--- /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
--- /dev/null
+/*****************************************************************************\r
+ *\r
+ * XVID MPEG-4 VIDEO CODEC\r
+ * - Forward DCT header -\r
+ *\r
+ * Copyright(C) 2001-2011 Michael Militzer <michael@xvid.org>\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.h 1986 2011-05-18 09:07:40Z Isibaar $\r
+ *\r
+ ****************************************************************************/\r
+\r
+#ifndef _FDCT_H_\r
+#define _FDCT_H_\r
+\r
+#include "../portab.h"\r
+\r
+typedef void (fdctFunc) (short *const block);\r
+typedef fdctFunc *fdctFuncPtr;\r
+\r
+extern fdctFuncPtr fdct;\r
+\r
+fdctFunc fdct_int32;\r
+\r
+#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)\r
+fdctFunc fdct_mmx_ffmpeg;\r
+fdctFunc fdct_xmm_ffmpeg;\r
+fdctFunc fdct_mmx_skal;\r
+fdctFunc fdct_xmm_skal;\r
+fdctFunc fdct_sse2_skal;\r
+#endif\r
+\r
+#ifdef ARCH_IS_IA64\r
+fdctFunc fdct_ia64;\r
+#endif\r
+\r
+#endif /* _FDCT_H_ */\r
--- /dev/null
+/*****************************************************************************\r
+ *\r
+ * XVID MPEG-4 VIDEO CODEC\r
+ * - Inverse 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: idct.c 1986 2011-05-18 09:07:40Z Isibaar $\r
+ *\r
+ ****************************************************************************/\r
+\r
+/*\r
+ * Authors: Skal\r
+ *\r
+ * Walken IDCT\r
+ * Alternative idct implementations for decoding compatibility\r
+ *\r
+ * NOTE: this "C" version is not the original one,\r
+ * but is modified to yield the same error profile\r
+ * than the MMX version.\r
+ *\r
+ ************************************************************************/\r
+\r
+#include <stdint.h>\r
+\r
+//#include "idct.h"\r
+\r
+/* function pointer */\r
+//idctFuncPtr idct;\r
+\r
+#define XVID_DSP_CLIP_255(x) ( ((x)&~255) ? ((-(x)) >> (8*sizeof((x))-1))&0xff : (x) )\r
+\r
+#define ROW_SHIFT 11\r
+#define COL_SHIFT 6\r
+\r
+// #define FIX(x) (int)((x) * (1<<ROW_SHIFT))\r
+#define Rnd0 65536 // 1<<(COL_SHIFT+ROW_SHIFT-1);\r
+#define Rnd1 3597 // FIX (1.75683487303);\r
+#define Rnd2 2260 // FIX (1.10355339059);\r
+#define Rnd3 1203 // FIX (0.587788325588);\r
+#define Rnd4 0\r
+#define Rnd5 120 // FIX (0.058658283817);\r
+#define Rnd6 512 // FIX (0.25);\r
+#define Rnd7 512 // FIX (0.25);\r
+#undef FIX\r
+\r
+static const int Tab04[] = { 22725, 21407, 19266, 16384, 12873, 8867, 4520 };\r
+static const int Tab17[] = { 31521, 29692, 26722, 22725, 17855, 12299, 6270 };\r
+static const int Tab26[] = { 29692, 27969, 25172, 21407, 16819, 11585, 5906 };\r
