Back to home page

OSCL-LXR
 
 

     

0001 /* SPDX-License-Identifier: GPL-2.0 */
0002 #ifndef _VIDEO_ATAFB_UTILS_H
0003 #define _VIDEO_ATAFB_UTILS_H
0004 
0005 /* ================================================================= */
0006 /*                      Utility Assembler Functions                  */
0007 /* ================================================================= */
0008 
0009 /* ====================================================================== */
0010 
0011 /* Those of a delicate disposition might like to skip the next couple of
0012  * pages.
0013  *
0014  * These functions are drop in replacements for memmove and
0015  * memset(_, 0, _). However their five instances add at least a kilobyte
0016  * to the object file. You have been warned.
0017  *
0018  * Not a great fan of assembler for the sake of it, but I think
0019  * that these routines are at least 10 times faster than their C
0020  * equivalents for large blits, and that's important to the lowest level of
0021  * a graphics driver. Question is whether some scheme with the blitter
0022  * would be faster. I suspect not for simple text system - not much
0023  * asynchrony.
0024  *
0025  * Code is very simple, just gruesome expansion. Basic strategy is to
0026  * increase data moved/cleared at each step to 16 bytes to reduce
0027  * instruction per data move overhead. movem might be faster still
0028  * For more than 15 bytes, we try to align the write direction on a
0029  * longword boundary to get maximum speed. This is even more gruesome.
0030  * Unaligned read/write used requires 68020+ - think this is a problem?
0031  *
0032  * Sorry!
0033  */
0034 
0035 
0036 /* ++roman: I've optimized Robert's original versions in some minor
0037  * aspects, e.g. moveq instead of movel, let gcc choose the registers,
0038  * use movem in some places...
0039  * For other modes than 1 plane, lots of more such assembler functions
0040  * were needed (e.g. the ones using movep or expanding color values).
0041  */
0042 
0043 /* ++andreas: more optimizations:
0044    subl #65536,d0 replaced by clrw d0; subql #1,d0 for dbcc
0045    addal is faster than addaw
0046    movep is rather expensive compared to ordinary move's
0047    some functions rewritten in C for clarity, no speed loss */
0048 
0049 static inline void *fb_memclear_small(void *s, size_t count)
0050 {
0051     if (!count)
0052         return 0;
0053 
0054     asm volatile ("\n"
0055         "   lsr.l   #1,%1 ; jcc 1f ; move.b %2,-(%0)\n"
0056         "1: lsr.l   #1,%1 ; jcc 1f ; move.w %2,-(%0)\n"
0057         "1: lsr.l   #1,%1 ; jcc 1f ; move.l %2,-(%0)\n"
0058         "1: lsr.l   #1,%1 ; jcc 1f ; move.l %2,-(%0) ; move.l %2,-(%0)\n"
0059         "1:"
0060         : "=a" (s), "=d" (count)
0061         : "d" (0), "0" ((char *)s + count), "1" (count));
0062     asm volatile ("\n"
0063         "   subq.l  #1,%1\n"
0064         "   jcs 3f\n"
0065         "   move.l  %2,%%d4; move.l %2,%%d5; move.l %2,%%d6\n"
0066         "2: movem.l %2/%%d4/%%d5/%%d6,-(%0)\n"
