media/libyuv/source/rotate_neon.cc@b8a032363ba2
media/libyuv/source/rotate_neon.cc
Thu, 22 Jan 2015 13:21:57 +0100
- author
- Michael Schloh von Bennewitz <michael@schloh.com>
- date
- Thu, 22 Jan 2015 13:21:57 +0100
- branch
- TOR_BUG_9701
- changeset 15
- b8a032363ba2
- permissions
- -rw-r--r--
Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6
1 /*
2 * Copyright 2011 The LibYuv Project Authors. All rights reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
11 #include "libyuv/row.h"
13 #include "libyuv/basic_types.h"
15 #ifdef __cplusplus
16 namespace libyuv {
17 extern "C" {
18 #endif
20 #if !defined(LIBYUV_DISABLE_NEON) && defined(__ARM_NEON__)
21 static uvec8 kVTbl4x4Transpose =
22 { 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15 };
24 void TransposeWx8_NEON(const uint8* src, int src_stride,
25 uint8* dst, int dst_stride,
26 int width) {
27 asm volatile (
28 // loops are on blocks of 8. loop will stop when
29 // counter gets to or below 0. starting the counter
30 // at w-8 allow for this
31 "sub %4, #8 \n"
33 // handle 8x8 blocks. this should be the majority of the plane
34 ".p2align 2 \n"
35 "1: \n"
36 "mov r9, %0 \n"
38 "vld1.8 {d0}, [r9], %1 \n"
39 "vld1.8 {d1}, [r9], %1 \n"
40 "vld1.8 {d2}, [r9], %1 \n"
41 "vld1.8 {d3}, [r9], %1 \n"
42 "vld1.8 {d4}, [r9], %1 \n"
43 "vld1.8 {d5}, [r9], %1 \n"
44 "vld1.8 {d6}, [r9], %1 \n"
45 "vld1.8 {d7}, [r9] \n"
47 "vtrn.8 d1, d0 \n"
48 "vtrn.8 d3, d2 \n"
49 "vtrn.8 d5, d4 \n"
50 "vtrn.8 d7, d6 \n"
52 "vtrn.16 d1, d3 \n"
53 "vtrn.16 d0, d2 \n"
54 "vtrn.16 d5, d7 \n"
55 "vtrn.16 d4, d6 \n"
57 "vtrn.32 d1, d5 \n"
58 "vtrn.32 d0, d4 \n"
59 "vtrn.32 d3, d7 \n"
60 "vtrn.32 d2, d6 \n"
62 "vrev16.8 q0, q0 \n"
63 "vrev16.8 q1, q1 \n"
64 "vrev16.8 q2, q2 \n"
65 "vrev16.8 q3, q3 \n"
67 "mov r9, %2 \n"
69 "vst1.8 {d1}, [r9], %3 \n"
70 "vst1.8 {d0}, [r9], %3 \n"
71 "vst1.8 {d3}, [r9], %3 \n"
72 "vst1.8 {d2}, [r9], %3 \n"
73 "vst1.8 {d5}, [r9], %3 \n"
74 "vst1.8 {d4}, [r9], %3 \n"
75 "vst1.8 {d7}, [r9], %3 \n"
76 "vst1.8 {d6}, [r9] \n"
78 "add %0, #8 \n" // src += 8
79 "add %2, %2, %3, lsl #3 \n" // dst += 8 * dst_stride
80 "subs %4, #8 \n" // w -= 8
81 "bge 1b \n"
83 // add 8 back to counter. if the result is 0 there are
84 // no residuals.
85 "adds %4, #8 \n"
86 "beq 4f \n"
88 // some residual, so between 1 and 7 lines left to transpose
89 "cmp %4, #2 \n"
90 "blt 3f \n"
92 "cmp %4, #4 \n"
93 "blt 2f \n"
95 // 4x8 block
96 "mov r9, %0 \n"
97 "vld1.32 {d0[0]}, [r9], %1 \n"
98 "vld1.32 {d0[1]}, [r9], %1 \n"
99 "vld1.32 {d1[0]}, [r9], %1 \n"
100 "vld1.32 {d1[1]}, [r9], %1 \n"
101 "vld1.32 {d2[0]}, [r9], %1 \n"
102 "vld1.32 {d2[1]}, [r9], %1 \n"
103 "vld1.32 {d3[0]}, [r9], %1 \n"
104 "vld1.32 {d3[1]}, [r9] \n"
106 "mov r9, %2 \n"
108 "vld1.8 {q3}, [%5] \n"
110 "vtbl.8 d4, {d0, d1}, d6 \n"
111 "vtbl.8 d5, {d0, d1}, d7 \n"
112 "vtbl.8 d0, {d2, d3}, d6 \n"
113 "vtbl.8 d1, {d2, d3}, d7 \n"
115 // TODO(frkoenig): Rework shuffle above to
116 // write out with 4 instead of 8 writes.
