1 /*
2 * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
3 * Copyright (c) 2011 Daniel Kang
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 #include "libavutil/attributes.h"
23 #include "libavutil/cpu.h"
24 #include "libavutil/mem_internal.h"
25 #include "libavutil/x86/asm.h"
26 #include "libavutil/x86/cpu.h"
27 #include "libavcodec/h264dec.h"
28 #include "libavcodec/h264qpel.h"
29 #include "libavcodec/pixels.h"
30 #include "fpel.h"
31
32 #if HAVE_X86ASM
33 void ff_put_pixels4_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
34 int dstStride, int src1Stride, int h);
35 void ff_avg_pixels4_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
36 int dstStride, int src1Stride, int h);
37 void ff_put_pixels8_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
38 int dstStride, int src1Stride, int h);
39 void ff_avg_pixels8_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
40 int dstStride, int src1Stride, int h);
41 void ff_put_pixels16_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
42 int dstStride, int src1Stride, int h);
43 void ff_avg_pixels16_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
44 int dstStride, int src1Stride, int h);
45 #define ff_put_pixels8_l2_sse2 ff_put_pixels8_l2_mmxext
46 #define ff_avg_pixels8_l2_sse2 ff_avg_pixels8_l2_mmxext
47 #define ff_put_pixels16_l2_sse2 ff_put_pixels16_l2_mmxext
48 #define ff_avg_pixels16_l2_sse2 ff_avg_pixels16_l2_mmxext
49 #define ff_put_pixels16_mmxext ff_put_pixels16_mmx
50 #define ff_put_pixels8_mmxext ff_put_pixels8_mmx
51 #define ff_put_pixels4_mmxext ff_put_pixels4_mmx
52
53 #define DEF_QPEL(OPNAME)\
54 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_mmxext(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\
55 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_mmxext(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\
56 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_ssse3(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\
57 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_l2_mmxext(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);\
58 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_l2_mmxext(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);\
59 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_l2_ssse3(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);\
60 void ff_ ## OPNAME ## _h264_qpel4_v_lowpass_mmxext(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\
61 void ff_ ## OPNAME ## _h264_qpel8or16_v_lowpass_op_mmxext(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h);\
62 void ff_ ## OPNAME ## _h264_qpel8or16_v_lowpass_sse2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h);\
63 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_v_mmxext(const uint8_t *src, int16_t *tmp, int srcStride);\
64 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_h_mmxext(int16_t *tmp, uint8_t *dst, int dstStride);\
65 void ff_ ## OPNAME ## _h264_qpel8or16_hv1_lowpass_op_mmxext(const uint8_t *src, int16_t *tmp, int srcStride, int size);\
66 void ff_ ## OPNAME ## _h264_qpel8or16_hv1_lowpass_op_sse2(const uint8_t *src, int16_t *tmp, int srcStride, int size);\
67 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_op_mmxext(uint8_t *dst, int16_t *tmp, int dstStride, int unused, int h);\
68 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_ssse3(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size);\
69 void ff_ ## OPNAME ## _pixels4_l2_shift5_mmxext(uint8_t *dst, const int16_t *src16, const uint8_t *src8, int dstStride, int src8Stride, int h);\
70 void ff_ ## OPNAME ## _pixels8_l2_shift5_mmxext(uint8_t *dst, const int16_t *src16, const uint8_t *src8, int dstStride, int src8Stride, int h);
71
72 DEF_QPEL(avg)
DEF_QPEL(put)73 DEF_QPEL(put)
74
75 static av_always_inline void ff_put_h264_qpel8or16_hv1_lowpass_mmxext(int16_t *tmp, const uint8_t *src, int tmpStride, int srcStride, int size)
76 {
77 int w = (size + 8) >> 2;
78 src -= 2 * srcStride + 2;
79 while (w--) {
80 ff_put_h264_qpel8or16_hv1_lowpass_op_mmxext(src, tmp, srcStride, size);
81 tmp += 4;
82 src += 4;
83 }
84 }
85
86 #define QPEL_H264(OPNAME, OP, MMX)\
87 static av_always_inline void ff_ ## OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, const uint8_t *src, int dstStride, int tmpStride, int srcStride){\
