1 | /* SPDX-License-Identifier: GPL-2.0 */ |
2 | #ifndef _VIDEO_ATAFB_UTILS_H |
3 | #define _VIDEO_ATAFB_UTILS_H |
4 | |
5 | /* ================================================================= */ |
6 | /* Utility Assembler Functions */ |
7 | /* ================================================================= */ |
8 | |
9 | /* ====================================================================== */ |
10 | |
11 | /* Those of a delicate disposition might like to skip the next couple of |
12 | * pages. |
13 | * |
14 | * These functions are drop in replacements for memmove and |
15 | * memset(_, 0, _). However their five instances add at least a kilobyte |
16 | * to the object file. You have been warned. |
17 | * |
18 | * Not a great fan of assembler for the sake of it, but I think |
19 | * that these routines are at least 10 times faster than their C |
20 | * equivalents for large blits, and that's important to the lowest level of |
21 | * a graphics driver. Question is whether some scheme with the blitter |
22 | * would be faster. I suspect not for simple text system - not much |
23 | * asynchrony. |
24 | * |
25 | * Code is very simple, just gruesome expansion. Basic strategy is to |
26 | * increase data moved/cleared at each step to 16 bytes to reduce |
27 | * instruction per data move overhead. movem might be faster still |
28 | * For more than 15 bytes, we try to align the write direction on a |
29 | * longword boundary to get maximum speed. This is even more gruesome. |
30 | * Unaligned read/write used requires 68020+ - think this is a problem? |
31 | * |
32 | * Sorry! |
33 | */ |
34 | |
35 | |
36 | /* ++roman: I've optimized Robert's original versions in some minor |
37 | * aspects, e.g. moveq instead of movel, let gcc choose the registers, |
38 | * use movem in some places... |
39 | * For other modes than 1 plane, lots of more such assembler functions |
40 | * were needed (e.g. the ones using movep or expanding color values). |
41 | */ |
42 | |
43 | /* ++andreas: more optimizations: |
44 | subl #65536,d0 replaced by clrw d0; subql #1,d0 for dbcc |
45 | addal is faster than addaw |
46 | movep is rather expensive compared to ordinary move's |
47 | some functions rewritten in C for clarity, no speed loss */ |
48 | |
49 | static inline void *fb_memclear_small(void *s, size_t count) |
50 | { |
51 | if (!count) |
52 | return 0; |
53 | |
54 | asm volatile ("\n" |
55 | " lsr.l #1,%1 ; jcc 1f ; move.b %2,-(%0)\n" |
56 | "1: lsr.l #1,%1 ; jcc 1f ; move.w %2,-(%0)\n" |
57 | "1: lsr.l #1,%1 ; jcc 1f ; move.l %2,-(%0)\n" |
58 | "1: lsr.l #1,%1 ; jcc 1f ; move.l %2,-(%0) ; move.l %2,-(%0)\n" |
59 | "1:" |
60 | : "=a" (s), "=d" (count) |
61 | : "d" (0), "0" ((char *)s + count), "1" (count)); |
