33 ptrdiff_t line_size,
int h);
35 ptrdiff_t line_size,
int h);
37 int dstStride,
int src1Stride,
int h);
39 int dstStride,
int src1Stride,
int h);
41 int dstStride,
int src1Stride,
int h);
43 int dstStride,
int src1Stride,
int h);
45 int dstStride,
int src1Stride,
int h);
47 int dstStride,
int src1Stride,
int h);
48 #define ff_put_pixels8_l2_sse2 ff_put_pixels8_l2_mmxext 49 #define ff_avg_pixels8_l2_sse2 ff_avg_pixels8_l2_mmxext 50 #define ff_put_pixels16_l2_sse2 ff_put_pixels16_l2_mmxext 51 #define ff_avg_pixels16_l2_sse2 ff_avg_pixels16_l2_mmxext 56 #define DEF_QPEL(OPNAME)\ 57 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_mmxext(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\ 58 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_mmxext(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\ 59 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_ssse3(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\ 60 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_l2_mmxext(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);\ 61 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_l2_mmxext(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);\ 62 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_l2_ssse3(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);\ 63 void ff_ ## OPNAME ## _h264_qpel4_v_lowpass_mmxext(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\ 64 void ff_ ## OPNAME ## _h264_qpel8or16_v_lowpass_op_mmxext(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h);\ 65 void ff_ ## OPNAME ## _h264_qpel8or16_v_lowpass_sse2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h);\ 66 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_v_mmxext(const uint8_t *src, int16_t *tmp, int srcStride);\ 67 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_h_mmxext(int16_t *tmp, uint8_t *dst, int dstStride);\ 68 void ff_ ## OPNAME ## _h264_qpel8or16_hv1_lowpass_op_mmxext(const uint8_t *src, int16_t *tmp, int srcStride, int size);\ 69 void ff_ ## OPNAME ## _h264_qpel8or16_hv1_lowpass_op_sse2(const uint8_t *src, int16_t *tmp, int srcStride, int size);\ 70 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_op_mmxext(uint8_t *dst, int16_t *tmp, int dstStride, int unused, int h);\ 71 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_ssse3(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size);\ 72 void ff_ ## OPNAME ## _pixels4_l2_shift5_mmxext(uint8_t *dst, const int16_t *src16, const uint8_t *src8, int dstStride, int src8Stride, int h);\ 73 void ff_ ## OPNAME ## _pixels8_l2_shift5_mmxext(uint8_t *dst, const int16_t *src16, const uint8_t *src8, int dstStride, int src8Stride, int h); 78 #define QPEL_H264(OPNAME, OP, MMX)\ 79 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){\ 81 src -= 2*srcStride+2;\ 83 ff_ ## OPNAME ## h264_qpel4_hv_lowpass_v_mmxext(src, tmp, srcStride);\ 88 ff_ ## OPNAME ## h264_qpel4_hv_lowpass_h_mmxext(tmp, dst, dstStride);\ 91 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h){\ 93 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_op_mmxext(dst, src, dstStride, srcStride, h);\ 96 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_op_mmxext(dst, src, dstStride, srcStride, h);\ 98 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, const uint8_t *src, int tmpStride, int srcStride, int size){\ 100 src -= 2*srcStride+2;\ 102 ff_ ## OPNAME ## h264_qpel8or16_hv1_lowpass_op_mmxext(src, tmp, srcStride, size);\ 107 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\ 110 ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_op_mmxext(dst, tmp, dstStride, 0, size);\ 116 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\ 117 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\ 119 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\ 120 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\ 121 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\ 124 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\ 125 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\ 126 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\ 129 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\ 130 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\ 133 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){\ 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);\ 138 src2 += 8*src2Stride;\ 139 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\ 140 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\ 143 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){\ 144 ff_put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\ 145 ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\ 147 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){\ 148 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 8);\ 151 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){\ 152 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 16);\ 155 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)\ 157 ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst , src16 , src8 , dstStride, src8Stride, h);\ 158 ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\ 163 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\ 165 void ff_avg_h264_qpel16_h_lowpass_l2_ssse3(