88 int w=3;\
89 src -= 2*srcStride+2;\
90 while(w--){\
91 ff_ ## OPNAME ## h264_qpel4_hv_lowpass_v_mmxext(src, tmp, srcStride);\
92 tmp += 4;\
93 src += 4;\
94 }\
95 tmp -= 3*4;\
96 ff_ ## OPNAME ## h264_qpel4_hv_lowpass_h_mmxext(tmp, dst, dstStride);\
97 }\
98 \
99 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h){\
100 src -= 2*srcStride;\
101 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_op_mmxext(dst, src, dstStride, srcStride, h);\
102 src += 4;\
103 dst += 4;\
104 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_op_mmxext(dst, src, dstStride, srcStride, h);\
105 }\
106 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
107 int w = size>>4;\
108 do{\
109 ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_op_mmxext(dst, tmp, dstStride, 0, size);\
110 tmp += 8;\
111 dst += 8;\
112 }while(w--);\
113 }\
114 \
115 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
116 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
117 }\
118 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
119 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\
120 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
121 }\
122 \
123 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
124 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
125 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
126 src += 8*srcStride;\
127 dst += 8*dstStride;\
128 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
129 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
130 }\
131 \
132 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride){\
133 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
134 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
135 src += 8*dstStride;\
136 dst += 8*dstStride;\
137 src2 += 8*src2Stride;\
138 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
139 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
140 }\
141 \
142 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, const uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
143 ff_put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\
144 ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
145 }\
146 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, const uint8_t *src, int dstStride, int tmpStride, int srcStride){\
147 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 8);\
148 }\
149 \
150 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, const uint8_t *src, int dstStride, int tmpStride, int srcStride){\
151 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 16);\
152 }\
153 \
154 static av_always_inline void ff_ ## OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, const int16_t *src16, const uint8_t *src8, int dstStride, int src8Stride, int h)\
155 {\
156 ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst , src16 , src8 , dstStride, src8Stride, h);\
157 ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
158 }\
159
160
161 #if ARCH_X86_64
162 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
163
164 void ff_avg_h264_qpel16_h_lowpass_l2_ssse3(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);
165 void ff_put_h264_qpel16_h_lowpass_l2_ssse3(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);
166
167 #else // ARCH_X86_64
168 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
169 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride){\
170 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
171 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
172 src += 8*dstStride;\
173 dst += 8*dstStride;\
174 src2 += 8*src2Stride;\
175 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
176 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
177 }
178 #endif // ARCH_X86_64
179
180 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
181 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
182 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
183 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
184 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
185 src += 8*srcStride;\
186 dst += 8*dstStride;\
187 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
188 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
189 }\
190
191 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
