2 * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
3 * Copyright (c) 2011 Daniel Kang
5 * This file is part of FFmpeg.
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.
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.
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
22 #include "libavutil/attributes.h"
23 #include "libavutil/cpu.h"
24 #include "libavutil/x86/asm.h"
25 #include "libavutil/x86/cpu.h"
26 #include "libavcodec/h264qpel.h"
27 #include "libavcodec/mpegvideo.h"
28 #include "dsputil_mmx.h"
31 void ff_put_pixels4_mmxext(uint8_t *block, const uint8_t *pixels, int line_size, int h);
32 void ff_avg_pixels4_mmxext(uint8_t *block, const uint8_t *pixels, int line_size, int h);
33 void ff_put_pixels8_mmxext(uint8_t *block, const uint8_t *pixels, int line_size, int h);
34 void ff_avg_pixels8_mmxext(uint8_t *block, const uint8_t *pixels, int line_size, int h);
35 static void ff_put_pixels16_mmxext(uint8_t *block, const uint8_t *pixels,
38 ff_put_pixels8_mmxext(block, pixels, line_size, h);
39 ff_put_pixels8_mmxext(block + 8, pixels + 8, line_size, h);
41 static void ff_avg_pixels16_mmxext(uint8_t *block, const uint8_t *pixels,
44 ff_avg_pixels8_mmxext(block, pixels, line_size, h);
45 ff_avg_pixels8_mmxext(block + 8, pixels + 8, line_size, h);
47 void ff_put_pixels4_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
48 int dstStride, int src1Stride, int h);
49 void ff_avg_pixels4_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
50 int dstStride, int src1Stride, int h);
51 void ff_put_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
52 int dstStride, int src1Stride, int h);
53 void ff_avg_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
54 int dstStride, int src1Stride, int h);
55 void ff_put_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
56 int dstStride, int src1Stride, int h);
57 void ff_avg_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
58 int dstStride, int src1Stride, int h);
59 void ff_put_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
60 int line_size, int h);
61 void ff_avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
62 int line_size, int h);
63 #define ff_put_pixels8_l2_sse2 ff_put_pixels8_l2_mmxext
64 #define ff_avg_pixels8_l2_sse2 ff_avg_pixels8_l2_mmxext
65 #define ff_put_pixels16_l2_sse2 ff_put_pixels16_l2_mmxext
66 #define ff_avg_pixels16_l2_sse2 ff_avg_pixels16_l2_mmxext
68 #define DEF_QPEL(OPNAME)\
69 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
70 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
71 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_ssse3(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
72 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_l2_mmxext(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);\
73 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_l2_mmxext(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);\
74 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_l2_ssse3(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);\
75 void ff_ ## OPNAME ## _h264_qpel4_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
76 void ff_ ## OPNAME ## _h264_qpel8or16_v_lowpass_op_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h);\
77 void ff_ ## OPNAME ## _h264_qpel8or16_v_lowpass_sse2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h);\
78 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_v_mmxext(uint8_t *src, int16_t *tmp, int srcStride);\
79 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_h_mmxext(int16_t *tmp, uint8_t *dst, int dstStride);\
80 void ff_ ## OPNAME ## _h264_qpel8or16_hv1_lowpass_op_mmxext(uint8_t *src, int16_t *tmp, int srcStride, int size);\
81 void ff_ ## OPNAME ## _h264_qpel8or16_hv1_lowpass_op_sse2(uint8_t *src, int16_t *tmp, int srcStride, int size);\
82 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_op_mmxext(uint8_t *dst, int16_t *tmp, int dstStride, int unused, int h);\
83 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_ssse3(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size);\
84 void ff_ ## OPNAME ## _pixels4_l2_shift5_mmxext(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h);\
85 void ff_ ## OPNAME ## _pixels8_l2_shift5_mmxext(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h);
90 #define QPEL_H264(OPNAME, OP, MMX)\
91 static av_always_inline void ff_ ## OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
93 src -= 2*srcStride+2;\
95 ff_ ## OPNAME ## h264_qpel4_hv_lowpass_v_mmxext(src, tmp, srcStride);\
100 ff_ ## OPNAME ## h264_qpel4_hv_lowpass_h_mmxext(tmp, dst, dstStride);\
103 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
