port to updated Cyclone, debug menu
[picodrive.git] / platform / uiq2 / blit.s
CommitLineData
cc68a136 1@ assembly "optimized" blitter and copy functions\r
2@ all pointers must be word-aligned\r
3\r
4@ (c) Copyright 2006, notaz\r
5@ All Rights Reserved\r
6\r
7\r
8@ Convert 0000bbb0 ggg0rrr0\r
9@ to 0000rrr0 ggg0bbb0\r
10\r
11@ r2,r3 - scratch, lr = 0x000F000F\r
12.macro convRGB444 reg\r
13 and r2, \reg, lr @ r2=red\r
14 and r3, \reg, lr, lsl #8 @ r3=blue\r
15 and \reg, \reg, lr, lsl #4 @ green stays in place\r
16 orr \reg, \reg, r2, lsl #8 @ add red back\r
17 orr \reg, \reg, r3, lsr #8 @ add blue back\r
18.endm\r
19\r
20.global vidConvCpyRGB444 @ void *to, void *from, int pixels\r
21\r
22vidConvCpyRGB444:\r
23 stmfd sp!, {r4-r11,lr}\r
24\r
25 mov r12, r2, lsr #4 @ repeats\r
26 mov lr, #0xF0000\r
27 orr lr, lr, #0xF @ lr == pattern 0x000F000F\r
28\r
29\r
30.loopRGB444:\r
31 subs r12, r12, #1\r
32\r
33 @ I first thought storing multiple registers would be faster,\r
34 @ but this doesn't seem to be the case, probably because of\r
35 @ slow video memory we are dealing with\r
36 ldmia r1!, {r4-r11}\r
37 convRGB444 r4\r
38 str r4, [r0], #4\r
39 convRGB444 r5\r
40 str r5, [r0], #4\r
41 convRGB444 r6\r
42 str r6, [r0], #4\r
43 convRGB444 r7\r
44 str r7, [r0], #4\r
45 convRGB444 r8\r
46 str r8, [r0], #4\r
47 convRGB444 r9\r
48 str r9, [r0], #4\r
49 convRGB444 r10\r
50 str r10, [r0], #4\r
51 convRGB444 r11\r
52 str r11, [r0], #4\r
53\r
54 bgt .loopRGB444\r
55\r
56\r
57 ldmfd sp!, {r4-r11,lr}\r
58 bx lr\r
59\r
60\r
61@ Convert 0000bbb0 ggg0rrr0\r
62@ to rrr00ggg 000bbb00\r
63\r
64@ r2,r3 - scratch, lr = 0x07800780\r
65.macro convRGB565 reg\r
66 and r2, \reg, lr, lsr #7 @ r2=red\r
67 and r3, \reg, lr, lsl #1 @ r3=blue\r
68 and \reg, lr, \reg,lsl #3 @ green stays, but needs shifting\r
69 orr \reg, \reg, r2, lsl #12 @ add red back\r
70 orr \reg, \reg, r3, lsr #7 @ add blue back\r
71.endm\r
72\r
73.global vidConvCpyRGB565 @ void *to, void *from, int pixels\r
74\r
75vidConvCpyRGB565:\r
76 stmfd sp!, {r4-r11,lr}\r
77\r
78 mov r12, r2, lsr #4 @ repeats\r
79 mov lr, #0x07800000\r
80 orr lr, lr, #0x780 @ lr == pattern 0x07800780\r
81\r
82.loopRGB565:\r
83 subs r12, r12, #1\r
84\r
85 ldmia r1!, {r4-r11}\r
86 convRGB565 r4\r
87 str r4, [r0], #4\r
88 convRGB565 r5\r
89 str r5, [r0], #4\r
90 convRGB565 r6\r
91 str r6, [r0], #4\r
92 convRGB565 r7\r
93 str r7, [r0], #4\r
94 convRGB565 r8\r
95 str r8, [r0], #4\r
96 convRGB565 r9\r
97 str r9, [r0], #4\r
98 convRGB565 r10\r
99 str r10, [r0], #4\r
100 convRGB565 r11\r
101 str r11, [r0], #4\r
102\r
103 bgt .loopRGB565\r
104\r
105 ldmfd sp!, {r4-r11,lr}\r
106 bx lr\r
107\r
108\r
109@ Convert 0000bbb0 ggg0rrr0 0000bbb0 ggg0rrr0\r
110@ to 00000000 rrr00000 ggg00000 bbb00000 ...\r
111\r
112@ r2,r3 - scratch, lr = 0x0000F000\r
113@ rin - src reg, rout - dest reg (can be same for both; rout can be r3)\r
