FFmpeg: libavcodec/x86/fdct

00001 /*
00002  * MMX optimized forward DCT
00003  * The gcc porting is Copyright (c) 2001 Fabrice Bellard.
00004  * cleanup/optimizations are Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
00005  * SSE2 optimization is Copyright (c) 2004 Denes Balatoni.
00006  *
00007  * from  fdctam32.c - AP922 MMX(3D-Now) forward-DCT
00008  *
00009  *  Intel Application Note AP-922 - fast, precise implementation of DCT
00010  *        http://developer.intel.com/vtune/cbts/appnotes.htm
00011  *
00012  * Also of inspiration:
00013  * a page about fdct at http://www.geocities.com/ssavekar/dct.htm
00014  * Skal's fdct at http://skal.planet-d.net/coding/dct.html
00015  *
00016  * This file is part of FFmpeg.
00017  *
00018  * FFmpeg is free software; you can redistribute it and/or
00019  * modify it under the terms of the GNU Lesser General Public
00020  * License as published by the Free Software Foundation; either
00021  * version 2.1 of the License, or (at your option) any later version.
00022  *
00023  * FFmpeg is distributed in the hope that it will be useful,
00024  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00025  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00026  * Lesser General Public License for more details.
00027  *
00028  * You should have received a copy of the GNU Lesser General Public
00029  * License along with FFmpeg; if not, write to the Free Software
00030  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00031  */
00032 
00033 #include "libavutil/common.h"
00034 #include "libavcodec/dsputil.h"
00035 
00036 #define ATTR_ALIGN(align) __attribute__ ((__aligned__ (align)))
00037 
00039 //
00040 // constants for the forward DCT
00041 // -----------------------------
00042 //
00043 // Be sure to check that your compiler is aligning all constants to QWORD
00044 // (8-byte) memory boundaries!  Otherwise the unaligned memory access will
00045 // severely stall MMX execution.
00046 //
00048 
00049 #define BITS_FRW_ACC   3 //; 2 or 3 for accuracy
00050 #define SHIFT_FRW_COL  BITS_FRW_ACC
00051 #define SHIFT_FRW_ROW  (BITS_FRW_ACC + 17 - 3)
00052 #define RND_FRW_ROW    (1 << (SHIFT_FRW_ROW-1))
00053 //#define RND_FRW_COL    (1 << (SHIFT_FRW_COL-1))
00054 
00055 #define X8(x) x,x,x,x,x,x,x,x
00056 
00057 //concatenated table, for forward DCT transformation
00058 static const int16_t fdct_tg_all_16[24] ATTR_ALIGN(16) = {
00059     X8(13036),  // tg * (2<<16) + 0.5
00060     X8(27146),  // tg * (2<<16) + 0.5
00061     X8(-21746)  // tg * (2<<16) + 0.5
00062 };
00063 
00064 static const int16_t ocos_4_16[8] ATTR_ALIGN(16) = {
00065     X8(23170)   //cos * (2<<15) + 0.5
00066 };
00067 
00068 static const int16_t fdct_one_corr[8] ATTR_ALIGN(16) = { X8(1) };
00069 
00070 static const int32_t fdct_r_row[2] ATTR_ALIGN(8) = {RND_FRW_ROW, RND_FRW_ROW };
00071 
00072 static struct
00073 {
00074  const int32_t fdct_r_row_sse2[4] ATTR_ALIGN(16);
00075 } fdct_r_row_sse2 ATTR_ALIGN(16)=
00076 {{
00077  RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW
