1 /*****************************************************************************
2 * copy.c: Fast YV12/NV12 copy
3 *****************************************************************************
4 * Copyright (C) 2010 Laurent Aimar
7 * Authors: Laurent Aimar <fenrir _AT_ videolan _DOT_ org>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
22 *****************************************************************************/
28 #include <vlc_common.h>
29 #include <vlc_picture.h>
35 int CopyInitCache(copy_cache_t *cache, unsigned width)
37 #ifdef CAN_COMPILE_SSE2
38 cache->size = __MAX((width + 0x0f) & ~ 0x0f, 4096);
39 cache->buffer = vlc_memalign(16, cache->size);
43 (void) cache; (void) width;
48 void CopyCleanCache(copy_cache_t *cache)
50 #ifdef CAN_COMPILE_SSE2
51 vlc_free(cache->buffer);
59 #ifdef CAN_COMPILE_SSE2
60 /* Copy 64 bytes from srcp to dstp loading data with the SSE>=2 instruction
61 * load and storing data with the SSE>=2 instruction store.
63 #define COPY64(dstp, srcp, load, store) \
65 load " 0(%[src]), %%xmm1\n" \
66 load " 16(%[src]), %%xmm2\n" \
67 load " 32(%[src]), %%xmm3\n" \
68 load " 48(%[src]), %%xmm4\n" \
69 store " %%xmm1, 0(%[dst])\n" \
70 store " %%xmm2, 16(%[dst])\n" \
71 store " %%xmm3, 32(%[dst])\n" \
72 store " %%xmm4, 48(%[dst])\n" \
73 : : [dst]"r"(dstp), [src]"r"(srcp) : "memory")
76 # undef vlc_CPU_SSE4_1
77 # define vlc_CPU_SSE4_1() ((cpu & VLC_CPU_SSE4_1) != 0)
82 # define vlc_CPU_SSSE3() ((cpu & VLC_CPU_SSSE3) != 0)
87 # define vlc_CPU_SSE2() ((cpu & VLC_CPU_SSE2) != 0)
90 /* Optimized copy from "Uncacheable Speculative Write Combining" memory
91 * as used by some video surface.
92 * XXX It is really efficient only when SSE4.1 is available.
94 static void CopyFromUswc(uint8_t *dst, size_t dst_pitch,
95 const uint8_t *src, size_t src_pitch,
96 unsigned width, unsigned height,
99 assert(((intptr_t)dst & 0x0f) == 0 && (dst_pitch & 0x0f) == 0);
101 asm volatile ("mfence");
103 for (unsigned y = 0; y < height; y++) {
104 const unsigned unaligned = (-(uintptr_t)src) & 0x0f;
107 for (; x < unaligned; x++)
110 #ifdef CAN_COMPILE_SSE4_1
111 if (vlc_CPU_SSE4_1()) {
113 for (; x+63 < width; x += 64)
114 COPY64(&dst[x], &src[x], "movntdqa", "movdqa");
116 for (; x+63 < width; x += 64)
117 COPY64(&dst[x], &src[x], "movntdqa", "movdqu");
123 for (; x+63 < width; x += 64)
124 COPY64(&dst[x], &src[x], "movdqa", "movdqa");
126 for (; x+63 < width; x += 64)
127 COPY64(&dst[x], &src[x], "movdqa", "movdqu");
131 for (; x < width; x++)
139 static void Copy2d(uint8_t *dst, size_t dst_pitch,
140 const uint8_t *src, size_t src_pitch,
141 unsigned width, unsigned height)
143 assert(((intptr_t)src & 0x0f) == 0 && (src_pitch & 0x0f) == 0);
145 asm volatile ("mfence");
147 for (unsigned y = 0; y < height; y++) {
150 bool unaligned = ((intptr_t)dst & 0x0f) != 0;
152 for (; x+63 < width; x += 64)
153 COPY64(&dst[x], &src[x], "movdqa", "movntdq");
155 for (; x+63 < width; x += 64)