192 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
193 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
194 }\
195 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
196 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\
197 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
198 }
199
put_h264_qpel8or16_hv1_lowpass_sse2(int16_t * tmp,const uint8_t * src,int tmpStride,int srcStride,int size)200 static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp,
201 const uint8_t *src,
202 int tmpStride,
203 int srcStride,
204 int size)
205 {
206 int w = (size+8)>>3;
207 src -= 2*srcStride+2;
208 while(w--){
209 ff_put_h264_qpel8or16_hv1_lowpass_op_sse2(src, tmp, srcStride, size);
210 tmp += 8;
211 src += 8;
212 }
213 }
214
215 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
216 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, const uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
217 put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
218 ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
219 }\
220 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, const uint8_t *src, int dstStride, int tmpStride, int srcStride){\
221 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
222 }\
223 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, const uint8_t *src, int dstStride, int tmpStride, int srcStride){\
224 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
225 }\
226
227 #define ff_put_h264_qpel8_h_lowpass_l2_sse2 ff_put_h264_qpel8_h_lowpass_l2_mmxext
228 #define ff_avg_h264_qpel8_h_lowpass_l2_sse2 ff_avg_h264_qpel8_h_lowpass_l2_mmxext
229 #define ff_put_h264_qpel16_h_lowpass_l2_sse2 ff_put_h264_qpel16_h_lowpass_l2_mmxext
230 #define ff_avg_h264_qpel16_h_lowpass_l2_sse2 ff_avg_h264_qpel16_h_lowpass_l2_mmxext
231
232 #define ff_put_h264_qpel8_v_lowpass_ssse3 ff_put_h264_qpel8_v_lowpass_sse2
233 #define ff_avg_h264_qpel8_v_lowpass_ssse3 ff_avg_h264_qpel8_v_lowpass_sse2
234 #define ff_put_h264_qpel16_v_lowpass_ssse3 ff_put_h264_qpel16_v_lowpass_sse2
235 #define ff_avg_h264_qpel16_v_lowpass_ssse3 ff_avg_h264_qpel16_v_lowpass_sse2
236
237 #define ff_put_h264_qpel8or16_hv2_lowpass_sse2 ff_put_h264_qpel8or16_hv2_lowpass_mmxext
238 #define ff_avg_h264_qpel8or16_hv2_lowpass_sse2 ff_avg_h264_qpel8or16_hv2_lowpass_mmxext
239
240 #define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
241 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
242 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
243 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
244 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
245
put_h264_qpel16_mc00_sse2(uint8_t * dst,const uint8_t * src,ptrdiff_t stride)246 static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, const uint8_t *src,
247 ptrdiff_t stride)
248 {
249 ff_put_pixels16_sse2(dst, src, stride, 16);
250 }
avg_h264_qpel16_mc00_sse2(uint8_t * dst,const uint8_t * src,ptrdiff_t stride)251 static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, const uint8_t *src,
252 ptrdiff_t stride)
253 {
254 ff_avg_pixels16_sse2(dst, src, stride, 16);
255 }
256 #define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmxext
257 #define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmxext
258
259 #define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
260 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
261 {\
262 ff_ ## OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
263 }\
264
265 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
266 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
267 {\
268 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
269 }\
270 \
271 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
272 {\
273 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
274 }\
275 \
276 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
277 {\
278 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
279 }\
280
281 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
282 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
283 {\
284 LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*SIZE]);\
285 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
286 ff_ ## OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
287 }\
288 \
289 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
290 {\
