2 * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
3 * Copyright (c) 2011 Daniel Kang
5 * This file is part of Libav.
7 * Libav 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 * Libav 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 Libav; 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 static void ff_put_pixels16_mmxext(uint8_t *block, const uint8_t *pixels,
37 ff_put_pixels8_mmxext(block, pixels, line_size, h);
38 ff_put_pixels8_mmxext(block + 8, pixels + 8, line_size, h);
40 static void ff_avg_pixels16_mmxext(uint8_t *block, const uint8_t *pixels,
43 ff_avg_pixels8_mmxext(block, pixels, line_size, h);
44 ff_avg_pixels8_mmxext(block + 8, pixels + 8, line_size, h);
46 void ff_put_pixels4_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
47 int dstStride, int src1Stride, int h);
48 void ff_avg_pixels4_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
49 int dstStride, int src1Stride, int h);
50 void ff_put_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
51 int dstStride, int src1Stride, int h);
52 void ff_avg_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
53 int dstStride, int src1Stride, int h);
54 void ff_put_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
55 int dstStride, int src1Stride, int h);
56 void ff_avg_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
57 int dstStride, int src1Stride, int h);
58 void ff_put_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
59 int line_size, int h);
60 void ff_avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
61 int line_size, int h);
62 #define ff_put_pixels8_l2_sse2 ff_put_pixels8_l2_mmxext
63 #define ff_avg_pixels8_l2_sse2 ff_avg_pixels8_l2_mmxext
64 #define ff_put_pixels16_l2_sse2 ff_put_pixels16_l2_mmxext
65 #define ff_avg_pixels16_l2_sse2 ff_avg_pixels16_l2_mmxext
67 #define DEF_QPEL(OPNAME)\
68 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
69 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
70 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_ssse3(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
71 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_l2_mmxext(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);\
72 void ff_ ## OPNAME ## _h264_qpel8_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_ssse3(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);\
74 void ff_ ## OPNAME ## _h264_qpel4_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
75 void ff_ ## OPNAME ## _h264_qpel8or16_v_lowpass_op_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h);\
76 void ff_ ## OPNAME ## _h264_qpel8or16_v_lowpass_sse2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h);\
77 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_v_mmxext(uint8_t *src, int16_t *tmp, int srcStride);\
78 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_h_mmxext(int16_t *tmp, uint8_t *dst, int dstStride);\
79 void ff_ ## OPNAME ## _h264_qpel8or16_hv1_lowpass_op_mmxext(uint8_t *src, int16_t *tmp, int srcStride, int size);\
80 void ff_ ## OPNAME ## _h264_qpel8or16_hv1_lowpass_op_sse2(uint8_t *src, int16_t *tmp, int srcStride, int size);\
81 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_op_mmxext(uint8_t *dst, int16_t *tmp, int dstStride, int unused, int h);\
82 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_ssse3(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size);\
83 void ff_ ## OPNAME ## _pixels4_l2_shift5_mmxext(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h);\
84 void ff_ ## OPNAME ## _pixels8_l2_shift5_mmxext(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h);
89 #define QPEL_H264(OPNAME, OP, MMX)\
90 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){\
92 src -= 2*srcStride+2;\
94 ff_ ## OPNAME ## h264_qpel4_hv_lowpass_v_mmxext(src, tmp, srcStride);\
99 ff_ ## OPNAME ## h264_qpel4_hv_lowpass_h_mmxext(tmp, dst, dstStride);\
102 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
104 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_op_mmxext(dst, src, dstStride, srcStride, h);\
107 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_op_mmxext(dst, src, dstStride, srcStride, h);\
109 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){\
110 int w = (size+8)>>2;\
111 src -= 2*srcStride+2;\
113 ff_ ## OPNAME ## h264_qpel8or16_hv1_lowpass_op_mmxext(src, tmp, srcStride, size);\
118 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
121 ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_op_mmxext(dst, tmp, dstStride, 0, size);\
127 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