62 | asm volatile ("\n" |
63 | " subq.l #1,%1\n" |
64 | " jcs 3f\n" |
65 | " move.l %2,%%d4; move.l %2,%%d5; move.l %2,%%d6\n" |
66 | "2: movem.l %2/%%d4/%%d5/%%d6,-(%0)\n" |
67 | " dbra %1,2b\n" |
68 | "3:" |
69 | : "=a" (s), "=d" (count) |
70 | : "d" (0), "0" (s), "1" (count) |
71 | : "d4" , "d5" , "d6" |
72 | ); |
73 | |
74 | return 0; |
75 | } |
76 | |
77 | |
78 | static inline void *fb_memclear(void *s, size_t count) |
79 | { |
80 | if (!count) |
81 | return 0; |
82 | |
83 | if (count < 16) { |
84 | asm volatile ("\n" |
85 | " lsr.l #1,%1 ; jcc 1f ; clr.b (%0)+\n" |
86 | "1: lsr.l #1,%1 ; jcc 1f ; clr.w (%0)+\n" |
87 | "1: lsr.l #1,%1 ; jcc 1f ; clr.l (%0)+\n" |
88 | "1: lsr.l #1,%1 ; jcc 1f ; clr.l (%0)+ ; clr.l (%0)+\n" |
89 | "1:" |
90 | : "=a" (s), "=d" (count) |
91 | : "0" (s), "1" (count)); |
92 | } else { |
93 | long tmp; |
94 | asm volatile ("\n" |
95 | " move.l %1,%2\n" |
96 | " lsr.l #1,%2 ; jcc 1f ; clr.b (%0)+ ; subq.w #1,%1\n" |
97 | " lsr.l #1,%2 ; jcs 2f\n" /* %0 increased=>bit 2 switched*/ |
98 | " clr.w (%0)+ ; subq.w #2,%1 ; jra 2f\n" |
99 | "1: lsr.l #1,%2 ; jcc 2f\n" |
100 | " clr.w (%0)+ ; subq.w #2,%1\n" |
101 | "2: move.w %1,%2; lsr.l #2,%1 ; jeq 6f\n" |
102 | " lsr.l #1,%1 ; jcc 3f ; clr.l (%0)+\n" |
103 | "3: lsr.l #1,%1 ; jcc 4f ; clr.l (%0)+ ; clr.l (%0)+\n" |
104 | "4: subq.l #1,%1 ; jcs 6f\n" |
105 | "5: clr.l (%0)+; clr.l (%0)+ ; clr.l (%0)+ ; clr.l (%0)+\n" |
106 | " dbra %1,5b ; clr.w %1; subq.l #1,%1; jcc 5b\n" |
107 | "6: move.w %2,%1; btst #1,%1 ; jeq 7f ; clr.w (%0)+\n" |
108 | "7: btst #0,%1 ; jeq 8f ; clr.b (%0)+\n" |
109 | "8:" |
110 | : "=a" (s), "=d" (count), "=d" (tmp) |
111 | : "0" (s), "1" (count)); |
112 | } |
113 | |
114 | return 0; |
115 | } |
116 | |
117 | |
118 | static inline void *fb_memset255(void *s, size_t count) |
119 | { |
120 | if (!count) |
121 | return 0; |
122 | |
123 | asm volatile ("\n" |
124 | " lsr.l #1,%1 ; jcc 1f ; move.b %2,-(%0)\n" |
125 | "1: lsr.l #1,%1 ; jcc 1f ; move.w %2,-(%0)\n" |
126 | "1: lsr.l #1,%1 ; jcc 1f ; move.l %2,-(%0)\n" |
127 | "1: lsr.l #1,%1 ; jcc 1f ; move.l %2,-(%0) ; move.l %2,-(%0)\n" |
128 | "1:" |
129 | : "=a" (s), "=d" (count) |
130 | : "d" (-1), "0" ((char *)s+count), "1" (count)); |
131 | asm volatile ("\n" |
132 | " subq.l #1,%1 ; jcs 3f\n" |
133 | " move.l %2,%%d4; move.l %2,%%d5; move.l %2,%%d6\n" |
134 | "2: movem.l %2/%%d4/%%d5/%%d6,-(%0)\n" |
135 | " dbra %1,2b\n" |
136 | "3:" |
137 | : "=a" (s), "=d" (count) |
138 | : "d" (-1), "0" (s), "1" (count) |
139 | : "d4" , "d5" , "d6" ); |
140 | |
141 | return 0; |
142 | } |
143 | |
144 | |
145 | static inline void *fb_memmove(void *d, const void *s, size_t count) |
146 | { |