291 ff_ ## OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
292 }\
293 \
294 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
295 {\
296 LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*SIZE]);\
297 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
298 ff_ ## OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
299 }\
300
301 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
302 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
303 {\
304 LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*SIZE]);\
305 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
306 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
307 }\
308 \
309 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
310 {\
311 LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*SIZE]);\
312 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
313 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
314 }\
315 \
316 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
317 {\
318 LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*SIZE]);\
319 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
320 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
321 }\
322 \
323 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
324 {\
325 LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*SIZE]);\
326 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
327 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
328 }\
329 \
330 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
331 {\
332 LOCAL_ALIGNED(ALIGN, uint16_t, temp, [SIZE*(SIZE<8?12:24)]);\
333 ff_ ## OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
334 }\
335 \
336 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
337 {\
338 LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
339 uint8_t * const halfHV= temp;\
340 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
341 av_assert2(((int)temp & 7) == 0);\
342 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
343 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
344 }\
345 \
346 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
347 {\
348 LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
349 uint8_t * const halfHV= temp;\
350 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
351 av_assert2(((int)temp & 7) == 0);\
352 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
353 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
354 }\
355 \
356 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
357 {\
358 LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
359 uint8_t * const halfHV= temp;\
360 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
361 av_assert2(((int)temp & 7) == 0);\
362 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
363 ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
364 }\
365 \
366 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
367 {\
368 LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
369 uint8_t * const halfHV= temp;\
370 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
371 av_assert2(((int)temp & 7) == 0);\
372 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
373 ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
374 }\
375
376 #define H264_MC_4816(MMX)\
377 H264_MC(put_, 4, MMX, 8)\
378 H264_MC(put_, 8, MMX, 8)\
379 H264_MC(put_, 16,MMX, 8)\
380 H264_MC(avg_, 4, MMX, 8)\
381 H264_MC(avg_, 8, MMX, 8)\
382 H264_MC(avg_, 16,MMX, 8)\
383
384 #define H264_MC_816(QPEL, XMM)\
385 QPEL(put_, 8, XMM, 16)\
386 QPEL(put_, 16,XMM, 16)\
387 QPEL(avg_, 8, XMM, 16)\
388 QPEL(avg_, 16,XMM, 16)\
389
QPEL_H264(put_,PUT_OP,mmxext)390 QPEL_H264(put_, PUT_OP, mmxext)
391 QPEL_H264(avg_, AVG_MMXEXT_OP, mmxext)
392 QPEL_H264_V_XMM(put_, PUT_OP, sse2)
393 QPEL_H264_V_XMM(avg_,AVG_MMXEXT_OP, sse2)
394 QPEL_H264_HV_XMM(put_, PUT_OP, sse2)
395 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, sse2)
396 QPEL_H264_H_XMM(put_, PUT_OP, ssse3)
397 QPEL_H264_H_XMM(avg_,AVG_MMXEXT_OP, ssse3)
398 QPEL_H264_HV_XMM(put_, PUT_OP, ssse3)
399 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, ssse3)
400
401 H264_MC_4816(mmxext)
402 H264_MC_816(H264_MC_V, sse2)
403 H264_MC_816(H264_MC_HV, sse2)
404 H264_MC_816(H264_MC_H, ssse3)
405 H264_MC_816(H264_MC_HV, ssse3)
406
407
408 //10bit
409 #define LUMA_MC_OP(OP, NUM, DEPTH, TYPE, OPT) \
