Cinelerra: hvirtual/quicktime/ffmpeg/libavcodec/i386/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 #include "../common.h"
00017 #include "../dsputil.h"
00018 #include "mmx.h"
00019 
00020 #define ATTR_ALIGN(align) __attribute__ ((__aligned__ (align)))
00021 
00023 //
00024 // constants for the forward DCT
00025 // -----------------------------
00026 //
00027 // Be sure to check that your compiler is aligning all constants to QWORD
00028 // (8-byte) memory boundaries!  Otherwise the unaligned memory access will
00029 // severely stall MMX execution.
00030 //
00032 
00033 #define BITS_FRW_ACC    3 //; 2 or 3 for accuracy
00034 #define SHIFT_FRW_COL   BITS_FRW_ACC
00035 #define SHIFT_FRW_ROW   (BITS_FRW_ACC + 17 - 3)
00036 #define RND_FRW_ROW             (1 << (SHIFT_FRW_ROW-1))
00037 //#define RND_FRW_COL           (1 << (SHIFT_FRW_COL-1))
00038 
00039 //concatenated table, for forward DCT transformation
00040 static const int16_t fdct_tg_all_16[] ATTR_ALIGN(8) = {
00041     13036, 13036, 13036, 13036,         // tg * (2<<16) + 0.5
00042     27146, 27146, 27146, 27146,         // tg * (2<<16) + 0.5
00043     -21746, -21746, -21746, -21746,     // tg * (2<<16) + 0.5
00044 };
00045 
00046 static const int16_t ocos_4_16[4] ATTR_ALIGN(8) = {
00047     23170, 23170, 23170, 23170, //cos * (2<<15) + 0.5
00048 };
00049 
00050 static const int64_t fdct_one_corr ATTR_ALIGN(8) = 0x0001000100010001LL;
00051 
00052 static const int32_t fdct_r_row[2] ATTR_ALIGN(8) = {RND_FRW_ROW, RND_FRW_ROW };
00053 
00054 struct 
00055 {
00056  const int32_t fdct_r_row_sse2[4] ATTR_ALIGN(16);
00057 } fdct_r_row_sse2 ATTR_ALIGN(16)=
00058 {{
00059  RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW
00060 }};
00061 //static const long fdct_r_row_sse2[4] ATTR_ALIGN(16) = {RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW};
00062 
00063 static const int16_t tab_frw_01234567[] ATTR_ALIGN(8) = {  // forward_dct coeff table
00064   16384,   16384,   22725,   19266, 
00065   16384,   16384,   12873,    4520, 
00066   21407,    8867,   19266,   -4520, 
00067   -8867,  -21407,  -22725,  -12873, 
00068   16384,  -16384,   12873,  -22725, 
00069  -16384,   16384,    4520,   19266, 
00070    8867,  -21407,    4520,  -12873, 
00071   21407,   -8867,   19266,  -22725, 
00072 
00073   22725,   22725,   31521,   26722, 
00074   22725,   22725,   17855,    6270, 
00075   29692,   12299,   26722,   -6270, 
00076  -12299,  -29692,  -31521,  -17855, 
00077   22725,  -22725,   17855,  -31521, 
00078  -22725,   22725,    6270,   26722, 
00079   12299,  -29692,    6270,  -17855, 
00080   29692,  -12299,   26722,  -31521, 
00081 
00082   21407,   21407,   29692,   25172, 
00083   21407,   21407,   16819,    5906, 
00084   27969,   11585,   25172,   -5906, 
00085  -11585,  -27969,  -29692,  -16819, 
00086   21407,  -21407,   16819,  -29692, 
00087  -21407,   21407,    5906,   25172, 
00088   11585,  -27969,    5906,  -16819, 