156 COPY64(&dst[x], &src[x], "movdqa", "movdqu");
159 for (; x < width; x++)
167 static void SSE_SplitUV(uint8_t *dstu, size_t dstu_pitch,
168 uint8_t *dstv, size_t dstv_pitch,
169 const uint8_t *src, size_t src_pitch,
170 unsigned width, unsigned height, unsigned cpu)
172 const uint8_t shuffle[] = { 0, 2, 4, 6, 8, 10, 12, 14,
173 1, 3, 5, 7, 9, 11, 13, 15 };
174 const uint8_t mask[] = { 0xff, 0x00, 0xff, 0x00, 0xff, 0x00, 0xff, 0x00,
175 0xff, 0x00, 0xff, 0x00, 0xff, 0x00, 0xff, 0x00 };
177 assert(((intptr_t)src & 0x0f) == 0 && (src_pitch & 0x0f) == 0);
179 asm volatile ("mfence");
181 for (unsigned y = 0; y < height; y++) {
185 "movdqa 0(%[src]), %%xmm0\n" \
186 "movdqa 16(%[src]), %%xmm1\n" \
187 "movdqa 32(%[src]), %%xmm2\n" \
188 "movdqa 48(%[src]), %%xmm3\n"
191 "movq %%xmm0, 0(%[dst1])\n" \
192 "movq %%xmm1, 8(%[dst1])\n" \
193 "movhpd %%xmm0, 0(%[dst2])\n" \
194 "movhpd %%xmm1, 8(%[dst2])\n" \
195 "movq %%xmm2, 16(%[dst1])\n" \
196 "movq %%xmm3, 24(%[dst1])\n" \
197 "movhpd %%xmm2, 16(%[dst2])\n" \
198 "movhpd %%xmm3, 24(%[dst2])\n"
200 #ifdef CAN_COMPILE_SSSE3
203 for (x = 0; x < (width & ~31); x += 32) {
205 "movdqu (%[shuffle]), %%xmm7\n"
207 "pshufb %%xmm7, %%xmm0\n"
208 "pshufb %%xmm7, %%xmm1\n"
209 "pshufb %%xmm7, %%xmm2\n"
210 "pshufb %%xmm7, %%xmm3\n"
212 : : [dst1]"r"(&dstu[x]), [dst2]"r"(&dstv[x]), [src]"r"(&src[2*x]), [shuffle]"r"(shuffle) : "memory");
217 for (x = 0; x < (width & ~31); x += 32) {
219 "movdqu (%[mask]), %%xmm7\n"
221 "movdqa %%xmm0, %%xmm4\n"
222 "movdqa %%xmm1, %%xmm5\n"
223 "movdqa %%xmm2, %%xmm6\n"
226 "pand %%xmm7, %%xmm4\n"
227 "pand %%xmm7, %%xmm5\n"
228 "pand %%xmm7, %%xmm6\n"
229 "packuswb %%xmm4, %%xmm0\n"
230 "packuswb %%xmm5, %%xmm1\n"
231 "pand %%xmm3, %%xmm7\n"
234 "packuswb %%xmm6, %%xmm2\n"
235 "packuswb %%xmm7, %%xmm3\n"
237 : : [dst2]"r"(&dstu[x]), [dst1]"r"(&dstv[x]), [src]"r"(&src[2*x]), [mask]"r"(mask) : "memory");
243 for (; x < width; x++) {
244 dstu[x] = src[2*x+0];
245 dstv[x] = src[2*x+1];
253 static void SSE_CopyPlane(uint8_t *dst, size_t dst_pitch,
254 const uint8_t *src, size_t src_pitch,
255 uint8_t *cache, size_t cache_size,
256 unsigned width, unsigned height, unsigned cpu)
258 const unsigned w16 = (width+15) & ~15;
259 const unsigned hstep = cache_size / w16;
262 for (unsigned y = 0; y < height; y += hstep) {
263 const unsigned hblock = __MIN(hstep, height - y);
265 /* Copy a bunch of line into our cache */
266 CopyFromUswc(cache, w16,
270 /* Copy from our cache to the destination */
271 Copy2d(dst, dst_pitch,
276 src += src_pitch * hblock;
277 dst += dst_pitch * hblock;
279 asm volatile ("mfence");
282 static void SSE_SplitPlanes(uint8_t *dstu, size_t dstu_pitch,
283 uint8_t *dstv, size_t dstv_pitch,
284 const uint8_t *src, size_t src_pitch,
285 uint8_t *cache, size_t cache_size,
286 unsigned width, unsigned height, unsigned cpu)
288 const unsigned w2_16 = (2*width+15) & ~15;
289 const unsigned hstep = cache_size / w2_16;