105 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_op_mmxext(dst, src, dstStride, srcStride, h);\
108 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_op_mmxext(dst, src, dstStride, srcStride, h);\
110 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){\
111 int w = (size+8)>>2;\
112 src -= 2*srcStride+2;\
114 ff_ ## OPNAME ## h264_qpel8or16_hv1_lowpass_op_mmxext(src, tmp, srcStride, size);\
119 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
122 ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_op_mmxext(dst, tmp, dstStride, 0, size);\
128 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
129 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
131 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
132 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\
133 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
136 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
137 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
138 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
141 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
142 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
145 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
146 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
147 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
150 src2 += 8*src2Stride;\
151 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
152 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
155 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
156 ff_put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\
157 ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
159 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
160 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 8);\
163 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
164 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 16);\
167 static av_always_inline void ff_ ## OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
169 ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst , src16 , src8 , dstStride, src8Stride, h);\
170 ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
175 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
177 void ff_avg_h264_qpel16_h_lowpass_l2_ssse3(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);
178 void ff_put_h264_qpel16_h_lowpass_l2_ssse3(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);
181 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
182 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
183 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
184 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
187 src2 += 8*src2Stride;\
188 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
189 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
191 #endif // ARCH_X86_64
193 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
194 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
195 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
196 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
197 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
200 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
201 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
204 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
205 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
206 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
208 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
209 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\
210 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
213 static av_always_inline void ff_put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
215 src -= 2*srcStride+2;
217 ff_put_h264_qpel8or16_hv1_lowpass_op_sse2(src, tmp, srcStride, size);
223 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
224 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
225 ff_put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
226 ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
228 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
229 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
231 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
232 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
235 #define ff_put_h264_qpel8_h_lowpass_l2_sse2 ff_put_h264_qpel8_h_lowpass_l2_mmxext
236 #define ff_avg_h264_qpel8_h_lowpass_l2_sse2 ff_avg_h264_qpel8_h_lowpass_l2_mmxext
237 #define ff_put_h264_qpel16_h_lowpass_l2_sse2 ff_put_h264_qpel16_h_lowpass_l2_mmxext
238 #define ff_avg_h264_qpel16_h_lowpass_l2_sse2 ff_avg_h264_qpel16_h_lowpass_l2_mmxext
240 #define ff_put_h264_qpel8_v_lowpass_ssse3 ff_put_h264_qpel8_v_lowpass_sse2
241 #define ff_avg_h264_qpel8_v_lowpass_ssse3 ff_avg_h264_qpel8_v_lowpass_sse2