410 void ff_ ## OP ## _h264_qpel ## NUM ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT \
411 (uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
412
413 #define LUMA_MC_ALL(DEPTH, TYPE, OPT) \
414 LUMA_MC_OP(put, 4, DEPTH, TYPE, OPT) \
415 LUMA_MC_OP(avg, 4, DEPTH, TYPE, OPT) \
416 LUMA_MC_OP(put, 8, DEPTH, TYPE, OPT) \
417 LUMA_MC_OP(avg, 8, DEPTH, TYPE, OPT) \
418 LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
419 LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
420
421 #define LUMA_MC_816(DEPTH, TYPE, OPT) \
422 LUMA_MC_OP(put, 8, DEPTH, TYPE, OPT) \
423 LUMA_MC_OP(avg, 8, DEPTH, TYPE, OPT) \
424 LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
425 LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
426
427 LUMA_MC_ALL(10, mc00, mmxext)
428 LUMA_MC_ALL(10, mc10, mmxext)
429 LUMA_MC_ALL(10, mc20, mmxext)
430 LUMA_MC_ALL(10, mc30, mmxext)
431 LUMA_MC_ALL(10, mc01, mmxext)
432 LUMA_MC_ALL(10, mc11, mmxext)
433 LUMA_MC_ALL(10, mc21, mmxext)
434 LUMA_MC_ALL(10, mc31, mmxext)
435 LUMA_MC_ALL(10, mc02, mmxext)
436 LUMA_MC_ALL(10, mc12, mmxext)
437 LUMA_MC_ALL(10, mc22, mmxext)
438 LUMA_MC_ALL(10, mc32, mmxext)
439 LUMA_MC_ALL(10, mc03, mmxext)
440 LUMA_MC_ALL(10, mc13, mmxext)
441 LUMA_MC_ALL(10, mc23, mmxext)
442 LUMA_MC_ALL(10, mc33, mmxext)
443
444 LUMA_MC_816(10, mc00, sse2)
445 LUMA_MC_816(10, mc10, sse2)
446 LUMA_MC_816(10, mc10, sse2_cache64)
447 LUMA_MC_816(10, mc10, ssse3_cache64)
448 LUMA_MC_816(10, mc20, sse2)
449 LUMA_MC_816(10, mc20, sse2_cache64)
450 LUMA_MC_816(10, mc20, ssse3_cache64)
451 LUMA_MC_816(10, mc30, sse2)
452 LUMA_MC_816(10, mc30, sse2_cache64)
453 LUMA_MC_816(10, mc30, ssse3_cache64)
454 LUMA_MC_816(10, mc01, sse2)
455 LUMA_MC_816(10, mc11, sse2)
456 LUMA_MC_816(10, mc21, sse2)
457 LUMA_MC_816(10, mc31, sse2)
458 LUMA_MC_816(10, mc02, sse2)
459 LUMA_MC_816(10, mc12, sse2)
460 LUMA_MC_816(10, mc22, sse2)
461 LUMA_MC_816(10, mc32, sse2)
462 LUMA_MC_816(10, mc03, sse2)
463 LUMA_MC_816(10, mc13, sse2)
464 LUMA_MC_816(10, mc23, sse2)
465 LUMA_MC_816(10, mc33, sse2)
466
467 #define QPEL16_OPMC(OP, MC, MMX)\
468 void ff_ ## OP ## _h264_qpel16_ ## MC ## _10_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride){\
469 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst , src , stride);\
470 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
471 src += 8*stride;\
472 dst += 8*stride;\
473 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst , src , stride);\
474 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
475 }
476
477 #define QPEL16_OP(MC, MMX)\
478 QPEL16_OPMC(put, MC, MMX)\
479 QPEL16_OPMC(avg, MC, MMX)
480
481 #define QPEL16(MMX)\
482 QPEL16_OP(mc00, MMX)\
483 QPEL16_OP(mc01, MMX)\
484 QPEL16_OP(mc02, MMX)\
485 QPEL16_OP(mc03, MMX)\
486 QPEL16_OP(mc10, MMX)\
487 QPEL16_OP(mc11, MMX)\
488 QPEL16_OP(mc12, MMX)\
489 QPEL16_OP(mc13, MMX)\
490 QPEL16_OP(mc20, MMX)\
491 QPEL16_OP(mc21, MMX)\
492 QPEL16_OP(mc22, MMX)\
493 QPEL16_OP(mc23, MMX)\
494 QPEL16_OP(mc30, MMX)\
495 QPEL16_OP(mc31, MMX)\
496 QPEL16_OP(mc32, MMX)\
497 QPEL16_OP(mc33, MMX)
498
499 #if ARCH_X86_32 // ARCH_X86_64 implies SSE2+
500 QPEL16(mmxext)
501 #endif
502
503 #endif /* HAVE_X86ASM */
504
505 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX) \
506 do { \
507 c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
508 c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
509 c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
510 c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
511 c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
512 c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
513 c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
514 c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
515 c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
516 c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
517 c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
518 c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
519 c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
520 c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
521 c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
522 c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
523 } while (0)
524
525 #define H264_QPEL_FUNCS(x, y, CPU) \