00078 }};
00079 //static const long fdct_r_row_sse2[4] ATTR_ALIGN(16) = {RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW};
00080 
00081 static const int16_t tab_frw_01234567[] ATTR_ALIGN(8) = {  // forward_dct coeff table
00082   16384,   16384,   22725,   19266,
00083   16384,   16384,   12873,    4520,
00084   21407,    8867,   19266,   -4520,
00085   -8867,  -21407,  -22725,  -12873,
00086   16384,  -16384,   12873,  -22725,
00087  -16384,   16384,    4520,   19266,
00088    8867,  -21407,    4520,  -12873,
00089   21407,   -8867,   19266,  -22725,
00090 
00091   22725,   22725,   31521,   26722,
00092   22725,   22725,   17855,    6270,
00093   29692,   12299,   26722,   -6270,
00094  -12299,  -29692,  -31521,  -17855,
00095   22725,  -22725,   17855,  -31521,
00096  -22725,   22725,    6270,   26722,
00097   12299,  -29692,    6270,  -17855,
00098   29692,  -12299,   26722,  -31521,
00099 
00100   21407,   21407,   29692,   25172,
00101   21407,   21407,   16819,    5906,
00102   27969,   11585,   25172,   -5906,
00103  -11585,  -27969,  -29692,  -16819,
00104   21407,  -21407,   16819,  -29692,
00105  -21407,   21407,    5906,   25172,
00106   11585,  -27969,    5906,  -16819,
00107   27969,  -11585,   25172,  -29692,
00108 
00109   19266,   19266,   26722,   22654,
00110   19266,   19266,   15137,    5315,
00111   25172,   10426,   22654,   -5315,
00112  -10426,  -25172,  -26722,  -15137,
00113   19266,  -19266,   15137,  -26722,
00114  -19266,   19266,    5315,   22654,
00115   10426,  -25172,    5315,  -15137,
00116   25172,  -10426,   22654,  -26722,
00117 
00118   16384,   16384,   22725,   19266,
00119   16384,   16384,   12873,    4520,
00120   21407,    8867,   19266,   -4520,
00121   -8867,  -21407,  -22725,  -12873,
00122   16384,  -16384,   12873,  -22725,
00123  -16384,   16384,    4520,   19266,
00124    8867,  -21407,    4520,  -12873,
00125   21407,   -8867,   19266,  -22725,
00126 
00127   19266,   19266,   26722,   22654,
00128   19266,   19266,   15137,    5315,
00129   25172,   10426,   22654,   -5315,
00130  -10426,  -25172,  -26722,  -15137,
00131   19266,  -19266,   15137,  -26722,
00132  -19266,   19266,    5315,   22654,
00133   10426,  -25172,    5315,  -15137,
00134   25172,  -10426,   22654,  -26722,
00135 
00136   21407,   21407,   29692,   25172,
00137   21407,   21407,   16819,    5906,
00138   27969,   11585,   25172,   -5906,
00139  -11585,  -27969,  -29692,  -16819,
00140   21407,  -21407,   16819,  -29692,
00141  -21407,   21407,    5906,   25172,
00142   11585,  -27969,    5906,  -16819,
00143   27969,  -11585,   25172,  -29692,
00144 
00145   22725,   22725,   31521,   26722,
00146   22725,   22725,   17855,    6270,
00147   29692,   12299,   26722,   -6270,
00148  -12299,  -29692,  -31521,  -17855,
00149   22725,  -22725,   17855,  -31521,
00150  -22725,   22725,    6270,   26722,
00151   12299,  -29692,    6270,  -17855,
00152   29692,  -12299,   26722,  -31521,
00153 };
00154 
