1 /*****************************************************************************
2 * transforms_yuvmmx.h: MMX YUV transformation assembly
3 *****************************************************************************
4 * Copyright (C) 1999-2004 the VideoLAN team
7 * Authors: Olie Lho <ollie@sis.com.tw>
8 * Gaƫl Hendryckx <jimmy@via.ecp.fr>
9 * Samuel Hocevar <sam@zoy.org>
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2 of the License, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; if not, write to the Free Software
23 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
24 *****************************************************************************/
26 /* hope these constant values are cache line aligned */
27 #if __GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ >= 3)
28 #define USED_U64(foo) \
29 static const uint64_t foo __asm__ (#foo) __attribute__((used))
31 #define USED_U64(foo) \
32 static const uint64_t foo __asm__ (#foo) __attribute__((unused))
34 USED_U64(mmx_80w) = 0x0080008000800080ULL;
35 USED_U64(mmx_10w) = 0x1010101010101010ULL;
36 USED_U64(mmx_00ffw) = 0x00ff00ff00ff00ffULL;
37 USED_U64(mmx_Y_coeff) = 0x253f253f253f253fULL;
39 USED_U64(mmx_U_green) = 0xf37df37df37df37dULL;
40 USED_U64(mmx_U_blue) = 0x4093409340934093ULL;
41 USED_U64(mmx_V_red) = 0x3312331233123312ULL;
42 USED_U64(mmx_V_green) = 0xe5fce5fce5fce5fcULL;
44 USED_U64(mmx_mask_f8) = 0xf8f8f8f8f8f8f8f8ULL;
45 USED_U64(mmx_mask_fc) = 0xfcfcfcfcfcfcfcfcULL;
48 /* Use RIP-relative code in PIC mode on amd64 */
49 #if defined(__x86_64__) && defined(__PIC__)
55 #define MMX_INIT_16 " \n\
56 movd (%1), %%mm0 # Load 4 Cb 00 00 00 00 u3 u2 u1 u0 \n\
57 movd (%2), %%mm1 # Load 4 Cr 00 00 00 00 v3 v2 v1 v0 \n\
58 pxor %%mm4, %%mm4 # zero mm4 \n\
59 movq (%0), %%mm6 # Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 \n\
60 #movl $0, (%3) # cache preload for image \n\
63 #define INTRINSICS_INIT_16 \
64 tmp64 = *(uint32_t *)p_u; \
66 tmp64 = *(uint32_t *)p_v; \
68 mm4 = (__m64)(uint64_t)0; \
69 mm6 = (__m64)*(uint64_t *)p_y; \
70 /* *(uint16_t *)p_buffer = 0; */
72 #define MMX_INIT_16_GRAY " \n\
73 movq (%0), %%mm6 # Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 \n\
74 #movl $0, (%3) # cache preload for image \n\
77 #define MMX_INIT_32 " \n\
78 movd (%1), %%mm0 # Load 4 Cb 00 00 00 00 u3 u2 u1 u0 \n\
79 movl $0, (%3) # cache preload for image \n\
80 movd (%2), %%mm1 # Load 4 Cr 00 00 00 00 v3 v2 v1 v0 \n\
81 pxor %%mm4, %%mm4 # zero mm4 \n\
82 movq (%0), %%mm6 # Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 \n\