00089   27969,  -11585,   25172,  -29692, 
00090 
00091   19266,   19266,   26722,   22654, 
00092   19266,   19266,   15137,    5315, 
00093   25172,   10426,   22654,   -5315, 
00094  -10426,  -25172,  -26722,  -15137, 
00095   19266,  -19266,   15137,  -26722, 
00096  -19266,   19266,    5315,   22654, 
00097   10426,  -25172,    5315,  -15137, 
00098   25172,  -10426,   22654,  -26722, 
00099 
00100   16384,   16384,   22725,   19266, 
00101   16384,   16384,   12873,    4520, 
00102   21407,    8867,   19266,   -4520, 
00103   -8867,  -21407,  -22725,  -12873, 
00104   16384,  -16384,   12873,  -22725, 
00105  -16384,   16384,    4520,   19266, 
00106    8867,  -21407,    4520,  -12873, 
00107   21407,   -8867,   19266,  -22725, 
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   21407,   21407,   29692,   25172, 
00119   21407,   21407,   16819,    5906, 
00120   27969,   11585,   25172,   -5906, 
00121  -11585,  -27969,  -29692,  -16819, 
00122   21407,  -21407,   16819,  -29692, 
00123  -21407,   21407,    5906,   25172, 
00124   11585,  -27969,    5906,  -16819, 
00125   27969,  -11585,   25172,  -29692, 
00126 
00127   22725,   22725,   31521,   26722, 
00128   22725,   22725,   17855,    6270, 
00129   29692,   12299,   26722,   -6270, 
00130  -12299,  -29692,  -31521,  -17855, 
00131   22725,  -22725,   17855,  -31521, 
00132  -22725,   22725,    6270,   26722, 
00133   12299,  -29692,    6270,  -17855, 
00134   29692,  -12299,   26722,  -31521, 
00135 };
00136 
00137 struct 
00138 {
00139  const int16_t tab_frw_01234567_sse2[256] ATTR_ALIGN(16);
00140 } tab_frw_01234567_sse2 ATTR_ALIGN(16) =
00141 {{
00142 //static const int16_t tab_frw_01234567_sse2[] ATTR_ALIGN(16) = {  // forward_dct coeff table  
00143 #define TABLE_SSE2 C4,  C4,  C1,  C3, -C6, -C2, -C1, -C5, \
00144                    C4,  C4,  C5,  C7,  C2,  C6,  C3, -C7, \
00145                   -C4,  C4,  C7,  C3,  C6, -C2,  C7, -C5, \
00146                    C4, -C4,  C5, -C1,  C2, -C6,  C3, -C1, 
00147 // c1..c7 * cos(pi/4) * 2^15 
00148 #define C1 22725
00149 #define C2 21407
00150 #define C3 19266
00151 #define C4 16384
00152 #define C5 12873
00153 #define C6 8867
00154 #define C7 4520
00155 TABLE_SSE2
00156 
00157 #undef C1
00158 #undef C2
00159 #undef C3
00160 #undef C4
00161 #undef C5
00162 #undef C6
00163 #undef C7
00164 #define C1 31521
00165 #define C2 29692
00166 #define C3 26722
00167 #define C4 22725
00168 #define C5 17855
00169 #define C6 12299
00170 #define C7 6270
00171 TABLE_SSE2
00172 
00173 #undef C1
00174 #undef C2
00175 #undef C3
00176 #undef C4
00177 #undef C5
00178 #undef C6
00179 #undef C7
00180 #define C1 29692
00181 #define C2 27969
00182 #define C3 25172
00183 #define C4 21407
00184 #define C5 16819
00185 #define C6 11585
00186 #define C7 5906
00187 TABLE_SSE2
00188 
00189 #undef C1
00190 #undef C2
00191 #undef C3
00192 #undef C4
00193 #undef C5
00194 #undef C6
00195 #undef C7
00196 #define C1 26722
00197 #define C2 25172