128 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
130 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
131 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\
132 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
135 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
136 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
137 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
140 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
141 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
144 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){\
145 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
146 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
149 src2 += 8*src2Stride;\
150 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
151 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
154 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){\
155 ff_put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\
156 ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
158 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){\
159 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 8);\
162 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){\
163 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 16);\
166 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)\
168 ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst , src16 , src8 , dstStride, src8Stride, h);\
169 ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
174 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
176 void ff_avg_h264_qpel16_h_lowpass_l2_ssse3(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);
177 void ff_put_h264_qpel16_h_lowpass_l2_ssse3(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);
180 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
181 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){\
182 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
183 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
186 src2 += 8*src2Stride;\
187 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
188 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
190 #endif // ARCH_X86_64
192 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
193 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
194 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
195 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
196 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
199 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
200 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
203 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
204 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
205 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
207 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
208 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\
209 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
212 static av_always_inline void ff_put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
214 src -= 2*srcStride+2;
216 ff_put_h264_qpel8or16_hv1_lowpass_op_sse2(src, tmp, srcStride, size);
222 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
223 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){\
224 ff_put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
225 ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
227 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){\
228 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
230 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){\
231 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
234 #define ff_put_h264_qpel8_h_lowpass_l2_sse2 ff_put_h264_qpel8_h_lowpass_l2_mmxext
235 #define ff_avg_h264_qpel8_h_lowpass_l2_sse2 ff_avg_h264_qpel8_h_lowpass_l2_mmxext
236 #define ff_put_h264_qpel16_h_lowpass_l2_sse2 ff_put_h264_qpel16_h_lowpass_l2_mmxext
237 #define ff_avg_h264_qpel16_h_lowpass_l2_sse2 ff_avg_h264_qpel16_h_lowpass_l2_mmxext
239 #define ff_put_h264_qpel8_v_lowpass_ssse3 ff_put_h264_qpel8_v_lowpass_sse2
240 #define ff_avg_h264_qpel8_v_lowpass_ssse3 ff_avg_h264_qpel8_v_lowpass_sse2
241 #define ff_put_h264_qpel16_v_lowpass_ssse3 ff_put_h264_qpel16_v_lowpass_sse2
242 #define ff_avg_h264_qpel16_v_lowpass_ssse3 ff_avg_h264_qpel16_v_lowpass_sse2
244 #define ff_put_h264_qpel8or16_hv2_lowpass_sse2 ff_put_h264_qpel8or16_hv2_lowpass_mmxext
245 #define ff_avg_h264_qpel8or16_hv2_lowpass_sse2 ff_avg_h264_qpel8or16_hv2_lowpass_mmxext
247 #define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