+static const int Tab35[] = { 26722, 25172, 22654, 19266, 15137, 10426, 5315 };\r
+\r
+static int Idct_Row(short * In, const int * const Tab, int Rnd)\r
+{\r
+ const int C1 = Tab[0];\r
+ const int C2 = Tab[1];\r
+ const int C3 = Tab[2];\r
+ const int C4 = Tab[3];\r
+ const int C5 = Tab[4];\r
+ const int C6 = Tab[5];\r
+ const int C7 = Tab[6];\r
+\r
+ const int Right = In[5]|In[6]|In[7];\r
+ const int Left = In[1]|In[2]|In[3];\r
+ if (!(Right | In[4]))\r
+ {\r
+ const int K = C4*In[0] + Rnd;\r
+ if (Left)\r
+ {\r
+ const int a0 = K + C2*In[2];\r
+ const int a1 = K + C6*In[2];\r
+ const int a2 = K - C6*In[2];\r
+ const int a3 = K - C2*In[2];\r
+\r
+ const int b0 = C1*In[1] + C3*In[3];\r
+ const int b1 = C3*In[1] - C7*In[3];\r
+ const int b2 = C5*In[1] - C1*In[3];\r
+ const int b3 = C7*In[1] - C5*In[3];\r
+\r
+ In[0] = (a0 + b0) >> ROW_SHIFT;\r
+ In[1] = (a1 + b1) >> ROW_SHIFT;\r
+ In[2] = (a2 + b2) >> ROW_SHIFT;\r
+ In[3] = (a3 + b3) >> ROW_SHIFT;\r
+ In[4] = (a3 - b3) >> ROW_SHIFT;\r
+ In[5] = (a2 - b2) >> ROW_SHIFT;\r
+ In[6] = (a1 - b1) >> ROW_SHIFT;\r
+ In[7] = (a0 - b0) >> ROW_SHIFT;\r
+ }\r
+ else\r
+ {\r
+ const int a0 = K >> ROW_SHIFT;\r
+ if (a0) {\r
+ In[0] = In[1] = In[2] = In[3] =\r
+ In[4] = In[5] = In[6] = In[7] = a0;\r
+ }\r
+ else return 0;\r
+ }\r
+ }\r
+ else if (!(Left|Right))\r
+ {\r
+ const int a0 = (Rnd + C4*(In[0]+In[4])) >> ROW_SHIFT;\r
+ const int a1 = (Rnd + C4*(In[0]-In[4])) >> ROW_SHIFT;\r
+\r
+ In[0] = a0;\r
+ In[3] = a0;\r
+ In[4] = a0;\r
+ In[7] = a0;\r
+ In[1] = a1;\r
+ In[2] = a1;\r
+ In[5] = a1;\r
+ In[6] = a1;\r
+ }\r
+ else\r
+ {\r
+ const int K = C4*In[0] + Rnd;\r
+ const int a0 = K + C2*In[2] + C4*In[4] + C6*In[6];\r
+ const int a1 = K + C6*In[2] - C4*In[4] - C2*In[6];\r
+ const int a2 = K - C6*In[2] - C4*In[4] + C2*In[6];\r
+ const int a3 = K - C2*In[2] + C4*In[4] - C6*In[6];\r
+\r
+ const int b0 = C1*In[1] + C3*In[3] + C5*In[5] + C7*In[7];\r
+ const int b1 = C3*In[1] - C7*In[3] - C1*In[5] - C5*In[7];\r
+ const int b2 = C5*In[1] - C1*In[3] + C7*In[5] + C3*In[7];\r
+ const int b3 = C7*In[1] - C5*In[3] + C3*In[5] - C1*In[7];\r
+\r
+ In[0] = (a0 + b0) >> ROW_SHIFT;\r
+ In[1] = (a1 + b1) >> ROW_SHIFT;\r
+ In[2] = (a2 + b2) >> ROW_SHIFT;\r
+ In[3] = (a3 + b3) >> ROW_SHIFT;\r
+ In[4] = (a3 - b3) >> ROW_SHIFT;\r
+ In[5] = (a2 - b2) >> ROW_SHIFT;\r
+ In[6] = (a1 - b1) >> ROW_SHIFT;\r
+ In[7] = (a0 - b0) >> ROW_SHIFT;\r
+ }\r
+ return 1;\r
+}\r
+\r
+#define Tan1 0x32ec\r
+#define Tan2 0x6a0a\r
+#define Tan3 0xab0e\r
