2 * VC-1 and WMV3 - DSP functions MMX-optimized
3 * Copyright (c) 2007 Christophe GISQUET <christophe.gisquet@free.fr>
5 * Permission is hereby granted, free of charge, to any person
6 * obtaining a copy of this software and associated documentation
7 * files (the "Software"), to deal in the Software without
8 * restriction, including without limitation the rights to use,
9 * copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following
14 * The above copyright notice and this permission notice shall be
15 * included in all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
18 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
19 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
20 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
21 * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
22 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
23 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
24 * OTHER DEALINGS IN THE SOFTWARE.
27 #include "libavutil/cpu.h"
28 #include "libavutil/mem.h"
29 #include "libavutil/x86/asm.h"
30 #include "libavutil/x86/cpu.h"
31 #include "dsputil_mmx.h"
32 #include "libavcodec/vc1dsp.h"
38 #define OP_AVG(S,D) "pavgb " #S ", " #D " \n\t"
40 /** Add rounder from mm7 to mm3 and pack result at destination */
41 #define NORMALIZE_MMX(SHIFT) \
42 "paddw %%mm7, %%mm3 \n\t" /* +bias-r */ \
43 "paddw %%mm7, %%mm4 \n\t" /* +bias-r */ \
44 "psraw "SHIFT", %%mm3 \n\t" \
45 "psraw "SHIFT", %%mm4 \n\t"
47 #define TRANSFER_DO_PACK(OP) \
48 "packuswb %%mm4, %%mm3 \n\t" \
50 "movq %%mm3, (%2) \n\t"
52 #define TRANSFER_DONT_PACK(OP) \
55 "movq %%mm3, 0(%2) \n\t" \
56 "movq %%mm4, 8(%2) \n\t"
58 /** @see MSPEL_FILTER13_CORE for use as UNPACK macro */
59 #define DO_UNPACK(reg) "punpcklbw %%mm0, " reg "\n\t"
60 #define DONT_UNPACK(reg)
62 /** Compute the rounder 32-r or 8-r and unpacks it to mm7 */
63 #define LOAD_ROUNDER_MMX(ROUND) \
64 "movd "ROUND", %%mm7 \n\t" \
65 "punpcklwd %%mm7, %%mm7 \n\t" \
66 "punpckldq %%mm7, %%mm7 \n\t"
68 #define SHIFT2_LINE(OFF, R0,R1,R2,R3) \
69 "paddw %%mm"#R2", %%mm"#R1" \n\t" \
70 "movd (%0,%3), %%mm"#R0" \n\t" \
71 "pmullw %%mm6, %%mm"#R1" \n\t" \
72 "punpcklbw %%mm0, %%mm"#R0" \n\t" \
73 "movd (%0,%2), %%mm"#R3" \n\t" \
74 "psubw %%mm"#R0", %%mm"#R1" \n\t" \
75 "punpcklbw %%mm0, %%mm"#R3" \n\t" \
76 "paddw %%mm7, %%mm"#R1" \n\t" \
77 "psubw %%mm"#R3", %%mm"#R1" \n\t" \
78 "psraw %4, %%mm"#R1" \n\t" \
79 "movq %%mm"#R1", "#OFF"(%1) \n\t" \
82 /** Sacrifying mm6 allows to pipeline loads from src */
83 static void vc1_put_ver_16b_shift2_mmx(int16_t *dst,
84 const uint8_t *src, x86_reg stride,
85 int rnd, int64_t shift)
88 "mov $3, %%"REG_c" \n\t"
89 LOAD_ROUNDER_MMX("%5")
90 "movq "MANGLE(ff_pw_9)", %%mm6 \n\t"
92 "movd (%0), %%mm2 \n\t"
94 "movd (%0), %%mm3 \n\t"
95 "punpcklbw %%mm0, %%mm2 \n\t"
96 "punpcklbw %%mm0, %%mm3 \n\t"
97 SHIFT2_LINE( 0, 1, 2, 3, 4)
