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/h264qpel.h"
28 #include "libavcodec/pixels.h"
29 #include "fpel.h"
30 
31 #if HAVE_X86ASM
32 void ff_put_pixels4_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
33                               int dstStride, int src1Stride, int h);
34 void ff_avg_pixels4_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
35                               int dstStride, int src1Stride, int h);
36 void ff_put_pixels8_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
37                               int dstStride, int src1Stride, int h);
38 void ff_avg_pixels8_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
39                               int dstStride, int src1Stride, int h);
40 void ff_put_pixels16_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
41                                int dstStride, int src1Stride, int h);
42 void ff_avg_pixels16_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
43                                int dstStride, int src1Stride, int h);
44 #define ff_put_pixels8_l2_sse2  ff_put_pixels8_l2_mmxext
45 #define ff_avg_pixels8_l2_sse2  ff_avg_pixels8_l2_mmxext
46 #define ff_put_pixels16_l2_sse2 ff_put_pixels16_l2_mmxext
47 #define ff_avg_pixels16_l2_sse2 ff_avg_pixels16_l2_mmxext
48 #define ff_put_pixels16_mmxext  ff_put_pixels16_mmx
49 #define ff_put_pixels8_mmxext   ff_put_pixels8_mmx
50 #define ff_put_pixels4_mmxext   ff_put_pixels4_mmx
51 
52 #define DEF_QPEL(OPNAME)\
53 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_mmxext(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\
54 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_mmxext(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\
55 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_ssse3(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\
56 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_l2_mmxext(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);\
57 void ff_ ## OPNAME ## _h264_qpel8_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_ssse3(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);\
59 void ff_ ## OPNAME ## _h264_qpel4_v_lowpass_mmxext(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\
60 void ff_ ## OPNAME ## _h264_qpel8or16_v_lowpass_sse2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h);\
61 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_v_mmxext(const uint8_t *src, int16_t *tmp, int srcStride);\
62 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_h_mmxext(int16_t *tmp, uint8_t *dst, int dstStride);\
63 void ff_ ## OPNAME ## _h264_qpel8or16_hv1_lowpass_op_sse2(const uint8_t *src, int16_t *tmp, int srcStride, int size);\
64 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_op_mmxext(uint8_t *dst, int16_t *tmp, int dstStride, int unused, int h);\
65 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_ssse3(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size);\
66 void ff_ ## OPNAME ## _pixels4_l2_shift5_mmxext(uint8_t *dst, const int16_t *src16, const uint8_t *src8, int dstStride, int src8Stride, int h);\
67 void ff_ ## OPNAME ## _pixels8_l2_shift5_mmxext(uint8_t *dst, const int16_t *src16, const uint8_t *src8, int dstStride, int src8Stride, int h);
68 
69 DEF_QPEL(avg)
70 DEF_QPEL(put)
71 
72 #define QPEL_H264(OPNAME, OP, MMX)\
73 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){\
74     int w=3;\
75     src -= 2*srcStride+2;\
76     while(w--){\
77         ff_ ## OPNAME ## h264_qpel4_hv_lowpass_v_mmxext(src, tmp, srcStride);\
78         tmp += 4;\
79         src += 4;\
80     }\
81     tmp -= 3*4;\
82     ff_ ## OPNAME ## h264_qpel4_hv_lowpass_h_mmxext(tmp, dst, dstStride);\
83 }\
84 \
85 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
86     int w = size>>4;\
87     do{\
88     ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_op_mmxext(dst, tmp, dstStride, 0, size);\
89     tmp += 8;\
90     dst += 8;\
91     }while(w--);\
92 }\
93 \
94 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
95     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
96     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
97     src += 8*srcStride;\
98     dst += 8*dstStride;\
99     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
100     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
101 }\
102 \
103 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){\
104     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
105     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
106     src += 8*dstStride;\
107     dst += 8*dstStride;\
108     src2 += 8*src2Stride;\
109     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
110     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
111 }\
112 \
113 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)\
