2 * 240x160 -> 320x213 upscaler for ARM with interpolation
4 * Written by GraÅžvydas "notaz" Ignotas
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions are met:
9 * * Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * * Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * * Neither the name of the organization nor the
15 * names of its contributors may be used to endorse or promote products
16 * derived from this software without specific prior written permission.
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
19 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
24 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
25 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
26 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
27 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 * 8 9 10 11 : 12 13 14 15
35 * 16 17 18 19 : 20 21 22 23
36 * 24 25 26 27 : 28 29 30 31
39 .macro unpack_hi dst, src
40 mov \dst, \src, lsr #16
41 orr \dst, \dst, \dst, lsl #16
45 .macro unpack_lo dst, src
46 mov \dst, \src, lsl #16
47 orr \dst, \dst, \dst, lsr #16
51 @ do 3:5 summing: r2 = (s1*3 + s2*5 + 4) / 8
54 add r2,\s1,\s1, lsl #1 @ r2 = s1 * 3
55 add r2, r2,\s2, lsl #2
56 add r2, r2,\s2 @ r2 += s2 * 5
57 add r2, r2, r12,lsl #2 @ sum += round * 4
58 and r2, lr, r2, lsr #3 @ mask_to_unpacked(sum / 8)
61 @ do 14:7:7:4: r2 = (s1*14 + s2*7 + s3*7 + s4*4 + 16) / 32
63 .macro do_14_7_7_4 s1, s2, s3, s4
65 sub r2, r2,\s1, lsl #1 @ r2 = s1 * 14
66 add r2, r2,\s2, lsl #3
67 sub r2, r2,\s2 @ r2 += s2 * 7
68 add r2, r2,\s3, lsl #3
69 sub r2, r2,\s3 @ r2 += s3 * 7
70 add r2, r2,\s4, lsl #2 @ r2 += s4 * 4
71 add r2, r2, r12,lsl #3 @ sum += round * 16
72 and r2, lr, r2, lsr #5 @ mask_to_unpacked(sum / 32)
75 .global upscale_aspect @ u16 *dst, u16 *src
79 orr lr, lr, #0x0000f800 @ for "unpacked" form of
80 orr lr, lr, #0x07e00000 @ 00000ggg'ggg00000'rrrrr000'000bbbbb
82 orr r12,r12,#0x00000800
83 orr r12,r12,#0x00200000 @ rounding constant
85 mov r8, #((240/6)-1) << 24 @ cols
86 orr r8, r8, #160/3 @ rows
91 ldr r11,[r1, #320*2*1]
101 orr r2, r2, r2, lsr #16
104 orr r2, r3, r2, lsl #16
107 ldr r10,[r1, #320*2*2]
110 orr r3, r2, r2, lsl #16
111 mov r3, r3, lsr #16 @ 8
113 do_14_7_7_4 r7, r5, r6, r4
114 orr r2, r2, r2, lsr #16
115 orr r2, r3, r2, lsl #16
116 str r2, [r0, #320*2*1] @ 8,9
122 orr r3, r2, r2, lsl #16
125 do_14_7_7_4 r7, r9, r6, r4
126 orr r2, r2, r2, lsr #16
127 orr r2, r3, r2, lsl #16
128 str r2, [r0, #320*2*2] @ 16,17
131 orr r2, r2, r2, lsr #16