85 #define INTRINSICS_INIT_32 \
86 tmp64 = *(uint32_t *)p_u; \
88 *(uint16_t *)p_buffer = 0; \
89 tmp64 = *(uint32_t *)p_v; \
91 mm4 = (__m64)(uint64_t)0; \
92 mm6 = (__m64)*(uint64_t *)p_y;
95 * Do the multiply part of the conversion for even and odd pixels,
97 * mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
98 * mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
99 * mm6 -> Y even, mm7 -> Y odd
102 #define MMX_YUV_MUL " \n\
103 # convert the chroma part \n\
104 punpcklbw %%mm4, %%mm0 # scatter 4 Cb 00 u3 00 u2 00 u1 00 u0 \n\
105 punpcklbw %%mm4, %%mm1 # scatter 4 Cr 00 v3 00 v2 00 v1 00 v0 \n\
106 psubsw mmx_80w"G", %%mm0 # Cb -= 128 \n\
107 psubsw mmx_80w"G", %%mm1 # Cr -= 128 \n\
108 psllw $3, %%mm0 # Promote precision \n\
109 psllw $3, %%mm1 # Promote precision \n\
110 movq %%mm0, %%mm2 # Copy 4 Cb 00 u3 00 u2 00 u1 00 u0 \n\
111 movq %%mm1, %%mm3 # Copy 4 Cr 00 v3 00 v2 00 v1 00 v0 \n\
112 pmulhw mmx_U_green"G", %%mm2 # Mul Cb with green coeff -> Cb green \n\
113 pmulhw mmx_V_green"G", %%mm3 # Mul Cr with green coeff -> Cr green \n\
114 pmulhw mmx_U_blue"G", %%mm0 # Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0 \n\
115 pmulhw mmx_V_red"G", %%mm1 # Mul Cr -> Cred 00 r3 00 r2 00 r1 00 r0 \n\
116 paddsw %%mm3, %%mm2 # Cb green + Cr green -> Cgreen \n\
118 # convert the luma part \n\
119 psubusb mmx_10w"G", %%mm6 # Y -= 16 \n\
120 movq %%mm6, %%mm7 # Copy 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 \n\
121 pand mmx_00ffw"G", %%mm6 # get Y even 00 Y6 00 Y4 00 Y2 00 Y0 \n\
122 psrlw $8, %%mm7 # get Y odd 00 Y7 00 Y5 00 Y3 00 Y1 \n\
123 psllw $3, %%mm6 # Promote precision \n\
124 psllw $3, %%mm7 # Promote precision \n\
125 pmulhw mmx_Y_coeff"G", %%mm6 # Mul 4 Y even 00 y6 00 y4 00 y2 00 y0 \n\
126 pmulhw mmx_Y_coeff"G", %%mm7 # Mul 4 Y odd 00 y7 00 y5 00 y3 00 y1 \n\
129 #define INTRINSICS_YUV_MUL \
130 mm0 = _mm_unpacklo_pi8(mm0, mm4); \
131 mm1 = _mm_unpacklo_pi8(mm1, mm4); \
132 mm0 = _mm_subs_pi16(mm0, (__m64)mmx_80w); \
133 mm1 = _mm_subs_pi16(mm1, (__m64)mmx_80w); \
134 mm0 = _mm_slli_pi16(mm0, 3); \
135 mm1 = _mm_slli_pi16(mm1, 3); \
138 mm2 = _mm_mulhi_pi16(mm2, (__m64)mmx_U_green); \
139 mm3 = _mm_mulhi_pi16(mm3, (__m64)mmx_V_green); \
140 mm0 = _mm_mulhi_pi16(mm0, (__m64)mmx_U_blue); \
141 mm1 = _mm_mulhi_pi16(mm1, (__m64)mmx_V_red); \
142 mm2 = _mm_adds_pi16(mm2, mm3); \
144 mm6 = _mm_subs_pu8(mm6, (__m64)mmx_10w); \
146 mm6 = _mm_and_si64(mm6, (__m64)mmx_00ffw); \
147 mm7 = _mm_srli_pi16(mm7, 8); \
148 mm6 = _mm_slli_pi16(mm6, 3); \