114 {\
115     ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\
116     ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
117 }\
118 
119 
120 #if ARCH_X86_64
121 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
122 
123 void ff_avg_h264_qpel16_h_lowpass_l2_ssse3(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);
124 void ff_put_h264_qpel16_h_lowpass_l2_ssse3(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);
125 
126 #else // ARCH_X86_64
127 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
128 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){\
129     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
130     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
131     src += 8*dstStride;\
132     dst += 8*dstStride;\
133     src2 += 8*src2Stride;\
134     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
135     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
136 }
137 #endif // ARCH_X86_64
138 
139 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
140 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
141 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
142     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
143     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
144     src += 8*srcStride;\
145     dst += 8*dstStride;\
146     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
147     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
148 }\
149 
150 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
151 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
152     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
153 }\
154 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
155     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
156     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
157 }
158 
put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, const uint8_t *src, int tmpStride, int srcStride, int size)159 static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp,
160                                                                  const uint8_t *src,
161                                                                  int tmpStride,
162                                                                  int srcStride,
163                                                                  int size)
164 {
165     int w = (size+8)>>3;
166     src -= 2*srcStride+2;
167     while(w--){
168         ff_put_h264_qpel8or16_hv1_lowpass_op_sse2(src, tmp, srcStride, size);
169         tmp += 8;
170         src += 8;
171     }
172 }
173 
174 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
175 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){\
176     put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
177     ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
178 }\
179 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){\
180     ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
181 }\
182 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){\
183     ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
184 }\
185 
186 #define ff_put_h264_qpel8_h_lowpass_l2_sse2  ff_put_h264_qpel8_h_lowpass_l2_mmxext
187 #define ff_avg_h264_qpel8_h_lowpass_l2_sse2  ff_avg_h264_qpel8_h_lowpass_l2_mmxext
188 #define ff_put_h264_qpel16_h_lowpass_l2_sse2 ff_put_h264_qpel16_h_lowpass_l2_mmxext
189 #define ff_avg_h264_qpel16_h_lowpass_l2_sse2 ff_avg_h264_qpel16_h_lowpass_l2_mmxext
190 
191 #define ff_put_h264_qpel8_v_lowpass_ssse3  ff_put_h264_qpel8_v_lowpass_sse2
192 #define ff_avg_h264_qpel8_v_lowpass_ssse3  ff_avg_h264_qpel8_v_lowpass_sse2
193 #define ff_put_h264_qpel16_v_lowpass_ssse3 ff_put_h264_qpel16_v_lowpass_sse2
194 #define ff_avg_h264_qpel16_v_lowpass_ssse3 ff_avg_h264_qpel16_v_lowpass_sse2
195 
196 #define ff_put_h264_qpel8or16_hv2_lowpass_sse2 ff_put_h264_qpel8or16_hv2_lowpass_mmxext
197 #define ff_avg_h264_qpel8or16_hv2_lowpass_sse2 ff_avg_h264_qpel8or16_hv2_lowpass_mmxext
198 
199 #define H264_MC_C_H(OPNAME, SIZE, MMX, ALIGN) \
200 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
201 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
202 
203 #define H264_MC_C_V_H_HV(OPNAME, SIZE, MMX, ALIGN) \
204 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
205 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
206 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
207 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
208 
put_h264_qpel16_mc00_sse2(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)209 static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, const uint8_t *src,
210                                        ptrdiff_t stride)
211 {
212     ff_put_pixels16_sse2(dst, src, stride, 16);
213 }