134 orr r2, r3, r2, lsl #16
135 str r2, [r0, #320*2*3] @ 24,25
137 ldr r10,[r1, #320*2*1+4]
143 orr r2, r2, r2, lsl #16
145 orr r2, r3, r2, lsr #16
146 str r2, [r0, #4] @ 2,3
148 do_14_7_7_4 r7, r4, r5, r6
149 orr r2, r2, r2, lsl #16
152 ldr r5, [r1, #320*2*2+4]
155 orr r2, r2, r2, lsr #16
156 orr r2, r3, r2, lsl #16
157 str r2, [r0, #320*2*1+4] @ 10,11
161 do_14_7_7_4 r7, r4, r9, r6
162 orr r2, r2, r2, lsl #16
166 orr r2, r2, r2, lsr #16
167 orr r2, r3, r2, lsl #16
168 str r2, [r0, #320*2*2+4] @ 18,19
175 orr r2, r2, r2, lsl #16
177 orr r2, r2, r5, lsl #16
178 str r2, [r0, #320*2*3+4] @ 26,27
184 orr r2, r2, r2, lsr #16
186 orr r2, r2, r11,lsr #16
187 str r2, [r0, #8] @ 4,5
189 ldr r11,[r1, #320*2*1+8]
194 orr r2, r2, r2, lsr #16
196 orr r2, r2, r10,lsr #16
198 str r2, [r0, #12] @ 6,7
203 orr r2, r2, r2, lsl #16
206 do_14_7_7_4 r10, r4, r7, r6
207 orr r2, r2, r2, lsr #16
208 orr r2, r3, r2, lsl #16
209 str r2, [r0, #320*2*1+8] @ 12,13
213 ldr r11,[r1, #320*2*2+8]
215 do_14_7_7_4 r10, r6, r7, r9
216 orr r2, r2, r2, lsl #16
220 orr r2, r2, r2, lsr #16
221 orr r2, r3, r2, lsl #16
222 str r2, [r0, #320*2*1+12] @ 14,15
228 orr r2, r2, r2, lsl #16
231 do_14_7_7_4 r10, r4, r9, r7
232 orr r2, r2, r2, lsr #16
233 orr r2, r3, r2, lsl #16
234 str r2, [r0, #320*2*2+8] @ 20,21
237 orr r2, r2, r2, lsr #16
239 orr r2, r2, r5, lsr #16
240 str r2, [r0, #320*2*3+8] @ 28,29
244 do_14_7_7_4 r10, r6, r9, r5
245 orr r2, r2, r2, lsl #16
249 orr r2, r2, r2, lsr #16
250 orr r2, r3, r2, lsl #16
251 str r2, [r0, #320*2*2+12] @ 22,23
254 orr r2, r2, r2, lsr #16
256 orr r2, r3, r2, lsl #16
258 str r2, [r0, #320*2*3+12] @ 30,31
267 add r1, r1, #(320*2+80)*2
270 add r8, r8, #(240/6) << 24 @ cols
277 ldmia r1!,{r9,r10,r11}
283 orr r2, r2, r2, lsr #16
286 orr r2, r3, r2, lsl #16
293 orr r2, r2, r2, lsl #16
295 orr r2, r2, r10,lsl #16
302 orr r2, r2, r2, lsr #16
304 orr r2, r3, r2, lsl #16
308 orr r2, r2, r2, lsr #16
310 orr r2, r3, r2, lsl #16
317 ldmfd sp!,{r4-r11,pc}
320 .global upscale_aspect_row @ void *dst, void *linesx4, u32 row
322 stmfd sp!,{r4-r11,lr}
324 orr lr, lr, #0x0000f800 @ for "unpacked" form of
325 orr lr, lr, #0x07e00000 @ 00000ggg'ggg00000'rrrrr000'000bbbbb
327 orr r12,r12,#0x00000800
328 orr r12,r12,#0x00200000 @ rounding constant
330 mov r8, #(240/6) @ cols
332 add r0, r0, #(240*320)*2
334 add r0, r0, r2, lsl #3
338 ldr r11,[r1, #240*2*1]
345 ldr r11,[r1, #240*2*2]
348 orr r2, r2, r2, lsr #16
351 orr r2, r3, r2, lsl #16
352 str r2, [r0, #-240*2]! @ 0,8
358 orr r2, r2, r2, lsl #16