242 #define ff_put_h264_qpel16_v_lowpass_ssse3 ff_put_h264_qpel16_v_lowpass_sse2
243 #define ff_avg_h264_qpel16_v_lowpass_ssse3 ff_avg_h264_qpel16_v_lowpass_sse2
245 #define ff_put_h264_qpel8or16_hv2_lowpass_sse2 ff_put_h264_qpel8or16_hv2_lowpass_mmxext
246 #define ff_avg_h264_qpel8or16_hv2_lowpass_sse2 ff_avg_h264_qpel8or16_hv2_lowpass_mmxext
248 #define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
249 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
250 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
251 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
252 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
254 static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
255 ff_put_pixels16_sse2(dst, src, stride, 16);
257 static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
258 ff_avg_pixels16_sse2(dst, src, stride, 16);
260 #define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmxext
261 #define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmxext
263 #define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
264 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
265 ff_ ## OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
268 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
269 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
270 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
273 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
274 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
277 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
278 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
281 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
282 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
283 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
284 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
285 ff_ ## OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
288 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
289 ff_ ## OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
292 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
293 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
294 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
295 ff_ ## OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
298 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
299 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
300 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
301 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
302 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
305 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
306 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
307 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
308 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
311 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
312 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
313 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
314 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
317 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
318 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
319 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
320 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
323 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
324 DECLARE_ALIGNED(ALIGN, uint16_t, temp)[SIZE*(SIZE<8?12:24)];\
325 ff_ ## OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
328 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
329 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
330 uint8_t * const halfHV= temp;\
331 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
332 assert(((int)temp & 7) == 0);\
333 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
334 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
337 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
338 DECLARE_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 assert(((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+stride, halfHV, stride, SIZE);\
346 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
347 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
348 uint8_t * const halfHV= temp;\
349 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
350 assert(((int)temp & 7) == 0);\
351 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
352 ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
355 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
356 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
357 uint8_t * const halfHV= temp;\
358 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
359 assert(((int)temp & 7) == 0);\