526 do { \
527 c->put_h264_qpel_pixels_tab[0][x + y * 4] = put_h264_qpel16_mc ## x ## y ## _ ## CPU; \
528 c->put_h264_qpel_pixels_tab[1][x + y * 4] = put_h264_qpel8_mc ## x ## y ## _ ## CPU; \
529 c->avg_h264_qpel_pixels_tab[0][x + y * 4] = avg_h264_qpel16_mc ## x ## y ## _ ## CPU; \
530 c->avg_h264_qpel_pixels_tab[1][x + y * 4] = avg_h264_qpel8_mc ## x ## y ## _ ## CPU; \
531 } while (0)
532
533 #define H264_QPEL_FUNCS_10(x, y, CPU) \
534 do { \
535 c->put_h264_qpel_pixels_tab[0][x + y * 4] = ff_put_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
536 c->put_h264_qpel_pixels_tab[1][x + y * 4] = ff_put_h264_qpel8_mc ## x ## y ## _10_ ## CPU; \
537 c->avg_h264_qpel_pixels_tab[0][x + y * 4] = ff_avg_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
538 c->avg_h264_qpel_pixels_tab[1][x + y * 4] = ff_avg_h264_qpel8_mc ## x ## y ## _10_ ## CPU; \
539 } while (0)
540
541 av_cold void ff_h264qpel_init_x86(H264QpelContext *c, int bit_depth)
542 {
543 #if HAVE_X86ASM
544 int high_bit_depth = bit_depth > 8;
545 int cpu_flags = av_get_cpu_flags();
546
547 if (EXTERNAL_MMXEXT(cpu_flags)) {
548 if (!high_bit_depth) {
549 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmxext, );
550 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, mmxext, );
551 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, mmxext, );
552 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmxext, );
553 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, mmxext, );
554 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, mmxext, );
555 } else if (bit_depth == 10) {
556 #if ARCH_X86_32
557 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_mmxext, ff_);
558 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_mmxext, ff_);
559 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 10_mmxext, ff_);
560 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 10_mmxext, ff_);
561 #endif
562 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, 10_mmxext, ff_);
563 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, 10_mmxext, ff_);
564 }
565 }
566
567 if (EXTERNAL_SSE2(cpu_flags)) {
568 if (!high_bit_depth) {
569 H264_QPEL_FUNCS(0, 1, sse2);
570 H264_QPEL_FUNCS(0, 2, sse2);
571 H264_QPEL_FUNCS(0, 3, sse2);
572 H264_QPEL_FUNCS(1, 1, sse2);
573 H264_QPEL_FUNCS(1, 2, sse2);
574 H264_QPEL_FUNCS(1, 3, sse2);
575 H264_QPEL_FUNCS(2, 1, sse2);
576 H264_QPEL_FUNCS(2, 2, sse2);
577 H264_QPEL_FUNCS(2, 3, sse2);
578 H264_QPEL_FUNCS(3, 1, sse2);
579 H264_QPEL_FUNCS(3, 2, sse2);
580 H264_QPEL_FUNCS(3, 3, sse2);
581 }
582
583 if (bit_depth == 10) {
584 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_sse2, ff_);
585 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 10_sse2, ff_);
586 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_sse2, ff_);
587 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 10_sse2, ff_);
588 H264_QPEL_FUNCS_10(1, 0, sse2_cache64);
589 H264_QPEL_FUNCS_10(2, 0, sse2_cache64);
590 H264_QPEL_FUNCS_10(3, 0, sse2_cache64);
591 }
592 }
593
594 if (EXTERNAL_SSE2_FAST(cpu_flags)) {
595 if (!high_bit_depth) {
596 H264_QPEL_FUNCS(0, 0, sse2);
597 }
598 }
599
600 if (EXTERNAL_SSSE3(cpu_flags)) {
601 if (!high_bit_depth) {
602 H264_QPEL_FUNCS(1, 0, ssse3);
603 H264_QPEL_FUNCS(1, 1, ssse3);
604 H264_QPEL_FUNCS(1, 2, ssse3);
605 H264_QPEL_FUNCS(1, 3, ssse3);
606 H264_QPEL_FUNCS(2, 0, ssse3);
607 H264_QPEL_FUNCS(2, 1, ssse3);
608 H264_QPEL_FUNCS(2, 2, ssse3);
609 H264_QPEL_FUNCS(2, 3, ssse3);
610 H264_QPEL_FUNCS(3, 0, ssse3);
611 H264_QPEL_FUNCS(3, 1, ssse3);
612 H264_QPEL_FUNCS(3, 2, ssse3);
613 H264_QPEL_FUNCS(3, 3, ssse3);
614 }
615
616 if (bit_depth == 10) {
617 H264_QPEL_FUNCS_10(1, 0, ssse3_cache64);
618 H264_QPEL_FUNCS_10(2, 0, ssse3_cache64);
619 H264_QPEL_FUNCS_10(3, 0, ssse3_cache64);
620 }
621 }
622
623 if (EXTERNAL_AVX(cpu_flags)) {
624 /* AVX implies 64 byte cache lines without the need to avoid unaligned
625 * memory accesses that cross the boundary between two cache lines.
626 * TODO: Port X264_CPU_CACHELINE_32/64 detection from x264 to avoid
627 * having to treat SSE2 functions with such properties as AVX. */
628 if (bit_depth == 10) {
629 H264_QPEL_FUNCS_10(1, 0, sse2);
630 H264_QPEL_FUNCS_10(2, 0, sse2);
631 H264_QPEL_FUNCS_10(3, 0, sse2);
632 }
633 }
634 #endif
635 }
636