147 | if (d < s) { |
148 | if (count < 16) { |
149 | asm volatile ("\n" |
150 | " lsr.l #1,%2 ; jcc 1f ; move.b (%1)+,(%0)+\n" |
151 | "1: lsr.l #1,%2 ; jcc 1f ; move.w (%1)+,(%0)+\n" |
152 | "1: lsr.l #1,%2 ; jcc 1f ; move.l (%1)+,(%0)+\n" |
153 | "1: lsr.l #1,%2 ; jcc 1f ; move.l (%1)+,(%0)+ ; move.l (%1)+,(%0)+\n" |
154 | "1:" |
155 | : "=a" (d), "=a" (s), "=d" (count) |
156 | : "0" (d), "1" (s), "2" (count)); |
157 | } else { |
158 | long tmp; |
159 | asm volatile ("\n" |
160 | " move.l %0,%3\n" |
161 | " lsr.l #1,%3 ; jcc 1f ; move.b (%1)+,(%0)+ ; subqw #1,%2\n" |
162 | " lsr.l #1,%3 ; jcs 2f\n" /* %0 increased=>bit 2 switched*/ |
163 | " move.w (%1)+,(%0)+ ; subqw #2,%2 ; jra 2f\n" |
164 | "1: lsr.l #1,%3 ; jcc 2f\n" |
165 | " move.w (%1)+,(%0)+ ; subqw #2,%2\n" |
166 | "2: move.w %2,%-; lsr.l #2,%2 ; jeq 6f\n" |
167 | " lsr.l #1,%2 ; jcc 3f ; move.l (%1)+,(%0)+\n" |
168 | "3: lsr.l #1,%2 ; jcc 4f ; move.l (%1)+,(%0)+ ; move.l (%1)+,(%0)+\n" |
169 | "4: subq.l #1,%2 ; jcs 6f\n" |
170 | "5: move.l (%1)+,(%0)+; move.l (%1)+,(%0)+\n" |
171 | " move.l (%1)+,(%0)+; move.l (%1)+,(%0)+\n" |
172 | " dbra %2,5b ; clr.w %2; subq.l #1,%2; jcc 5b\n" |
173 | "6: move.w %+,%2; btst #1,%2 ; jeq 7f ; move.w (%1)+,(%0)+\n" |
174 | "7: btst #0,%2 ; jeq 8f ; move.b (%1)+,(%0)+\n" |
175 | "8:" |
176 | : "=a" (d), "=a" (s), "=d" (count), "=d" (tmp) |
177 | : "0" (d), "1" (s), "2" (count)); |
178 | } |
179 | } else { |
180 | if (count < 16) { |
181 | asm volatile ("\n" |
182 | " lsr.l #1,%2 ; jcc 1f ; move.b -(%1),-(%0)\n" |
183 | "1: lsr.l #1,%2 ; jcc 1f ; move.w -(%1),-(%0)\n" |
184 | "1: lsr.l #1,%2 ; jcc 1f ; move.l -(%1),-(%0)\n" |
185 | "1: lsr.l #1,%2 ; jcc 1f ; move.l -(%1),-(%0) ; move.l -(%1),-(%0)\n" |
186 | "1:" |
187 | : "=a" (d), "=a" (s), "=d" (count) |
188 | : "0" ((char *) d + count), "1" ((char *) s + count), "2" (count)); |
189 | } else { |
190 | long tmp; |
191 | |
192 | asm volatile ("\n" |
193 | " move.l %0,%3\n" |
194 | " lsr.l #1,%3 ; jcc 1f ; move.b -(%1),-(%0) ; subqw #1,%2\n" |
195 | " lsr.l #1,%3 ; jcs 2f\n" /* %0 increased=>bit 2 switched*/ |
196 | " move.w -(%1),-(%0) ; subqw #2,%2 ; jra 2f\n" |
197 | "1: lsr.l #1,%3 ; jcc 2f\n" |
198 | " move.w -(%1),-(%0) ; subqw #2,%2\n" |
199 | "2: move.w %2,%-; lsr.l #2,%2 ; jeq 6f\n" |
200 | " lsr.l #1,%2 ; jcc 3f ; move.l -(%1),-(%0)\n" |
201 | "3: lsr.l #1,%2 ; jcc 4f ; move.l -(%1),-(%0) ; move.l -(%1),-(%0)\n" |
202 | "4: subq.l #1,%2 ; jcs 6f\n" |
203 | "5: move.l -(%1),-(%0); move.l -(%1),-(%0)\n" |
204 | " move.l -(%1),-(%0); move.l -(%1),-(%0)\n" |
205 | " dbra %2,5b ; clr.w %2; subq.l #1,%2; jcc 5b\n" |
206 | "6: move.w %+,%2; btst #1,%2 ; jeq 7f ; move.w -(%1),-(%0)\n" |
207 | "7: btst #0,%2 ; jeq 8f ; move.b -(%1),-(%0)\n" |