98 SHIFT2_LINE( 24, 2, 3, 4, 1)
99 SHIFT2_LINE( 48, 3, 4, 1, 2)
100 SHIFT2_LINE( 72, 4, 1, 2, 3)
101 SHIFT2_LINE( 96, 1, 2, 3, 4)
102 SHIFT2_LINE(120, 2, 3, 4, 1)
103 SHIFT2_LINE(144, 3, 4, 1, 2)
104 SHIFT2_LINE(168, 4, 1, 2, 3)
109 : "+r"(src), "+r"(dst)
110 : "r"(stride), "r"(-2*stride),
111 "m"(shift), "m"(rnd), "r"(9*stride-4)
117 * Data is already unpacked, so some operations can directly be made from
120 #define VC1_HOR_16b_SHIFT2(OP, OPNAME)\
121 static void OPNAME ## vc1_hor_16b_shift2_mmx(uint8_t *dst, x86_reg stride,\
122 const int16_t *src, int rnd)\
127 rnd -= (-1+9+9-1)*1024; /* Add -1024 bias */\
129 LOAD_ROUNDER_MMX("%4")\
130 "movq "MANGLE(ff_pw_128)", %%mm6\n\t"\
131 "movq "MANGLE(ff_pw_9)", %%mm5 \n\t"\
133 "movq 2*0+0(%1), %%mm1 \n\t"\
134 "movq 2*0+8(%1), %%mm2 \n\t"\
135 "movq 2*1+0(%1), %%mm3 \n\t"\
136 "movq 2*1+8(%1), %%mm4 \n\t"\
137 "paddw 2*3+0(%1), %%mm1 \n\t"\
138 "paddw 2*3+8(%1), %%mm2 \n\t"\
139 "paddw 2*2+0(%1), %%mm3 \n\t"\
140 "paddw 2*2+8(%1), %%mm4 \n\t"\
141 "pmullw %%mm5, %%mm3 \n\t"\
142 "pmullw %%mm5, %%mm4 \n\t"\
143 "psubw %%mm1, %%mm3 \n\t"\
144 "psubw %%mm2, %%mm4 \n\t"\
147 "paddw %%mm6, %%mm3 \n\t"\
148 "paddw %%mm6, %%mm4 \n\t"\
149 TRANSFER_DO_PACK(OP)\
154 : "+r"(h), "+r" (src), "+r" (dst)\
155 : "r"(stride), "m"(rnd)\
160 VC1_HOR_16b_SHIFT2(OP_PUT, put_)
161 VC1_HOR_16b_SHIFT2(OP_AVG, avg_)
165 * Purely vertical or horizontal 1/2 shift interpolation.
166 * Sacrify mm6 for *9 factor.
168 #define VC1_SHIFT2(OP, OPNAME)\
169 static void OPNAME ## vc1_shift2_mmx(uint8_t *dst, const uint8_t *src,\
170 x86_reg stride, int rnd, x86_reg offset)\
174 "mov $8, %%"REG_c" \n\t"\
175 LOAD_ROUNDER_MMX("%5")\
176 "movq "MANGLE(ff_pw_9)", %%mm6\n\t"\
178 "movd 0(%0 ), %%mm3 \n\t"\
179 "movd 4(%0 ), %%mm4 \n\t"\
180 "movd 0(%0,%2), %%mm1 \n\t"\
181 "movd 4(%0,%2), %%mm2 \n\t"\
183 "punpcklbw %%mm0, %%mm3 \n\t"\
184 "punpcklbw %%mm0, %%mm4 \n\t"\
185 "punpcklbw %%mm0, %%mm1 \n\t"\
186 "punpcklbw %%mm0, %%mm2 \n\t"\
187 "paddw %%mm1, %%mm3 \n\t"\
188 "paddw %%mm2, %%mm4 \n\t"\
189 "movd 0(%0,%3), %%mm1 \n\t"\
190 "movd 4(%0,%3), %%mm2 \n\t"\
191 "pmullw %%mm6, %%mm3 \n\t" /* 0,9,9,0*/\
192 "pmullw %%mm6, %%mm4 \n\t" /* 0,9,9,0*/\
193 "punpcklbw %%mm0, %%mm1 \n\t"\
194 "punpcklbw %%mm0, %%mm2 \n\t"\
195 "psubw %%mm1, %%mm3 \n\t" /*-1,9,9,0*/\
196 "psubw %%mm2, %%mm4 \n\t" /*-1,9,9,0*/\
197 "movd 0(%0,%2), %%mm1 \n\t"\
198 "movd 4(%0,%2), %%mm2 \n\t"\
199 "punpcklbw %%mm0, %%mm1 \n\t"\
200 "punpcklbw %%mm0, %%mm2 \n\t"\
201 "psubw %%mm1, %%mm3 \n\t" /*-1,9,9,-1*/\
202 "psubw %%mm2, %%mm4 \n\t" /*-1,9,9,-1*/\
204 "packuswb %%mm4, %%mm3 \n\t"\
206 "movq %%mm3, (%1) \n\t"\
209 "dec %%"REG_c" \n\t"\
211 : "+r"(src), "+r"(dst)\
212 : "r"(offset), "r"(-2*offset), "g"(stride), "m"(rnd),\
214 : "%"REG_c, "memory"\
218 VC1_SHIFT2(OP_PUT, put_)
219 VC1_SHIFT2(OP_AVG, avg_)
222 * Core of the 1/4 and 3/4 shift bicubic interpolation.