360 orr r2, r3, r2, lsr #16
361 str r2, [r0, #4] @ 16,24
364 orr r3, r2, r2, lsl #16
366 do_14_7_7_4 r7, r5, r6, r4
367 orr r2, r2, r2, lsr #16
369 orr r2, r3, r2, lsl #16
370 str r2, [r0, #-240*2]! @ 1,9
374 do_14_7_7_4 r7, r6, r9, r10
375 orr r3, r2, r2, lsl #16
378 orr r2, r2, r2, lsr #16
380 orr r2, r3, r2, lsl #16
381 str r2, [r0, #4] @ 17,25
383 ldr r10,[r1, #240*2*1+4]
389 orr r3, r2, r2, lsl #16
391 do_14_7_7_4 r7, r5, r6, r4
392 orr r2, r2, r2, lsr #16
394 orr r2, r3, r2, lsl #16
395 str r2, [r0, #-240*2]! @ 2,10
399 ldr r4, [r1, #240*2*2+4]
401 orr r2, r2, r2, lsr #16
404 orr r2, r3, r2, lsl #16
405 str r2, [r0, #-240*2] @ 3,11
409 do_14_7_7_4 r7, r6, r9, r5
410 orr r3, r2, r2, lsl #16
413 orr r2, r2, r2, lsr #16
415 orr r2, r2, r3, lsr #16
416 str r2, [r0, #4] @ 18,26
419 orr r2, r2, r2, lsl #16
421 orr r2, r3, r2, lsr #16
422 str r2, [r0, #-240*2+4] @ 19,27
431 orr r2, r2, r2, lsr #16
433 orr r2, r3, r2, lsl #16
434 str r2, [r0, #-240*2*2]! @ 4,12
436 ldr r11,[r1, #240*2*1+8]
439 orr r2, r2, r2, lsl #16
442 orr r2, r3, r2, lsr #16
443 str r2, [r0, #4] @ 20,28
448 ldr r11,[r1, #240*2*2+8]
451 orr r3, r2, r2, lsl #16
453 do_14_7_7_4 r9, r4, r6, r5
454 orr r2, r2, r2, lsr #16
456 orr r2, r2, r3, lsr #16
457 str r2, [r0, #-240*2]! @ 5,13
461 do_14_7_7_4 r9, r5, r6, r7
462 orr r3, r2, r2, lsl #16
465 orr r2, r2, r2, lsr #16
467 orr r2, r3, r2, lsl #16
468 str r2, [r0, #4] @ 21,29
470 ldr r7, [r1, #240*2*1+8]
476 orr r3, r2, r2, lsl #16
478 do_14_7_7_4 r9, r4, r7, r6
479 orr r2, r2, r2, lsr #16
481 orr r2, r2, r3, lsr #16
482 str r2, [r0, #-240*2]! @ 6,14
486 do_14_7_7_4 r9, r5, r7, r4
487 orr r3, r2, r2, lsl #16
490 orr r2, r2, r2, lsr #16
492 orr r2, r3, r2, lsl #16
493 str r2, [r0, #4] @ 22,30
496 orr r2, r2, r2, lsr #16
498 orr r2, r3, r2, lsl #16
499 str r2, [r0, #-240*2]! @ 7,15
502 orr r2, r2, r2, lsl #16
505 orr r2, r3, r2, lsr #16
506 str r2, [r0, #4] @ 23,31
512 ldmfd sp!,{r4-r11,pc}
523 orr r7, r7, r6, lsr #16
526 orr r8, r8, r6, lsr #16
532 orr r7, r6, r7, lsl #16
535 orr r8, r6, r8, lsl #16
538 .global do_rotated_blit @ void *dst, void *linesx4, u32 y
542 add r0, r0, #(240*320)*2
543 sub r0, r0, #(240*40)*2
544 sub r0, r0, #(240-40+4)*2 @ y starts from 4
545 add r0, r0, r2, lsl #1
550 ldr r2, [r1, #240*0*2]
551 ldr r3, [r1, #240*1*2]
552 ldr r4, [r1, #240*2*2]
553 ldr r5, [r1, #240*3*2]
561 ldr r2, [r1, #240*0*2+4]
562 ldr r3, [r1, #240*1*2+4]
563 ldr r4, [r1, #240*2*2+4]
564 ldr r5, [r1, #240*3*2+4]
574 bne rotated_blit_loop
578 @ vim:filetype=armasm