0067         "   dbra    %1,2b\n"
0068         "3:"
0069         : "=a" (s), "=d" (count)
0070         : "d" (0), "0" (s), "1" (count)
0071         : "d4", "d5", "d6"
0072         );
0073 
0074     return 0;
0075 }
0076 
0077 
0078 static inline void *fb_memclear(void *s, size_t count)
0079 {
0080     if (!count)
0081         return 0;
0082 
0083     if (count < 16) {
0084         asm volatile ("\n"
0085             "   lsr.l   #1,%1 ; jcc 1f ; clr.b (%0)+\n"
0086             "1: lsr.l   #1,%1 ; jcc 1f ; clr.w (%0)+\n"
0087             "1: lsr.l   #1,%1 ; jcc 1f ; clr.l (%0)+\n"
0088             "1: lsr.l   #1,%1 ; jcc 1f ; clr.l (%0)+ ; clr.l (%0)+\n"
0089             "1:"
0090             : "=a" (s), "=d" (count)
0091             : "0" (s), "1" (count));
0092     } else {
0093         long tmp;
0094         asm volatile ("\n"
0095             "   move.l  %1,%2\n"
0096             "   lsr.l   #1,%2 ; jcc 1f ; clr.b (%0)+ ; subq.w #1,%1\n"
0097             "   lsr.l   #1,%2 ; jcs 2f\n"  /* %0 increased=>bit 2 switched*/
0098             "   clr.w   (%0)+  ; subq.w  #2,%1 ; jra 2f\n"
0099             "1: lsr.l   #1,%2 ; jcc 2f\n"
0100             "   clr.w   (%0)+  ; subq.w  #2,%1\n"
0101             "2: move.w  %1,%2; lsr.l #2,%1 ; jeq 6f\n"
0102             "   lsr.l   #1,%1 ; jcc 3f ; clr.l (%0)+\n"
0103             "3: lsr.l   #1,%1 ; jcc 4f ; clr.l (%0)+ ; clr.l (%0)+\n"
0104             "4: subq.l  #1,%1 ; jcs 6f\n"
0105             "5: clr.l   (%0)+; clr.l (%0)+ ; clr.l (%0)+ ; clr.l (%0)+\n"
0106             "   dbra    %1,5b ; clr.w %1; subq.l #1,%1; jcc 5b\n"
0107             "6: move.w  %2,%1; btst #1,%1 ; jeq 7f ; clr.w (%0)+\n"
0108             "7: btst    #0,%1 ; jeq 8f ; clr.b (%0)+\n"
0109             "8:"
0110             : "=a" (s), "=d" (count), "=d" (tmp)
0111             : "0" (s), "1" (count));
0112     }
0113 
0114     return 0;
0115 }
0116 
0117 
0118 static inline void *fb_memset255(void *s, size_t count)
0119 {
0120     if (!count)
0121         return 0;
0122 
0123     asm volatile ("\n"
0124         "   lsr.l   #1,%1 ; jcc 1f ; move.b %2,-(%0)\n"
0125         "1: lsr.l   #1,%1 ; jcc 1f ; move.w %2,-(%0)\n"
0126         "1: lsr.l   #1,%1 ; jcc 1f ; move.l %2,-(%0)\n"
0127         "1: lsr.l   #1,%1 ; jcc 1f ; move.l %2,-(%0) ; move.l %2,-(%0)\n"
0128         "1:"
0129         : "=a" (s), "=d" (count)
0130         : "d" (-1), "0" ((char *)s+count), "1" (count));
0131     asm volatile ("\n"
0132         "   subq.l  #1,%1 ; jcs 3f\n"
0133         "   move.l  %2,%%d4; move.l %2,%%d5; move.l %2,%%d6\n"
0134         "2: movem.l %2/%%d4/%%d5/%%d6,-(%0)\n"
0135         "   dbra    %1,2b\n"
0136         "3:"
0137         : "=a" (s), "=d" (count)
0138         : "d" (-1), "0" (s), "1" (count)
0139         : "d4", "d5", "d6");
0140 
0141     return 0;
0142 }
0143 
0144 
0145 static inline void *fb_memmove(void *d, const void *s, size_t count)
0146 {
0147     if (d < s) {
0148         if (count < 16) {
0149             asm volatile ("\n"