224 * @param UNPACK Macro unpacking arguments from 8 to 16bits (can be empty).
225 * @param MOVQ "movd 1" or "movq 2", if data read is already unpacked.
226 * @param A1 Address of 1st tap (beware of unpacked/packed).
227 * @param A2 Address of 2nd tap
228 * @param A3 Address of 3rd tap
229 * @param A4 Address of 4th tap
231 #define MSPEL_FILTER13_CORE(UNPACK, MOVQ, A1, A2, A3, A4) \
232 MOVQ "*0+"A1", %%mm1 \n\t" \
233 MOVQ "*4+"A1", %%mm2 \n\t" \
236 "pmullw "MANGLE(ff_pw_3)", %%mm1\n\t" \
237 "pmullw "MANGLE(ff_pw_3)", %%mm2\n\t" \
238 MOVQ "*0+"A2", %%mm3 \n\t" \
239 MOVQ "*4+"A2", %%mm4 \n\t" \
242 "pmullw %%mm6, %%mm3 \n\t" /* *18 */ \
243 "pmullw %%mm6, %%mm4 \n\t" /* *18 */ \
244 "psubw %%mm1, %%mm3 \n\t" /* 18,-3 */ \
245 "psubw %%mm2, %%mm4 \n\t" /* 18,-3 */ \
246 MOVQ "*0+"A4", %%mm1 \n\t" \
247 MOVQ "*4+"A4", %%mm2 \n\t" \
250 "psllw $2, %%mm1 \n\t" /* 4* */ \
251 "psllw $2, %%mm2 \n\t" /* 4* */ \
252 "psubw %%mm1, %%mm3 \n\t" /* -4,18,-3 */ \
253 "psubw %%mm2, %%mm4 \n\t" /* -4,18,-3 */ \
254 MOVQ "*0+"A3", %%mm1 \n\t" \
255 MOVQ "*4+"A3", %%mm2 \n\t" \
258 "pmullw %%mm5, %%mm1 \n\t" /* *53 */ \
259 "pmullw %%mm5, %%mm2 \n\t" /* *53 */ \
260 "paddw %%mm1, %%mm3 \n\t" /* 4,53,18,-3 */ \
261 "paddw %%mm2, %%mm4 \n\t" /* 4,53,18,-3 */
264 * Macro to build the vertical 16bits version of vc1_put_shift[13].
265 * Here, offset=src_stride. Parameters passed A1 to A4 must use
266 * %3 (src_stride) and %4 (3*src_stride).
268 * @param NAME Either 1 or 3
269 * @see MSPEL_FILTER13_CORE for information on A1->A4
271 #define MSPEL_FILTER13_VER_16B(NAME, A1, A2, A3, A4) \
273 vc1_put_ver_16b_ ## NAME ## _mmx(int16_t *dst, const uint8_t *src, \
274 x86_reg src_stride, \
275 int rnd, int64_t shift) \
280 LOAD_ROUNDER_MMX("%5") \
281 "movq "MANGLE(ff_pw_53)", %%mm5\n\t" \
282 "movq "MANGLE(ff_pw_18)", %%mm6\n\t" \
285 MSPEL_FILTER13_CORE(DO_UNPACK, "movd 1", A1, A2, A3, A4) \
286 NORMALIZE_MMX("%6") \
287 TRANSFER_DONT_PACK(OP_PUT) \
288 /* Last 3 (in fact 4) bytes on the line */ \
289 "movd 8+"A1", %%mm1 \n\t" \
291 "movq %%mm1, %%mm3 \n\t" \
292 "paddw %%mm1, %%mm1 \n\t" \
293 "paddw %%mm3, %%mm1 \n\t" /* 3* */ \
294 "movd 8+"A2", %%mm3 \n\t" \
296 "pmullw %%mm6, %%mm3 \n\t" /* *18 */ \
297 "psubw %%mm1, %%mm3 \n\t" /*18,-3 */ \
298 "movd 8+"A3", %%mm1 \n\t" \
300 "pmullw %%mm5, %%mm1 \n\t" /* *53 */ \
301 "paddw %%mm1, %%mm3 \n\t" /*53,18,-3 */ \
302 "movd 8+"A4", %%mm1 \n\t" \
304 "psllw $2, %%mm1 \n\t" /* 4* */ \
305 "psubw %%mm1, %%mm3 \n\t" \
306 "paddw %%mm7, %%mm3 \n\t" \
307 "psraw %6, %%mm3 \n\t" \
308 "movq %%mm3, 16(%2) \n\t" \
313 : "+r"(h), "+r" (src), "+r" (dst) \
314 : "r"(src_stride), "r"(3*src_stride), \
315 "m"(rnd), "m"(shift) \
321 * Macro to build the horizontal 16bits version of vc1_put_shift[13].