149 mm7 = _mm_slli_pi16(mm7, 3); \
150 mm6 = _mm_mulhi_pi16(mm6, (__m64)mmx_Y_coeff); \
151 mm7 = _mm_mulhi_pi16(mm7, (__m64)mmx_Y_coeff);
154 * Do the addition part of the conversion for even and odd pixels,
156 * mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
157 * mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
158 * mm6 -> Y even, mm7 -> Y odd
161 #define MMX_YUV_ADD " \n\
162 # Do horizontal and vertical scaling \n\
163 movq %%mm0, %%mm3 # Copy Cblue \n\
164 movq %%mm1, %%mm4 # Copy Cred \n\
165 movq %%mm2, %%mm5 # Copy Cgreen \n\
166 paddsw %%mm6, %%mm0 # Y even + Cblue 00 B6 00 B4 00 B2 00 B0 \n\
167 paddsw %%mm7, %%mm3 # Y odd + Cblue 00 B7 00 B5 00 B3 00 B1 \n\
168 paddsw %%mm6, %%mm1 # Y even + Cred 00 R6 00 R4 00 R2 00 R0 \n\
169 paddsw %%mm7, %%mm4 # Y odd + Cred 00 R7 00 R5 00 R3 00 R1 \n\
170 paddsw %%mm6, %%mm2 # Y even + Cgreen 00 G6 00 G4 00 G2 00 G0 \n\
171 paddsw %%mm7, %%mm5 # Y odd + Cgreen 00 G7 00 G5 00 G3 00 G1 \n\
173 # Limit RGB even to 0..255 \n\
174 packuswb %%mm0, %%mm0 # B6 B4 B2 B0 / B6 B4 B2 B0 \n\
175 packuswb %%mm1, %%mm1 # R6 R4 R2 R0 / R6 R4 R2 R0 \n\
176 packuswb %%mm2, %%mm2 # G6 G4 G2 G0 / G6 G4 G2 G0 \n\
178 # Limit RGB odd to 0..255 \n\
179 packuswb %%mm3, %%mm3 # B7 B5 B3 B1 / B7 B5 B3 B1 \n\
180 packuswb %%mm4, %%mm4 # R7 R5 R3 R1 / R7 R5 R3 R1 \n\
181 packuswb %%mm5, %%mm5 # G7 G5 G3 G1 / G7 G5 G3 G1 \n\
183 # Interleave RGB even and odd \n\
184 punpcklbw %%mm3, %%mm0 # B7 B6 B5 B4 B3 B2 B1 B0 \n\
185 punpcklbw %%mm4, %%mm1 # R7 R6 R5 R4 R3 R2 R1 R0 \n\
186 punpcklbw %%mm5, %%mm2 # G7 G6 G5 G4 G3 G2 G1 G0 \n\
189 #define INTRINSICS_YUV_ADD \
193 mm0 = _mm_adds_pi16(mm0, mm6); \
194 mm3 = _mm_adds_pi16(mm3, mm7); \
195 mm1 = _mm_adds_pi16(mm1, mm6); \
196 mm4 = _mm_adds_pi16(mm4, mm7); \
197 mm2 = _mm_adds_pi16(mm2, mm6); \
198 mm5 = _mm_adds_pi16(mm5, mm7); \
200 mm0 = _mm_packs_pu16(mm0, mm0); \
201 mm1 = _mm_packs_pu16(mm1, mm1); \
202 mm2 = _mm_packs_pu16(mm2, mm2); \
204 mm3 = _mm_packs_pu16(mm3, mm3); \
205 mm4 = _mm_packs_pu16(mm4, mm4); \
206 mm5 = _mm_packs_pu16(mm5, mm5); \
208 mm0 = _mm_unpacklo_pi8(mm0, mm3); \
209 mm1 = _mm_unpacklo_pi8(mm1, mm4); \
210 mm2 = _mm_unpacklo_pi8(mm2, mm5);
213 * Grayscale case, only use Y
216 #define MMX_YUV_GRAY " \n\
217 # convert the luma part \n\
218 psubusb mmx_10w"G", %%mm6 \n\
219 movq %%mm6, %%mm7 \n\
220 pand mmx_00ffw"G", %%mm6 \n\
224 pmulhw mmx_Y_coeff"G", %%mm6 \n\
225 pmulhw mmx_Y_coeff"G", %%mm7 \n\
226 packuswb %%mm6, %%mm6 \n\
227 packuswb %%mm7, %%mm7 \n\
