a1548a77175ade4ab40a609761a7941f39439439
2 * AltiVec-enhanced yuv2yuvX
4 * Copyright (C) 2004 Romain Dolbeau <romain@dolbeau.org>
5 * based on the equivalent C code in swscale.c
7 * This file is part of FFmpeg.
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * FFmpeg 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 GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
27 #include "libswscale/swscale.h"
28 #include "libswscale/swscale_internal.h"
29 #include "libavutil/attributes.h"
30 #include "libavutil/cpu.h"
31 #include "yuv2rgb_altivec.h"
32 #include "libavutil/ppc/util_altivec.h"
35 #define vzero vec_splat_s32(0)
38 #define GET_LS(a,b,c,s) {\
39 vector signed short l2 = vec_ld(((b) << 1) + 16, s);\
40 ls = vec_perm(a, l2, c);\
43 #define GET_VF(a, b, c,d) {\
44 a = vec_mergeh(c, d);\
45 b = vec_mergel(c, d);\
48 #define GET_LS(a,b,c,s) {\
50 a = vec_vsx_ld(((b) << 1) + 16, s);\
52 #define GET_VF(a, b, c, d) {\
53 a = vec_mergel(d, c);\
54 b = vec_mergeh(d, c);\
58 #define yuv2planeX_8(d1, d2, l1, src, x, perm, filter) do {\
59 vector signed short ls;\
60 GET_LS(l1, x, perm, src);\
61 vector signed int i1 = vec_mule(filter, ls);\
62 vector signed int i2 = vec_mulo(filter, ls);\
63 vector signed int vf1, vf2;\
64 GET_VF(vf1, vf2, i1, i2);\
65 d1 = vec_add(d1, vf1);\
66 d2 = vec_add(d2, vf2);\
70 #define LOAD_FILTER(vf,f) {\
71 vector unsigned char perm0 = vec_lvsl(joffset, f);\
72 vf = vec_ld(joffset, f);\
73 vf = vec_perm(vf, vf, perm0);\
75 #define LOAD_L1(ll1,s,p){\
76 p = vec_lvsl(xoffset, s);\
77 ll1 = vec_ld(xoffset, s);\
80 #define LOAD_FILTER(vf,f) {\
81 vf = vec_vsx_ld(joffset, f);\
83 #define LOAD_L1(ll1,s,p){\
84 ll1 = vec_vsx_ld(xoffset, s);\
88 static void yuv2planeX_16_altivec(const int16_t *filter
, int filterSize
,
89 const int16_t **src
, uint8_t *dest
,
90 const uint8_t *dither
, int offset
, int x
)
93 DECLARE_ALIGNED(16, int, val
)[16];
94 vector
signed int vo1
, vo2
, vo3
, vo4
;
95 vector
unsigned short vs1
, vs2
;
96 vector
unsigned char vf
;
97 vector
unsigned int altivec_vectorShiftInt19
=
98 vec_add(vec_splat_u32(10), vec_splat_u32(9));
100 for (i
= 0; i
< 16; i
++)
101 val
[i
] = dither
[(x
+ i
+ offset
) & 7] << 12;
103 vo1
= vec_ld(0, val
);
104 vo2
= vec_ld(16, val
);
105 vo3
= vec_ld(32, val
);
106 vo4
= vec_ld(48, val
);
108 for (j
= 0; j
< filterSize
; j
++) {
109 unsigned int joffset
=j
<<1;
110 unsigned int xoffset
=x
<<1;
111 vector
unsigned char perm
;
112 vector
signed short l1
,vLumFilter
;
113 LOAD_FILTER(vLumFilter
,filter
);
114 vLumFilter
= vec_splat(vLumFilter
, 0);
115 LOAD_L1(l1
,src
[j
],perm
);
116 yuv2planeX_8(vo1
, vo2
, l1
, src
[j
], x
, perm
, vLumFilter
);
117 yuv2planeX_8(vo3
, vo4
, l1
, src
[j
], x
+ 8, perm
, vLumFilter
);
120 vo1
= vec_sra(vo1
, altivec_vectorShiftInt19
);
121 vo2
= vec_sra(vo2
, altivec_vectorShiftInt19
);
122 vo3
= vec_sra(vo3
, altivec_vectorShiftInt19
);
123 vo4
= vec_sra(vo4
, altivec_vectorShiftInt19
);
124 vs1
= vec_packsu(vo1
, vo2
);
125 vs2
= vec_packsu(vo3
, vo4
);
126 vf
= vec_packsu(vs1
, vs2
);
131 static inline void yuv2planeX_u(const int16_t *filter
, int filterSize
,
132 const int16_t **src
, uint8_t *dest
, int dstW
,