117 "vst1.32 {d4[0]}, [r9], %3 \n"
118 "vst1.32 {d4[1]}, [r9], %3 \n"
119 "vst1.32 {d5[0]}, [r9], %3 \n"
120 "vst1.32 {d5[1]}, [r9] \n"
122 "add r9, %2, #4 \n"
123 "vst1.32 {d0[0]}, [r9], %3 \n"
124 "vst1.32 {d0[1]}, [r9], %3 \n"
125 "vst1.32 {d1[0]}, [r9], %3 \n"
126 "vst1.32 {d1[1]}, [r9] \n"
128 "add %0, #4 \n" // src += 4
129 "add %2, %2, %3, lsl #2 \n" // dst += 4 * dst_stride
130 "subs %4, #4 \n" // w -= 4
131 "beq 4f \n"
133 // some residual, check to see if it includes a 2x8 block,
134 // or less
135 "cmp %4, #2 \n"
136 "blt 3f \n"
138 // 2x8 block
139 "2: \n"
140 "mov r9, %0 \n"
141 "vld1.16 {d0[0]}, [r9], %1 \n"
142 "vld1.16 {d1[0]}, [r9], %1 \n"
143 "vld1.16 {d0[1]}, [r9], %1 \n"
144 "vld1.16 {d1[1]}, [r9], %1 \n"
145 "vld1.16 {d0[2]}, [r9], %1 \n"
146 "vld1.16 {d1[2]}, [r9], %1 \n"
147 "vld1.16 {d0[3]}, [r9], %1 \n"
148 "vld1.16 {d1[3]}, [r9] \n"
150 "vtrn.8 d0, d1 \n"
152 "mov r9, %2 \n"
154 "vst1.64 {d0}, [r9], %3 \n"
155 "vst1.64 {d1}, [r9] \n"
157 "add %0, #2 \n" // src += 2
158 "add %2, %2, %3, lsl #1 \n" // dst += 2 * dst_stride
159 "subs %4, #2 \n" // w -= 2
160 "beq 4f \n"
162 // 1x8 block
163 "3: \n"
164 "vld1.8 {d0[0]}, [%0], %1 \n"
165 "vld1.8 {d0[1]}, [%0], %1 \n"
166 "vld1.8 {d0[2]}, [%0], %1 \n"
167 "vld1.8 {d0[3]}, [%0], %1 \n"
168 "vld1.8 {d0[4]}, [%0], %1 \n"
169 "vld1.8 {d0[5]}, [%0], %1 \n"
170 "vld1.8 {d0[6]}, [%0], %1 \n"
171 "vld1.8 {d0[7]}, [%0] \n"
173 "vst1.64 {d0}, [%2] \n"
175 "4: \n"
177 : "+r"(src), // %0
178 "+r"(src_stride), // %1
179 "+r"(dst), // %2
180 "+r"(dst_stride), // %3
181 "+r"(width) // %4
182 : "r"(&kVTbl4x4Transpose) // %5
183 : "memory", "cc", "r9", "q0", "q1", "q2", "q3"
184 );
185 }
187 static uvec8 kVTbl4x4TransposeDi =
188 { 0, 8, 1, 9, 2, 10, 3, 11, 4, 12, 5, 13, 6, 14, 7, 15 };
190 void TransposeUVWx8_NEON(const uint8* src, int src_stride,
191 uint8* dst_a, int dst_stride_a,
192 uint8* dst_b, int dst_stride_b,
193 int width) {
194 asm volatile (
195 // loops are on blocks of 8. loop will stop when
196 // counter gets to or below 0. starting the counter
197 // at w-8 allow for this
198 "sub %6, #8 \n"
200 // handle 8x8 blocks. this should be the majority of the plane
201 ".p2align 2 \n"
202 "1: \n"
203 "mov r9, %0 \n"
205 "vld2.8 {d0, d1}, [r9], %1 \n"
206 "vld2.8 {d2, d3}, [r9], %1 \n"
207 "vld2.8 {d4, d5}, [r9], %1 \n"
208 "vld2.8 {d6, d7}, [r9], %1 \n"
209 "vld2.8 {d16, d17}, [r9], %1 \n"
210 "vld2.8 {d18, d19}, [r9], %1 \n"