00155 static struct
00156 {
00157  const int16_t tab_frw_01234567_sse2[256] ATTR_ALIGN(16);
00158 } tab_frw_01234567_sse2 ATTR_ALIGN(16) =
00159 {{
00160 //static const int16_t tab_frw_01234567_sse2[] ATTR_ALIGN(16) = {  // forward_dct coeff table
00161 #define TABLE_SSE2 C4,  C4,  C1,  C3, -C6, -C2, -C1, -C5, \
00162                    C4,  C4,  C5,  C7,  C2,  C6,  C3, -C7, \
00163                   -C4,  C4,  C7,  C3,  C6, -C2,  C7, -C5, \
00164                    C4, -C4,  C5, -C1,  C2, -C6,  C3, -C1,
00165 // c1..c7 * cos(pi/4) * 2^15
00166 #define C1 22725
00167 #define C2 21407
00168 #define C3 19266
00169 #define C4 16384
00170 #define C5 12873
00171 #define C6 8867
00172 #define C7 4520
00173 TABLE_SSE2
00174 
00175 #undef C1
00176 #undef C2
00177 #undef C3
00178 #undef C4
00179 #undef C5
00180 #undef C6
00181 #undef C7
00182 #define C1 31521
00183 #define C2 29692
00184 #define C3 26722
00185 #define C4 22725
00186 #define C5 17855
00187 #define C6 12299
00188 #define C7 6270
00189 TABLE_SSE2
00190 
00191 #undef C1
00192 #undef C2
00193 #undef C3
00194 #undef C4
00195 #undef C5
00196 #undef C6
00197 #undef C7
00198 #define C1 29692
00199 #define C2 27969
00200 #define C3 25172
00201 #define C4 21407
00202 #define C5 16819
00203 #define C6 11585
00204 #define C7 5906
00205 TABLE_SSE2
00206 
00207 #undef C1
00208 #undef C2
00209 #undef C3
00210 #undef C4
00211 #undef C5
00212 #undef C6
00213 #undef C7
00214 #define C1 26722
00215 #define C2 25172
00216 #define C3 22654
00217 #define C4 19266
00218 #define C5 15137
00219 #define C6 10426
00220 #define C7 5315
00221 TABLE_SSE2
00222 
00223 #undef C1
00224 #undef C2
00225 #undef C3
00226 #undef C4
00227 #undef C5
00228 #undef C6
00229 #undef C7
00230 #define C1 22725
00231 #define C2 21407
00232 #define C3 19266
00233 #define C4 16384
00234 #define C5 12873
00235 #define C6 8867
00236 #define C7 4520
00237 TABLE_SSE2
00238 
00239 #undef C1
00240 #undef C2
00241 #undef C3
00242 #undef C4
00243 #undef C5
00244 #undef C6
00245 #undef C7
00246 #define C1 26722
00247 #define C2 25172
00248 #define C3 22654
00249 #define C4 19266
00250 #define C5 15137
00251 #define C6 10426
00252 #define C7 5315
00253 TABLE_SSE2
00254 
00255 #undef C1
00256 #undef C2
00257 #undef C3
00258 #undef C4
00259 #undef C5
00260 #undef C6
00261 #undef C7
00262 #define C1 29692
00263 #define C2 27969
00264 #define C3 25172
00265 #define C4 21407
00266 #define C5 16819
00267 #define C6 11585
00268 #define C7 5906
00269 TABLE_SSE2
00270 
00271 #undef C1
00272 #undef C2
00273 #undef C3
00274 #undef C4
00275 #undef C5
00276 #undef C6
00277 #undef C7
00278 #define C1 31521
00279 #define C2 29692
00280 #define C3 26722
00281 #define C4 22725
00282 #define C5 17855
00283 #define C6 12299
00284 #define C7 6270
00285 TABLE_SSE2
00286 }};
00287 
00288 #define S(s) AV_TOSTRING(s) //AV_STRINGIFY is too long
00289 
00290 #define FDCT_COL(cpu, mm, mov)\
00291 static av_always_inline void fdct_col_##cpu(const int16_t *in, int16_t *out, int offset)\
00292 {\