208 | "8:" |
209 | : "=a" (d), "=a" (s), "=d" (count), "=d" (tmp) |
210 | : "0" ((char *) d + count), "1" ((char *) s + count), "2" (count)); |
211 | } |
212 | } |
213 | |
214 | return 0; |
215 | } |
216 | |
217 | |
218 | /* ++andreas: Simple and fast version of memmove, assumes size is |
219 | divisible by 16, suitable for moving the whole screen bitplane */ |
220 | static inline void fast_memmove(char *dst, const char *src, size_t size) |
221 | { |
222 | if (!size) |
223 | return; |
224 | if (dst < src) |
225 | asm volatile ("\n" |
226 | "1: movem.l (%0)+,%%d0/%%d1/%%a0/%%a1\n" |
227 | " movem.l %%d0/%%d1/%%a0/%%a1,%1@\n" |
228 | " addq.l #8,%1; addq.l #8,%1\n" |
229 | " dbra %2,1b\n" |
230 | " clr.w %2; subq.l #1,%2\n" |
231 | " jcc 1b" |
232 | : "=a" (src), "=a" (dst), "=d" (size) |
233 | : "0" (src), "1" (dst), "2" (size / 16 - 1) |
234 | : "d0" , "d1" , "a0" , "a1" , "memory" ); |
235 | else |
236 | asm volatile ("\n" |
237 | "1: subq.l #8,%0; subq.l #8,%0\n" |
238 | " movem.l %0@,%%d0/%%d1/%%a0/%%a1\n" |
239 | " movem.l %%d0/%%d1/%%a0/%%a1,-(%1)\n" |
240 | " dbra %2,1b\n" |
241 | " clr.w %2; subq.l #1,%2\n" |
242 | " jcc 1b" |
243 | : "=a" (src), "=a" (dst), "=d" (size) |
244 | : "0" (src + size), "1" (dst + size), "2" (size / 16 - 1) |
245 | : "d0" , "d1" , "a0" , "a1" , "memory" ); |
246 | } |
247 | |
248 | #ifdef BPL |
249 | |
250 | /* |
251 | * This expands a up to 8 bit color into two longs |
252 | * for movel operations. |
253 | */ |
254 | static const u32 four2long[] = { |
255 | 0x00000000, 0x000000ff, 0x0000ff00, 0x0000ffff, |
256 | 0x00ff0000, 0x00ff00ff, 0x00ffff00, 0x00ffffff, |
257 | 0xff000000, 0xff0000ff, 0xff00ff00, 0xff00ffff, |
258 | 0xffff0000, 0xffff00ff, 0xffffff00, 0xffffffff, |
259 | }; |
260 | |
261 | static inline void expand8_col2mask(u8 c, u32 m[]) |
262 | { |
263 | m[0] = four2long[c & 15]; |
264 | #if BPL > 4 |
265 | m[1] = four2long[c >> 4]; |
266 | #endif |
267 | } |
268 | |
269 | static inline void expand8_2col2mask(u8 fg, u8 bg, u32 fgm[], u32 bgm[]) |
270 | { |
271 | fgm[0] = four2long[fg & 15] ^ (bgm[0] = four2long[bg & 15]); |
272 | #if BPL > 4 |
273 | fgm[1] = four2long[fg >> 4] ^ (bgm[1] = four2long[bg >> 4]); |
274 | #endif |
275 | } |
276 | |
277 | /* |
278 | * set an 8bit value to a color |
279 | */ |
280 | static inline void fill8_col(u8 *dst, u32 m[]) |
281 | { |
282 | u32 tmp = m[0]; |
283 | dst[0] = tmp; |
284 | dst[2] = (tmp >>= 8); |
285 | #if BPL > 2 |
286 | dst[4] = (tmp >>= 8); |
287 | dst[6] = tmp >> 8; |
288 | #endif |
289 | #if BPL > 4 |
290 | tmp = m[1]; |
291 | dst[8] = tmp; |
292 | dst[10] = (tmp >>= 8); |
293 | dst[12] = (tmp >>= 8); |
294 | dst[14] = tmp >> 8; |
295 | #endif |