248 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
249 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
250 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
251 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
253 static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src,
256 ff_put_pixels16_sse2(dst, src, stride, 16);
258 static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src,
261 ff_avg_pixels16_sse2(dst, src, stride, 16);
263 #define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmxext
264 #define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmxext
266 #define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
267 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
269 ff_ ## OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
272 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
273 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
275 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
278 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
280 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
283 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
285 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
288 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
289 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
291 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
292 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
293 ff_ ## OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
296 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
298 ff_ ## OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
301 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
303 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
304 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
305 ff_ ## OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
308 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
309 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
311 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
312 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
313 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
316 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t 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, temp, stride, SIZE);\
323 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
325 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
326 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
327 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
330 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
332 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
333 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
334 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
337 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
339 DECLARE_ALIGNED(ALIGN, uint16_t, temp)[SIZE*(SIZE<8?12:24)];\
340 ff_ ## OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
343 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
345 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
346 uint8_t * const halfHV= temp;\
347 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
348 assert(((int)temp & 7) == 0);\
349 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
350 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
353 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
355 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
356 uint8_t * const halfHV= temp;\
357 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
358 assert(((int)temp & 7) == 0);\
359 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
360 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
363 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
365 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
366 uint8_t * const halfHV= temp;\
367 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
368 assert(((int)temp & 7) == 0);\
369 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
370 ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
373 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
375 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
376 uint8_t * const halfHV= temp;\
377 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
378 assert(((int)temp & 7) == 0);\
379 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
380 ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
383 #define H264_MC_4816(MMX)\
384 H264_MC(put_, 4, MMX, 8)\
385 H264_MC(put_, 8, MMX, 8)\
386 H264_MC(put_, 16,MMX, 8)\
387 H264_MC(avg_, 4, MMX, 8)\
388 H264_MC(avg_, 8, MMX, 8)\