uint8_t *dst,
const uint8_t *src,
const uint8_t *src2,
int dstStride,
int src2Stride);
166 void ff_put_h264_qpel16_h_lowpass_l2_ssse3(
uint8_t *dst,
const uint8_t *src,
const uint8_t *src2,
int dstStride,
int src2Stride);
169 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\ 170 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){\ 171 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\ 172 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\ 175 src2 += 8*src2Stride;\ 176 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\ 177 ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\ 179 #endif // ARCH_X86_64 181 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\ 182 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\ 183 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\ 184 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\ 185 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\ 188 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\ 189 ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\ 192 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\ 193 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\ 194 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\ 196 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\ 197 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\ 198 ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\ 201 static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp,
208 src -= 2*srcStride+2;
210 ff_put_h264_qpel8or16_hv1_lowpass_op_sse2(src, tmp, srcStride, size);
216 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\ 217 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){\ 218 put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\ 219 ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\ 221 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){\ 222 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\ 224 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){\ 225 ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\ 228 #define ff_put_h264_qpel8_h_lowpass_l2_sse2 ff_put_h264_qpel8_h_lowpass_l2_mmxext 229 #define ff_avg_h264_qpel8_h_lowpass_l2_sse2 ff_avg_h264_qpel8_h_lowpass_l2_mmxext 230 #define ff_put_h264_qpel16_h_lowpass_l2_sse2 ff_put_h264_qpel16_h_lowpass_l2_mmxext 231 #define ff_avg_h264_qpel16_h_lowpass_l2_sse2 ff_avg_h264_qpel16_h_lowpass_l2_mmxext 233 #define ff_put_h264_qpel8_v_lowpass_ssse3 ff_put_h264_qpel8_v_lowpass_sse2 234 #define ff_avg_h264_qpel8_v_lowpass_ssse3 ff_avg_h264_qpel8_v_lowpass_sse2 235 #define ff_put_h264_qpel16_v_lowpass_ssse3 ff_put_h264_qpel16_v_lowpass_sse2 236 #define ff_avg_h264_qpel16_v_lowpass_ssse3 ff_avg_h264_qpel16_v_lowpass_sse2 238 #define ff_put_h264_qpel8or16_hv2_lowpass_sse2 ff_put_h264_qpel8or16_hv2_lowpass_mmxext 239 #define ff_avg_h264_qpel8or16_hv2_lowpass_sse2 ff_avg_h264_qpel8or16_hv2_lowpass_mmxext 241 #define H264_MC(OPNAME, SIZE, MMX, ALIGN) \ 242 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\ 243 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\ 244 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\ 245 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\ 247 static void put_h264_qpel16_mc00_sse2 (
uint8_t *dst,
const uint8_t *src,
252 static void avg_h264_qpel16_mc00_sse2 (
uint8_t *dst,
const uint8_t *src,
257 #define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmxext 258 #define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmxext 260 #define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \ 261 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 263 ff_ ## OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\ 266 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \ 267 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 269 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\ 272 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t 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, const uint8_t *src, ptrdiff_t stride)\ 279 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\ 282 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \ 283 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 285 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\ 286 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\ 287 ff_ ## OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\ 290 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 292 ff_ ## OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\ 295 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 297 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\ 298 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\ 299 ff_ ## OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\ 302 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \ 303 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 305 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\ 306 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\ 307 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\ 310 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 312 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\ 313 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\ 314 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\ 317 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 319 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\ 320 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\ 321 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\ 324 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 326 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\ 327 ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\ 328 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\ 331 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 333 DECLARE_ALIGNED(ALIGN, uint16_t, temp)[SIZE*(SIZE<8?12:24)];\ 334 ff_ ## OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\ 337 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 339 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\ 340 uint8_t * const halfHV= temp;\ 341 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\ 342 assert(((int)temp & 7) == 0);\ 343 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\ 344 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\ 347 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 349 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\ 350 uint8_t * const halfHV= temp;\ 351 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\ 352 assert(((int)temp & 7) == 0);\ 353 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\ 354 ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\ 357 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 359 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\ 360 uint8_t * const halfHV= temp;\ 361 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\ 362 assert(((int)temp & 7) == 0);\ 363 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\ 364 ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+2, halfHV, stride, SIZE, SIZE);\ 367 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 369 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\ 370 uint8_t * const halfHV= temp;\ 371 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\ 372 assert(((int)temp & 7) == 0);\ 373 ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\ 374 ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+3, halfHV, stride, SIZE, SIZE);\ 377 #define H264_MC_4816(MMX)\ 378 H264_MC(put_, 4, MMX, 8)\ 379 H264_MC(put_, 8, MMX, 8)\ 380 H264_MC(put_, 16,MMX, 8)\ 381 H264_MC(avg_, 4, MMX, 8)\ 382 H264_MC(avg_, 8, MMX, 8)\ 383 H264_MC(avg_, 16,MMX, 8)\ 385 #define H264_MC_816(QPEL, XMM)\ 386 QPEL(put_, 8, XMM, 16)\ 387 QPEL(put_, 16,XMM, 16)\ 388 QPEL(avg_, 8, XMM, 16)\ 389 QPEL(avg_, 16,XMM, 16)\ 391 QPEL_H264(put_, PUT_OP, mmxext)
392 QPEL_H264(avg_, AVG_MMXEXT_OP, mmxext)
393 QPEL_H264_V_XMM(put_, PUT_OP, sse2)
394 QPEL_H264_V_XMM(avg_,AVG_MMXEXT_OP, sse2)
395 QPEL_H264_HV_XMM(put_, PUT_OP, sse2)
396 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, sse2)
397 QPEL_H264_H_XMM(put_, PUT_OP, ssse3)
398 QPEL_H264_H_XMM(avg_,AVG_MMXEXT_OP, ssse3)
399 QPEL_H264_HV_XMM(put_, PUT_OP, ssse3)
400 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, ssse3)
403 H264_MC_816(H264_MC_V, sse2)
404 H264_MC_816(H264_MC_HV, sse2)
405 H264_MC_816(H264_MC_H, ssse3)
406 H264_MC_816(H264_MC_HV, ssse3)
410 #define LUMA_MC_OP(OP, NUM, DEPTH, TYPE, OPT) \ 411 void ff_ ## OP ## _h264_qpel ## NUM ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT \ 412 (uint8_t *dst, const uint8_t *src, ptrdiff_t stride); 414 #define LUMA_MC_ALL(DEPTH, TYPE, OPT) \ 415 LUMA_MC_OP(put, 4, DEPTH, TYPE, OPT) \ 416 LUMA_MC_OP(avg, 4, DEPTH, TYPE, OPT) \ 417 LUMA_MC_OP(put, 8, DEPTH, TYPE, OPT) \ 418 LUMA_MC_OP(avg, 8, DEPTH, TYPE, OPT) \ 419 LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \ 420 LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT) 422 #define LUMA_MC_816(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 LUMA_MC_ALL(10, mc00, mmxext)
429 LUMA_MC_ALL(10, mc10, mmxext)
430 LUMA_MC_ALL(10, mc20, mmxext)
431 LUMA_MC_ALL(10, mc30, mmxext)
432 LUMA_MC_ALL(10, mc01, mmxext)
433 LUMA_MC_ALL(10, mc11, mmxext)
434 LUMA_MC_ALL(10, mc21, mmxext)
435 LUMA_MC_ALL(10, mc31, mmxext)
436 LUMA_MC_ALL(10, mc02, mmxext)
437 LUMA_MC_ALL(10, mc12, mmxext)
438 LUMA_MC_ALL(10, mc22, mmxext)
439 LUMA_MC_ALL(10, mc32, mmxext)
440 LUMA_MC_ALL(10, mc03, mmxext)
441 LUMA_MC_ALL(10, mc13, mmxext)
442 LUMA_MC_ALL(10, mc23, mmxext)
443 LUMA_MC_ALL(10, mc33, mmxext)
445 LUMA_MC_816(10, mc00, sse2)
446 LUMA_MC_816(10, mc10, sse2)