avg_h264_qpel16_mc00_sse2(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)214 static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, const uint8_t *src,
215                                        ptrdiff_t stride)
216 {
217     ff_avg_pixels16_sse2(dst, src, stride, 16);
218 }
219 #define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmxext
220 #define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmxext
221 
222 #define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
223 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
224 {\
225     ff_ ## OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
226 }\
227 
228 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
229 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
230 {\
231     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
232 }\
233 \
234 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
235 {\
236     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
237 }\
238 \
239 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
240 {\
241     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
242 }\
243 
244 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
245 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
246 {\
247     LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*SIZE]);\
248     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
249     ff_ ## OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
250 }\
251 \
252 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
253 {\
254     ff_ ## OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
255 }\
256 \
257 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
258 {\
259     LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*SIZE]);\
260     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
261     ff_ ## OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
262 }\
263 
264 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
265 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
266 {\
267     LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*SIZE]);\
268     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
269     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
270 }\
271 \
272 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
273 {\
274     LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*SIZE]);\
275     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
276     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
277 }\
278 \
279 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
280 {\
281     LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*SIZE]);\
282     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
283     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
284 }\
285 \
286 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
287 {\
288     LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*SIZE]);\
289     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
290     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
291 }\
292 \
293 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
294 {\
295     LOCAL_ALIGNED(ALIGN, uint16_t, temp, [SIZE*(SIZE<8?12:24)]);\
296     ff_ ## OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
297 }\
298 \
299 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
300 {\
301     LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
302     uint8_t * const halfHV= temp;\
303     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
304     av_assert2(((uintptr_t)temp & 7) == 0);\
305     ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
306     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
307 }\
308 \
309 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
310 {\
311     LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
312     uint8_t * const halfHV= temp;\
313     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
314     av_assert2(((uintptr_t)temp & 7) == 0);\
315     ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
316     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
317 }\
318 \
319 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
320 {\
321     LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
322     uint8_t * const halfHV= temp;\
323     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
324     av_assert2(((uintptr_t)temp & 7) == 0);\
325     ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
326     ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
327 }\
328 \
329 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
330 {\
331     LOCAL_ALIGNED(ALIGN, uint8_t, temp, [SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
332     uint8_t * const halfHV= temp;\
333     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
334     av_assert2(((uintptr_t)temp & 7) == 0);\