00293     __asm__ volatile (\
00294         #mov"      16(%0),  %%"#mm"0 \n\t" \
00295         #mov"      96(%0),  %%"#mm"1 \n\t" \
00296         #mov"    %%"#mm"0,  %%"#mm"2 \n\t" \
00297         #mov"      32(%0),  %%"#mm"3 \n\t" \
00298         "paddsw  %%"#mm"1,  %%"#mm"0 \n\t" \
00299         #mov"      80(%0),  %%"#mm"4 \n\t" \
00300         "psllw  $"S(SHIFT_FRW_COL)", %%"#mm"0 \n\t" \
00301         #mov"        (%0),  %%"#mm"5 \n\t" \
00302         "paddsw  %%"#mm"3,  %%"#mm"4 \n\t" \
00303         "paddsw   112(%0),  %%"#mm"5 \n\t" \
00304         "psllw  $"S(SHIFT_FRW_COL)", %%"#mm"4 \n\t" \
00305         #mov"    %%"#mm"0,  %%"#mm"6 \n\t" \
00306         "psubsw  %%"#mm"1,  %%"#mm"2 \n\t" \
00307         #mov"      16(%1),  %%"#mm"1 \n\t" \
00308         "psubsw  %%"#mm"4,  %%"#mm"0 \n\t" \
00309         #mov"      48(%0),  %%"#mm"7 \n\t" \
00310         "pmulhw  %%"#mm"0,  %%"#mm"1 \n\t" \
00311         "paddsw    64(%0),  %%"#mm"7 \n\t" \
00312         "psllw  $"S(SHIFT_FRW_COL)", %%"#mm"5 \n\t" \
00313         "paddsw  %%"#mm"4,  %%"#mm"6 \n\t" \
00314         "psllw  $"S(SHIFT_FRW_COL)", %%"#mm"7 \n\t" \
00315         #mov"    %%"#mm"5,  %%"#mm"4 \n\t" \
00316         "psubsw  %%"#mm"7,  %%"#mm"5 \n\t" \
00317         "paddsw  %%"#mm"5,  %%"#mm"1 \n\t" \
00318         "paddsw  %%"#mm"7,  %%"#mm"4 \n\t" \
00319         "por         (%2),  %%"#mm"1 \n\t" \
00320         "psllw  $"S(SHIFT_FRW_COL)"+1, %%"#mm"2 \n\t" \
00321         "pmulhw    16(%1),  %%"#mm"5 \n\t" \
00322         #mov"    %%"#mm"4,  %%"#mm"7 \n\t" \
00323         "psubsw    80(%0),  %%"#mm"3 \n\t" \
00324         "psubsw  %%"#mm"6,  %%"#mm"4 \n\t" \
00325         #mov"    %%"#mm"1,    32(%3) \n\t" \
00326         "paddsw  %%"#mm"6,  %%"#mm"7 \n\t" \
00327         #mov"      48(%0),  %%"#mm"1 \n\t" \
00328         "psllw  $"S(SHIFT_FRW_COL)"+1, %%"#mm"3 \n\t" \
00329         "psubsw    64(%0),  %%"#mm"1 \n\t" \
00330         #mov"    %%"#mm"2,  %%"#mm"6 \n\t" \
00331         #mov"    %%"#mm"4,    64(%3) \n\t" \
00332         "paddsw  %%"#mm"3,  %%"#mm"2 \n\t" \
00333         "pmulhw      (%4),  %%"#mm"2 \n\t" \
00334         "psubsw  %%"#mm"3,  %%"#mm"6 \n\t" \
00335         "pmulhw      (%4),  %%"#mm"6 \n\t" \
00336         "psubsw  %%"#mm"0,  %%"#mm"5 \n\t" \
00337         "por         (%2),  %%"#mm"5 \n\t" \
00338         "psllw  $"S(SHIFT_FRW_COL)", %%"#mm"1 \n\t" \
00339         "por         (%2),  %%"#mm"2 \n\t" \
00340         #mov"    %%"#mm"1,  %%"#mm"4 \n\t" \
00341         #mov"        (%0),  %%"#mm"3 \n\t" \
00342         "paddsw  %%"#mm"6,  %%"#mm"1 \n\t" \
00343         "psubsw   112(%0),  %%"#mm"3 \n\t" \
00344         "psubsw  %%"#mm"6,  %%"#mm"4 \n\t" \
00345         #mov"        (%1),  %%"#mm"0 \n\t" \
00346         "psllw  $"S(SHIFT_FRW_COL)", %%"#mm"3 \n\t" \
00347         #mov"      32(%1),  %%"#mm"6 \n\t" \
00348         "pmulhw  %%"#mm"1,  %%"#mm"0 \n\t" \
00349         #mov"    %%"#mm"7,      (%3) \n\t" \
00350         "pmulhw  %%"#mm"4,  %%"#mm"6 \n\t" \