447 LUMA_MC_816(10, mc10, sse2_cache64)
448 LUMA_MC_816(10, mc10, ssse3_cache64)
449 LUMA_MC_816(10, mc20, sse2)
450 LUMA_MC_816(10, mc20, sse2_cache64)
451 LUMA_MC_816(10, mc20, ssse3_cache64)
452 LUMA_MC_816(10, mc30, sse2)
453 LUMA_MC_816(10, mc30, sse2_cache64)
454 LUMA_MC_816(10, mc30, ssse3_cache64)
455 LUMA_MC_816(10, mc01, sse2)
456 LUMA_MC_816(10, mc11, sse2)
457 LUMA_MC_816(10, mc21, sse2)
458 LUMA_MC_816(10, mc31, sse2)
459 LUMA_MC_816(10, mc02, sse2)
460 LUMA_MC_816(10, mc12, sse2)
461 LUMA_MC_816(10, mc22, sse2)
462 LUMA_MC_816(10, mc32, sse2)
463 LUMA_MC_816(10, mc03, sse2)
464 LUMA_MC_816(10, mc13, sse2)
465 LUMA_MC_816(10, mc23, sse2)
466 LUMA_MC_816(10, mc33, sse2)
468 #define QPEL16_OPMC(OP, MC, MMX)\ 469 void ff_ ## OP ## _h264_qpel16_ ## MC ## _10_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride){\ 470 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst , src , stride);\ 471 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\ 474 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst , src , stride);\ 475 ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\ 478 #define QPEL16_OP(MC, MMX)\ 479 QPEL16_OPMC(put, MC, MMX)\ 480 QPEL16_OPMC(avg, MC, MMX) 483 QPEL16_OP(mc00, MMX)\ 484 QPEL16_OP(mc01, MMX)\ 485 QPEL16_OP(mc02, MMX)\ 486 QPEL16_OP(mc03, MMX)\ 487 QPEL16_OP(mc10, MMX)\ 488 QPEL16_OP(mc11, MMX)\ 489 QPEL16_OP(mc12, MMX)\ 490 QPEL16_OP(mc13, MMX)\ 491 QPEL16_OP(mc20, MMX)\ 492 QPEL16_OP(mc21, MMX)\ 493 QPEL16_OP(mc22, MMX)\ 494 QPEL16_OP(mc23, MMX)\ 495 QPEL16_OP(mc30, MMX)\ 496 QPEL16_OP(mc31, MMX)\ 497 QPEL16_OP(mc32, MMX)\ 500 #if ARCH_X86_32 // ARCH_X86_64 implies SSE2+ 506 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX) \ 508 c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \ 509 c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \ 510 c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \ 511 c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \ 512 c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \ 513 c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \ 514 c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \ 515 c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \ 516 c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \ 517 c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \ 518 c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \ 519 c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \ 520 c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \ 521 c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \ 522 c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \ 523 c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \ 526 #define H264_QPEL_FUNCS(x, y, CPU) \ 528 c->put_h264_qpel_pixels_tab[0][x + y * 4] = put_h264_qpel16_mc ## x ## y ## _ ## CPU; \ 529 c->put_h264_qpel_pixels_tab[1][x + y * 4] = put_h264_qpel8_mc ## x ## y ## _ ## CPU; \ 530 c->avg_h264_qpel_pixels_tab[0][x + y * 4] = avg_h264_qpel16_mc ## x ## y ## _ ## CPU; \ 531 c->avg_h264_qpel_pixels_tab[1][x + y * 4] = avg_h264_qpel8_mc ## x ## y ## _ ## CPU; \ 534 #define H264_QPEL_FUNCS_10(x, y, CPU) \ 536 c->put_h264_qpel_pixels_tab[0][x + y * 4] = ff_put_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \ 537 c->put_h264_qpel_pixels_tab[1][x + y * 4] = ff_put_h264_qpel8_mc ## x ## y ## _10_ ## CPU; \ 538 c->avg_h264_qpel_pixels_tab[0][x + y * 4] = ff_avg_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \ 539 c->avg_h264_qpel_pixels_tab[1][x + y * 4] = ff_avg_h264_qpel8_mc ## x ## y ## _10_ ## CPU; \ 545 int high_bit_depth = bit_depth > 8;
549 if (!high_bit_depth) {
556 }
else if (bit_depth == 10) {
574 if (!high_bit_depth) {
589 if (bit_depth == 10) {
601 if (!high_bit_depth) {
616 if (bit_depth == 10) {
628 if (bit_depth == 10) {
void ff_avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
Macro definitions for various function/variable attributes.
#define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX)
#define AV_CPU_FLAG_SSE2SLOW
SSE2 supported, but usually not faster.
#define CALL_2X_PIXELS(a, b, n)
#define EXTERNAL_SSE2(flags)
void ff_put_pixels8_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dstStride, int src1Stride, int h)
void ff_avg_pixels8_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dstStride, int src1Stride, int h)
av_cold void ff_h264qpel_init_x86(H264QpelContext *c, int bit_depth)
void ff_put_pixels16_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dstStride, int src1Stride, int h)
#define EXTERNAL_SSSE3(flags)
#define H264_QPEL_FUNCS_10(x, y, CPU)
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
void ff_avg_pixels8_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
#define EXTERNAL_MMXEXT(flags)
void ff_put_pixels8_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
#define H264_QPEL_FUNCS(x, y, CPU)
#define EXTERNAL_AVX(flags)
void ff_avg_pixels16_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dstStride, int src1Stride, int h)