389 H264_MC(avg_, 16,MMX, 8)\
391 #define H264_MC_816(QPEL, XMM)\
392 QPEL(put_, 8, XMM, 16)\
393 QPEL(put_, 16,XMM, 16)\
394 QPEL(avg_, 8, XMM, 16)\
395 QPEL(avg_, 16,XMM, 16)\
398 #define PAVGB "pavgb"
399 QPEL_H264(put_, PUT_OP, mmxext)
400 QPEL_H264(avg_, AVG_MMXEXT_OP, mmxext)
401 QPEL_H264_V_XMM(put_, PUT_OP, sse2)
402 QPEL_H264_V_XMM(avg_,AVG_MMXEXT_OP, sse2)
403 QPEL_H264_HV_XMM(put_, PUT_OP, sse2)
404 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, sse2)
405 QPEL_H264_H_XMM(put_, PUT_OP, ssse3)
406 QPEL_H264_H_XMM(avg_,AVG_MMXEXT_OP, ssse3)
407 QPEL_H264_HV_XMM(put_, PUT_OP, ssse3)
408 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, ssse3)
412 H264_MC_816(H264_MC_V, sse2)
413 H264_MC_816(H264_MC_HV, sse2)
414 H264_MC_816(H264_MC_H, ssse3)
415 H264_MC_816(H264_MC_HV, ssse3)
419 #define LUMA_MC_OP(OP, NUM, DEPTH, TYPE, OPT) \
420 void ff_ ## OP ## _h264_qpel ## NUM ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT \
421 (uint8_t *dst, uint8_t *src, ptrdiff_t stride);
423 #define LUMA_MC_ALL(DEPTH, TYPE, OPT) \
424 LUMA_MC_OP(put, 4, DEPTH, TYPE, OPT) \
425 LUMA_MC_OP(avg, 4, DEPTH, TYPE, OPT) \
426 LUMA_MC_OP(put, 8, DEPTH, TYPE, OPT) \
427 LUMA_MC_OP(avg, 8, DEPTH, TYPE, OPT) \
428 LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
429 LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
431 #define LUMA_MC_816(DEPTH, TYPE, OPT) \
432 LUMA_MC_OP(put, 8, DEPTH, TYPE, OPT) \
433 LUMA_MC_OP(avg, 8, DEPTH, TYPE, OPT) \
434 LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
435 LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
437 LUMA_MC_ALL(10, mc00, mmxext)
438 LUMA_MC_ALL(10, mc10, mmxext)
439 LUMA_MC_ALL(10, mc20, mmxext)
440 LUMA_MC_ALL(10, mc30, mmxext)
441 LUMA_MC_ALL(10, mc01, mmxext)
442 LUMA_MC_ALL(10, mc11, mmxext)
443 LUMA_MC_ALL(10, mc21, mmxext)
444 LUMA_MC_ALL(10, mc31, mmxext)
445 LUMA_MC_ALL(10, mc02, mmxext)
446 LUMA_MC_ALL(10, mc12, mmxext)
447 LUMA_MC_ALL(10, mc22, mmxext)
448 LUMA_MC_ALL(10, mc32, mmxext)
449 LUMA_MC_ALL(10, mc03, mmxext)
450 LUMA_MC_ALL(10, mc13, mmxext)
451 LUMA_MC_ALL(10, mc23, mmxext)
452 LUMA_MC_ALL(10, mc33, mmxext)
454 LUMA_MC_816(10, mc00, sse2)
455 LUMA_MC_816(10, mc10, sse2)
456 LUMA_MC_816(10, mc10, sse2_cache64)
457 LUMA_MC_816(10, mc10, ssse3_cache64)
458 LUMA_MC_816(10, mc20, sse2)
459 LUMA_MC_816(10, mc20, sse2_cache64)
460 LUMA_MC_816(10, mc20, ssse3_cache64)
461 LUMA_MC_816(10, mc30, sse2)
462 LUMA_MC_816(10, mc30, sse2_cache64)
463 LUMA_MC_816(10, mc30, ssse3_cache64)
464 LUMA_MC_816(10, mc01, sse2)
465 LUMA_MC_816(10, mc11, sse2)
466 LUMA_MC_816(10, mc21, sse2)
467 LUMA_MC_816(10, mc31, sse2)
468 LUMA_MC_816(10, mc02, sse2)
469 LUMA_MC_816(10, mc12, sse2)
470 LUMA_MC_816(10, mc22, sse2)
471 LUMA_MC_816(10, mc32, sse2)
472 LUMA_MC_816(10, mc03, sse2)
473 LUMA_MC_816(10, mc13, sse2)
474 LUMA_MC_816(10, mc23, sse2)
475 LUMA_MC_816(10, mc33, sse2)
477 #define QPEL16_OPMC(OP, MC, MMX)\
478 void ff_ ## OP ## _h264_qpel16_ ## MC ## _10_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride){\
479 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst , src , stride);\
480 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
483 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst , src , stride);\
484 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
487 #define QPEL16_OP(MC, MMX)\
488 QPEL16_OPMC(put, MC, MMX)\
489 QPEL16_OPMC(avg, MC, MMX)
492 QPEL16_OP(mc00, MMX)\
493 QPEL16_OP(mc01, MMX)\
494 QPEL16_OP(mc02, MMX)\
495 QPEL16_OP(mc03, MMX)\
496 QPEL16_OP(mc10, MMX)\
497 QPEL16_OP(mc11, MMX)\
498 QPEL16_OP(mc12, MMX)\
499 QPEL16_OP(mc13, MMX)\
500 QPEL16_OP(mc20, MMX)\
501 QPEL16_OP(mc21, MMX)\
502 QPEL16_OP(mc22, MMX)\
503 QPEL16_OP(mc23, MMX)\
504 QPEL16_OP(mc30, MMX)\
505 QPEL16_OP(mc31, MMX)\
506 QPEL16_OP(mc32, MMX)\
509 #if ARCH_X86_32 && HAVE_YASM && CONFIG_H264QPEL // ARCH_X86_64 implies SSE2+
513 #endif /* HAVE_YASM */
515 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX) \
517 c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
518 c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
519 c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
520 c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
521 c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
522 c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
523 c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
524 c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
525 c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