360 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
361 ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
364 #define H264_MC_4816(MMX)\
365 H264_MC(put_, 4, MMX, 8)\
366 H264_MC(put_, 8, MMX, 8)\
367 H264_MC(put_, 16,MMX, 8)\
368 H264_MC(avg_, 4, MMX, 8)\
369 H264_MC(avg_, 8, MMX, 8)\
370 H264_MC(avg_, 16,MMX, 8)\
372 #define H264_MC_816(QPEL, XMM)\
373 QPEL(put_, 8, XMM, 16)\
374 QPEL(put_, 16,XMM, 16)\
375 QPEL(avg_, 8, XMM, 16)\
376 QPEL(avg_, 16,XMM, 16)\
379 #define PAVGB "pavgb"
380 QPEL_H264(put_, PUT_OP, mmxext)
381 QPEL_H264(avg_, AVG_MMXEXT_OP, mmxext)
382 QPEL_H264_V_XMM(put_, PUT_OP, sse2)
383 QPEL_H264_V_XMM(avg_,AVG_MMXEXT_OP, sse2)
384 QPEL_H264_HV_XMM(put_, PUT_OP, sse2)
385 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, sse2)
386 QPEL_H264_H_XMM(put_, PUT_OP, ssse3)
387 QPEL_H264_H_XMM(avg_,AVG_MMXEXT_OP, ssse3)
388 QPEL_H264_HV_XMM(put_, PUT_OP, ssse3)
389 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, ssse3)
393 H264_MC_816(H264_MC_V, sse2)
394 H264_MC_816(H264_MC_HV, sse2)
395 H264_MC_816(H264_MC_H, ssse3)
396 H264_MC_816(H264_MC_HV, ssse3)
400 #define LUMA_MC_OP(OP, NUM, DEPTH, TYPE, OPT) \
401 void ff_ ## OP ## _h264_qpel ## NUM ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT \
402 (uint8_t *dst, uint8_t *src, int stride);
404 #define LUMA_MC_ALL(DEPTH, TYPE, OPT) \
405 LUMA_MC_OP(put, 4, DEPTH, TYPE, OPT) \
406 LUMA_MC_OP(avg, 4, DEPTH, TYPE, OPT) \
407 LUMA_MC_OP(put, 8, DEPTH, TYPE, OPT) \
408 LUMA_MC_OP(avg, 8, DEPTH, TYPE, OPT) \
409 LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
410 LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
412 #define LUMA_MC_816(DEPTH, TYPE, OPT) \
413 LUMA_MC_OP(put, 8, DEPTH, TYPE, OPT) \
414 LUMA_MC_OP(avg, 8, DEPTH, TYPE, OPT) \
415 LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
416 LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
418 LUMA_MC_ALL(10, mc00, mmxext)
419 LUMA_MC_ALL(10, mc10, mmxext)
420 LUMA_MC_ALL(10, mc20, mmxext)
421 LUMA_MC_ALL(10, mc30, mmxext)
422 LUMA_MC_ALL(10, mc01, mmxext)
423 LUMA_MC_ALL(10, mc11, mmxext)
424 LUMA_MC_ALL(10, mc21, mmxext)
425 LUMA_MC_ALL(10, mc31, mmxext)
426 LUMA_MC_ALL(10, mc02, mmxext)
427 LUMA_MC_ALL(10, mc12, mmxext)
428 LUMA_MC_ALL(10, mc22, mmxext)
429 LUMA_MC_ALL(10, mc32, mmxext)
430 LUMA_MC_ALL(10, mc03, mmxext)
431 LUMA_MC_ALL(10, mc13, mmxext)
432 LUMA_MC_ALL(10, mc23, mmxext)
433 LUMA_MC_ALL(10, mc33, mmxext)
435 LUMA_MC_816(10, mc00, sse2)
436 LUMA_MC_816(10, mc10, sse2)
437 LUMA_MC_816(10, mc10, sse2_cache64)
438 LUMA_MC_816(10, mc10, ssse3_cache64)
439 LUMA_MC_816(10, mc20, sse2)
440 LUMA_MC_816(10, mc20, sse2_cache64)
441 LUMA_MC_816(10, mc20, ssse3_cache64)
442 LUMA_MC_816(10, mc30, sse2)
443 LUMA_MC_816(10, mc30, sse2_cache64)
444 LUMA_MC_816(10, mc30, ssse3_cache64)
445 LUMA_MC_816(10, mc01, sse2)
446 LUMA_MC_816(10, mc11, sse2)
447 LUMA_MC_816(10, mc21, sse2)
448 LUMA_MC_816(10, mc31, sse2)
449 LUMA_MC_816(10, mc02, sse2)
450 LUMA_MC_816(10, mc12, sse2)
451 LUMA_MC_816(10, mc22, sse2)
452 LUMA_MC_816(10, mc32, sse2)
453 LUMA_MC_816(10, mc03, sse2)
454 LUMA_MC_816(10, mc13, sse2)
455 LUMA_MC_816(10, mc23, sse2)
456 LUMA_MC_816(10, mc33, sse2)
458 #define QPEL16_OPMC(OP, MC, MMX)\
459 void ff_ ## OP ## _h264_qpel16_ ## MC ## _10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
460 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst , src , stride);\
461 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
464 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst , src , stride);\
465 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
468 #define QPEL16_OP(MC, MMX)\
469 QPEL16_OPMC(put, MC, MMX)\
470 QPEL16_OPMC(avg, MC, MMX)
473 QPEL16_OP(mc00, MMX)\
474 QPEL16_OP(mc01, MMX)\
475 QPEL16_OP(mc02, MMX)\
476 QPEL16_OP(mc03, MMX)\
477 QPEL16_OP(mc10, MMX)\
478 QPEL16_OP(mc11, MMX)\
479 QPEL16_OP(mc12, MMX)\
480 QPEL16_OP(mc13, MMX)\
481 QPEL16_OP(mc20, MMX)\
482 QPEL16_OP(mc21, MMX)\
483 QPEL16_OP(mc22, MMX)\
484 QPEL16_OP(mc23, MMX)\
485 QPEL16_OP(mc30, MMX)\
486 QPEL16_OP(mc31, MMX)\
487 QPEL16_OP(mc32, MMX)\
490 #if ARCH_X86_32 && HAVE_YASM && CONFIG_H264QPEL // ARCH_X86_64 implies SSE2+
494 #endif /* HAVE_YASM */
496 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX) \
498 c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
499 c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
500 c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
501 c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
502 c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
503 c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
504 c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