228 punpcklbw %%mm7, %%mm6 \n\
231 #define MMX_UNPACK_16_GRAY " \n\
232 movq %%mm6, %%mm5 \n\
233 pand mmx_mask_f8"G", %%mm6 \n\
234 pand mmx_mask_fc"G", %%mm5 \n\
235 movq %%mm6, %%mm7 \n\
237 pxor %%mm3, %%mm3 \n\
238 movq %%mm7, %%mm2 \n\
239 movq %%mm5, %%mm0 \n\
240 punpcklbw %%mm3, %%mm5 \n\
241 punpcklbw %%mm6, %%mm7 \n\
245 punpckhbw %%mm3, %%mm0 \n\
246 punpckhbw %%mm6, %%mm2 \n\
248 movq 8(%0), %%mm6 \n\
250 movq %%mm2, 8(%3) \n\
255 * convert RGB plane to RGB 15 bits,
256 * mm0 -> B, mm1 -> R, mm2 -> G,
257 * mm4 -> GB, mm5 -> AR pixel 4-7,
258 * mm6 -> GB, mm7 -> AR pixel 0-3
261 #define MMX_UNPACK_15 " \n\
262 # mask unneeded bits off \n\
263 pand mmx_mask_f8"G", %%mm0 # b7b6b5b4 b3______ b7b6b5b4 b3______ \n\
264 psrlw $3,%%mm0 # ______b7 b6b5b4b3 ______b7 b6b5b4b3 \n\
265 pand mmx_mask_f8"G", %%mm2 # g7g6g5g4 g3______ g7g6g5g4 g3______ \n\
266 pand mmx_mask_f8"G", %%mm1 # r7r6r5r4 r3______ r7r6r5r4 r3______ \n\
267 psrlw $1,%%mm1 # __r7r6r5 r4r3____ __r7r6r5 r4r3____ \n\
268 pxor %%mm4, %%mm4 # zero mm4 \n\
269 movq %%mm0, %%mm5 # Copy B7-B0 \n\
270 movq %%mm2, %%mm7 # Copy G7-G0 \n\
272 # convert rgb24 plane to rgb15 pack for pixel 0-3 \n\
273 punpcklbw %%mm4, %%mm2 # ________ ________ g7g6g5g4 g3______ \n\
274 punpcklbw %%mm1, %%mm0 # r7r6r5r4 r3______ ______b7 b6b5b4b3 \n\
275 psllw $2,%%mm2 # ________ ____g7g6 g5g4g3__ ________ \n\
276 por %%mm2, %%mm0 # r7r6r5r4 r3__g7g6 g5g4g3b7 b6b5b4b3 \n\
277 movq 8(%0), %%mm6 # Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 \n\
278 movq %%mm0, (%3) # store pixel 0-3 \n\
280 # convert rgb24 plane to rgb16 pack for pixel 0-3 \n\
281 punpckhbw %%mm4, %%mm7 # ________ ________ g7g6g5g4 g3______ \n\
282 punpckhbw %%mm1, %%mm5 # r7r6r5r4 r3______ ______b7 b6b5b4b3 \n\
283 psllw $2,%%mm7 # ________ ____g7g6 g5g4g3__ ________ \n\
284 movd 4(%1), %%mm0 # Load 4 Cb __ __ __ __ u3 u2 u1 u0 \n\
285 por %%mm7, %%mm5 # r7r6r5r4 r3__g7g6 g5g4g3b7 b6b5b4b3 \n\
286 movd 4(%2), %%mm1 # Load 4 Cr __ __ __ __ v3 v2 v1 v0 \n\
287 movq %%mm5, 8(%3) # store pixel 4-7 \n\
290 #define INTRINSICS_UNPACK_15 \
291 mm0 = _mm_and_si64(mm0, (__m64)mmx_mask_f8); \
292 mm0 = _mm_srli_pi16(mm0, 3); \
293 mm2 = _mm_and_si64(mm2, (__m64)mmx_mask_f8); \
294 mm1 = _mm_and_si64(mm1, (__m64)mmx_mask_f8); \
295 mm1 = _mm_srli_pi16(mm1, 1); \
296 mm4 = (__m64)(uint64_t)0; \
300 mm2 = _mm_unpacklo_pi8(mm2, mm4); \
301 mm0 = _mm_unpacklo_pi8(mm0, mm1); \
302 mm2 = _mm_slli_pi16(mm2, 2); \
303 mm0 = _mm_or_si64(mm0, mm2); \
304 tmp64 = *(uint64_t *)(p_y + 8); \