322 * Here, offset=16bits, so parameters passed A1 to A4 should be simple.
324 * @param NAME Either 1 or 3
325 * @see MSPEL_FILTER13_CORE for information on A1->A4
327 #define MSPEL_FILTER13_HOR_16B(NAME, A1, A2, A3, A4, OP, OPNAME) \
329 OPNAME ## vc1_hor_16b_ ## NAME ## _mmx(uint8_t *dst, x86_reg stride, \
330 const int16_t *src, int rnd) \
334 rnd -= (-4+58+13-3)*256; /* Add -256 bias */ \
336 LOAD_ROUNDER_MMX("%4") \
337 "movq "MANGLE(ff_pw_18)", %%mm6 \n\t" \
338 "movq "MANGLE(ff_pw_53)", %%mm5 \n\t" \
341 MSPEL_FILTER13_CORE(DONT_UNPACK, "movq 2", A1, A2, A3, A4) \
342 NORMALIZE_MMX("$7") \
344 "paddw "MANGLE(ff_pw_128)", %%mm3 \n\t" \
345 "paddw "MANGLE(ff_pw_128)", %%mm4 \n\t" \
346 TRANSFER_DO_PACK(OP) \
351 : "+r"(h), "+r" (src), "+r" (dst) \
352 : "r"(stride), "m"(rnd) \
358 * Macro to build the 8bits, any direction, version of vc1_put_shift[13].
359 * Here, offset=src_stride. Parameters passed A1 to A4 must use
360 * %3 (offset) and %4 (3*offset).
362 * @param NAME Either 1 or 3
363 * @see MSPEL_FILTER13_CORE for information on A1->A4
365 #define MSPEL_FILTER13_8B(NAME, A1, A2, A3, A4, OP, OPNAME) \
367 OPNAME ## vc1_## NAME ## _mmx(uint8_t *dst, const uint8_t *src, \
368 x86_reg stride, int rnd, x86_reg offset) \
374 LOAD_ROUNDER_MMX("%6") \
375 "movq "MANGLE(ff_pw_53)", %%mm5 \n\t" \
376 "movq "MANGLE(ff_pw_18)", %%mm6 \n\t" \
379 MSPEL_FILTER13_CORE(DO_UNPACK, "movd 1", A1, A2, A3, A4) \
380 NORMALIZE_MMX("$6") \
381 TRANSFER_DO_PACK(OP) \
386 : "+r"(h), "+r" (src), "+r" (dst) \
387 : "r"(offset), "r"(3*offset), "g"(stride), "m"(rnd) \
392 /** 1/4 shift bicubic interpolation */
393 MSPEL_FILTER13_8B (shift1, "0(%1,%4 )", "0(%1,%3,2)", "0(%1,%3 )", "0(%1 )", OP_PUT, put_)
394 MSPEL_FILTER13_8B (shift1, "0(%1,%4 )", "0(%1,%3,2)", "0(%1,%3 )", "0(%1 )", OP_AVG, avg_)
395 MSPEL_FILTER13_VER_16B(shift1, "0(%1,%4 )", "0(%1,%3,2)", "0(%1,%3 )", "0(%1 )")
396 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_PUT, put_)
397 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_AVG, avg_)
399 /** 3/4 shift bicubic interpolation */
400 MSPEL_FILTER13_8B (shift3, "0(%1 )", "0(%1,%3 )", "0(%1,%3,2)", "0(%1,%4 )", OP_PUT, put_)
401 MSPEL_FILTER13_8B (shift3, "0(%1 )", "0(%1,%3 )", "0(%1,%3,2)", "0(%1,%4 )", OP_AVG, avg_)
402 MSPEL_FILTER13_VER_16B(shift3, "0(%1 )", "0(%1,%3 )", "0(%1,%3,2)", "0(%1,%4 )")
403 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_PUT, put_)
404 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_AVG, avg_)
406 typedef void (*vc1_mspel_mc_filter_ver_16bits)(int16_t *dst, const uint8_t *src, x86_reg src_stride, int rnd, int64_t shift);
407 typedef void (*vc1_mspel_mc_filter_hor_16bits)(uint8_t *dst, x86_reg dst_stride, const int16_t *src, int rnd);
408 typedef void (*vc1_mspel_mc_filter_8bits)(uint8_t *dst, const uint8_t *src, x86_reg stride, int rnd, x86_reg offset);
411 * Interpolate fractional pel values by applying proper vertical then
414 * @param dst Destination buffer for interpolated pels.
