	 #1,%3\n\t"
		  "scs	 %0\n\t"
		  "lsrb	 #1,%3\n\t"
		  "scs	 %1\n\t"
		  "lsll	 #8,%1\n\t"
		  "lsrb	 #1,%3\n\t"
		  "scs	 %1\n\t"
		  "lsll	 #8,%1\n\t"
		  "lsrb	 #1,%3\n\t"
		  "scs	 %1\n\t"
		  "lsll	 #8,%1\n\t"
		  "lsrb	 #1,%3\n\t"
		  "scs	 %1"
		  : "=&d" (rv1), "=&d" (rv2),"=d" (c)
		  : "2" (c)
		);

	*ret1 = rv1;
	*ret2 = rv2;
}

/* This expands a 8 bit color into four longs for four movel operations
 * (8 planes).
 */

/* ++andreas: use macro to avoid taking address of return values */
#define expand8ql(c, rv1, rv2, rv3, rv4) \
do {	u_char tmp = c;						\
	__asm__ __volatile__					\
		("lsrb	 #1,%5\n\t"				\
		  "scs	 %0\n\t"				\
		  "extw	 %0\n\t"				\
		  "swap	 %0\n\t"				\
		  "lsrb	 #1,%5\n\t"				\
		  "scs	 %0\n\t"				\
		  "extw	 %0\n\t"				\
		  "lsrb	 #1,%5\n\t"				\
		  "scs	 %1\n\t"				\
		  "extw	 %1\n\t"				\
		  "swap	 %1\n\t"				\
		  "lsrb	 #1,%5\n\t"				\
		  "scs   %1\n\t"				\
		  "extw	 %1\n\t"				\
		  "lsrb	 #1,%5\n\t"				\
		  "scs	 %2\n\t"				\
		  "extw	 %2\n\t"				\
		  "swap	 %2\n\t"				\
		  "lsrb	 #1,%5\n\t"				\
		  "scs	 %2\n\t"				\
		  "extw	 %2\n\t"				\
		  "lsrb	 #1,%5\n\t"				\
		  "scs	 %3\n\t"				\
		  "extw	 %3\n\t"				\
		  "swap	 %3\n\t"				\
		  "lsrb	 #1,%5\n\t"				\
		  "scs   %3\n\t"				\
		  "extw	 %3"					\
		  : "=&d" (rv1), "=&d" (rv2), "=&d" (rv3),	\
		    "=&d" (rv4), "=d" (tmp)			\
		  : "4" (tmp)					\
		);						\
} while (0)


/* Sets the bytes in the visible column at d, height h, to the value
 * val for a 2 plane screen. The the bis of the color in 'color' are
 * moved (8 times) to the respective bytes. This means:
 *
 * for(h times; d += bpr)
 *   *d     = (color & 1) ? 0xff : 0;
 *   *(d+2) = (color & 2) ? 0xff : 0;
 */

static __inline__ void memclear_2p_col(void *d, size_t h, u_short val, int bpr)
{
	__asm__ __volatile__
		("1: movepw %4,%0@(0)\n\t"
		  "addal  %5,%0\n\t"
		  "dbra	  %1,1b"
		  : "=a" (d), "=d" (h)
		  : "0" (d), "1" (h - 1), "d" (val), "r" (bpr)
		);
}

/* Sets a 2 plane region from 'd', length 'count' bytes, to the color
 * in val1. 'd' has to be an even address and count must be divisible
 * by 8, because only whole words and all planes are accessed. I.e.:
 *
 * for(count/4 times)
 *   *d     = *(d+1) = (color & 1) ? 0xff : 0;
 *   *(d+2) = *(d+3) = (color & 2) ? 0xff : 0;
 */

static __inline__ void memset_even_2p(void *d, size_t count, u_long val)
{
  u_long *dd = d;

  count /= 4;
  while (count--)
    *dd++ = val;
}

/* Copies a 2 plane column from 's', height 'h', to 'd'. */

static __inline__ void memmove_2p_col (void *d, void *s, int h, int bpr)
{
  u_char *dd = d, *ss = s;

  while (h--)
    {
      dd[0] = ss[0];
      dd[2] = ss[2];
      dd += bpr;
      ss += bpr;
    }
}


/* This expands a 2 bit color into a short for movepw (2 plane) operations. */

static __inline__ u_short expand2w(u_char c)
{
	u_short	rv;

	__asm__ __volatile__
		("lsrb	 #1,%2\n\t"
		  "scs	 %0\n\t"
		  "lsll	 #8,%0\n\t"
		  "lsrb	 #1,%2\n\t"
		  "scs	 %0\n\t"
		  : "=&d" (rv), "=d" (c)
		  : "1" (c)
		);
	return(rv);
}

/* This expands a 2 bit color into one long for a movel operation
 * (2 planes).
 */

static __inline__ u_long expand2l(u_char c)
{
	u_long	rv;

	__asm__ __volatile__
		("lsrb	 #1,%2\n\t"
		  "scs	 %0\n\t"
		  "extw	 %0\n\t"
		  "swap	 %0\n\t"
		  "lsrb	 #1,%2\n\t"
		  "scs	 %0\n\t"
		  "extw	 %0\n\t"
		  : "=&d" (rv), "=d" (c)
		  : "1" (c)
		);

	return rv;
}


/* This duplicates a byte 2 times into a short. */

static __inline__ u_short dup2w(u_char c)
{
    ushort  rv;

    __asm__ __volatile__
        ( "moveb  %1,%0\n\t"
          "lslw   #8,%0\n\t"
          "moveb  %1,%0\n\t"
          : "=&d" (rv)
          : "d" (c)
        );

    return( rv );
}


/* ====================================================================== */

/* fbcon_XXX routines - interface used by the world
 *
 * This system is now divided into two levels because of complications
 * caused by hardware scrolling. Top level functions:
 *
 *    fbcon_bmove(), fbcon_clear(), fbcon_putc()
 *
 * handles y values in range [0, scr_height-1] that correspond to real
 * screen positions. y_wrap shift means that first line of bitmap may be
 * anywhere on this display. These functions convert lineoffsets to
 * bitmap offsets and deal with the wrap-around case by splitting blits.
 *
 *    fbcon_bmove_physical_8()   -- These functions fast implementations
 *    fbcon_clear_physical_8()   -- of original fbcon_XXX fns.
 *    fbcon_putc_physical_8()    -- (fontwidth != 8) may be added later
 *
 * WARNING:
 *