292 for (unsigned y = 0; y < height; y += hstep) {
293 const unsigned hblock = __MIN(hstep, height - y);
295 /* Copy a bunch of line into our cache */
296 CopyFromUswc(cache, w2_16, src, src_pitch,
297 2*width, hblock, cpu);
299 /* Copy from our cache to the destination */
300 SSE_SplitUV(dstu, dstu_pitch, dstv, dstv_pitch,
301 cache, w2_16, width, hblock, cpu);
304 src += src_pitch * hblock;
305 dstu += dstu_pitch * hblock;
306 dstv += dstv_pitch * hblock;
308 asm volatile ("mfence");
311 static void SSE_CopyFromNv12(picture_t *dst,
312 uint8_t *src[2], size_t src_pitch[2],
313 unsigned width, unsigned height,
316 const unsigned cpu = vlc_CPU();
319 SSE_CopyPlane(dst->p[0].p_pixels, dst->p[0].i_pitch,
320 src[0], src_pitch[0],
321 cache->buffer, cache->size,
323 SSE_SplitPlanes(dst->p[2].p_pixels, dst->p[2].i_pitch,
324 dst->p[1].p_pixels, dst->p[1].i_pitch,
325 src[1], src_pitch[1],
326 cache->buffer, cache->size,
327 width/2, height/2, cpu);
328 asm volatile ("emms");
331 static void SSE_CopyFromYv12(picture_t *dst,
332 uint8_t *src[3], size_t src_pitch[3],
333 unsigned width, unsigned height,
336 const unsigned cpu = vlc_CPU();
339 for (unsigned n = 0; n < 3; n++) {
340 const unsigned d = n > 0 ? 2 : 1;
341 SSE_CopyPlane(dst->p[n].p_pixels, dst->p[n].i_pitch,
342 src[n], src_pitch[n],
343 cache->buffer, cache->size,
344 width/d, height/d, cpu);
346 asm volatile ("emms");
349 #endif /* CAN_COMPILE_SSE2 */
351 static void CopyPlane(uint8_t *dst, size_t dst_pitch,
352 const uint8_t *src, size_t src_pitch,
353 unsigned width, unsigned height)
355 for (unsigned y = 0; y < height; y++) {
356 memcpy(dst, src, width);
362 static void SplitPlanes(uint8_t *dstu, size_t dstu_pitch,
363 uint8_t *dstv, size_t dstv_pitch,
364 const uint8_t *src, size_t src_pitch,
365 unsigned width, unsigned height)
367 for (unsigned y = 0; y < height; y++) {
368 for (unsigned x = 0; x < width; x++) {
369 dstu[x] = src[2*x+0];
370 dstv[x] = src[2*x+1];
378 void CopyFromNv12(picture_t *dst, uint8_t *src[2], size_t src_pitch[2],
379 unsigned width, unsigned height,
382 #ifdef CAN_COMPILE_SSE2
383 unsigned cpu = vlc_CPU();
385 return SSE_CopyFromNv12(dst, src, src_pitch, width, height, cache);
390 CopyPlane(dst->p[0].p_pixels, dst->p[0].i_pitch,
391 src[0], src_pitch[0],
393 SplitPlanes(dst->p[2].p_pixels, dst->p[2].i_pitch,
394 dst->p[1].p_pixels, dst->p[1].i_pitch,
395 src[1], src_pitch[1],
399 void CopyFromYv12(picture_t *dst, uint8_t *src[3], size_t src_pitch[3],
400 unsigned width, unsigned height,
403 #ifdef CAN_COMPILE_SSE2
404 unsigned cpu = vlc_CPU();
406 return SSE_CopyFromYv12(dst, src, src_pitch, width, height, cache);
411 CopyPlane(dst->p[0].p_pixels, dst->p[0].i_pitch,
412 src[0], src_pitch[0], width, height);
413 CopyPlane(dst->p[1].p_pixels, dst->p[1].i_pitch,
414 src[1], src_pitch[1], width / 2, height / 2);
415 CopyPlane(dst->p[2].p_pixels, dst->p[2].i_pitch,
416 src[1], src_pitch[2], width / 2, height / 2);