415 * @param src Source buffer.
416 * @param stride Stride for both src and dst buffers.
417 * @param hmode Horizontal filter (expressed in quarter pixels shift).
418 * @param hmode Vertical filter.
419 * @param rnd Rounding bias.
421 #define VC1_MSPEL_MC(OP)\
422 static void OP ## vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride,\
423 int hmode, int vmode, int rnd)\
425 static const vc1_mspel_mc_filter_ver_16bits vc1_put_shift_ver_16bits[] =\
426 { NULL, vc1_put_ver_16b_shift1_mmx, vc1_put_ver_16b_shift2_mmx, vc1_put_ver_16b_shift3_mmx };\
427 static const vc1_mspel_mc_filter_hor_16bits vc1_put_shift_hor_16bits[] =\
428 { NULL, OP ## vc1_hor_16b_shift1_mmx, OP ## vc1_hor_16b_shift2_mmx, OP ## vc1_hor_16b_shift3_mmx };\
429 static const vc1_mspel_mc_filter_8bits vc1_put_shift_8bits[] =\
430 { NULL, OP ## vc1_shift1_mmx, OP ## vc1_shift2_mmx, OP ## vc1_shift3_mmx };\
433 "pxor %%mm0, %%mm0 \n\t"\
437 if (vmode) { /* Vertical filter to apply */\
438 if (hmode) { /* Horizontal filter to apply, output to tmp */\
439 static const int shift_value[] = { 0, 5, 1, 5 };\
440 int shift = (shift_value[hmode]+shift_value[vmode])>>1;\
442 DECLARE_ALIGNED(16, int16_t, tmp)[12*8];\
444 r = (1<<(shift-1)) + rnd-1;\
445 vc1_put_shift_ver_16bits[vmode](tmp, src-1, stride, r, shift);\
447 vc1_put_shift_hor_16bits[hmode](dst, stride, tmp+1, 64-rnd);\
450 else { /* No horizontal filter, output 8 lines to dst */\
451 vc1_put_shift_8bits[vmode](dst, src, stride, 1-rnd, stride);\
456 /* Horizontal mode with no vertical mode */\
457 vc1_put_shift_8bits[hmode](dst, src, stride, rnd, 1);\
463 /** Macro to ease bicubic filter interpolation functions declarations */
464 #define DECLARE_FUNCTION(a, b) \
465 static void put_vc1_mspel_mc ## a ## b ## _mmx(uint8_t *dst, \
466 const uint8_t *src, \
470 put_vc1_mspel_mc(dst, src, stride, a, b, rnd); \
472 static void avg_vc1_mspel_mc ## a ## b ## _mmxext(uint8_t *dst, \
473 const uint8_t *src, \
477 avg_vc1_mspel_mc(dst, src, stride, a, b, rnd); \
480 DECLARE_FUNCTION(0, 1)
481 DECLARE_FUNCTION(0, 2)
482 DECLARE_FUNCTION(0, 3)
484 DECLARE_FUNCTION(1, 0)
485 DECLARE_FUNCTION(1, 1)
486 DECLARE_FUNCTION(1, 2)
487 DECLARE_FUNCTION(1, 3)
489 DECLARE_FUNCTION(2, 0)
490 DECLARE_FUNCTION(2, 1)
491 DECLARE_FUNCTION(2, 2)
492 DECLARE_FUNCTION(2, 3)
494 DECLARE_FUNCTION(3, 0)
495 DECLARE_FUNCTION(3, 1)
496 DECLARE_FUNCTION(3, 2)
497 DECLARE_FUNCTION(3, 3)
499 static void vc1_inv_trans_4x4_dc_mmxext(uint8_t *dest, int linesize,