00351         #mov"    %%"#mm"5,    96(%3) \n\t" \
00352         #mov"    %%"#mm"3,  %%"#mm"7 \n\t" \
00353         #mov"      32(%1),  %%"#mm"5 \n\t" \
00354         "psubsw  %%"#mm"2,  %%"#mm"7 \n\t" \
00355         "paddsw  %%"#mm"2,  %%"#mm"3 \n\t" \
00356         "pmulhw  %%"#mm"7,  %%"#mm"5 \n\t" \
00357         "paddsw  %%"#mm"3,  %%"#mm"0 \n\t" \
00358         "paddsw  %%"#mm"4,  %%"#mm"6 \n\t" \
00359         "pmulhw      (%1),  %%"#mm"3 \n\t" \
00360         "por         (%2),  %%"#mm"0 \n\t" \
00361         "paddsw  %%"#mm"7,  %%"#mm"5 \n\t" \
00362         "psubsw  %%"#mm"6,  %%"#mm"7 \n\t" \
00363         #mov"    %%"#mm"0,    16(%3) \n\t" \
00364         "paddsw  %%"#mm"4,  %%"#mm"5 \n\t" \
00365         #mov"    %%"#mm"7,    48(%3) \n\t" \
00366         "psubsw  %%"#mm"1,  %%"#mm"3 \n\t" \
00367         #mov"    %%"#mm"5,    80(%3) \n\t" \
00368         #mov"    %%"#mm"3,   112(%3) \n\t" \
00369         : \
00370         : "r" (in  + offset), "r" (fdct_tg_all_16), "r" (fdct_one_corr), \
00371           "r" (out + offset), "r" (ocos_4_16)); \
00372 }
00373 
00374 FDCT_COL(mmx, mm, movq)
00375 FDCT_COL(sse2, xmm, movdqa)
00376 
00377 static av_always_inline void fdct_row_sse2(const int16_t *in, int16_t *out)
00378 {
00379     __asm__ volatile(
00380 #define FDCT_ROW_SSE2_H1(i,t)                    \
00381         "movq      " #i "(%0), %%xmm2      \n\t" \
00382         "movq      " #i "+8(%0), %%xmm0    \n\t" \
00383         "movdqa    " #t "+32(%1), %%xmm3   \n\t" \
00384         "movdqa    " #t "+48(%1), %%xmm7   \n\t" \
00385         "movdqa    " #t "(%1), %%xmm4      \n\t" \
00386         "movdqa    " #t "+16(%1), %%xmm5   \n\t"
00387 
00388 #define FDCT_ROW_SSE2_H2(i,t)                    \
00389         "movq      " #i "(%0), %%xmm2      \n\t" \
00390         "movq      " #i "+8(%0), %%xmm0    \n\t" \
00391         "movdqa    " #t "+32(%1), %%xmm3   \n\t" \
00392         "movdqa    " #t "+48(%1), %%xmm7   \n\t"
00393 
00394 #define FDCT_ROW_SSE2(i)                      \
00395         "movq      %%xmm2, %%xmm1       \n\t" \
00396         "pshuflw   $27, %%xmm0, %%xmm0  \n\t" \
00397         "paddsw    %%xmm0, %%xmm1       \n\t" \
00398         "psubsw    %%xmm0, %%xmm2       \n\t" \
00399         "punpckldq %%xmm2, %%xmm1       \n\t" \
00400         "pshufd    $78, %%xmm1, %%xmm2  \n\t" \
00401         "pmaddwd   %%xmm2, %%xmm3       \n\t" \
00402         "pmaddwd   %%xmm1, %%xmm7       \n\t" \
00403         "pmaddwd   %%xmm5, %%xmm2       \n\t" \
00404         "pmaddwd   %%xmm4, %%xmm1       \n\t" \
00405         "paddd     %%xmm7, %%xmm3       \n\t" \
00406         "paddd     %%xmm2, %%xmm1       \n\t" \
00407         "paddd     %%xmm6, %%xmm3       \n\t" \
00408         "paddd     %%xmm6, %%xmm1       \n\t" \
00409         "psrad     %3, %%xmm3           \n\t" \
00410         "psrad     %3, %%xmm1           \n\t" \
00411         "packssdw  %%xmm3, %%xmm1       \n\t" \
00412         "movdqa    %%xmm1, " #i "(%4)   \n\t"
00413 