335     ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
336     ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
337 }\
338 
339 #define H264_MC(QPEL, SIZE, MMX, ALIGN)\
340 QPEL(put_, SIZE, MMX, ALIGN) \
341 QPEL(avg_, SIZE, MMX, ALIGN) \
342 
343 #define H264_MC_816(QPEL, XMM)\
344 QPEL(put_, 8, XMM, 16)\
345 QPEL(put_, 16,XMM, 16)\
346 QPEL(avg_, 8, XMM, 16)\
347 QPEL(avg_, 16,XMM, 16)\
348 
QPEL_H264null349 QPEL_H264(put_,        PUT_OP, mmxext)
350 QPEL_H264(avg_, AVG_MMXEXT_OP, mmxext)
351 QPEL_H264_V_XMM(put_,       PUT_OP, sse2)
352 QPEL_H264_V_XMM(avg_,AVG_MMXEXT_OP, sse2)
353 QPEL_H264_HV_XMM(put_,       PUT_OP, sse2)
354 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, sse2)
355 QPEL_H264_H_XMM(put_,       PUT_OP, ssse3)
356 QPEL_H264_H_XMM(avg_,AVG_MMXEXT_OP, ssse3)
357 QPEL_H264_HV_XMM(put_,       PUT_OP, ssse3)
358 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, ssse3)
359 
360 H264_MC(H264_MC_C_V_H_HV, 4, mmxext, 8)
361 H264_MC(H264_MC_C_H, 8, mmxext, 8)
362 H264_MC(H264_MC_C_H, 16, mmxext, 8)
363 H264_MC_816(H264_MC_V, sse2)
364 H264_MC_816(H264_MC_HV, sse2)
365 H264_MC_816(H264_MC_H, ssse3)
366 H264_MC_816(H264_MC_HV, ssse3)
367 
368 
369 //10bit
370 #define LUMA_MC_OP(OP, NUM, DEPTH, TYPE, OPT) \
371 void ff_ ## OP ## _h264_qpel ## NUM ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT \
372     (uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
373 
374 #define LUMA_MC_4(DEPTH, TYPE, OPT) \
375     LUMA_MC_OP(put,  4, DEPTH, TYPE, OPT) \
376     LUMA_MC_OP(avg,  4, DEPTH, TYPE, OPT)
377 
378 #define LUMA_MC_816(DEPTH, TYPE, OPT) \
379     LUMA_MC_OP(put,  8, DEPTH, TYPE, OPT) \
380     LUMA_MC_OP(avg,  8, DEPTH, TYPE, OPT) \
381     LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
382     LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
383 
384 LUMA_MC_4(10, mc00, mmxext)
385 LUMA_MC_4(10, mc10, mmxext)
386 LUMA_MC_4(10, mc20, mmxext)
387 LUMA_MC_4(10, mc30, mmxext)
388 LUMA_MC_4(10, mc01, mmxext)
389 LUMA_MC_4(10, mc11, mmxext)
390 LUMA_MC_4(10, mc21, mmxext)
391 LUMA_MC_4(10, mc31, mmxext)
392 LUMA_MC_4(10, mc02, mmxext)
393 LUMA_MC_4(10, mc12, mmxext)
394 LUMA_MC_4(10, mc22, mmxext)
395 LUMA_MC_4(10, mc32, mmxext)
396 LUMA_MC_4(10, mc03, mmxext)
397 LUMA_MC_4(10, mc13, mmxext)
398 LUMA_MC_4(10, mc23, mmxext)
399 LUMA_MC_4(10, mc33, mmxext)
400 
401 LUMA_MC_816(10, mc00, sse2)
402 LUMA_MC_816(10, mc10, sse2)
403 LUMA_MC_816(10, mc10, sse2_cache64)
404 LUMA_MC_816(10, mc10, ssse3_cache64)
405 LUMA_MC_816(10, mc20, sse2)
406 LUMA_MC_816(10, mc20, sse2_cache64)
407 LUMA_MC_816(10, mc20, ssse3_cache64)
408 LUMA_MC_816(10, mc30, sse2)
409 LUMA_MC_816(10, mc30, sse2_cache64)
410 LUMA_MC_816(10, mc30, ssse3_cache64)
411 LUMA_MC_816(10, mc01, sse2)
412 LUMA_MC_816(10, mc11, sse2)
413 LUMA_MC_816(10, mc21, sse2)
414 LUMA_MC_816(10, mc31, sse2)
415 LUMA_MC_816(10, mc02, sse2)
416 LUMA_MC_816(10, mc12, sse2)
417 LUMA_MC_816(10, mc22, sse2)
418 LUMA_MC_816(10, mc32, sse2)
419 LUMA_MC_816(10, mc03, sse2)
420 LUMA_MC_816(10, mc13, sse2)
421 LUMA_MC_816(10, mc23, sse2)
422 LUMA_MC_816(10, mc33, sse2)
423 
424 #endif /* HAVE_X86ASM */
425 
426 #define SET_QPEL_FUNCS0123(PFX, IDX, SIZE, CPU, PREFIX)                      \
427     do {                                                                     \
428     c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
429     c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
430     c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
431     c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
432     } while (0)
433 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX)                          \
434     do {                                                                     \
435     SET_QPEL_FUNCS0123(PFX, IDX, SIZE, CPU, PREFIX);                         \
436     c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
437     c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
438     c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
439     c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
440     c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
441     c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
442     c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
443     c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
444     c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
445     c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
446     c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
447     c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
448     } while (0)
449 
450 #define H264_QPEL_FUNCS(x, y, CPU)                                                            \
451     do {                                                                                      \