305 mm6 = (__m64)tmp64; \
306 *(uint64_t *)p_buffer = (uint64_t)mm0; \
308 mm7 = _mm_unpackhi_pi8(mm7, mm4); \
309 mm5 = _mm_unpackhi_pi8(mm5, mm1); \
310 mm7 = _mm_slli_pi16(mm7, 2); \
311 tmp64 = (uint64_t)*(uint32_t *)(p_u + 4); \
312 mm0 = (__m64)tmp64; \
313 mm5 = _mm_or_si64(mm5, mm7); \
314 tmp64 = (uint64_t)*(uint32_t *)(p_v + 4); \
315 mm1 = (__m64)tmp64; \
316 *(uint64_t *)(p_buffer + 4) = (uint64_t)mm5;
319 * convert RGB plane to RGB 16 bits,
320 * mm0 -> B, mm1 -> R, mm2 -> G,
321 * mm4 -> GB, mm5 -> AR pixel 4-7,
322 * mm6 -> GB, mm7 -> AR pixel 0-3
325 #define MMX_UNPACK_16 " \n\
326 # mask unneeded bits off \n\
327 pand mmx_mask_f8"G", %%mm0 # b7b6b5b4 b3______ b7b6b5b4 b3______ \n\
328 pand mmx_mask_fc"G", %%mm2 # g7g6g5g4 g3g2____ g7g6g5g4 g3g2____ \n\
329 pand mmx_mask_f8"G", %%mm1 # r7r6r5r4 r3______ r7r6r5r4 r3______ \n\
330 psrlw $3,%%mm0 # ______b7 b6b5b4b3 ______b7 b6b5b4b3 \n\
331 pxor %%mm4, %%mm4 # zero mm4 \n\
332 movq %%mm0, %%mm5 # Copy B7-B0 \n\
333 movq %%mm2, %%mm7 # Copy G7-G0 \n\
335 # convert rgb24 plane to rgb16 pack for pixel 0-3 \n\
336 punpcklbw %%mm4, %%mm2 # ________ ________ g7g6g5g4 g3g2____ \n\
337 punpcklbw %%mm1, %%mm0 # r7r6r5r4 r3______ ______b7 b6b5b4b3 \n\
338 psllw $3,%%mm2 # ________ __g7g6g5 g4g3g2__ ________ \n\
339 por %%mm2, %%mm0 # r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 \n\
340 movq 8(%0), %%mm6 # Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 \n\
341 movq %%mm0, (%3) # store pixel 0-3 \n\
343 # convert rgb24 plane to rgb16 pack for pixel 0-3 \n\
344 punpckhbw %%mm4, %%mm7 # ________ ________ g7g6g5g4 g3g2____ \n\
345 punpckhbw %%mm1, %%mm5 # r7r6r5r4 r3______ ______b7 b6b5b4b3 \n\
346 psllw $3,%%mm7 # ________ __g7g6g5 g4g3g2__ ________ \n\
347 movd 4(%1), %%mm0 # Load 4 Cb __ __ __ __ u3 u2 u1 u0 \n\
348 por %%mm7, %%mm5 # r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 \n\
349 movd 4(%2), %%mm1 # Load 4 Cr __ __ __ __ v3 v2 v1 v0 \n\
350 movq %%mm5, 8(%3) # store pixel 4-7 \n\
353 #define INTRINSICS_UNPACK_16 \
354 mm0 = _mm_and_si64(mm0, (__m64)mmx_mask_f8); \
355 mm2 = _mm_and_si64(mm2, (__m64)mmx_mask_fc); \
356 mm1 = _mm_and_si64(mm1, (__m64)mmx_mask_f8); \
357 mm0 = _mm_srli_pi16(mm0, 3); \
358 mm4 = (__m64)(uint64_t)0; \
362 mm2 = _mm_unpacklo_pi8(mm2, mm4); \
363 mm0 = _mm_unpacklo_pi8(mm0, mm1); \
364 mm2 = _mm_slli_pi16(mm2, 3); \
365 mm0 = _mm_or_si64(mm0, mm2); \
366 tmp64 = *(uint64_t *)(p_y + 8); \
367 mm6 = (__m64)tmp64; \
368 *(uint64_t *)p_buffer = (uint64_t)mm0; \
370 mm7 = _mm_unpackhi_pi8(mm7, mm4); \
