--- /dev/null
+/*****************************************************************************\r
+ *\r
+ * XVID MPEG-4 VIDEO CODEC\r
+ * - Forward DCT -\r
+ *\r
+ * Copyright (C) 2006-2011 Xvid Solutions GmbH\r
+ *\r
+ * This program is free software ; you can redistribute it and/or modify\r
+ * it under the terms of the GNU General Public License as published by\r
+ * the Free Software Foundation ; either version 2 of the License, or\r
+ * (at your option) any later version.\r
+ *\r
+ * This program is distributed in the hope that it will be useful,\r
+ * but WITHOUT ANY WARRANTY ; without even the implied warranty of\r
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the\r
+ * GNU General Public License for more details.\r
+ *\r
+ * You should have received a copy of the GNU General Public License\r
+ * along with this program ; if not, write to the Free Software\r
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA\r
+ *\r
+ * $Id: fdct.c 1986 2011-05-18 09:07:40Z Isibaar $\r
+ *\r
+ ****************************************************************************/\r
+\r
+/*\r
+ * Authors: Skal\r
+ *\r
+ * "Fast and precise" LLM implementation of FDCT/IDCT, where\r
+ * rotations are decomposed using:\r
+ * tmp = (x+y).cos t\r
+ * x' = tmp + y.(sin t - cos t)\r
+ * y' = tmp - x.(sin t + cos t)\r
+ *\r
+ * See details at the end of this file...\r
+ *\r
+ * Reference (e.g.):\r
+ * Loeffler C., Ligtenberg A., and Moschytz C.S.:\r
+ * Practical Fast 1D DCT Algorithm with Eleven Multiplications,\r
+ * Proc. ICASSP 1989, 988-991.\r
+ *\r
+ * IEEE-1180-like error specs for FDCT:\r
+ * Peak error: 1.0000\r
+ * Peak MSE: 0.0340\r
+ * Overall MSE: 0.0200\r
+ * Peak ME: 0.0191\r
+ * Overall ME: -0.0033\r
+ *\r
+ ********************************************************/\r
+\r
+#include <stdint.h>\r
+\r
+//#include "fdct.h"\r
+\r
+/* function pointer */\r
+//fdctFuncPtr fdct;\r
+\r
+/*\r
+//////////////////////////////////////////////////////////\r
+*/\r
+\r
+#define BUTF(a, b, tmp) \\r
+ (tmp) = (a)+(b); \\r
+ (b) = (a)-(b); \\r
+ (a) = (tmp)\r
+\r
+#define LOAD_BUTF(m1, m2, a, b, tmp, S) \\r
+ (m1) = (S)[(a)] + (S)[(b)]; \\r
+ (m2) = (S)[(a)] - (S)[(b)]\r
+\r
+#define ROTATE(m1,m2,c,k1,k2,tmp,Fix,Rnd) \\r
+ (tmp) = ( (m1) + (m2) )*(c); \\r
+ (m1) *= k1; \\r
+ (m2) *= k2; \\r
+ (tmp) += (Rnd); \\r
+ (m1) = ((m1)+(tmp))>>(Fix); \\r
+ (m2) = ((m2)+(tmp))>>(Fix);\r
+\r
+#define ROTATE2(m1,m2,c,k1,k2,tmp) \\r
+ (tmp) = ( (m1) + (m2) )*(c); \\r
+ (m1) *= k1; \\r
+ (m2) *= k2; \\r
+ (m1) = (m1)+(tmp); \\r
+ (m2) = (m2)+(tmp);\r
+\r
+#define ROTATE0(m1,m2,c,k1,k2,tmp) \\r
+ (m1) = ( (m2) )*(c); \\r
+ (m2) = (m2)*k2+(m1);\r
+\r
+#define SHIFTL(x,n) ((x)<<(n))\r
+#define SHIFTR(x, n) ((x)>>(n))\r
+#define HALF(n) (1<<((n)-1))\r
+\r
+#define IPASS 3\r