526 c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
527 c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
528 c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
529 c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
530 c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
531 c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
532 c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
535 #define H264_QPEL_FUNCS(x, y, CPU) \
537 c->put_h264_qpel_pixels_tab[0][x + y * 4] = put_h264_qpel16_mc ## x ## y ## _ ## CPU; \
538 c->put_h264_qpel_pixels_tab[1][x + y * 4] = put_h264_qpel8_mc ## x ## y ## _ ## CPU; \
539 c->avg_h264_qpel_pixels_tab[0][x + y * 4] = avg_h264_qpel16_mc ## x ## y ## _ ## CPU; \
540 c->avg_h264_qpel_pixels_tab[1][x + y * 4] = avg_h264_qpel8_mc ## x ## y ## _ ## CPU; \
543 #define H264_QPEL_FUNCS_10(x, y, CPU) \
545 c->put_h264_qpel_pixels_tab[0][x + y * 4] = ff_put_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
546 c->put_h264_qpel_pixels_tab[1][x + y * 4] = ff_put_h264_qpel8_mc ## x ## y ## _10_ ## CPU; \
547 c->avg_h264_qpel_pixels_tab[0][x + y * 4] = ff_avg_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
548 c->avg_h264_qpel_pixels_tab[1][x + y * 4] = ff_avg_h264_qpel8_mc ## x ## y ## _10_ ## CPU; \
551 av_cold void ff_h264qpel_init_x86(H264QpelContext *c, int bit_depth)
554 int high_bit_depth = bit_depth > 8;
555 int mm_flags = av_get_cpu_flags();
557 if (EXTERNAL_MMXEXT(mm_flags)) {
558 if (!high_bit_depth) {
559 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmxext, );
560 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, mmxext, );
561 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, mmxext, );
562 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmxext, );
563 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, mmxext, );
564 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, mmxext, );
565 } else if (bit_depth == 10) {
567 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_mmxext, ff_);
568 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_mmxext, ff_);
569 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 10_mmxext, ff_);
570 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 10_mmxext, ff_);
572 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, 10_mmxext, ff_);
573 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, 10_mmxext, ff_);
577 if (EXTERNAL_SSE2(mm_flags)) {
578 if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW) && !high_bit_depth) {
579 // these functions are slower than mmx on AMD, but faster on Intel
580 H264_QPEL_FUNCS(0, 0, sse2);
583 if (!high_bit_depth) {
584 H264_QPEL_FUNCS(0, 1, sse2);
585 H264_QPEL_FUNCS(0, 2, sse2);
586 H264_QPEL_FUNCS(0, 3, sse2);
587 H264_QPEL_FUNCS(1, 1, sse2);
588 H264_QPEL_FUNCS(1, 2, sse2);
589 H264_QPEL_FUNCS(1, 3, sse2);
590 H264_QPEL_FUNCS(2, 1, sse2);
591 H264_QPEL_FUNCS(2, 2, sse2);
592 H264_QPEL_FUNCS(2, 3, sse2);
593 H264_QPEL_FUNCS(3, 1, sse2);
594 H264_QPEL_FUNCS(3, 2, sse2);
595 H264_QPEL_FUNCS(3, 3, sse2);
598 if (bit_depth == 10) {
599 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_sse2, ff_);
600 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 10_sse2, ff_);
601 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_sse2, ff_);
602 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 10_sse2, ff_);
603 H264_QPEL_FUNCS_10(1, 0, sse2_cache64);
604 H264_QPEL_FUNCS_10(2, 0, sse2_cache64);
605 H264_QPEL_FUNCS_10(3, 0, sse2_cache64);
609 if (EXTERNAL_SSSE3(mm_flags)) {
610 if (!high_bit_depth) {
611 H264_QPEL_FUNCS(1, 0, ssse3);
612 H264_QPEL_FUNCS(1, 1, ssse3);
613 H264_QPEL_FUNCS(1, 2, ssse3);
614 H264_QPEL_FUNCS(1, 3, ssse3);
615 H264_QPEL_FUNCS(2, 0, ssse3);
616 H264_QPEL_FUNCS(2, 1, ssse3);
617 H264_QPEL_FUNCS(2, 2, ssse3);
618 H264_QPEL_FUNCS(2, 3, ssse3);
619 H264_QPEL_FUNCS(3, 0, ssse3);
620 H264_QPEL_FUNCS(3, 1, ssse3);
621 H264_QPEL_FUNCS(3, 2, ssse3);
622 H264_QPEL_FUNCS(3, 3, ssse3);
625 if (bit_depth == 10) {
626 H264_QPEL_FUNCS_10(1, 0, ssse3_cache64);
627 H264_QPEL_FUNCS_10(2, 0, ssse3_cache64);
628 H264_QPEL_FUNCS_10(3, 0, ssse3_cache64);
632 if (EXTERNAL_AVX(mm_flags)) {
633 /* AVX implies 64 byte cache lines without the need to avoid unaligned
634 * memory accesses that cross the boundary between two cache lines.
635 * TODO: Port X264_CPU_CACHELINE_32/64 detection from x264 to avoid
636 * having to treat SSE2 functions with such properties as AVX. */
637 if (bit_depth == 10) {
638 H264_QPEL_FUNCS_10(1, 0, sse2);
639 H264_QPEL_FUNCS_10(2, 0, sse2);
640 H264_QPEL_FUNCS_10(3, 0, sse2);