0150                 "   lsr.l   #1,%2 ; jcc 1f ; move.b (%1)+,(%0)+\n"
0151                 "1: lsr.l   #1,%2 ; jcc 1f ; move.w (%1)+,(%0)+\n"
0152                 "1: lsr.l   #1,%2 ; jcc 1f ; move.l (%1)+,(%0)+\n"
0153                 "1: lsr.l   #1,%2 ; jcc 1f ; move.l (%1)+,(%0)+ ; move.l (%1)+,(%0)+\n"
0154                 "1:"
0155                 : "=a" (d), "=a" (s), "=d" (count)
0156                 : "0" (d), "1" (s), "2" (count));
0157         } else {
0158             long tmp;
0159             asm volatile ("\n"
0160                 "   move.l  %0,%3\n"
0161                 "   lsr.l   #1,%3 ; jcc 1f ; move.b (%1)+,(%0)+ ; subqw #1,%2\n"
0162                 "   lsr.l   #1,%3 ; jcs 2f\n"  /* %0 increased=>bit 2 switched*/
0163                 "   move.w  (%1)+,(%0)+  ; subqw  #2,%2 ; jra 2f\n"
0164                 "1: lsr.l   #1,%3 ; jcc 2f\n"
0165                 "   move.w  (%1)+,(%0)+  ; subqw  #2,%2\n"
0166                 "2: move.w  %2,%-; lsr.l #2,%2 ; jeq 6f\n"
0167                 "   lsr.l   #1,%2 ; jcc 3f ; move.l (%1)+,(%0)+\n"
0168                 "3: lsr.l   #1,%2 ; jcc 4f ; move.l (%1)+,(%0)+ ; move.l (%1)+,(%0)+\n"
0169                 "4: subq.l  #1,%2 ; jcs 6f\n"
0170                 "5: move.l  (%1)+,(%0)+; move.l (%1)+,(%0)+\n"
0171                 "   move.l  (%1)+,(%0)+; move.l (%1)+,(%0)+\n"
0172                 "   dbra    %2,5b ; clr.w %2; subq.l #1,%2; jcc 5b\n"
0173                 "6: move.w  %+,%2; btst #1,%2 ; jeq 7f ; move.w (%1)+,(%0)+\n"
0174                 "7: btst    #0,%2 ; jeq 8f ; move.b (%1)+,(%0)+\n"
0175                 "8:"
0176                 : "=a" (d), "=a" (s), "=d" (count), "=d" (tmp)
0177                 : "0" (d), "1" (s), "2" (count));
0178         }
0179     } else {
0180         if (count < 16) {
0181             asm volatile ("\n"
0182                 "   lsr.l   #1,%2 ; jcc 1f ; move.b -(%1),-(%0)\n"
0183                 "1: lsr.l   #1,%2 ; jcc 1f ; move.w -(%1),-(%0)\n"
0184                 "1: lsr.l   #1,%2 ; jcc 1f ; move.l -(%1),-(%0)\n"
0185                 "1: lsr.l   #1,%2 ; jcc 1f ; move.l -(%1),-(%0) ; move.l -(%1),-(%0)\n"
0186                 "1:"
0187                 : "=a" (d), "=a" (s), "=d" (count)
0188                 : "0" ((char *) d + count), "1" ((char *) s + count), "2" (count));
0189         } else {
0190             long tmp;
0191 
0192             asm volatile ("\n"
0193                 "   move.l  %0,%3\n"
0194                 "   lsr.l   #1,%3 ; jcc 1f ; move.b -(%1),-(%0) ; subqw #1,%2\n"
0195                 "   lsr.l   #1,%3 ; jcs 2f\n"  /* %0 increased=>bit 2 switched*/
0196                 "   move.w  -(%1),-(%0) ; subqw  #2,%2 ; jra 2f\n"
0197                 "1: lsr.l   #1,%3 ; jcc 2f\n"
0198                 "   move.w  -(%1),-(%0) ; subqw  #2,%2\n"
0199                 "2: move.w  %2,%-; lsr.l #2,%2 ; jeq 6f\n"
0200                 "   lsr.l   #1,%2 ; jcc 3f ; move.l -(%1),-(%0)\n"
0201                 "3: lsr.l   #1,%2 ; jcc 4f ; move.l -(%1),-(%0) ; move.l -(%1),-(%0)\n"