00198 #define C3 22654
00199 #define C4 19266
00200 #define C5 15137
00201 #define C6 10426
00202 #define C7 5315
00203 TABLE_SSE2
00204 
00205 #undef C1
00206 #undef C2
00207 #undef C3
00208 #undef C4
00209 #undef C5
00210 #undef C6
00211 #undef C7
00212 #define C1 22725
00213 #define C2 21407
00214 #define C3 19266
00215 #define C4 16384
00216 #define C5 12873
00217 #define C6 8867
00218 #define C7 4520
00219 TABLE_SSE2
00220 
00221 #undef C1
00222 #undef C2
00223 #undef C3
00224 #undef C4
00225 #undef C5
00226 #undef C6
00227 #undef C7
00228 #define C1 26722
00229 #define C2 25172
00230 #define C3 22654
00231 #define C4 19266
00232 #define C5 15137
00233 #define C6 10426
00234 #define C7 5315
00235 TABLE_SSE2
00236 
00237 #undef C1
00238 #undef C2
00239 #undef C3
00240 #undef C4
00241 #undef C5
00242 #undef C6
00243 #undef C7
00244 #define C1 29692
00245 #define C2 27969
00246 #define C3 25172
00247 #define C4 21407
00248 #define C5 16819
00249 #define C6 11585
00250 #define C7 5906
00251 TABLE_SSE2
00252 
00253 #undef C1
00254 #undef C2
00255 #undef C3
00256 #undef C4
00257 #undef C5
00258 #undef C6
00259 #undef C7
00260 #define C1 31521
00261 #define C2 29692
00262 #define C3 26722
00263 #define C4 22725
00264 #define C5 17855
00265 #define C6 12299
00266 #define C7 6270
00267 TABLE_SSE2
00268 }};
00269 
00270 
00271 static always_inline void fdct_col(const int16_t *in, int16_t *out, int offset)
00272 {
00273     movq_m2r(*(in + offset + 1 * 8), mm0);
00274     movq_m2r(*(in + offset + 6 * 8), mm1);
00275     movq_r2r(mm0, mm2);
00276     movq_m2r(*(in + offset + 2 * 8), mm3);
00277     paddsw_r2r(mm1, mm0);
00278     movq_m2r(*(in + offset + 5 * 8), mm4);
00279     psllw_i2r(SHIFT_FRW_COL, mm0);
00280     movq_m2r(*(in + offset + 0 * 8), mm5);
00281     paddsw_r2r(mm3, mm4);
00282     paddsw_m2r(*(in + offset + 7 * 8), mm5);
00283     psllw_i2r(SHIFT_FRW_COL, mm4);
00284     movq_r2r(mm0, mm6);
00285     psubsw_r2r(mm1, mm2);
00286     movq_m2r(*(fdct_tg_all_16 + 4), mm1);
00287     psubsw_r2r(mm4, mm0);
00288     movq_m2r(*(in + offset + 3 * 8), mm7);
00289     pmulhw_r2r(mm0, mm1);
00290     paddsw_m2r(*(in + offset + 4 * 8), mm7);
00291     psllw_i2r(SHIFT_FRW_COL, mm5);
00292     paddsw_r2r(mm4, mm6);
00293     psllw_i2r(SHIFT_FRW_COL, mm7);
00294     movq_r2r(mm5, mm4);
00295     psubsw_r2r(mm7, mm5);
00296     paddsw_r2r(mm5, mm1);
00297     paddsw_r2r(mm7, mm4);
00298     por_m2r(fdct_one_corr, mm1);
00299     psllw_i2r(SHIFT_FRW_COL + 1, mm2);
00300     pmulhw_m2r(*(fdct_tg_all_16 + 4), mm5);
00301     movq_r2r(mm4, mm7);
00302     psubsw_m2r(*(in + offset + 5 * 8), mm3);
00303     psubsw_r2r(mm6, mm4);
00304     movq_r2m(mm1, *(out + offset + 2 * 8));
00305     paddsw_r2r(mm6, mm7);
00306     movq_m2r(*(in + offset + 3 * 8), mm1);
00307     psllw_i2r(SHIFT_FRW_COL + 1, mm3);
00308     psubsw_m2r(*(in + offset + 4 * 8), mm1);
00309     movq_r2r(mm2, mm6);
00310     movq_r2m(mm4, *(out + offset + 4 * 8));