114.macro convRGB32_l rout rin\r
115 and r2, \rin, lr, lsr #12 @ r2=red\r
116 and r3, \rin, lr, lsr #4 @ r3=blue\r
117 orr r2, r3, r2, lsl #24\r
118 and \rout, lr, \rin, lsl #8 @ green stays, but needs shifting\r
119 orr \rout, \rout, r2, lsr #4 @ add red+blue back\r
120.endm\r
121\r
122@ r2,r3 - scratch, lr = 0x0000F000\r
123@ rin - src reg, rout - dest reg (can be same for both; rout can be r3)\r
124.macro convRGB32_h rout rin\r
125 and r2, \rin, lr, lsl #4 @ r2=red\r
126 mov r3, \rin, lsr #24 @ r3=blue\r
127 orr r2, r3, r2\r
128 and \rout, lr, \rin, lsr #8 @ green\r
129 orr \rout, \rout, r2, lsl #4\r
130.endm\r
131\r
132@ slightly faster conversion, saves 1 opcode, writes output\r
133@ lr = 0x00F000F0, out: r3=lower_pix, r2=higher_pix; trashes rin\r
134.macro convRGB32_2 rin rethigh=0\r
135 and r2, lr, \rin, lsr #4 @ blue\r
136 and r3, \rin, lr\r
137 orr r2, r2, r3, lsl #8 @ g0b0g0b0\r
138\r
139 mov r3, r2, lsl #16 @ g0b00000\r
140 and \rin,lr, \rin, ror #12 @ 00r000r0 (reversed)\r
141 orr r3, r3, \rin, lsr #16 @ g0b000r0\r
142 mov r3, r3, ror #16 @ r3=low\r
143\r
144 str r3, [r0], #4\r
145\r
146 mov r2, r2, lsr #16\r
147.if \rethigh\r
148 orr \rin,r2, \rin, lsl #16\r
149.else\r
150 orr r2, r2, \rin, lsl #16\r
151 str r2, [r0], #4\r
152.endif\r
153.endm\r
154\r
155\r
156.global vidConvCpyRGB32 @ void *to, void *from, int pixels\r
157\r
158vidConvCpyRGB32:\r
159 stmfd sp!, {r4-r7,lr}\r
160\r
161 mov r12, r2, lsr #3 @ repeats\r
162 mov lr, #0x00F00000\r
163 orr lr, lr, #0x00F0\r
164\r
165.loopRGB32:\r
166 subs r12, r12, #1\r
167\r
168 ldmia r1!, {r4-r7}\r
169 convRGB32_2 r4\r
170 convRGB32_2 r5\r
171 convRGB32_2 r6\r
172 convRGB32_2 r7\r
173\r
174 bgt .loopRGB32\r
175\r
176 ldmfd sp!, {r4-r7,lr}\r
177 bx lr\r
178\r
179\r
180@ -------- M2 stuff ---------\r
181\r
182.bss\r
183tmpstore1d: .long\r
184\r
185.text\r
186tmpstore1: .long tmpstore1d\r
187\r
188\r
189@ r3 - scratch, ru - reg with 2 pixels from upper col, rl - ... lower col\r
190.macro rot_str16_90 ru rl\r
191 mov r3, \rl,lsl #16\r
192 mov r3, r3, lsr #16\r
193 orr r3, r3, \ru, lsl #16\r
194 str r3, [r0], #208*2\r
195 mov r3, \ru,lsr #16\r
196 mov r3, r3, lsl #16\r
197 orr r3, r3, \rl, lsr #16\r
198 str r3, [r0], #208*2\r
199.endm\r
200\r
201\r
202.global vidConvCpyM2_16_90 @ void *to, void *from, int width\r
203\r
204vidConvCpyM2_16_90:\r
205 stmfd sp!, {r4-r11,lr}\r
206\r
207 ldr r4, =tmpstore1\r
208 str sp, [r4] @ save sp, we will need sp reg..\r
209 mov sp, r0 @ .. to store our dst\r
210\r
211 @ crashing beyond this point will be fatal (phone reboots), as Symbian OS expects sp to always point to stack\r
212\r
213 sub r2, r2, #1\r
214 mov r12, #0x00670000\r
215 orr r12, r12, r2, lsl #24\r
216 orr r12, r12, r2 @ r12 == ((208-2)/2 << 16) | ((width-1)<<24) | (width-1)\r
217\r
218 add r0, r0, #206*2\r