0202                 "4: subq.l  #1,%2 ; jcs 6f\n"
0203                 "5: move.l  -(%1),-(%0); move.l -(%1),-(%0)\n"
0204                 "   move.l  -(%1),-(%0); move.l -(%1),-(%0)\n"
0205                 "   dbra    %2,5b ; clr.w %2; subq.l #1,%2; jcc 5b\n"
0206                 "6: move.w  %+,%2; btst #1,%2 ; jeq 7f ; move.w -(%1),-(%0)\n"
0207                 "7: btst    #0,%2 ; jeq 8f ; move.b -(%1),-(%0)\n"
0208                 "8:"
0209                 : "=a" (d), "=a" (s), "=d" (count), "=d" (tmp)
0210                 : "0" ((char *) d + count), "1" ((char *) s + count), "2" (count));
0211         }
0212     }
0213 
0214     return 0;
0215 }
0216 
0217 
0218 /* ++andreas: Simple and fast version of memmove, assumes size is
0219    divisible by 16, suitable for moving the whole screen bitplane */
0220 static inline void fast_memmove(char *dst, const char *src, size_t size)
0221 {
0222     if (!size)
0223         return;
0224     if (dst < src)
0225         asm volatile ("\n"
0226             "1: movem.l (%0)+,%%d0/%%d1/%%a0/%%a1\n"
0227             "   movem.l %%d0/%%d1/%%a0/%%a1,%1@\n"
0228             "   addq.l  #8,%1; addq.l #8,%1\n"
0229             "   dbra    %2,1b\n"
0230             "   clr.w   %2; subq.l #1,%2\n"
0231             "   jcc 1b"
0232             : "=a" (src), "=a" (dst), "=d" (size)
0233             : "0" (src), "1" (dst), "2" (size / 16 - 1)
0234             : "d0", "d1", "a0", "a1", "memory");
0235     else
0236         asm volatile ("\n"
0237             "1: subq.l  #8,%0; subq.l #8,%0\n"
0238             "   movem.l %0@,%%d0/%%d1/%%a0/%%a1\n"
0239             "   movem.l %%d0/%%d1/%%a0/%%a1,-(%1)\n"
0240             "   dbra    %2,1b\n"
0241             "   clr.w   %2; subq.l #1,%2\n"
0242             "   jcc 1b"
0243             : "=a" (src), "=a" (dst), "=d" (size)
0244             : "0" (src + size), "1" (dst + size), "2" (size / 16 - 1)
0245             : "d0", "d1", "a0", "a1", "memory");
0246 }
0247 
0248 #ifdef BPL
0249 
0250 /*
0251  * This expands a up to 8 bit color into two longs
0252  * for movel operations.
0253  */
0254 static const u32 four2long[] = {
0255     0x00000000, 0x000000ff, 0x0000ff00, 0x0000ffff,
0256     0x00ff0000, 0x00ff00ff, 0x00ffff00, 0x00ffffff,
0257     0xff000000, 0xff0000ff, 0xff00ff00, 0xff00ffff,
0258     0xffff0000, 0xffff00ff, 0xffffff00, 0xffffffff,
0259 };
0260 
0261 static inline void expand8_col2mask(u8 c, u32 m[])
0262 {
0263     m[0] = four2long[c & 15];
0264 #if BPL > 4
0265     m[1] = four2long[c >> 4];
0266 #endif
0267 }
0268 
0269 static inline void expand8_2col2mask(u8 fg, u8 bg, u32 fgm[], u32 bgm[])
0270 {
0271     fgm[0] = four2long[fg & 15] ^ (bgm[0] = four2long[bg & 15]);
0272 #if BPL > 4
0273     fgm[1] = four2long[fg >> 4] ^ (bgm[1] = four2long[bg >> 4]);
0274 #endif
0275 }
0276 
0277 /*
0278  * set an 8bit value to a color
0279  */
0280 static inline void fill8_col(u8 *dst, u32 m[])
0281 {
0282     u32 tmp = m[0];
0283     dst[0] = tmp;
0284     dst[2] = (tmp >>= 8);
0285 #if BPL > 2
0286     dst[4] = (tmp >>= 8);