00311     paddsw_r2r(mm3, mm2);
00312     pmulhw_m2r(*ocos_4_16, mm2);
00313     psubsw_r2r(mm3, mm6);
00314     pmulhw_m2r(*ocos_4_16, mm6);
00315     psubsw_r2r(mm0, mm5);
00316     por_m2r(fdct_one_corr, mm5);
00317     psllw_i2r(SHIFT_FRW_COL, mm1);
00318     por_m2r(fdct_one_corr, mm2);
00319     movq_r2r(mm1, mm4);
00320     movq_m2r(*(in + offset + 0 * 8), mm3);
00321     paddsw_r2r(mm6, mm1);
00322     psubsw_m2r(*(in + offset + 7 * 8), mm3);
00323     psubsw_r2r(mm6, mm4);
00324     movq_m2r(*(fdct_tg_all_16 + 0), mm0);
00325     psllw_i2r(SHIFT_FRW_COL, mm3);
00326     movq_m2r(*(fdct_tg_all_16 + 8), mm6);
00327     pmulhw_r2r(mm1, mm0);
00328     movq_r2m(mm7, *(out + offset + 0 * 8));
00329     pmulhw_r2r(mm4, mm6);
00330     movq_r2m(mm5, *(out + offset + 6 * 8));
00331     movq_r2r(mm3, mm7);
00332     movq_m2r(*(fdct_tg_all_16 + 8), mm5);
00333     psubsw_r2r(mm2, mm7);
00334     paddsw_r2r(mm2, mm3);
00335     pmulhw_r2r(mm7, mm5);
00336     paddsw_r2r(mm3, mm0);
00337     paddsw_r2r(mm4, mm6);
00338     pmulhw_m2r(*(fdct_tg_all_16 + 0), mm3);
00339     por_m2r(fdct_one_corr, mm0);
00340     paddsw_r2r(mm7, mm5);
00341     psubsw_r2r(mm6, mm7);
00342     movq_r2m(mm0, *(out + offset + 1 * 8));
00343     paddsw_r2r(mm4, mm5);
00344     movq_r2m(mm7, *(out + offset + 3 * 8));
00345     psubsw_r2r(mm1, mm3);
00346     movq_r2m(mm5, *(out + offset + 5 * 8));
00347     movq_r2m(mm3, *(out + offset + 7 * 8));
00348 }
00349 
00350 
00351 static always_inline void fdct_row_sse2(const int16_t *in, int16_t *out)
00352 {
00353     asm volatile(
00354         ".macro FDCT_ROW_SSE2_H1 i t   \n\t"
00355         "movq      \\i(%0), %%xmm2     \n\t"
00356         "movq      \\i+8(%0), %%xmm0   \n\t"
00357         "movdqa    \\t+32(%1), %%xmm3  \n\t"
00358         "movdqa    \\t+48(%1), %%xmm7  \n\t"    
00359         "movdqa    \\t(%1), %%xmm4     \n\t"
00360         "movdqa    \\t+16(%1), %%xmm5  \n\t"    
00361         ".endm                         \n\t"
00362         ".macro FDCT_ROW_SSE2_H2 i t   \n\t"
00363         "movq      \\i(%0), %%xmm2     \n\t"
00364         "movq      \\i+8(%0), %%xmm0   \n\t"
00365         "movdqa    \\t+32(%1), %%xmm3  \n\t"
00366         "movdqa    \\t+48(%1), %%xmm7  \n\t"    
00367         ".endm                         \n\t"
00368         ".macro FDCT_ROW_SSE2 i        \n\t"    
00369         "movq      %%xmm2, %%xmm1      \n\t"
00370         "pshuflw   $27, %%xmm0, %%xmm0 \n\t"
00371         "paddsw    %%xmm0, %%xmm1      \n\t"
00372         "psubsw    %%xmm0, %%xmm2      \n\t"
00373         "punpckldq %%xmm2, %%xmm1      \n\t"
00374         "pshufd    $78, %%xmm1, %%xmm2 \n\t"
00375         "pmaddwd   %%xmm2, %%xmm3      \n\t"
00376         "pmaddwd   %%xmm1, %%xmm7      \n\t"
00377         "pmaddwd   %%xmm5, %%xmm2      \n\t"
00378         "pmaddwd   %%xmm4, %%xmm1      \n\t"
00379         "paddd     %%xmm7, %%xmm3      \n\t"    
