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