+#define FPASS 2\r
+#define FIX 16\r
+\r
+#if 1\r
+\r
+#define ROT6_C 35468\r
+#define ROT6_SmC 50159\r
+#define ROT6_SpC 121095\r
+#define ROT17_C 77062\r
+#define ROT17_SmC 25571\r
+#define ROT17_SpC 128553\r
+#define ROT37_C 58981\r
+#define ROT37_SmC 98391\r
+#define ROT37_SpC 19571\r
+#define ROT13_C 167963\r
+#define ROT13_SmC 134553\r
+#define ROT13_SpC 201373\r
+\r
+#else\r
+\r
+#include <math.h>\r
+#define FX(x) ( (int)floor((x)*(1<<FIX) + .5 ) )\r
+\r
+static const double c1 = cos(1.*M_PI/16);\r
+static const double c2 = cos(2.*M_PI/16);\r
+static const double c3 = cos(3.*M_PI/16);\r
+static const double c4 = cos(4.*M_PI/16);\r
+static const double c5 = cos(5.*M_PI/16);\r
+static const double c6 = cos(6.*M_PI/16);\r
+static const double c7 = cos(7.*M_PI/16);\r
+\r
+static const int ROT6_C = FX(c2-c6); // 0.541\r
+static const int ROT6_SmC = FX(2*c6); // 0.765\r
+static const int ROT6_SpC = FX(2*c2); // 1.847\r
+\r
+static const int ROT17_C = FX(c1+c7); // 1.175\r
+static const int ROT17_SmC = FX(2*c7); // 0.390\r
+static const int ROT17_SpC = FX(2*c1); // 1.961\r
+\r
+static const int ROT37_C = FX((c3-c7)/c4); // 0.899\r
+static const int ROT37_SmC = FX(2*(c5+c7)); // 1.501\r
+static const int ROT37_SpC = FX(2*(c1-c3)); // 0.298\r
+\r
+static const int ROT13_C = FX((c1+c3)/c4); // 2.562\r
+static const int ROT13_SmC = FX(2*(c3+c7)); // 2.053\r
+static const int ROT13_SpC = FX(2*(c1+c5)); // 3.072\r
+\r
+#endif\r
+\r
+/*\r
+//////////////////////////////////////////////////////////\r
+// Forward transform \r
+*/\r
+\r
+void fdct_int32( short *const In )\r
+{\r
+ short *pIn;\r
+ int i;\r
+\r
+ pIn = In;\r
+ for(i=8; i>0; --i)\r
+ {\r
+ int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, Spill;\r
+\r
+ // even\r
+\r
+ LOAD_BUTF(mm1,mm6, 1, 6, mm0, pIn);\r
+ LOAD_BUTF(mm2,mm5, 2, 5, mm0, pIn);\r
+ LOAD_BUTF(mm3,mm4, 3, 4, mm0, pIn);\r
+ LOAD_BUTF(mm0,mm7, 0, 7, Spill, pIn);\r
+\r
+ BUTF(mm1, mm2, Spill);\r
+ BUTF(mm0, mm3, Spill);\r
+\r
+ ROTATE(mm3, mm2, ROT6_C, ROT6_SmC, -ROT6_SpC, Spill, FIX-FPASS, HALF(FIX-FPASS));\r
+ pIn[2] = mm3;\r
+ pIn[6] = mm2;\r
+\r
+ BUTF(mm0, mm1, Spill);\r
+ pIn[0] = SHIFTL(mm0, FPASS);\r
+ pIn[4] = SHIFTL(mm1, FPASS);\r
+\r
+\r
+ // odd\r
+\r
+ mm3 = mm5 + mm7;\r
+ mm2 = mm4 + mm6;\r
+ ROTATE(mm2, mm3, ROT17_C, -ROT17_SpC, -ROT17_SmC, mm0, FIX-FPASS, HALF(FIX-FPASS));\r
+ ROTATE(mm4, mm7, -ROT37_C, ROT37_SpC, ROT37_SmC, mm0, FIX-FPASS, HALF(FIX-FPASS));\r
+ mm7 += mm3;\r
+ mm4 += mm2;\r
+ pIn[1] = mm7;\r
+ pIn[7] = mm4;\r
+\r
+ ROTATE(mm5, mm6, -ROT13_C, ROT13_SmC, ROT13_SpC, mm0, FIX-FPASS, HALF(FIX-FPASS));\r
+ mm5 += mm3;\r
+ mm6 += mm2;\r
+ pIn[3] = mm6;\r
+ pIn[5] = mm5;\r
+\r
+ pIn += 8;\r
+ }\r
+\r
+ pIn = In;\r
+ for(i=8; i>0; --i)\r
+ {\r