00414         "movdqa    (%2), %%xmm6         \n\t"
00415         FDCT_ROW_SSE2_H1(0,0)
00416         FDCT_ROW_SSE2(0)
00417         FDCT_ROW_SSE2_H2(64,0)
00418         FDCT_ROW_SSE2(64)
00419 
00420         FDCT_ROW_SSE2_H1(16,64)
00421         FDCT_ROW_SSE2(16)
00422         FDCT_ROW_SSE2_H2(112,64)
00423         FDCT_ROW_SSE2(112)
00424 
00425         FDCT_ROW_SSE2_H1(32,128)
00426         FDCT_ROW_SSE2(32)
00427         FDCT_ROW_SSE2_H2(96,128)
00428         FDCT_ROW_SSE2(96)
00429 
00430         FDCT_ROW_SSE2_H1(48,192)
00431         FDCT_ROW_SSE2(48)
00432         FDCT_ROW_SSE2_H2(80,192)
00433         FDCT_ROW_SSE2(80)
00434         :
00435         : "r" (in), "r" (tab_frw_01234567_sse2.tab_frw_01234567_sse2), "r" (fdct_r_row_sse2.fdct_r_row_sse2), "i" (SHIFT_FRW_ROW), "r" (out)
00436     );
00437 }
00438 
00439 static av_always_inline void fdct_row_mmx2(const int16_t *in, int16_t *out, const int16_t *table)
00440 {
00441     __asm__ volatile (
00442         "pshufw    $0x1B, 8(%0), %%mm5 \n\t"
00443         "movq       (%0), %%mm0 \n\t"
00444         "movq      %%mm0, %%mm1 \n\t"
00445         "paddsw    %%mm5, %%mm0 \n\t"
00446         "psubsw    %%mm5, %%mm1 \n\t"
00447         "movq      %%mm0, %%mm2 \n\t"
00448         "punpckldq %%mm1, %%mm0 \n\t"
00449         "punpckhdq %%mm1, %%mm2 \n\t"
00450         "movq       (%1), %%mm1 \n\t"
00451         "movq      8(%1), %%mm3 \n\t"
00452         "movq     16(%1), %%mm4 \n\t"
00453         "movq     24(%1), %%mm5 \n\t"
00454         "movq     32(%1), %%mm6 \n\t"
00455         "movq     40(%1), %%mm7 \n\t"
00456         "pmaddwd   %%mm0, %%mm1 \n\t"
00457         "pmaddwd   %%mm2, %%mm3 \n\t"
00458         "pmaddwd   %%mm0, %%mm4 \n\t"
00459         "pmaddwd   %%mm2, %%mm5 \n\t"
00460         "pmaddwd   %%mm0, %%mm6 \n\t"
00461         "pmaddwd   %%mm2, %%mm7 \n\t"
00462         "pmaddwd  48(%1), %%mm0 \n\t"
00463         "pmaddwd  56(%1), %%mm2 \n\t"
00464         "paddd     %%mm1, %%mm3 \n\t"
00465         "paddd     %%mm4, %%mm5 \n\t"
00466         "paddd     %%mm6, %%mm7 \n\t"
00467         "paddd     %%mm0, %%mm2 \n\t"
00468         "movq       (%2), %%mm0 \n\t"
00469         "paddd     %%mm0, %%mm3 \n\t"
00470         "paddd     %%mm0, %%mm5 \n\t"
00471         "paddd     %%mm0, %%mm7 \n\t"
00472         "paddd     %%mm0, %%mm2 \n\t"
00473         "psrad $"S(SHIFT_FRW_ROW)", %%mm3 \n\t"
00474         "psrad $"S(SHIFT_FRW_ROW)", %%mm5 \n\t"
00475         "psrad $"S(SHIFT_FRW_ROW)", %%mm7 \n\t"
00476         "psrad $"S(SHIFT_FRW_ROW)", %%mm2 \n\t"
00477         "packssdw  %%mm5, %%mm3 \n\t"
00478         "packssdw  %%mm2, %%mm7 \n\t"
00479         "movq      %%mm3,  (%3) \n\t"
00480         "movq      %%mm7, 8(%3) \n\t"
00481         :
00482         : "r" (in), "r" (table), "r" (fdct_r_row), "r" (out));
00483 }
00484 
00485 static av_always_inline void fdct_row_mmx(const int16_t *in, int16_t *out, const int16_t *table)
00486 {
00487     //FIXME reorder (I do not have an old MMX-only CPU here to benchmark ...)
