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,
32 ptrdiff_t line_size, int h);
33 void ff_avg_pixels4_mmxext(uint8_t *block, const uint8_t *pixels,
34 ptrdiff_t line_size, int h);
35 static void ff_put_pixels16_mmxext(uint8_t *block, const uint8_t *pixels,
36 ptrdiff_t line_size, int h)
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,
42 ptrdiff_t line_size, int h)
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 #define ff_put_pixels8_l2_sse2 ff_put_pixels8_l2_mmxext
60 #define ff_avg_pixels8_l2_sse2 ff_avg_pixels8_l2_mmxext
61 #define ff_put_pixels16_l2_sse2 ff_put_pixels16_l2_mmxext
62 #define ff_avg_pixels16_l2_sse2 ff_avg_pixels16_l2_mmxext
64 #define DEF_QPEL(OPNAME)\
65 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
66 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
67 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_ssse3(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
68 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_l2_mmxext(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);\
69 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_l2_mmxext(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);\
70 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_l2_ssse3(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);\
71 void ff_ ## OPNAME ## _h264_qpel4_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
72 void ff_ ## OPNAME ## _h264_qpel8or16_v_lowpass_op_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h);\
73 void ff_ ## OPNAME ## _h264_qpel8or16_v_lowpass_sse2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h);\
74 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_v_mmxext(uint8_t *src, int16_t *tmp, int srcStride);\
75 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_h_mmxext(int16_t *tmp, uint8_t *dst, int dstStride);\
76 void ff_ ## OPNAME ## _h264_qpel8or16_hv1_lowpass_op_mmxext(uint8_t *src, int16_t *tmp, int srcStride, int size);\
77 void ff_ ## OPNAME ## _h264_qpel8or16_hv1_lowpass_op_sse2(uint8_t *src, int16_t *tmp, int srcStride, int size);\
78 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_op_mmxext(uint8_t *dst, int16_t *tmp, int dstStride, int unused, int h);\
79 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_ssse3(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size);\
80 void ff_ ## OPNAME ## _pixels4_l2_shift5_mmxext(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h);\
81 void ff_ ## OPNAME ## _pixels8_l2_shift5_mmxext(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h);
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, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
89 src -= 2*srcStride+2;\
91 ff_ ## OPNAME ## h264_qpel4_hv_lowpass_v_mmxext(src, tmp, srcStride);\
96 ff_ ## OPNAME ## h264_qpel4_hv_lowpass_h_mmxext(tmp, dst, dstStride);\
99 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
101 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_op_mmxext(dst, src, dstStride, srcStride, h);\
104 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_op_mmxext(dst, src, dstStride, srcStride, h);\
106 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){\
107 int w = (size+8)>>2;\
108 src -= 2*srcStride+2;\
110 ff_ ## OPNAME ## h264_qpel8or16_hv1_lowpass_op_mmxext(src, tmp, srcStride, size);\
115 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
118 ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_op_mmxext(dst, tmp, dstStride, 0, size);\
124 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
125 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
127 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_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, 16);\
129 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
132 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
133 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
134 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, 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 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){\
142 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
143 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
146 src2 += 8*src2Stride;\
147 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
148 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
151 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){\
152 ff_put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\
153 ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
155 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){\
156 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 8);\
159 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){\
160 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 16);\
163 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)\
165 ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst , src16 , src8 , dstStride, src8Stride, h);\
166 ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
171 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
173 void ff_avg_h264_qpel16_h_lowpass_l2_ssse3(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);
174 void ff_put_h264_qpel16_h_lowpass_l2_ssse3(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);
177 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
178 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){\