00380         "paddd     %%xmm2, %%xmm1      \n\t"
00381         "paddd     %%xmm6, %%xmm3      \n\t"
00382         "paddd     %%xmm6, %%xmm1      \n\t"
00383         "psrad     %3, %%xmm3          \n\t"
00384         "psrad     %3, %%xmm1          \n\t"
00385         "packssdw  %%xmm3, %%xmm1      \n\t"
00386         "movdqa    %%xmm1, \\i(%4)     \n\t"
00387         ".endm                         \n\t"    
00388         "movdqa    (%2), %%xmm6        \n\t"            
00389         "FDCT_ROW_SSE2_H1 0 0 \n\t"
00390         "FDCT_ROW_SSE2 0 \n\t"
00391         "FDCT_ROW_SSE2_H2 64 0 \n\t"
00392         "FDCT_ROW_SSE2 64 \n\t"
00393 
00394         "FDCT_ROW_SSE2_H1 16 64 \n\t"
00395         "FDCT_ROW_SSE2 16 \n\t"
00396         "FDCT_ROW_SSE2_H2 112 64 \n\t"
00397         "FDCT_ROW_SSE2 112 \n\t"
00398 
00399         "FDCT_ROW_SSE2_H1 32 128 \n\t"
00400         "FDCT_ROW_SSE2 32 \n\t"
00401         "FDCT_ROW_SSE2_H2 96 128 \n\t"
00402         "FDCT_ROW_SSE2 96 \n\t"
00403 
00404         "FDCT_ROW_SSE2_H1 48 192 \n\t"
00405         "FDCT_ROW_SSE2 48 \n\t"
00406         "FDCT_ROW_SSE2_H2 80 192 \n\t"
00407         "FDCT_ROW_SSE2 80 \n\t"
00408         :
00409         : "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)
00410     );
00411 }
00412 
00413 static always_inline void fdct_row_mmx2(const int16_t *in, int16_t *out, const int16_t *table)
00414 { 
00415     pshufw_m2r(*(in + 4), mm5, 0x1B);
00416     movq_m2r(*(in + 0), mm0);
00417     movq_r2r(mm0, mm1);
00418     paddsw_r2r(mm5, mm0);
00419     psubsw_r2r(mm5, mm1);
00420     movq_r2r(mm0, mm2);
00421     punpckldq_r2r(mm1, mm0);
00422     punpckhdq_r2r(mm1, mm2);
00423     movq_m2r(*(table + 0), mm1);
00424     movq_m2r(*(table + 4), mm3);
00425     movq_m2r(*(table + 8), mm4);
00426     movq_m2r(*(table + 12), mm5);
00427     movq_m2r(*(table + 16), mm6);
00428     movq_m2r(*(table + 20), mm7);
00429     pmaddwd_r2r(mm0, mm1);
00430     pmaddwd_r2r(mm2, mm3);
00431     pmaddwd_r2r(mm0, mm4);
00432     pmaddwd_r2r(mm2, mm5);
00433     pmaddwd_r2r(mm0, mm6);
00434     pmaddwd_r2r(mm2, mm7);
00435     pmaddwd_m2r(*(table + 24), mm0);
00436     pmaddwd_m2r(*(table + 28), mm2);
00437     paddd_r2r(mm1, mm3);
00438     paddd_r2r(mm4, mm5);
00439     paddd_r2r(mm6, mm7);
00440     paddd_r2r(mm0, mm2);
00441     movq_m2r(*fdct_r_row, mm0);
00442     paddd_r2r(mm0, mm3);
00443     paddd_r2r(mm0, mm5);
00444     paddd_r2r(mm0, mm7);
00445     paddd_r2r(mm0, mm2);
00446     psrad_i2r(SHIFT_FRW_ROW, mm3);
00447     psrad_i2r(SHIFT_FRW_ROW, mm5);
00448     psrad_i2r(SHIFT_FRW_ROW, mm7);
00449     psrad_i2r(SHIFT_FRW_ROW, mm2);
00450     packssdw_r2r(mm5, mm3);
00451     packssdw_r2r(mm2, mm7);
00452     movq_r2m(mm3, *(out + 0));
00453     movq_r2m(mm7, *(out + 4));
00454 }
00455 
00456 static always_inline void fdct_row_mmx(const int16_t *in, int16_t *out, const int16_t *table)
00457 { 
00458 //FIXME reorder (i dont have a old mmx only cpu here to benchmark ...)