00488     __asm__ volatile(
00489         "movd     12(%0), %%mm1 \n\t"
00490         "punpcklwd 8(%0), %%mm1 \n\t"
00491         "movq      %%mm1, %%mm2 \n\t"
00492         "psrlq     $0x20, %%mm1 \n\t"
00493         "movq      0(%0), %%mm0 \n\t"
00494         "punpcklwd %%mm2, %%mm1 \n\t"
00495         "movq      %%mm0, %%mm5 \n\t"
00496         "paddsw    %%mm1, %%mm0 \n\t"
00497         "psubsw    %%mm1, %%mm5 \n\t"
00498         "movq      %%mm0, %%mm2 \n\t"
00499         "punpckldq %%mm5, %%mm0 \n\t"
00500         "punpckhdq %%mm5, %%mm2 \n\t"
00501         "movq      0(%1), %%mm1 \n\t"
00502         "movq      8(%1), %%mm3 \n\t"
00503         "movq     16(%1), %%mm4 \n\t"
00504         "movq     24(%1), %%mm5 \n\t"
00505         "movq     32(%1), %%mm6 \n\t"
00506         "movq     40(%1), %%mm7 \n\t"
00507         "pmaddwd   %%mm0, %%mm1 \n\t"
00508         "pmaddwd   %%mm2, %%mm3 \n\t"
00509         "pmaddwd   %%mm0, %%mm4 \n\t"
00510         "pmaddwd   %%mm2, %%mm5 \n\t"
00511         "pmaddwd   %%mm0, %%mm6 \n\t"
00512         "pmaddwd   %%mm2, %%mm7 \n\t"
00513         "pmaddwd  48(%1), %%mm0 \n\t"
00514         "pmaddwd  56(%1), %%mm2 \n\t"
00515         "paddd     %%mm1, %%mm3 \n\t"
00516         "paddd     %%mm4, %%mm5 \n\t"
00517         "paddd     %%mm6, %%mm7 \n\t"
00518         "paddd     %%mm0, %%mm2 \n\t"
00519         "movq       (%2), %%mm0 \n\t"
00520         "paddd     %%mm0, %%mm3 \n\t"
00521         "paddd     %%mm0, %%mm5 \n\t"
00522         "paddd     %%mm0, %%mm7 \n\t"
00523         "paddd     %%mm0, %%mm2 \n\t"
00524         "psrad $"S(SHIFT_FRW_ROW)", %%mm3 \n\t"
00525         "psrad $"S(SHIFT_FRW_ROW)", %%mm5 \n\t"
00526         "psrad $"S(SHIFT_FRW_ROW)", %%mm7 \n\t"
00527         "psrad $"S(SHIFT_FRW_ROW)", %%mm2 \n\t"
00528         "packssdw  %%mm5, %%mm3 \n\t"
00529         "packssdw  %%mm2, %%mm7 \n\t"
00530         "movq      %%mm3, 0(%3) \n\t"
00531         "movq      %%mm7, 8(%3) \n\t"
00532         :
00533         : "r" (in), "r" (table), "r" (fdct_r_row), "r" (out));
00534 }
00535 
00536 void ff_fdct_mmx(int16_t *block)
00537 {
00538     int64_t align_tmp[16] ATTR_ALIGN(8);
00539     int16_t * block1= (int16_t*)align_tmp;
00540     const int16_t *table= tab_frw_01234567;
00541     int i;
00542 
00543     fdct_col_mmx(block, block1, 0);
00544     fdct_col_mmx(block, block1, 4);
00545 
00546     for(i=8;i>0;i--) {
00547         fdct_row_mmx(block1, block, table);
00548         block1 += 8;
00549         table += 32;
00550         block += 8;
00551     }
00552 }
00553 
00554 void ff_fdct_mmx2(int16_t *block)
00555 {
00556     int64_t align_tmp[16] ATTR_ALIGN(8);
00557     int16_t *block1= (int16_t*)align_tmp;
00558     const int16_t *table= tab_frw_01234567;
00559     int i;
00560 
00561     fdct_col_mmx(block, block1, 0);
00562     fdct_col_mmx(block, block1, 4);
00563 
00564     for(i=8;i>0;i--) {
00565         fdct_row_mmx2(block1, block, table);
00566         block1 += 8;
00567         table += 32;
00568         block += 8;
00569     }
00570 }
00571 
00572 void ff_fdct_sse2(int16_t *block)
00573 {
00574     int64_t align_tmp[16] ATTR_ALIGN(16);
00575     int16_t * const block1= (int16_t*)align_tmp;
00576 
00577     fdct_col_sse2(block, block1, 0);
00578     fdct_row_sse2(block1, block);
00579 }
00580
libavcodec/x86/fdct_mmx.c