219 add r1, r1, #8*2 @ skip left border\r
220 add lr, r1, #328*2\r
221\r
222.loopM2_16_90:\r
223 subs r12, r12, #1<<24\r
224\r
225 ldmia r1!, {r4-r7}\r
226 ldmia lr!, {r8-r11}\r
227 rot_str16_90 r4 r8\r
228 rot_str16_90 r5 r9\r
229 rot_str16_90 r6 r10\r
230 rot_str16_90 r7 r11\r
231\r
232 bpl .loopM2_16_90\r
233\r
234 add r12, r12, #1<<24\r
235 subs r12, r12, #0x00010000\r
236 bmi .loopM2_16_90_end\r
237\r
238 add r0, sp, r12, lsr #14 @ calculate new dst pointer\r
239 orr r12, r12, r12, lsl #24 @ restore the width counter\r
240\r
241 @ skip remaining pixels on these 2 lines\r
242 mov r4, #328/8-1 @ width of mode2 in line_pixels/8\r
243 sub r4, r4, r12, lsr #24\r
244 add r1, lr, r4, lsl #4 @ skip src pixels\r
245 add lr, r1, #328*2\r
246 b .loopM2_16_90\r
247\r
248.loopM2_16_90_end:\r
249 @ restore sp\r
250 ldr r4, =tmpstore1\r
251 ldr sp, [r4]\r
252\r
253 ldmfd sp!, {r4-r11,lr}\r
254 bx lr\r
255\r
256\r
257\r
258@ r3 - scratch, ru - reg with 2 pixels from upper col, rl - ... lower col (for right-to-left copies)\r
259.macro rot_str16_270 ru rl\r
260 mov r3, \rl,lsr #16\r
261 mov r3, r3, lsl #16\r
262 orr r3, r3, \ru, lsr #16\r
263 str r3, [r0], #208*2\r
264 mov r3, \ru,lsl #16\r
265 mov r3, r3, lsr #16\r
266 orr r3, r3, \rl, lsl #16\r
267 str r3, [r0], #208*2\r
268.endm\r
269\r
270\r
271.global vidConvCpyM2_16_270 @ void *to, void *from, int width\r
272\r
273vidConvCpyM2_16_270:\r
274 stmfd sp!, {r4-r11,lr}\r
275\r
276 ldr r4, =tmpstore1\r
277 str sp, [r4] @ save sp, we will need sp reg to store our dst\r
278\r
279 sub r2, r2, #1\r
280 mov r12, #0x00670000\r
281 orr r12, r12, r2, lsl #24\r
282 orr r12, r12, r2 @ r12 == ((208-2)/2 << 16) | ((width-1)<<24) | (width-1)\r
283\r
284 add r1, r1, #328*2 @ skip left border+1line\r
285 add lr, r1, #328*2\r
286 add sp, r0, #206*2 @ adjust for algo\r
287\r
288.loopM2_16_270:\r
289 subs r12, r12, #1<<24\r
290\r
291 ldmdb r1!, {r4-r7}\r
292 ldmdb lr!, {r8-r11}\r
293 rot_str16_270 r7 r11 @ update the screen in incrementing direction, reduces tearing slightly\r
294 rot_str16_270 r6 r10\r
295 rot_str16_270 r5 r9\r
296 rot_str16_270 r4 r8\r
297\r
298 bpl .loopM2_16_270\r
299\r
300 add r12, r12, #1<<24\r
301 subs r12, r12, #0x00010000\r
302 bmi .loopM2_16_90_end @ same end as in 90\r
303\r
304 sub r0, sp, r12, lsr #14 @ calculate new dst pointer\r
305 orr r12, r12, r12, lsl #24 @ restore the width counter\r
306\r
307 @ skip remaining pixels on these 2 lines\r
308 mov r4, #328/8-1 @ width of mode2 in line_pixels/8\r
309 sub r4, r4, r12, lsr #24\r
310 sub r1, lr, r4, lsl #4 @ skip src pixels\r
311 add r1, r1, #328*2*2\r
312 add lr, r1, #328*2\r
313 b .loopM2_16_270\r
314\r
315\r
316\r
317.global vidConvCpyM2_RGB32_90 @ void *to, void *from, int width\r
318\r
319vidConvCpyM2_RGB32_90:\r
320 stmfd sp!, {r4-r10,lr}\r
321\r
322 mov lr, #0x00F00000\r