00459     movd_m2r(*(in + 6), mm1);
00460     punpcklwd_m2r(*(in + 4), mm1);
00461     movq_r2r(mm1, mm2);
00462     psrlq_i2r(0x20, mm1);
00463     movq_m2r(*(in + 0), mm0);
00464     punpcklwd_r2r(mm2, mm1);
00465     movq_r2r(mm0, mm5);
00466     paddsw_r2r(mm1, mm0);
00467     psubsw_r2r(mm1, mm5);
00468     movq_r2r(mm0, mm2);
00469     punpckldq_r2r(mm5, mm0);
00470     punpckhdq_r2r(mm5, mm2);
00471     movq_m2r(*(table + 0), mm1);
00472     movq_m2r(*(table + 4), mm3);
00473     movq_m2r(*(table + 8), mm4);
00474     movq_m2r(*(table + 12), mm5);
00475     movq_m2r(*(table + 16), mm6);
00476     movq_m2r(*(table + 20), mm7);
00477     pmaddwd_r2r(mm0, mm1);
00478     pmaddwd_r2r(mm2, mm3);
00479     pmaddwd_r2r(mm0, mm4);
00480     pmaddwd_r2r(mm2, mm5);
00481     pmaddwd_r2r(mm0, mm6);
00482     pmaddwd_r2r(mm2, mm7);
00483     pmaddwd_m2r(*(table + 24), mm0);
00484     pmaddwd_m2r(*(table + 28), mm2);
00485     paddd_r2r(mm1, mm3);
00486     paddd_r2r(mm4, mm5);
00487     paddd_r2r(mm6, mm7);
00488     paddd_r2r(mm0, mm2);
00489     movq_m2r(*fdct_r_row, mm0);
00490     paddd_r2r(mm0, mm3);
00491     paddd_r2r(mm0, mm5);
00492     paddd_r2r(mm0, mm7);
00493     paddd_r2r(mm0, mm2);
00494     psrad_i2r(SHIFT_FRW_ROW, mm3);
00495     psrad_i2r(SHIFT_FRW_ROW, mm5);
00496     psrad_i2r(SHIFT_FRW_ROW, mm7);
00497     psrad_i2r(SHIFT_FRW_ROW, mm2);
00498     packssdw_r2r(mm5, mm3);
00499     packssdw_r2r(mm2, mm7);
00500     movq_r2m(mm3, *(out + 0));
00501     movq_r2m(mm7, *(out + 4));
00502 }
00503 
00504 void ff_fdct_mmx(int16_t *block)
00505 {
00506     int64_t align_tmp[16] ATTR_ALIGN(8);
00507     int16_t * const block_tmp= (int16_t*)align_tmp;
00508     int16_t *block1, *out;
00509     const int16_t *table;
00510     int i;
00511 
00512     block1 = block_tmp;
00513     fdct_col(block, block1, 0);
00514     fdct_col(block, block1, 4);
00515 
00516     block1 = block_tmp;
00517     table = tab_frw_01234567;
00518     out = block;
00519     for(i=8;i>0;i--) {
00520         fdct_row_mmx(block1, out, table);
00521         block1 += 8;
00522         table += 32;
00523         out += 8;
00524     }
00525 }
00526 
00527 void ff_fdct_mmx2(int16_t *block)
00528 {
00529     int64_t align_tmp[16] ATTR_ALIGN(8);
00530     int16_t * const block_tmp= (int16_t*)align_tmp;
00531     int16_t *block1, *out;
00532     const int16_t *table;
00533     int i;
00534 
00535     block1 = block_tmp;
00536     fdct_col(block, block1, 0);
00537     fdct_col(block, block1, 4);
00538 
00539     block1 = block_tmp;
00540     table = tab_frw_01234567;
00541     out = block;
00542     for(i=8;i>0;i--) {
00543         fdct_row_mmx2(block1, out, table);
00544         block1 += 8;
00545         table += 32;
00546         out += 8;
00547     }
00548 }
00549 
00550 void ff_fdct_sse2(int16_t *block) 
00551 {
00552     int64_t align_tmp[16] ATTR_ALIGN(8);
00553     int16_t * const block_tmp= (int16_t*)align_tmp;
00554     int16_t *block1;
00555 
00556     block1 = block_tmp;
00557     fdct_col(block, block1, 0);
00558     fdct_col(block, block1, 4);
00559 
00560     fdct_row_sse2(block1, block);
00561 }
00562
hvirtual/quicktime/ffmpeg/libavcodec/i386/fdct_mmx.c