296 | } |
297 | |
298 | /* |
299 | * set an 8bit value according to foreground/background color |
300 | */ |
301 | static inline void fill8_2col(u8 *dst, u8 fg, u8 bg, u32 mask) |
302 | { |
303 | u32 fgm[2], bgm[2], tmp; |
304 | |
305 | expand8_2col2mask(fg, bg, fgm, bgm); |
306 | |
307 | mask |= mask << 8; |
308 | #if BPL > 2 |
309 | mask |= mask << 16; |
310 | #endif |
311 | tmp = (mask & fgm[0]) ^ bgm[0]; |
312 | dst[0] = tmp; |
313 | dst[2] = (tmp >>= 8); |
314 | #if BPL > 2 |
315 | dst[4] = (tmp >>= 8); |
316 | dst[6] = tmp >> 8; |
317 | #endif |
318 | #if BPL > 4 |
319 | tmp = (mask & fgm[1]) ^ bgm[1]; |
320 | dst[8] = tmp; |
321 | dst[10] = (tmp >>= 8); |
322 | dst[12] = (tmp >>= 8); |
323 | dst[14] = tmp >> 8; |
324 | #endif |
325 | } |
326 | |
327 | static const u32 two2word[] = { |
328 | 0x00000000, 0xffff0000, 0x0000ffff, 0xffffffff |
329 | }; |
330 | |
331 | static inline void expand16_col2mask(u8 c, u32 m[]) |
332 | { |
333 | m[0] = two2word[c & 3]; |
334 | #if BPL > 2 |
335 | m[1] = two2word[(c >> 2) & 3]; |
336 | #endif |
337 | #if BPL > 4 |
338 | m[2] = two2word[(c >> 4) & 3]; |
339 | m[3] = two2word[c >> 6]; |
340 | #endif |
341 | } |
342 | |
343 | static inline void expand16_2col2mask(u8 fg, u8 bg, u32 fgm[], u32 bgm[]) |
344 | { |
345 | bgm[0] = two2word[bg & 3]; |
346 | fgm[0] = two2word[fg & 3] ^ bgm[0]; |
347 | #if BPL > 2 |
348 | bgm[1] = two2word[(bg >> 2) & 3]; |
349 | fgm[1] = two2word[(fg >> 2) & 3] ^ bgm[1]; |
350 | #endif |
351 | #if BPL > 4 |
352 | bgm[2] = two2word[(bg >> 4) & 3]; |
353 | fgm[2] = two2word[(fg >> 4) & 3] ^ bgm[2]; |
354 | bgm[3] = two2word[bg >> 6]; |
355 | fgm[3] = two2word[fg >> 6] ^ bgm[3]; |
356 | #endif |
357 | } |
358 | |
359 | static inline u32 *fill16_col(u32 *dst, int rows, u32 m[]) |
360 | { |
361 | while (rows) { |
362 | *dst++ = m[0]; |
363 | #if BPL > 2 |
364 | *dst++ = m[1]; |
365 | #endif |
366 | #if BPL > 4 |
367 | *dst++ = m[2]; |
368 | *dst++ = m[3]; |
369 | #endif |
370 | rows--; |
371 | } |
372 | return dst; |
373 | } |
374 | |
375 | static inline void memmove32_col(void *dst, void *src, u32 mask, u32 h, u32 bytes) |
376 | { |
377 | u32 *s, *d, v; |
378 | |
379 | s = src; |
380 | d = dst; |
381 | do { |
382 | v = (*s++ & mask) | (*d & ~mask); |
383 | *d++ = v; |
384 | #if BPL > 2 |
385 | v = (*s++ & mask) | (*d & ~mask); |
386 | *d++ = v; |
387 | #endif |
388 | #if BPL > 4 |
389 | v = (*s++ & mask) | (*d & ~mask); |
390 | *d++ = v; |
391 | v = (*s++ & mask) | (*d & ~mask); |
392 | *d++ = v; |
393 | #endif |
394 | d = (u32 *)((u8 *)d + bytes); |
395 | s = (u32 *)((u8 *)s + bytes); |
396 | } while (--h); |
397 | } |
398 | |
399 | #endif |
400 | |
401 | #endif /* _VIDEO_ATAFB_UTILS_H */ |
402 | |