211 "vld2.8 {d20, d21}, [r9], %1 \n"
212 "vld2.8 {d22, d23}, [r9] \n"
214 "vtrn.8 q1, q0 \n"
215 "vtrn.8 q3, q2 \n"
216 "vtrn.8 q9, q8 \n"
217 "vtrn.8 q11, q10 \n"
219 "vtrn.16 q1, q3 \n"
220 "vtrn.16 q0, q2 \n"
221 "vtrn.16 q9, q11 \n"
222 "vtrn.16 q8, q10 \n"
224 "vtrn.32 q1, q9 \n"
225 "vtrn.32 q0, q8 \n"
226 "vtrn.32 q3, q11 \n"
227 "vtrn.32 q2, q10 \n"
229 "vrev16.8 q0, q0 \n"
230 "vrev16.8 q1, q1 \n"
231 "vrev16.8 q2, q2 \n"
232 "vrev16.8 q3, q3 \n"
233 "vrev16.8 q8, q8 \n"
234 "vrev16.8 q9, q9 \n"
235 "vrev16.8 q10, q10 \n"
236 "vrev16.8 q11, q11 \n"
238 "mov r9, %2 \n"
240 "vst1.8 {d2}, [r9], %3 \n"
241 "vst1.8 {d0}, [r9], %3 \n"
242 "vst1.8 {d6}, [r9], %3 \n"
243 "vst1.8 {d4}, [r9], %3 \n"
244 "vst1.8 {d18}, [r9], %3 \n"
245 "vst1.8 {d16}, [r9], %3 \n"
246 "vst1.8 {d22}, [r9], %3 \n"
247 "vst1.8 {d20}, [r9] \n"
249 "mov r9, %4 \n"
251 "vst1.8 {d3}, [r9], %5 \n"
252 "vst1.8 {d1}, [r9], %5 \n"
253 "vst1.8 {d7}, [r9], %5 \n"
254 "vst1.8 {d5}, [r9], %5 \n"
255 "vst1.8 {d19}, [r9], %5 \n"
256 "vst1.8 {d17}, [r9], %5 \n"
257 "vst1.8 {d23}, [r9], %5 \n"
258 "vst1.8 {d21}, [r9] \n"
260 "add %0, #8*2 \n" // src += 8*2
261 "add %2, %2, %3, lsl #3 \n" // dst_a += 8 * dst_stride_a
262 "add %4, %4, %5, lsl #3 \n" // dst_b += 8 * dst_stride_b
263 "subs %6, #8 \n" // w -= 8
264 "bge 1b \n"
266 // add 8 back to counter. if the result is 0 there are
267 // no residuals.
268 "adds %6, #8 \n"
269 "beq 4f \n"
271 // some residual, so between 1 and 7 lines left to transpose
272 "cmp %6, #2 \n"
273 "blt 3f \n"
275 "cmp %6, #4 \n"
276 "blt 2f \n"
278 //TODO(frkoenig): Clean this up
279 // 4x8 block
280 "mov r9, %0 \n"
281 "vld1.64 {d0}, [r9], %1 \n"
282 "vld1.64 {d1}, [r9], %1 \n"
283 "vld1.64 {d2}, [r9], %1 \n"
284 "vld1.64 {d3}, [r9], %1 \n"
285 "vld1.64 {d4}, [r9], %1 \n"
286 "vld1.64 {d5}, [r9], %1 \n"
287 "vld1.64 {d6}, [r9], %1 \n"
288 "vld1.64 {d7}, [r9] \n"
290 "vld1.8 {q15}, [%7] \n"
292 "vtrn.8 q0, q1 \n"
293 "vtrn.8 q2, q3 \n"
295 "vtbl.8 d16, {d0, d1}, d30 \n"
296 "vtbl.8 d17, {d0, d1}, d31 \n"
297 "vtbl.8 d18, {d2, d3}, d30 \n"
298 "vtbl.8 d19, {d2, d3}, d31 \n"
299 "vtbl.8 d20, {d4, d5}, d30 \n"
300 "vtbl.8 d21, {d4, d5}, d31 \n"
301 "vtbl.8 d22, {d6, d7}, d30 \n"
302 "vtbl.8 d23, {d6, d7}, d31 \n"
304 "mov r9, %2 \n"
306 "vst1.32 {d16[0]}, [r9], %3 \n"
307 "vst1.32 {d16[1]}, [r9], %3 \n"
308 "vst1.32 {d17[0]}, [r9], %3 \n"
309 "vst1.32 {d17[1]}, [r9], %3 \n"
311 "add r9, %2, #4 \n"