133 const uint8_t *dither
, int offset
, int x
)
137 for (i
= x
; i
< dstW
; i
++) {
138 int t
= dither
[(i
+ offset
) & 7] << 12;
139 for (j
= 0; j
< filterSize
; j
++)
140 t
+= src
[j
][i
] * filter
[j
];
141 dest
[i
] = av_clip_uint8(t
>> 19);
145 static void yuv2planeX_altivec(const int16_t *filter
, int filterSize
,
146 const int16_t **src
, uint8_t *dest
, int dstW
,
147 const uint8_t *dither
, int offset
)
149 int dst_u
= -(uintptr_t)dest
& 15;
152 yuv2planeX_u(filter
, filterSize
, src
, dest
, dst_u
, dither
, offset
, 0);
154 for (i
= dst_u
; i
< dstW
- 15; i
+= 16)
155 yuv2planeX_16_altivec(filter
, filterSize
, src
, dest
+ i
, dither
,
158 yuv2planeX_u(filter
, filterSize
, src
, dest
, dstW
, dither
, offset
, i
);
162 // The 3 above is 2 (filterSize == 4) + 1 (sizeof(short) == 2).
164 // The neat trick: We only care for half the elements,
165 // high or low depending on (i<<3)%16 (it's 0 or 8 here),
166 // and we're going to use vec_mule, so we choose
167 // carefully how to "unpack" the elements into the even slots.
168 #define GET_VF4(a, vf, f) {\
169 vf = vec_ld(a<< 3, f);\
171 vf = vec_mergel(vf, (vector signed short)vzero);\
173 vf = vec_mergeh(vf, (vector signed short)vzero);\
175 #define FIRST_LOAD(sv, pos, s, per) {\
176 sv = vec_ld(pos, s);\
177 per = vec_lvsl(pos, s);\
179 #define UPDATE_PTR(s0, d0, s1, d1) {\
183 #define LOAD_SRCV(pos, a, s, per, v0, v1, vf) {\
184 v1 = vec_ld(pos + a + 16, s);\
185 vf = vec_perm(v0, v1, per);\
187 #define LOAD_SRCV8(pos, a, s, per, v0, v1, vf) {\
188 if ((((uintptr_t)s + pos) % 16) > 8) {\
189 v1 = vec_ld(pos + a + 16, s);\
191 vf = vec_perm(v0, src_v1, per);\
193 #define GET_VFD(a, b, f, vf0, vf1, per, vf, off) {\
194 vf1 = vec_ld((a * 2 * filterSize) + (b * 2) + 16 + off, f);\
195 vf = vec_perm(vf0, vf1, per);\
197 #else /* else of #if HAVE_BIGENDIAN */
198 #define GET_VF4(a, vf, f) {\
199 vf = (vector signed short)vec_vsx_ld(a << 3, f);\
200 vf = vec_mergeh(vf, (vector signed short)vzero);\
202 #define FIRST_LOAD(sv, pos, s, per) {}
203 #define UPDATE_PTR(s0, d0, s1, d1) {}
204 #define LOAD_SRCV(pos, a, s, per, v0, v1, vf) {\
205 vf = vec_vsx_ld(pos + a, s);\
207 #define LOAD_SRCV8(pos, a, s, per, v0, v1, vf) LOAD_SRCV(pos, a, s, per, v0, v1, vf)
208 #define GET_VFD(a, b, f, vf0, vf1, per, vf, off) {\
209 vf = vec_vsx_ld((a * 2 * filterSize) + (b * 2) + off, f);\
211 #endif /* end of #if HAVE_BIGENDIAN */
213 static void hScale_altivec_real(SwsContext
*c
, int16_t *dst
, int dstW
,
214 const uint8_t *src
, const int16_t *filter
,
215 const int32_t *filterPos
, int filterSize
)
218 DECLARE_ALIGNED(16, int, tempo
)[4];
220 if (filterSize
% 4) {
221 for (i
= 0; i
< dstW
; i
++) {
223 register int srcPos
= filterPos
[i
];
224 register int val
= 0;
225 for (j
= 0; j
< filterSize
; j
++)
226 val
+= ((int)src
[srcPos
+ j
]) * filter
[filterSize
* i
+ j
];
227 dst
[i
] = FFMIN(val
>> 7, (1 << 15) - 1);
230 switch (filterSize
) {
232 for (i
= 0; i
< dstW
; i
++) {
233 register int srcPos
= filterPos
[i
];
235 vector
unsigned char src_vF
= unaligned_load(srcPos
, src
);
236 vector
signed short src_v
, filter_v
;
237 vector
signed int val_vEven
, val_s
;
238 src_v
= // vec_unpackh sign-extends...