505 c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
506 c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
507 c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
508 c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
509 c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
510 c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
511 c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
512 c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
513 c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
516 #define H264_QPEL_FUNCS(x, y, CPU) \
518 c->put_h264_qpel_pixels_tab[0][x + y * 4] = put_h264_qpel16_mc ## x ## y ## _ ## CPU; \
519 c->put_h264_qpel_pixels_tab[1][x + y * 4] = put_h264_qpel8_mc ## x ## y ## _ ## CPU; \
520 c->avg_h264_qpel_pixels_tab[0][x + y * 4] = avg_h264_qpel16_mc ## x ## y ## _ ## CPU; \
521 c->avg_h264_qpel_pixels_tab[1][x + y * 4] = avg_h264_qpel8_mc ## x ## y ## _ ## CPU; \
524 #define H264_QPEL_FUNCS_10(x, y, CPU) \
526 c->put_h264_qpel_pixels_tab[0][x + y * 4] = ff_put_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
527 c->put_h264_qpel_pixels_tab[1][x + y * 4] = ff_put_h264_qpel8_mc ## x ## y ## _10_ ## CPU; \
528 c->avg_h264_qpel_pixels_tab[0][x + y * 4] = ff_avg_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
529 c->avg_h264_qpel_pixels_tab[1][x + y * 4] = ff_avg_h264_qpel8_mc ## x ## y ## _10_ ## CPU; \
532 av_cold void ff_h264qpel_init_x86(H264QpelContext *c, int bit_depth)
535 int high_bit_depth = bit_depth > 8;
536 int mm_flags = av_get_cpu_flags();
538 if (EXTERNAL_MMXEXT(mm_flags)) {
539 if (!high_bit_depth) {
540 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmxext, );
541 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, mmxext, );
542 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, mmxext, );
543 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmxext, );
544 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, mmxext, );
545 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, mmxext, );
546 } else if (bit_depth == 10) {
548 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_mmxext, ff_);
549 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_mmxext, ff_);
550 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 10_mmxext, ff_);
551 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 10_mmxext, ff_);
553 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, 10_mmxext, ff_);
554 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, 10_mmxext, ff_);
558 if (EXTERNAL_SSE2(mm_flags)) {
559 if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW) && !high_bit_depth) {
560 // these functions are slower than mmx on AMD, but faster on Intel
561 H264_QPEL_FUNCS(0, 0, sse2);
564 if (!high_bit_depth) {
565 H264_QPEL_FUNCS(0, 1, sse2);
566 H264_QPEL_FUNCS(0, 2, sse2);
567 H264_QPEL_FUNCS(0, 3, sse2);
568 H264_QPEL_FUNCS(1, 1, sse2);
569 H264_QPEL_FUNCS(1, 2, sse2);
570 H264_QPEL_FUNCS(1, 3, sse2);
571 H264_QPEL_FUNCS(2, 1, sse2);
572 H264_QPEL_FUNCS(2, 2, sse2);
573 H264_QPEL_FUNCS(2, 3, sse2);
574 H264_QPEL_FUNCS(3, 1, sse2);
575 H264_QPEL_FUNCS(3, 2, sse2);
576 H264_QPEL_FUNCS(3, 3, sse2);
579 if (bit_depth == 10) {
580 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_sse2, ff_);
581 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 10_sse2, ff_);
582 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_sse2, ff_);
583 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 10_sse2, ff_);
584 H264_QPEL_FUNCS_10(1, 0, sse2_cache64);
585 H264_QPEL_FUNCS_10(2, 0, sse2_cache64);
586 H264_QPEL_FUNCS_10(3, 0, sse2_cache64);
590 if (EXTERNAL_SSSE3(mm_flags)) {
591 if (!high_bit_depth) {
592 H264_QPEL_FUNCS(1, 0, ssse3);
593 H264_QPEL_FUNCS(1, 1, ssse3);
594 H264_QPEL_FUNCS(1, 2, ssse3);
595 H264_QPEL_FUNCS(1, 3, ssse3);
596 H264_QPEL_FUNCS(2, 0, ssse3);
597 H264_QPEL_FUNCS(2, 1, ssse3);
598 H264_QPEL_FUNCS(2, 2, ssse3);
599 H264_QPEL_FUNCS(2, 3, ssse3);
600 H264_QPEL_FUNCS(3, 0, ssse3);
601 H264_QPEL_FUNCS(3, 1, ssse3);
602 H264_QPEL_FUNCS(3, 2, ssse3);
603 H264_QPEL_FUNCS(3, 3, ssse3);
606 if (bit_depth == 10) {
607 H264_QPEL_FUNCS_10(1, 0, ssse3_cache64);
608 H264_QPEL_FUNCS_10(2, 0, ssse3_cache64);
609 H264_QPEL_FUNCS_10(3, 0, ssse3_cache64);
613 if (EXTERNAL_AVX(mm_flags)) {
614 /* AVX implies 64 byte cache lines without the need to avoid unaligned
615 * memory accesses that cross the boundary between two cache lines.
616 * TODO: Port X264_CPU_CACHELINE_32/64 detection from x264 to avoid
617 * having to treat SSE2 functions with such properties as AVX. */
618 if (bit_depth == 10) {
619 H264_QPEL_FUNCS_10(1, 0, sse2);
620 H264_QPEL_FUNCS_10(2, 0, sse2);
621 H264_QPEL_FUNCS_10(3, 0, sse2);