503 dc = (17 * dc + 4) >> 3;
504 dc = (17 * dc + 64) >> 7;
506 "movd %0, %%mm0 \n\t"
507 "pshufw $0, %%mm0, %%mm0 \n\t"
508 "pxor %%mm1, %%mm1 \n\t"
509 "psubw %%mm0, %%mm1 \n\t"
510 "packuswb %%mm0, %%mm0 \n\t"
511 "packuswb %%mm1, %%mm1 \n\t"
515 "movd %0, %%mm2 \n\t"
516 "movd %1, %%mm3 \n\t"
517 "movd %2, %%mm4 \n\t"
518 "movd %3, %%mm5 \n\t"
519 "paddusb %%mm0, %%mm2 \n\t"
520 "paddusb %%mm0, %%mm3 \n\t"
521 "paddusb %%mm0, %%mm4 \n\t"
522 "paddusb %%mm0, %%mm5 \n\t"
523 "psubusb %%mm1, %%mm2 \n\t"
524 "psubusb %%mm1, %%mm3 \n\t"
525 "psubusb %%mm1, %%mm4 \n\t"
526 "psubusb %%mm1, %%mm5 \n\t"
527 "movd %%mm2, %0 \n\t"
528 "movd %%mm3, %1 \n\t"
529 "movd %%mm4, %2 \n\t"
530 "movd %%mm5, %3 \n\t"
531 :"+m"(*(uint32_t*)(dest+0*linesize)),
532 "+m"(*(uint32_t*)(dest+1*linesize)),
533 "+m"(*(uint32_t*)(dest+2*linesize)),
534 "+m"(*(uint32_t*)(dest+3*linesize))
538 static void vc1_inv_trans_4x8_dc_mmxext(uint8_t *dest, int linesize,
542 dc = (17 * dc + 4) >> 3;
543 dc = (12 * dc + 64) >> 7;
545 "movd %0, %%mm0 \n\t"
546 "pshufw $0, %%mm0, %%mm0 \n\t"
547 "pxor %%mm1, %%mm1 \n\t"
548 "psubw %%mm0, %%mm1 \n\t"
549 "packuswb %%mm0, %%mm0 \n\t"
550 "packuswb %%mm1, %%mm1 \n\t"
554 "movd %0, %%mm2 \n\t"
555 "movd %1, %%mm3 \n\t"
556 "movd %2, %%mm4 \n\t"
557 "movd %3, %%mm5 \n\t"
558 "paddusb %%mm0, %%mm2 \n\t"
559 "paddusb %%mm0, %%mm3 \n\t"
560 "paddusb %%mm0, %%mm4 \n\t"
561 "paddusb %%mm0, %%mm5 \n\t"
562 "psubusb %%mm1, %%mm2 \n\t"
563 "psubusb %%mm1, %%mm3 \n\t"
564 "psubusb %%mm1, %%mm4 \n\t"
565 "psubusb %%mm1, %%mm5 \n\t"
566 "movd %%mm2, %0 \n\t"
567 "movd %%mm3, %1 \n\t"
568 "movd %%mm4, %2 \n\t"
569 "movd %%mm5, %3 \n\t"
570 :"+m"(*(uint32_t*)(dest+0*linesize)),
571 "+m"(*(uint32_t*)(dest+1*linesize)),
572 "+m"(*(uint32_t*)(dest+2*linesize)),
573 "+m"(*(uint32_t*)(dest+3*linesize))
577 "movd %0, %%mm2 \n\t"
578 "movd %1, %%mm3 \n\t"
579 "movd %2, %%mm4 \n\t"
580 "movd %3, %%mm5 \n\t"
581 "paddusb %%mm0, %%mm2 \n\t"
582 "paddusb %%mm0, %%mm3 \n\t"
583 "paddusb %%mm0, %%mm4 \n\t"
584 "paddusb %%mm0, %%mm5 \n\t"
585 "psubusb %%mm1, %%mm2 \n\t"
586 "psubusb %%mm1, %%mm3 \n\t"
587 "psubusb %%mm1, %%mm4 \n\t"
588 "psubusb %%mm1, %%mm5 \n\t"
589 "movd %%mm2, %0 \n\t"
590 "movd %%mm3, %1 \n\t"
591 "movd %%mm4, %2 \n\t"
592 "movd %%mm5, %3 \n\t"
593 :"+m"(*(uint32_t*)(dest+0*linesize)),
594 "+m"(*(uint32_t*)(dest+1*linesize)),
595 "+m"(*(uint32_t*)(dest+2*linesize)),
596 "+m"(*(uint32_t*)(dest+3*linesize))