179 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
180 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
183 src2 += 8*src2Stride;\
184 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
185 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
187 #endif // ARCH_X86_64
189 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
190 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
191 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
192 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
193 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, 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 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
201 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
202 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
204 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_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, 16);\
206 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
209 static av_always_inline void ff_put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
211 src -= 2*srcStride+2;
213 ff_put_h264_qpel8or16_hv1_lowpass_op_sse2(src, tmp, srcStride, size);
219 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
220 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){\
221 ff_put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
222 ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
224 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){\
225 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
227 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){\
228 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
231 #define ff_put_h264_qpel8_h_lowpass_l2_sse2 ff_put_h264_qpel8_h_lowpass_l2_mmxext
232 #define ff_avg_h264_qpel8_h_lowpass_l2_sse2 ff_avg_h264_qpel8_h_lowpass_l2_mmxext
233 #define ff_put_h264_qpel16_h_lowpass_l2_sse2 ff_put_h264_qpel16_h_lowpass_l2_mmxext
234 #define ff_avg_h264_qpel16_h_lowpass_l2_sse2 ff_avg_h264_qpel16_h_lowpass_l2_mmxext
236 #define ff_put_h264_qpel8_v_lowpass_ssse3 ff_put_h264_qpel8_v_lowpass_sse2
237 #define ff_avg_h264_qpel8_v_lowpass_ssse3 ff_avg_h264_qpel8_v_lowpass_sse2
238 #define ff_put_h264_qpel16_v_lowpass_ssse3 ff_put_h264_qpel16_v_lowpass_sse2
239 #define ff_avg_h264_qpel16_v_lowpass_ssse3 ff_avg_h264_qpel16_v_lowpass_sse2
241 #define ff_put_h264_qpel8or16_hv2_lowpass_sse2 ff_put_h264_qpel8or16_hv2_lowpass_mmxext
242 #define ff_avg_h264_qpel8or16_hv2_lowpass_sse2 ff_avg_h264_qpel8or16_hv2_lowpass_mmxext
244 #define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
245 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
246 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
247 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
248 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
250 static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src,
253 ff_put_pixels16_sse2(dst, src, stride, 16);
255 static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src,
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, ptrdiff_t stride)\
266 ff_ ## OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
269 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
270 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
272 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
275 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
277 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
280 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
282 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
285 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
286 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
288 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
289 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
290 ff_ ## OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
293 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
295 ff_ ## OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
298 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t 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 ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
305 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
306 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
308 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
309 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
310 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
313 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
315 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
316 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
317 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
320 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
322 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
323 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
324 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
327 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
329 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
330 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
331 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
334 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
336 DECLARE_ALIGNED(ALIGN, uint16_t, temp)[SIZE*(SIZE<8?12:24)];\
337 ff_ ## OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