+ int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, Spill;\r
+\r
+ // even\r
+\r
+ LOAD_BUTF(mm1,mm6, 1*8, 6*8, mm0, pIn);\r
+ LOAD_BUTF(mm2,mm5, 2*8, 5*8, mm0, pIn);\r
+ BUTF(mm1, mm2, mm0);\r
+\r
+ LOAD_BUTF(mm3,mm4, 3*8, 4*8, mm0, pIn);\r
+ LOAD_BUTF(mm0,mm7, 0*8, 7*8, Spill, pIn);\r
+ BUTF(mm0, mm3, Spill);\r
+\r
+ ROTATE(mm3, mm2, ROT6_C, ROT6_SmC, -ROT6_SpC, Spill, 0, HALF(FIX+FPASS+3));\r
+ pIn[2*8] = (int16_t)SHIFTR(mm3,FIX+FPASS+3);\r
+ pIn[6*8] = (int16_t)SHIFTR(mm2,FIX+FPASS+3);\r
+\r
+ mm0 += HALF(FPASS+3) - 1;\r
+ BUTF(mm0, mm1, Spill);\r
+ pIn[0*8] = (int16_t)SHIFTR(mm0, FPASS+3);\r
+ pIn[4*8] = (int16_t)SHIFTR(mm1, FPASS+3);\r
+\r
+ // odd\r
+\r
+ mm3 = mm5 + mm7;\r
+ mm2 = mm4 + mm6;\r
+\r
+ ROTATE(mm2, mm3, ROT17_C, -ROT17_SpC, -ROT17_SmC, mm0, 0, HALF(FIX+FPASS+3));\r
+ ROTATE2(mm4, mm7, -ROT37_C, ROT37_SpC, ROT37_SmC, mm0);\r
+ mm7 += mm3;\r
+ mm4 += mm2;\r
+ pIn[7*8] = (int16_t)SHIFTR(mm4,FIX+FPASS+3);\r
+ pIn[1*8] = (int16_t)SHIFTR(mm7,FIX+FPASS+3);\r
+\r
+ ROTATE2(mm5, mm6, -ROT13_C, ROT13_SmC, ROT13_SpC, mm0);\r
+ mm5 += mm3;\r
+ mm6 += mm2;\r
+ pIn[5*8] = (int16_t)SHIFTR(mm5,FIX+FPASS+3);\r
+ pIn[3*8] = (int16_t)SHIFTR(mm6,FIX+FPASS+3);\r
+\r
+ pIn++;\r
+ }\r
+}\r
+#undef FIX\r
+#undef FPASS\r
+#undef IPASS\r
+\r
+#undef BUTF\r
+#undef LOAD_BUTF\r
+#undef ROTATE\r
+#undef ROTATE2\r
+#undef SHIFTL\r
+#undef SHIFTR\r
+\r
+//////////////////////////////////////////////////////////\r
+// - Data flow schematics for FDCT -\r
+// Output is scaled by 2.sqrt(2)\r
+// Initial butterflies (in0/in7, etc.) are not fully depicted.\r
+// Note: Rot6 coeffs are multiplied by sqrt(2).\r
+//////////////////////////////////////////////////////////\r
+/*\r
+ <---------Stage1 =even part=----------->\r
+\r
+ in3 mm3 +_____.___-___________.____* out6\r
+ x \ / |\r
+ in4 mm4 \ |\r
+ / \ |\r
+ in0 mm0 +_____o___+__.___-___ | ___* out4\r
+ x \ / |\r
+ in7 mm7 \ (Rot6)\r
+ / \ |\r
+ in1 mm1 +_____o___+__o___+___ | ___* out0\r
+ x \ / |\r
+ in6 mm6 / |\r
+ / \ |\r
+ in2 mm2 +_____.___-___________o____* out2\r
+ x\r
+ in5 mm5\r
+\r
+ <---------Stage2 =odd part=---------------->\r
+\r
+ mm7*___._________.___-___[xSqrt2]___* out3\r
+ | \ /\r
+ (Rot3) \\r
+ | / \\r
+ mm5*__ | ___o____o___+___.___-______* out7\r
+ | | \ /\r
+ | (Rot1) \\r
+ | | / \\r
+ mm6*__ |____.____o___+___o___+______* out1\r
+ | \ /\r
+ | /\r
+ | / \\r
+ mm4*___o_________.___-___[xSqrt2]___* out5\r
+\r
+\r
+\r
+ Alternative schematics for stage 2:\r
+ -----------------------------------\r
+\r
+ mm7 *___[xSqrt2]____o___+____o_______* out1\r
+ \ / |\r
+ / (Rot1)\r
+ / \ |\r
+ mm6 *____o___+______.___-___ | __.___* out5\r
+ \ / | |\r
+ / | |\r
+ / \ | |\r
+ mm5 *____.___-______.___-____.__ | __* out7\r
+ \ / |\r
+ \ (Rot3)\r
+ / \ |\r
+ mm4 *___[xSqrt2]____o___+________o___* out3\r
+\r
+*/\r