239 (vector
signed short)(VEC_MERGEH((vector
unsigned char)vzero
, src_vF
));
240 // now put our elements in the even slots
241 src_v
= vec_mergeh(src_v
, (vector
signed short)vzero
);
242 GET_VF4(i
, filter_v
, filter
);
243 val_vEven
= vec_mule(src_v
, filter_v
);
244 val_s
= vec_sums(val_vEven
, vzero
);
245 vec_st(val_s
, 0, tempo
);
246 dst
[i
] = FFMIN(tempo
[3] >> 7, (1 << 15) - 1);
250 for (i
= 0; i
< dstW
; i
++) {
251 register int srcPos
= filterPos
[i
];
252 vector
unsigned char src_vF
, src_v0
, src_v1
;
253 vector
unsigned char permS
;
254 vector
signed short src_v
, filter_v
;
255 vector
signed int val_v
, val_s
;
256 FIRST_LOAD(src_v0
, srcPos
, src
, permS
);
257 LOAD_SRCV8(srcPos
, 0, src
, permS
, src_v0
, src_v1
, src_vF
);
258 src_v
= // vec_unpackh sign-extends...
259 (vector
signed short)(VEC_MERGEH((vector
unsigned char)vzero
, src_vF
));
260 filter_v
= vec_ld(i
<< 4, filter
);
261 val_v
= vec_msums(src_v
, filter_v
, (vector
signed int)vzero
);
262 val_s
= vec_sums(val_v
, vzero
);
263 vec_st(val_s
, 0, tempo
);
264 dst
[i
] = FFMIN(tempo
[3] >> 7, (1 << 15) - 1);
269 for (i
= 0; i
< dstW
; i
++) {
270 register int srcPos
= filterPos
[i
];
272 vector
unsigned char src_vF
= unaligned_load(srcPos
, src
);
273 vector
signed short src_vA
= // vec_unpackh sign-extends...
274 (vector
signed short)(VEC_MERGEH((vector
unsigned char)vzero
, src_vF
));
275 vector
signed short src_vB
= // vec_unpackh sign-extends...
276 (vector
signed short)(VEC_MERGEL((vector
unsigned char)vzero
, src_vF
));
277 vector
signed short filter_v0
= vec_ld(i
<< 5, filter
);
278 vector
signed short filter_v1
= vec_ld((i
<< 5) + 16, filter
);
280 vector
signed int val_acc
= vec_msums(src_vA
, filter_v0
, (vector
signed int)vzero
);
281 vector
signed int val_v
= vec_msums(src_vB
, filter_v1
, val_acc
);
283 vector
signed int val_s
= vec_sums(val_v
, vzero
);
285 VEC_ST(val_s
, 0, tempo
);
286 dst
[i
] = FFMIN(tempo
[3] >> 7, (1 << 15) - 1);
291 for (i
= 0; i
< dstW
; i
++) {
292 register int j
, offset
= i
* 2 * filterSize
;
293 register int srcPos
= filterPos
[i
];
295 vector
signed int val_s
, val_v
= (vector
signed int)vzero
;
296 vector
signed short filter_v0R
;
297 vector
unsigned char permF
, src_v0
, permS
;
298 FIRST_LOAD(filter_v0R
, offset
, filter
, permF
);
299 FIRST_LOAD(src_v0
, srcPos
, src
, permS
);
301 for (j
= 0; j
< filterSize
- 15; j
+= 16) {
302 vector
unsigned char src_v1
, src_vF
;
303 vector
signed short filter_v1R
, filter_v2R
, filter_v0
, filter_v1
;
304 LOAD_SRCV(srcPos
, j
, src
, permS
, src_v0
, src_v1
, src_vF
);
305 vector
signed short src_vA
= // vec_unpackh sign-extends...