0287     dst[6] = tmp >> 8;
0288 #endif
0289 #if BPL > 4
0290     tmp = m[1];
0291     dst[8] = tmp;
0292     dst[10] = (tmp >>= 8);
0293     dst[12] = (tmp >>= 8);
0294     dst[14] = tmp >> 8;
0295 #endif
0296 }
0297 
0298 /*
0299  * set an 8bit value according to foreground/background color
0300  */
0301 static inline void fill8_2col(u8 *dst, u8 fg, u8 bg, u32 mask)
0302 {
0303     u32 fgm[2], bgm[2], tmp;
0304 
0305     expand8_2col2mask(fg, bg, fgm, bgm);
0306 
0307     mask |= mask << 8;
0308 #if BPL > 2
0309     mask |= mask << 16;
0310 #endif
0311     tmp = (mask & fgm[0]) ^ bgm[0];
0312     dst[0] = tmp;
0313     dst[2] = (tmp >>= 8);
0314 #if BPL > 2
0315     dst[4] = (tmp >>= 8);
0316     dst[6] = tmp >> 8;
0317 #endif
0318 #if BPL > 4
0319     tmp = (mask & fgm[1]) ^ bgm[1];
0320     dst[8] = tmp;
0321     dst[10] = (tmp >>= 8);
0322     dst[12] = (tmp >>= 8);
0323     dst[14] = tmp >> 8;
0324 #endif
0325 }
0326 
0327 static const u32 two2word[] = {
0328     0x00000000, 0xffff0000, 0x0000ffff, 0xffffffff
0329 };
0330 
0331 static inline void expand16_col2mask(u8 c, u32 m[])
0332 {
0333     m[0] = two2word[c & 3];
0334 #if BPL > 2
0335     m[1] = two2word[(c >> 2) & 3];
0336 #endif
0337 #if BPL > 4
0338     m[2] = two2word[(c >> 4) & 3];
0339     m[3] = two2word[c >> 6];
0340 #endif
0341 }
0342 
0343 static inline void expand16_2col2mask(u8 fg, u8 bg, u32 fgm[], u32 bgm[])
0344 {
0345     bgm[0] = two2word[bg & 3];
0346     fgm[0] = two2word[fg & 3] ^ bgm[0];
0347 #if BPL > 2
0348     bgm[1] = two2word[(bg >> 2) & 3];
0349     fgm[1] = two2word[(fg >> 2) & 3] ^ bgm[1];
0350 #endif
0351 #if BPL > 4
0352     bgm[2] = two2word[(bg >> 4) & 3];
0353     fgm[2] = two2word[(fg >> 4) & 3] ^ bgm[2];
0354     bgm[3] = two2word[bg >> 6];
0355     fgm[3] = two2word[fg >> 6] ^ bgm[3];
0356 #endif
0357 }
0358 
0359 static inline u32 *fill16_col(u32 *dst, int rows, u32 m[])
0360 {
0361     while (rows) {
0362         *dst++ = m[0];
0363 #if BPL > 2
0364         *dst++ = m[1];
0365 #endif
0366 #if BPL > 4
0367         *dst++ = m[2];
0368         *dst++ = m[3];
0369 #endif
0370         rows--;
0371     }
0372     return dst;
0373 }
0374 
0375 static inline void memmove32_col(void *dst, void *src, u32 mask, u32 h, u32 bytes)
0376 {
0377     u32 *s, *d, v;
0378 
0379         s = src;
0380         d = dst;
0381         do {
0382                 v = (*s++ & mask) | (*d  & ~mask);
0383                 *d++ = v;
0384 #if BPL > 2
0385                 v = (*s++ & mask) | (*d  & ~mask);
0386                 *d++ = v;
0387 #endif
0388 #if BPL > 4
0389                 v = (*s++ & mask) | (*d  & ~mask);
0390                 *d++ = v;
0391                 v = (*s++ & mask) | (*d  & ~mask);
0392                 *d++ = v;
0393 #endif
0394                 d = (u32 *)((u8 *)d + bytes);
0395                 s = (u32 *)((u8 *)s + bytes);
0396         } while (--h);
0397 }
0398 
0399 #endif
0400 
0401 #endif /* _VIDEO_ATAFB_UTILS_H */
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!