323 orr lr, lr, #0x00F0\r
324\r
325 mov r12, #208/4 @ row counter\r
326 mov r10, r2, lsl #2 @ we do 2 pixel wide copies\r
327\r
328 add r8, r0, #208*4 @ parallel line\r
329 add r1, r1, #0x21000\r
330 add r1, r1, #0x00280 @ r1+=328*207*2+8*2\r
331 mov r9, r1\r
332\r
333.loopM2RGB32_90:\r
334 subs r12, r12, #1\r
335\r
336 @ at first this loop was written differently: src pixels were fetched with ldm's and\r
337 @ dest was not sequential. It ran nearly 2 times slower. It seems it is very important\r
338 @ to do sequential memory access on those items, which we have more (to offload addressing bus?).\r
339\r
340 ldr r4, [r1], #-328*2\r
341 ldr r5, [r1], #-328*2\r
342 ldr r6, [r1], #-328*2\r
343 ldr r7, [r1], #-328*2\r
344\r
345 convRGB32_2 r4, 1\r
346 convRGB32_2 r5, 1\r
347 convRGB32_2 r6, 1\r
348 convRGB32_2 r7, 1\r
349\r
350 str r4, [r8], #4\r
351 str r5, [r8], #4\r
352 str r6, [r8], #4\r
353 str r7, [r8], #4\r
354\r
355 bne .loopM2RGB32_90\r
356\r
357 subs r10, r10, #1\r
358 ldmeqfd sp!, {r4-r10,pc} @ return\r
359\r
360 mov r12, #208/4 @ restore row counter\r
361 mov r0, r8 @ set new dst pointer\r
362 add r8, r0, #208*4\r
363 add r9, r9, #2*2 @ fix src pointer\r
364 mov r1, r9\r
365 b .loopM2RGB32_90\r
366\r
367\r
368\r
369@ converter for vidConvCpyM2_RGB32_270\r
370@ lr = 0x00F000F0, out: r3=lower_pix, r2=higher_pix; trashes rin\r
371.macro convRGB32_3 rin\r
372 and r2, lr, \rin, lsr #4 @ blue\r
373 and r3, \rin, lr\r
374 orr r2, r2, r3, lsl #8 @ g0b0g0b0\r
375\r
376 mov r3, r2, lsl #16 @ g0b00000\r
377 and \rin,lr, \rin, ror #12 @ 00r000r0 (reversed)\r
378 orr r3, r3, \rin, lsr #16 @ g0b000r0\r
379\r
380 mov r2, r2, lsr #16\r
381 orr r2, r2, \rin, lsl #16\r
382 str r2, [r0], #4\r
383\r
384 mov \rin,r3, ror #16 @ r3=low\r
385.endm\r
386\r
387\r
388.global vidConvCpyM2_RGB32_270 @ void *to, void *from, int width\r
389\r
390vidConvCpyM2_RGB32_270:\r
391 stmfd sp!, {r4-r10,lr}\r
392\r
393 mov lr, #0x00F00000\r
394 orr lr, lr, #0x00F0\r
395\r
396 mov r12, #208/4 @ row counter\r
397 mov r10, r2, lsl #2 @ we do 2 pixel wide copies (right to left)\r
398\r
399 add r8, r0, #208*4 @ parallel line\r
400 add r1, r1, #326*2\r
401 mov r9, r1\r
402\r
403.loopM2RGB32_270:\r
404 subs r12, r12, #1\r
405\r
406 ldr r4, [r1], #328*2\r
407 ldr r5, [r1], #328*2\r
408 ldr r6, [r1], #328*2\r
409 ldr r7, [r1], #328*2\r
410\r
411 convRGB32_3 r4\r
412 convRGB32_3 r5\r
413 convRGB32_3 r6\r
414 convRGB32_3 r7\r
415\r
416 str r4, [r8], #4\r
417 str r5, [r8], #4\r
418 str r6, [r8], #4\r
419 str r7, [r8], #4\r
420\r
421 bne .loopM2RGB32_270\r
422\r
423 subs r10, r10, #1\r
424 ldmeqfd sp!, {r4-r10,pc} @ return\r
425\r
426 mov r12, #208/4 @ restore row counter\r
427 mov r0, r8 @ set new dst pointer\r
428 add r8, r0, #208*4\r
429 sub r9, r9, #2*2 @ fix src pointer\r
430 mov r1, r9\r
431 b .loopM2RGB32_270\r
432\r