312 "vst1.32 {d20[0]}, [r9], %3 \n"
313 "vst1.32 {d20[1]}, [r9], %3 \n"
314 "vst1.32 {d21[0]}, [r9], %3 \n"
315 "vst1.32 {d21[1]}, [r9] \n"
317 "mov r9, %4 \n"
319 "vst1.32 {d18[0]}, [r9], %5 \n"
320 "vst1.32 {d18[1]}, [r9], %5 \n"
321 "vst1.32 {d19[0]}, [r9], %5 \n"
322 "vst1.32 {d19[1]}, [r9], %5 \n"
324 "add r9, %4, #4 \n"
325 "vst1.32 {d22[0]}, [r9], %5 \n"
326 "vst1.32 {d22[1]}, [r9], %5 \n"
327 "vst1.32 {d23[0]}, [r9], %5 \n"
328 "vst1.32 {d23[1]}, [r9] \n"
330 "add %0, #4*2 \n" // src += 4 * 2
331 "add %2, %2, %3, lsl #2 \n" // dst_a += 4 * dst_stride_a
332 "add %4, %4, %5, lsl #2 \n" // dst_b += 4 * dst_stride_b
333 "subs %6, #4 \n" // w -= 4
334 "beq 4f \n"
336 // some residual, check to see if it includes a 2x8 block,
337 // or less
338 "cmp %6, #2 \n"
339 "blt 3f \n"
341 // 2x8 block
342 "2: \n"
343 "mov r9, %0 \n"
344 "vld2.16 {d0[0], d2[0]}, [r9], %1 \n"
345 "vld2.16 {d1[0], d3[0]}, [r9], %1 \n"
346 "vld2.16 {d0[1], d2[1]}, [r9], %1 \n"
347 "vld2.16 {d1[1], d3[1]}, [r9], %1 \n"
348 "vld2.16 {d0[2], d2[2]}, [r9], %1 \n"
349 "vld2.16 {d1[2], d3[2]}, [r9], %1 \n"
350 "vld2.16 {d0[3], d2[3]}, [r9], %1 \n"
351 "vld2.16 {d1[3], d3[3]}, [r9] \n"
353 "vtrn.8 d0, d1 \n"
354 "vtrn.8 d2, d3 \n"
356 "mov r9, %2 \n"
358 "vst1.64 {d0}, [r9], %3 \n"
359 "vst1.64 {d2}, [r9] \n"
361 "mov r9, %4 \n"
363 "vst1.64 {d1}, [r9], %5 \n"
364 "vst1.64 {d3}, [r9] \n"
366 "add %0, #2*2 \n" // src += 2 * 2
367 "add %2, %2, %3, lsl #1 \n" // dst_a += 2 * dst_stride_a
368 "add %4, %4, %5, lsl #1 \n" // dst_b += 2 * dst_stride_b
369 "subs %6, #2 \n" // w -= 2
370 "beq 4f \n"
372 // 1x8 block
373 "3: \n"
374 "vld2.8 {d0[0], d1[0]}, [%0], %1 \n"
375 "vld2.8 {d0[1], d1[1]}, [%0], %1 \n"
376 "vld2.8 {d0[2], d1[2]}, [%0], %1 \n"
377 "vld2.8 {d0[3], d1[3]}, [%0], %1 \n"
378 "vld2.8 {d0[4], d1[4]}, [%0], %1 \n"
379 "vld2.8 {d0[5], d1[5]}, [%0], %1 \n"
380 "vld2.8 {d0[6], d1[6]}, [%0], %1 \n"
381 "vld2.8 {d0[7], d1[7]}, [%0] \n"
383 "vst1.64 {d0}, [%2] \n"
384 "vst1.64 {d1}, [%4] \n"
386 "4: \n"
388 : "+r"(src), // %0
389 "+r"(src_stride), // %1
390 "+r"(dst_a), // %2
391 "+r"(dst_stride_a), // %3
392 "+r"(dst_b), // %4
393 "+r"(dst_stride_b), // %5
394 "+r"(width) // %6
395 : "r"(&kVTbl4x4TransposeDi) // %7
396 : "memory", "cc", "r9",
397 "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11"
398 );
399 }
400 #endif
402 #ifdef __cplusplus
403 } // extern "C"
404 } // namespace libyuv
405 #endif