452         c->put_h264_qpel_pixels_tab[0][x + y * 4] = put_h264_qpel16_mc ## x ## y ## _ ## CPU; \
453         c->put_h264_qpel_pixels_tab[1][x + y * 4] = put_h264_qpel8_mc  ## x ## y ## _ ## CPU; \
454         c->avg_h264_qpel_pixels_tab[0][x + y * 4] = avg_h264_qpel16_mc ## x ## y ## _ ## CPU; \
455         c->avg_h264_qpel_pixels_tab[1][x + y * 4] = avg_h264_qpel8_mc  ## x ## y ## _ ## CPU; \
456     } while (0)
457 
458 #define H264_QPEL_FUNCS_10(x, y, CPU)                                                               \
459     do {                                                                                            \
460         c->put_h264_qpel_pixels_tab[0][x + y * 4] = ff_put_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
461         c->put_h264_qpel_pixels_tab[1][x + y * 4] = ff_put_h264_qpel8_mc  ## x ## y ## _10_ ## CPU; \
462         c->avg_h264_qpel_pixels_tab[0][x + y * 4] = ff_avg_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
463         c->avg_h264_qpel_pixels_tab[1][x + y * 4] = ff_avg_h264_qpel8_mc  ## x ## y ## _10_ ## CPU; \
464     } while (0)
465 
466 av_cold void ff_h264qpel_init_x86(H264QpelContext *c, int bit_depth)
467 {
468 #if HAVE_X86ASM
469     int high_bit_depth = bit_depth > 8;
470     int cpu_flags = av_get_cpu_flags();
471 
472     if (EXTERNAL_MMXEXT(cpu_flags)) {
473         if (!high_bit_depth) {
474             SET_QPEL_FUNCS0123(put_h264_qpel, 0, 16, mmxext, );
475             SET_QPEL_FUNCS0123(put_h264_qpel, 1,  8, mmxext, );
476             SET_QPEL_FUNCS(put_h264_qpel, 2,  4, mmxext, );
477             SET_QPEL_FUNCS0123(avg_h264_qpel, 0, 16, mmxext, );
478             SET_QPEL_FUNCS0123(avg_h264_qpel, 1,  8, mmxext, );
479             SET_QPEL_FUNCS(avg_h264_qpel, 2,  4, mmxext, );
480         } else if (bit_depth == 10) {
481             SET_QPEL_FUNCS(put_h264_qpel, 2, 4,  10_mmxext, ff_);
482             SET_QPEL_FUNCS(avg_h264_qpel, 2, 4,  10_mmxext, ff_);
483         }
484     }
485 
486     if (EXTERNAL_SSE2(cpu_flags)) {
487         if (!high_bit_depth) {
488             H264_QPEL_FUNCS(0, 1, sse2);
489             H264_QPEL_FUNCS(0, 2, sse2);
490             H264_QPEL_FUNCS(0, 3, sse2);
491             H264_QPEL_FUNCS(1, 1, sse2);
492             H264_QPEL_FUNCS(1, 2, sse2);
493             H264_QPEL_FUNCS(1, 3, sse2);
494             H264_QPEL_FUNCS(2, 1, sse2);
495             H264_QPEL_FUNCS(2, 2, sse2);
496             H264_QPEL_FUNCS(2, 3, sse2);
497             H264_QPEL_FUNCS(3, 1, sse2);
498             H264_QPEL_FUNCS(3, 2, sse2);
499             H264_QPEL_FUNCS(3, 3, sse2);
500         }
501 
502         if (bit_depth == 10) {
503             SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_sse2, ff_);
504             SET_QPEL_FUNCS(put_h264_qpel, 1,  8, 10_sse2, ff_);
505             SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_sse2, ff_);
506             SET_QPEL_FUNCS(avg_h264_qpel, 1,  8, 10_sse2, ff_);
507             H264_QPEL_FUNCS_10(1, 0, sse2_cache64);
508             H264_QPEL_FUNCS_10(2, 0, sse2_cache64);
509             H264_QPEL_FUNCS_10(3, 0, sse2_cache64);
510         }
511     }
512 
513     if (EXTERNAL_SSE2_FAST(cpu_flags)) {
514         if (!high_bit_depth) {
515             H264_QPEL_FUNCS(0, 0, sse2);
516         }
517     }
518 
519     if (EXTERNAL_SSSE3(cpu_flags)) {
520         if (!high_bit_depth) {
521             H264_QPEL_FUNCS(1, 0, ssse3);
522             H264_QPEL_FUNCS(1, 1, ssse3);
523             H264_QPEL_FUNCS(1, 2, ssse3);
524             H264_QPEL_FUNCS(1, 3, ssse3);
525             H264_QPEL_FUNCS(2, 0, ssse3);
526             H264_QPEL_FUNCS(2, 1, ssse3);
527             H264_QPEL_FUNCS(2, 2, ssse3);
528             H264_QPEL_FUNCS(2, 3, ssse3);
529             H264_QPEL_FUNCS(3, 0, ssse3);
530             H264_QPEL_FUNCS(3, 1, ssse3);
531             H264_QPEL_FUNCS(3, 2, ssse3);
532             H264_QPEL_FUNCS(3, 3, ssse3);
533         }
534 
535         if (bit_depth == 10) {
536             H264_QPEL_FUNCS_10(1, 0, ssse3_cache64);
537             H264_QPEL_FUNCS_10(2, 0, ssse3_cache64);
538             H264_QPEL_FUNCS_10(3, 0, ssse3_cache64);
539         }
540     }
541 
542     if (EXTERNAL_AVX(cpu_flags)) {
543         /* AVX implies 64 byte cache lines without the need to avoid unaligned
544          * memory accesses that cross the boundary between two cache lines.
545          * TODO: Port X264_CPU_CACHELINE_32/64 detection from x264 to avoid
546          * having to treat SSE2 functions with such properties as AVX. */
547         if (bit_depth == 10) {
548             H264_QPEL_FUNCS_10(1, 0, sse2);
549             H264_QPEL_FUNCS_10(2, 0, sse2);
550             H264_QPEL_FUNCS_10(3, 0, sse2);
551         }
552     }
553 #endif
554 }
555