340 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
342 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
343 uint8_t * const halfHV= temp;\
344 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
345 assert(((int)temp & 7) == 0);\
346 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
347 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
350 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
352 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
353 uint8_t * const halfHV= temp;\
354 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
355 assert(((int)temp & 7) == 0);\
356 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
357 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
360 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
362 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
363 uint8_t * const halfHV= temp;\
364 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
365 assert(((int)temp & 7) == 0);\
366 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
367 ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
370 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
372 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
373 uint8_t * const halfHV= temp;\
374 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
375 assert(((int)temp & 7) == 0);\
376 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
377 ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
380 #define H264_MC_4816(MMX)\
381 H264_MC(put_, 4, MMX, 8)\
382 H264_MC(put_, 8, MMX, 8)\
383 H264_MC(put_, 16,MMX, 8)\
384 H264_MC(avg_, 4, MMX, 8)\
385 H264_MC(avg_, 8, MMX, 8)\
386 H264_MC(avg_, 16,MMX, 8)\
388 #define H264_MC_816(QPEL, XMM)\
389 QPEL(put_, 8, XMM, 16)\
390 QPEL(put_, 16,XMM, 16)\
391 QPEL(avg_, 8, XMM, 16)\
392 QPEL(avg_, 16,XMM, 16)\
395 #define PAVGB "pavgb"
396 QPEL_H264(put_, PUT_OP, mmxext)
397 QPEL_H264(avg_, AVG_MMXEXT_OP, mmxext)
398 QPEL_H264_V_XMM(put_, PUT_OP, sse2)
399 QPEL_H264_V_XMM(avg_,AVG_MMXEXT_OP, sse2)
400 QPEL_H264_HV_XMM(put_, PUT_OP, sse2)
401 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, sse2)
402 QPEL_H264_H_XMM(put_, PUT_OP, ssse3)
403 QPEL_H264_H_XMM(avg_,AVG_MMXEXT_OP, ssse3)
404 QPEL_H264_HV_XMM(put_, PUT_OP, ssse3)
405 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, ssse3)
409 H264_MC_816(H264_MC_V, sse2)
410 H264_MC_816(H264_MC_HV, sse2)
411 H264_MC_816(H264_MC_H, ssse3)
412 H264_MC_816(H264_MC_HV, ssse3)
416 #define LUMA_MC_OP(OP, NUM, DEPTH, TYPE, OPT) \
417 void ff_ ## OP ## _h264_qpel ## NUM ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT \
418 (uint8_t *dst, uint8_t *src, ptrdiff_t stride);
420 #define LUMA_MC_ALL(DEPTH, TYPE, OPT) \
421 LUMA_MC_OP(put, 4, DEPTH, TYPE, OPT) \
422 LUMA_MC_OP(avg, 4, DEPTH, TYPE, OPT) \
423 LUMA_MC_OP(put, 8, DEPTH, TYPE, OPT) \
424 LUMA_MC_OP(avg, 8, DEPTH, TYPE, OPT) \
425 LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
426 LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
428 #define LUMA_MC_816(DEPTH, TYPE, OPT) \
429 LUMA_MC_OP(put, 8, DEPTH, TYPE, OPT) \
430 LUMA_MC_OP(avg, 8, DEPTH, TYPE, OPT) \
431 LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
432 LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
434 LUMA_MC_ALL(10, mc00, mmxext)
435 LUMA_MC_ALL(10, mc10, mmxext)
436 LUMA_MC_ALL(10, mc20, mmxext)
437 LUMA_MC_ALL(10, mc30, mmxext)
438 LUMA_MC_ALL(10, mc01, mmxext)
439 LUMA_MC_ALL(10, mc11, mmxext)
440 LUMA_MC_ALL(10, mc21, mmxext)
441 LUMA_MC_ALL(10, mc31, mmxext)
442 LUMA_MC_ALL(10, mc02, mmxext)
443 LUMA_MC_ALL(10, mc12, mmxext)
444 LUMA_MC_ALL(10, mc22, mmxext)
445 LUMA_MC_ALL(10, mc32, mmxext)
446 LUMA_MC_ALL(10, mc03, mmxext)
447 LUMA_MC_ALL(10, mc13, mmxext)
448 LUMA_MC_ALL(10, mc23, mmxext)
449 LUMA_MC_ALL(10, mc33, mmxext)
451 LUMA_MC_816(10, mc00, sse2)
452 LUMA_MC_816(10, mc10, sse2)
453 LUMA_MC_816(10, mc10, sse2_cache64)
454 LUMA_MC_816(10, mc10, ssse3_cache64)
455 LUMA_MC_816(10, mc20, sse2)
456 LUMA_MC_816(10, mc20, sse2_cache64)
457 LUMA_MC_816(10, mc20, ssse3_cache64)
458 LUMA_MC_816(10, mc30, sse2)
459 LUMA_MC_816(10, mc30, sse2_cache64)
460 LUMA_MC_816(10, mc30, ssse3_cache64)
461 LUMA_MC_816(10, mc01, sse2)
462 LUMA_MC_816(10, mc11, sse2)
463 LUMA_MC_816(10, mc21, sse2)
464 LUMA_MC_816(10, mc31, sse2)
465 LUMA_MC_816(10, mc02, sse2)
466 LUMA_MC_816(10, mc12, sse2)
467 LUMA_MC_816(10, mc22, sse2)
468 LUMA_MC_816(10, mc32, sse2)
469 LUMA_MC_816(10, mc03, sse2)
470 LUMA_MC_816(10, mc13, sse2)
471 LUMA_MC_816(10, mc23, sse2)
472 LUMA_MC_816(10, mc33, sse2)
474 #define QPEL16_OPMC(OP, MC, MMX)\
475 void ff_ ## OP ## _h264_qpel16_ ## MC ## _10_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride){\
476 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst , src , stride);\
477 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
480 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst , src , stride);\
481 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
484 #define QPEL16_OP(MC, MMX)\
485 QPEL16_OPMC(put, MC, MMX)\
486 QPEL16_OPMC(avg, MC, MMX)
489 QPEL16_OP(mc00, MMX)\
490 QPEL16_OP(mc01, MMX)\
491 QPEL16_OP(mc02, MMX)\
492 QPEL16_OP(mc03, MMX)\
493 QPEL16_OP(mc10, MMX)\
494 QPEL16_OP(mc11, MMX)\
495 QPEL16_OP(mc12, MMX)\
496 QPEL16_OP(mc13, MMX)\
497 QPEL16_OP(mc20, MMX)\
498 QPEL16_OP(mc21, MMX)\
499 QPEL16_OP(mc22, MMX)\
500 QPEL16_OP(mc23, MMX)\
501 QPEL16_OP(mc30, MMX)\