306 (vector
signed short)(VEC_MERGEH((vector
unsigned char)vzero
, src_vF
));
307 vector
signed short src_vB
= // vec_unpackh sign-extends...
308 (vector
signed short)(VEC_MERGEL((vector
unsigned char)vzero
, src_vF
));
309 GET_VFD(i
, j
, filter
, filter_v0R
, filter_v1R
, permF
, filter_v0
, 0);
310 GET_VFD(i
, j
, filter
, filter_v1R
, filter_v2R
, permF
, filter_v1
, 16);
312 vector
signed int val_acc
= vec_msums(src_vA
, filter_v0
, val_v
);
313 val_v
= vec_msums(src_vB
, filter_v1
, val_acc
);
314 UPDATE_PTR(filter_v2R
, filter_v0R
, src_v1
, src_v0
);
317 if (j
< filterSize
- 7) {
318 // loading src_v0 is useless, it's already done above
319 vector
unsigned char src_v1
, src_vF
;
320 vector
signed short src_v
, filter_v1R
, filter_v
;
321 LOAD_SRCV8(srcPos
, j
, src
, permS
, src_v0
, src_v1
, src_vF
);
322 src_v
= // vec_unpackh sign-extends...
323 (vector
signed short)(VEC_MERGEH((vector
unsigned char)vzero
, src_vF
));
324 GET_VFD(i
, j
, filter
, filter_v0R
, filter_v1R
, permF
, filter_v
, 0);
325 val_v
= vec_msums(src_v
, filter_v
, val_v
);
327 val_s
= vec_sums(val_v
, vzero
);
329 VEC_ST(val_s
, 0, tempo
);
330 dst
[i
] = FFMIN(tempo
[3] >> 7, (1 << 15) - 1);
334 #endif /* HAVE_ALTIVEC */
336 av_cold
void ff_sws_init_swscale_ppc(SwsContext
*c
)
339 enum AVPixelFormat dstFormat
= c
->dstFormat
;
341 if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC
))
344 if (c
->srcBpc
== 8 && c
->dstBpc
<= 14) {
345 c
->hyScale
= c
->hcScale
= hScale_altivec_real
;
347 if (!is16BPS(dstFormat
) && !is9_OR_10BPS(dstFormat
) &&
348 dstFormat
!= AV_PIX_FMT_NV12
&& dstFormat
!= AV_PIX_FMT_NV21
&&
350 c
->yuv2planeX
= yuv2planeX_altivec
;
353 /* The following list of supported dstFormat values should
354 * match what's found in the body of ff_yuv2packedX_altivec() */
355 if (!(c
->flags
& (SWS_BITEXACT
| SWS_FULL_CHR_H_INT
)) && !c
->alpPixBuf
) {
356 switch (c
->dstFormat
) {
357 case AV_PIX_FMT_ABGR
:
358 c
->yuv2packedX
= ff_yuv2abgr_X_altivec
;
360 case AV_PIX_FMT_BGRA
:
361 c
->yuv2packedX
= ff_yuv2bgra_X_altivec
;
363 case AV_PIX_FMT_ARGB
:
364 c
->yuv2packedX
= ff_yuv2argb_X_altivec
;
366 case AV_PIX_FMT_RGBA
:
367 c
->yuv2packedX
= ff_yuv2rgba_X_altivec
;
369 case AV_PIX_FMT_BGR24
:
370 c
->yuv2packedX
= ff_yuv2bgr24_X_altivec
;
372 case AV_PIX_FMT_RGB24
:
373 c
->yuv2packedX
= ff_yuv2rgb24_X_altivec
;
377 #endif /* HAVE_ALTIVEC */