600 static void vc1_inv_trans_8x4_dc_mmxext(uint8_t *dest, int linesize,
604 dc = ( 3 * dc + 1) >> 1;
605 dc = (17 * dc + 64) >> 7;
607 "movd %0, %%mm0 \n\t"
608 "pshufw $0, %%mm0, %%mm0 \n\t"
609 "pxor %%mm1, %%mm1 \n\t"
610 "psubw %%mm0, %%mm1 \n\t"
611 "packuswb %%mm0, %%mm0 \n\t"
612 "packuswb %%mm1, %%mm1 \n\t"
616 "movq %0, %%mm2 \n\t"
617 "movq %1, %%mm3 \n\t"
618 "movq %2, %%mm4 \n\t"
619 "movq %3, %%mm5 \n\t"
620 "paddusb %%mm0, %%mm2 \n\t"
621 "paddusb %%mm0, %%mm3 \n\t"
622 "paddusb %%mm0, %%mm4 \n\t"
623 "paddusb %%mm0, %%mm5 \n\t"
624 "psubusb %%mm1, %%mm2 \n\t"
625 "psubusb %%mm1, %%mm3 \n\t"
626 "psubusb %%mm1, %%mm4 \n\t"
627 "psubusb %%mm1, %%mm5 \n\t"
628 "movq %%mm2, %0 \n\t"
629 "movq %%mm3, %1 \n\t"
630 "movq %%mm4, %2 \n\t"
631 "movq %%mm5, %3 \n\t"
632 :"+m"(*(uint32_t*)(dest+0*linesize)),
633 "+m"(*(uint32_t*)(dest+1*linesize)),
634 "+m"(*(uint32_t*)(dest+2*linesize)),
635 "+m"(*(uint32_t*)(dest+3*linesize))
639 static void vc1_inv_trans_8x8_dc_mmxext(uint8_t *dest, int linesize,
643 dc = (3 * dc + 1) >> 1;
644 dc = (3 * dc + 16) >> 5;
646 "movd %0, %%mm0 \n\t"
647 "pshufw $0, %%mm0, %%mm0 \n\t"
648 "pxor %%mm1, %%mm1 \n\t"
649 "psubw %%mm0, %%mm1 \n\t"
650 "packuswb %%mm0, %%mm0 \n\t"
651 "packuswb %%mm1, %%mm1 \n\t"
655 "movq %0, %%mm2 \n\t"
656 "movq %1, %%mm3 \n\t"
657 "movq %2, %%mm4 \n\t"
658 "movq %3, %%mm5 \n\t"
659 "paddusb %%mm0, %%mm2 \n\t"
660 "paddusb %%mm0, %%mm3 \n\t"
661 "paddusb %%mm0, %%mm4 \n\t"
662 "paddusb %%mm0, %%mm5 \n\t"
663 "psubusb %%mm1, %%mm2 \n\t"
664 "psubusb %%mm1, %%mm3 \n\t"
665 "psubusb %%mm1, %%mm4 \n\t"
666 "psubusb %%mm1, %%mm5 \n\t"
667 "movq %%mm2, %0 \n\t"
668 "movq %%mm3, %1 \n\t"
669 "movq %%mm4, %2 \n\t"
670 "movq %%mm5, %3 \n\t"
671 :"+m"(*(uint32_t*)(dest+0*linesize)),
672 "+m"(*(uint32_t*)(dest+1*linesize)),
673 "+m"(*(uint32_t*)(dest+2*linesize)),
674 "+m"(*(uint32_t*)(dest+3*linesize))
678 "movq %0, %%mm2 \n\t"
679 "movq %1, %%mm3 \n\t"
680 "movq %2, %%mm4 \n\t"
681 "movq %3, %%mm5 \n\t"
682 "paddusb %%mm0, %%mm2 \n\t"
683 "paddusb %%mm0, %%mm3 \n\t"
684 "paddusb %%mm0, %%mm4 \n\t"
685 "paddusb %%mm0, %%mm5 \n\t"
686 "psubusb %%mm1, %%mm2 \n\t"
687 "psubusb %%mm1, %%mm3 \n\t"
688 "psubusb %%mm1, %%mm4 \n\t"
689 "psubusb %%mm1, %%mm5 \n\t"
690 "movq %%mm2, %0 \n\t"
691 "movq %%mm3, %1 \n\t"
692 "movq %%mm4, %2 \n\t"
693 "movq %%mm5, %3 \n\t"
694 :"+m"(*(uint32_t*)(dest+0*linesize)),
695 "+m"(*(uint32_t*)(dest+1*linesize)),
696 "+m"(*(uint32_t*)(dest+2*linesize)),