502 QPEL16_OP(mc31, MMX)\
503 QPEL16_OP(mc32, MMX)\
506 #if ARCH_X86_32 && HAVE_YASM && CONFIG_H264QPEL // ARCH_X86_64 implies SSE2+
510 #endif /* HAVE_YASM */
512 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX) \
514 c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
515 c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
516 c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
517 c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
518 c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
519 c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
520 c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
521 c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
522 c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
523 c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
524 c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
525 c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
526 c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
527 c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
528 c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
529 c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
532 #define H264_QPEL_FUNCS(x, y, CPU) \
534 c->put_h264_qpel_pixels_tab[0][x + y * 4] = put_h264_qpel16_mc ## x ## y ## _ ## CPU; \
535 c->put_h264_qpel_pixels_tab[1][x + y * 4] = put_h264_qpel8_mc ## x ## y ## _ ## CPU; \
536 c->avg_h264_qpel_pixels_tab[0][x + y * 4] = avg_h264_qpel16_mc ## x ## y ## _ ## CPU; \
537 c->avg_h264_qpel_pixels_tab[1][x + y * 4] = avg_h264_qpel8_mc ## x ## y ## _ ## CPU; \
540 #define H264_QPEL_FUNCS_10(x, y, CPU) \
542 c->put_h264_qpel_pixels_tab[0][x + y * 4] = ff_put_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
543 c->put_h264_qpel_pixels_tab[1][x + y * 4] = ff_put_h264_qpel8_mc ## x ## y ## _10_ ## CPU; \
544 c->avg_h264_qpel_pixels_tab[0][x + y * 4] = ff_avg_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
545 c->avg_h264_qpel_pixels_tab[1][x + y * 4] = ff_avg_h264_qpel8_mc ## x ## y ## _10_ ## CPU; \
548 av_cold void ff_h264qpel_init_x86(H264QpelContext *c, int bit_depth)
551 int high_bit_depth = bit_depth > 8;
552 int mm_flags = av_get_cpu_flags();
554 if (EXTERNAL_MMXEXT(mm_flags)) {
555 if (!high_bit_depth) {
556 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmxext, );
557 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, mmxext, );
558 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, mmxext, );
559 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmxext, );
560 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, mmxext, );
561 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, mmxext, );
562 } else if (bit_depth == 10) {
564 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_mmxext, ff_);
565 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_mmxext, ff_);
566 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 10_mmxext, ff_);
567 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 10_mmxext, ff_);
569 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, 10_mmxext, ff_);
570 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, 10_mmxext, ff_);
574 if (EXTERNAL_SSE2(mm_flags)) {
575 if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW) && !high_bit_depth) {
576 // these functions are slower than mmx on AMD, but faster on Intel
577 H264_QPEL_FUNCS(0, 0, sse2);
580 if (!high_bit_depth) {
581 H264_QPEL_FUNCS(0, 1, sse2);
582 H264_QPEL_FUNCS(0, 2, sse2);
583 H264_QPEL_FUNCS(0, 3, sse2);
584 H264_QPEL_FUNCS(1, 1, sse2);
585 H264_QPEL_FUNCS(1, 2, sse2);
586 H264_QPEL_FUNCS(1, 3, sse2);
587 H264_QPEL_FUNCS(2, 1, sse2);
588 H264_QPEL_FUNCS(2, 2, sse2);
589 H264_QPEL_FUNCS(2, 3, sse2);
590 H264_QPEL_FUNCS(3, 1, sse2);
591 H264_QPEL_FUNCS(3, 2, sse2);
592 H264_QPEL_FUNCS(3, 3, sse2);
595 if (bit_depth == 10) {
596 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_sse2, ff_);
597 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 10_sse2, ff_);
598 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_sse2, ff_);
599 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 10_sse2, ff_);
600 H264_QPEL_FUNCS_10(1, 0, sse2_cache64);
601 H264_QPEL_FUNCS_10(2, 0, sse2_cache64);
602 H264_QPEL_FUNCS_10(3, 0, sse2_cache64);
606 if (EXTERNAL_SSSE3(mm_flags)) {
607 if (!high_bit_depth) {
608 H264_QPEL_FUNCS(1, 0, ssse3);
609 H264_QPEL_FUNCS(1, 1, ssse3);
610 H264_QPEL_FUNCS(1, 2, ssse3);
611 H264_QPEL_FUNCS(1, 3, ssse3);
612 H264_QPEL_FUNCS(2, 0, ssse3);
613 H264_QPEL_FUNCS(2, 1, ssse3);
614 H264_QPEL_FUNCS(2, 2, ssse3);
615 H264_QPEL_FUNCS(2, 3, ssse3);
616 H264_QPEL_FUNCS(3, 0, ssse3);
617 H264_QPEL_FUNCS(3, 1, ssse3);
618 H264_QPEL_FUNCS(3, 2, ssse3);
619 H264_QPEL_FUNCS(3, 3, ssse3);
622 if (bit_depth == 10) {
623 H264_QPEL_FUNCS_10(1, 0, ssse3_cache64);
624 H264_QPEL_FUNCS_10(2, 0, ssse3_cache64);
625 H264_QPEL_FUNCS_10(3, 0, ssse3_cache64);
629 if (EXTERNAL_AVX(mm_flags)) {
630 /* AVX implies 64 byte cache lines without the need to avoid unaligned
631 * memory accesses that cross the boundary between two cache lines.
632 * TODO: Port X264_CPU_CACHELINE_32/64 detection from x264 to avoid
633 * having to treat SSE2 functions with such properties as AVX. */
634 if (bit_depth == 10) {
635 H264_QPEL_FUNCS_10(1, 0, sse2);
636 H264_QPEL_FUNCS_10(2, 0, sse2);
637 H264_QPEL_FUNCS_10(3, 0, sse2);