371 mm5 = _mm_unpackhi_pi8(mm5, mm1); \
372 mm7 = _mm_slli_pi16(mm7, 3); \
373 tmp64 = (uint64_t)*(uint32_t *)(p_u + 4); \
374 mm0 = (__m64)tmp64; \
375 mm5 = _mm_or_si64(mm5, mm7); \
376 tmp64 = (uint64_t)*(uint32_t *)(p_v + 4); \
377 mm1 = (__m64)tmp64; \
378 *(uint64_t *)(p_buffer + 4) = (uint64_t)mm5;
381 * convert RGB plane to RGB packed format,
382 * mm0 -> B, mm1 -> R, mm2 -> G, mm3 -> 0,
383 * mm4 -> GB, mm5 -> AR pixel 4-7,
384 * mm6 -> GB, mm7 -> AR pixel 0-3
387 #define MMX_UNPACK_32 " \n\
388 pxor %%mm3, %%mm3 # zero mm3 \n\
389 movq %%mm0, %%mm6 # B7 B6 B5 B4 B3 B2 B1 B0 \n\
390 movq %%mm1, %%mm7 # R7 R6 R5 R4 R3 R2 R1 R0 \n\
391 movq %%mm0, %%mm4 # B7 B6 B5 B4 B3 B2 B1 B0 \n\
392 movq %%mm1, %%mm5 # R7 R6 R5 R4 R3 R2 R1 R0 \n\
393 punpcklbw %%mm2, %%mm6 # G3 B3 G2 B2 G1 B1 G0 B0 \n\
394 punpcklbw %%mm3, %%mm7 # 00 R3 00 R2 00 R1 00 R0 \n\
395 punpcklwd %%mm7, %%mm6 # 00 R1 B1 G1 00 R0 B0 G0 \n\
396 movq %%mm6, (%3) # Store ARGB1 ARGB0 \n\
397 movq %%mm0, %%mm6 # B7 B6 B5 B4 B3 B2 B1 B0 \n\
398 punpcklbw %%mm2, %%mm6 # G3 B3 G2 B2 G1 B1 G0 B0 \n\
399 punpckhwd %%mm7, %%mm6 # 00 R3 G3 B3 00 R2 B3 G2 \n\
400 movq %%mm6, 8(%3) # Store ARGB3 ARGB2 \n\
401 punpckhbw %%mm2, %%mm4 # G7 B7 G6 B6 G5 B5 G4 B4 \n\
402 punpckhbw %%mm3, %%mm5 # 00 R7 00 R6 00 R5 00 R4 \n\
403 punpcklwd %%mm5, %%mm4 # 00 R5 B5 G5 00 R4 B4 G4 \n\
404 movq %%mm4, 16(%3) # Store ARGB5 ARGB4 \n\
405 movq %%mm0, %%mm4 # B7 B6 B5 B4 B3 B2 B1 B0 \n\
406 punpckhbw %%mm2, %%mm4 # G7 B7 G6 B6 G5 B5 G4 B4 \n\
407 punpckhwd %%mm5, %%mm4 # 00 R7 G7 B7 00 R6 B6 G6 \n\
408 movq %%mm4, 24(%3) # Store ARGB7 ARGB6 \n\
410 #movd 4(%1), %%mm0 # Load 4 Cb 00 00 00 00 u3 u2 u1 u0 \n\
411 #movd 4(%2), %%mm1 # Load 4 Cr 00 00 00 00 v3 v2 v1 v0 \n\
412 #pxor %%mm4, %%mm4 # zero mm4 \n\
413 #movq 8(%0), %%mm6 # Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 \n\
416 #define INTRINSICS_UNPACK_32 \
417 mm3 = (__m64)(uint64_t)0; \
422 mm6 = _mm_unpacklo_pi8(mm6, mm2); \
423 mm7 = _mm_unpacklo_pi8(mm7, mm3); \
424 mm6 = _mm_unpacklo_pi16(mm6, mm7); \
425 *(uint64_t *)p_buffer = (uint64_t)mm6; \
427 mm6 = _mm_unpacklo_pi8(mm6, mm2); \
428 mm6 = _mm_unpackhi_pi16(mm6, mm7); \
429 *(uint64_t *)(p_buffer + 2) = (uint64_t)mm6; \
430 mm4 = _mm_unpackhi_pi8(mm4, mm2); \
431 mm5 = _mm_unpackhi_pi8(mm5, mm3); \
432 mm4 = _mm_unpacklo_pi16(mm4, mm5); \
433 *(uint64_t *)(p_buffer + 4) = (uint64_t)mm4; \
435 mm4 = _mm_unpackhi_pi8(mm4, mm2); \
436 mm4 = _mm_unpackhi_pi16(mm4, mm5); \
437 *(uint64_t *)(p_buffer + 6) = (uint64_t)mm4; \