697 "+m"(*(uint32_t*)(dest+3*linesize))
701 av_cold void ff_vc1dsp_init_mmx(VC1DSPContext *dsp)
703 dsp->put_vc1_mspel_pixels_tab[ 0] = ff_put_vc1_mspel_mc00_mmx;
704 dsp->put_vc1_mspel_pixels_tab[ 4] = put_vc1_mspel_mc01_mmx;
705 dsp->put_vc1_mspel_pixels_tab[ 8] = put_vc1_mspel_mc02_mmx;
706 dsp->put_vc1_mspel_pixels_tab[12] = put_vc1_mspel_mc03_mmx;
708 dsp->put_vc1_mspel_pixels_tab[ 1] = put_vc1_mspel_mc10_mmx;
709 dsp->put_vc1_mspel_pixels_tab[ 5] = put_vc1_mspel_mc11_mmx;
710 dsp->put_vc1_mspel_pixels_tab[ 9] = put_vc1_mspel_mc12_mmx;
711 dsp->put_vc1_mspel_pixels_tab[13] = put_vc1_mspel_mc13_mmx;
713 dsp->put_vc1_mspel_pixels_tab[ 2] = put_vc1_mspel_mc20_mmx;
714 dsp->put_vc1_mspel_pixels_tab[ 6] = put_vc1_mspel_mc21_mmx;
715 dsp->put_vc1_mspel_pixels_tab[10] = put_vc1_mspel_mc22_mmx;
716 dsp->put_vc1_mspel_pixels_tab[14] = put_vc1_mspel_mc23_mmx;
718 dsp->put_vc1_mspel_pixels_tab[ 3] = put_vc1_mspel_mc30_mmx;
719 dsp->put_vc1_mspel_pixels_tab[ 7] = put_vc1_mspel_mc31_mmx;
720 dsp->put_vc1_mspel_pixels_tab[11] = put_vc1_mspel_mc32_mmx;
721 dsp->put_vc1_mspel_pixels_tab[15] = put_vc1_mspel_mc33_mmx;
724 av_cold void ff_vc1dsp_init_mmxext(VC1DSPContext *dsp)
726 dsp->avg_vc1_mspel_pixels_tab[ 4] = avg_vc1_mspel_mc01_mmxext;
727 dsp->avg_vc1_mspel_pixels_tab[ 8] = avg_vc1_mspel_mc02_mmxext;
728 dsp->avg_vc1_mspel_pixels_tab[12] = avg_vc1_mspel_mc03_mmxext;
730 dsp->avg_vc1_mspel_pixels_tab[ 1] = avg_vc1_mspel_mc10_mmxext;
731 dsp->avg_vc1_mspel_pixels_tab[ 5] = avg_vc1_mspel_mc11_mmxext;
732 dsp->avg_vc1_mspel_pixels_tab[ 9] = avg_vc1_mspel_mc12_mmxext;
733 dsp->avg_vc1_mspel_pixels_tab[13] = avg_vc1_mspel_mc13_mmxext;
735 dsp->avg_vc1_mspel_pixels_tab[ 2] = avg_vc1_mspel_mc20_mmxext;
736 dsp->avg_vc1_mspel_pixels_tab[ 6] = avg_vc1_mspel_mc21_mmxext;
737 dsp->avg_vc1_mspel_pixels_tab[10] = avg_vc1_mspel_mc22_mmxext;
738 dsp->avg_vc1_mspel_pixels_tab[14] = avg_vc1_mspel_mc23_mmxext;
740 dsp->avg_vc1_mspel_pixels_tab[ 3] = avg_vc1_mspel_mc30_mmxext;
741 dsp->avg_vc1_mspel_pixels_tab[ 7] = avg_vc1_mspel_mc31_mmxext;
742 dsp->avg_vc1_mspel_pixels_tab[11] = avg_vc1_mspel_mc32_mmxext;
743 dsp->avg_vc1_mspel_pixels_tab[15] = avg_vc1_mspel_mc33_mmxext;
745 dsp->vc1_inv_trans_8x8_dc = vc1_inv_trans_8x8_dc_mmxext;
746 dsp->vc1_inv_trans_4x8_dc = vc1_inv_trans_4x8_dc_mmxext;
747 dsp->vc1_inv_trans_8x4_dc = vc1_inv_trans_8x4_dc_mmxext;
748 dsp->vc1_inv_trans_4x4_dc = vc1_inv_trans_4x4_dc_mmxext;
750 #endif /* HAVE_INLINE_ASM */