+#define Sqrt2 0x5a82\r
+\r
+#define MULT(c,x, n) ( ((c) * (x)) >> (n) )\r
+// 12b version => #define MULT(c,x, n) ( (((c)>>3) * (x)) >> ((n)-3) )\r
+// 12b zero-testing version:\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
+static void Idct_Col_8(short * const In)\r
+{\r
+ int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, Spill;\r
+\r
+ // odd\r
+\r
+ mm4 = (int)In[7*8];\r
+ mm5 = (int)In[5*8];\r
+ mm6 = (int)In[3*8];\r
+ mm7 = (int)In[1*8];\r
+\r
+ mm0 = MULT(Tan1, mm4, 16) + mm7;\r
+ mm1 = MULT(Tan1, mm7, 16) - mm4;\r
+ mm2 = MULT(Tan3, mm5, 16) + mm6;\r
+ mm3 = MULT(Tan3, mm6, 16) - mm5;\r
+\r
+ mm7 = mm0 + mm2;\r
+ mm4 = mm1 - mm3;\r
+ mm0 = mm0 - mm2;\r
+ mm1 = mm1 + mm3;\r
+ mm6 = mm0 + mm1;\r
+ mm5 = mm0 - mm1;\r
+ mm5 = 2*MULT(Sqrt2, mm5, 16); // 2*sqrt2\r
+ mm6 = 2*MULT(Sqrt2, mm6, 16); // Watch out: precision loss but done to match\r
+ // the pmulhw used in mmx/sse versions\r
+ \r
+ // even\r
+\r
+ mm1 = (int)In[2*8];\r
+ mm2 = (int)In[6*8];\r
+ mm3 = MULT(Tan2,mm2, 16) + mm1;\r
+ mm2 = MULT(Tan2,mm1, 16) - mm2;\r
+\r
+ LOAD_BUTF(mm0, mm1, 0*8, 4*8, Spill, In);\r
+\r
+ BUTF(mm0, mm3, Spill);\r
+ BUTF(mm0, mm7, Spill);\r
+ In[8*0] = (int16_t) (mm0 >> COL_SHIFT);\r
+ In[8*7] = (int16_t) (mm7 >> COL_SHIFT);\r
+ BUTF(mm3, mm4, mm0);\r
+ In[8*3] = (int16_t) (mm3 >> COL_SHIFT);\r
+ In[8*4] = (int16_t) (mm4 >> COL_SHIFT);\r
+\r
+ BUTF(mm1, mm2, mm0);\r
+ BUTF(mm1, mm6, mm0);\r
+ In[8*1] = (int16_t) (mm1 >> COL_SHIFT);\r
+ In[8*6] = (int16_t) (mm6 >> COL_SHIFT);\r
+ BUTF(mm2, mm5, mm0);\r
+ In[8*2] = (int16_t) (mm2 >> COL_SHIFT);\r
+ In[8*5] = (int16_t) (mm5 >> COL_SHIFT);\r
+}\r
+\r
+static void Idct_Col_4(short * const In)\r
+{\r
+ int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, Spill;\r
+\r
+ // odd\r
+\r
+ mm0 = (int)In[1*8];\r
+ mm2 = (int)In[3*8];\r
+\r
+ mm1 = MULT(Tan1, mm0, 16);\r
+ mm3 = MULT(Tan3, mm2, 16);\r
+\r
+ mm7 = mm0 + mm2;\r
+ mm4 = mm1 - mm3;\r
+ mm0 = mm0 - mm2;\r
+ mm1 = mm1 + mm3;\r
+ mm6 = mm0 + mm1;\r
+ mm5 = mm0 - mm1;\r
+ mm6 = 2*MULT(Sqrt2, mm6, 16); // 2*sqrt2\r
+ mm5 = 2*MULT(Sqrt2, mm5, 16);\r
+\r
+ // even\r
+\r
+ mm0 = mm1 = (int)In[0*8];\r
+ mm3 = (int)In[2*8];\r
+ mm2 = MULT(Tan2,mm3, 16);\r
+\r
+ BUTF(mm0, mm3, Spill);\r
+ BUTF(mm0, mm7, Spill);\r
+ In[8*0] = (int16_t) (mm0 >> COL_SHIFT);\r
+ In[8*7] = (int16_t) (mm7 >> COL_SHIFT);\r
+ BUTF(mm3, mm4, mm0);\r
+ In[8*3] = (int16_t) (mm3 >> COL_SHIFT);\r
+ In[8*4] = (int16_t) (mm4 >> COL_SHIFT);\r
+\r
+ BUTF(mm1, mm2, mm0);\r
+ BUTF(mm1, mm6, mm0);\r
+ In[8*1] = (int16_t) (mm1 >> COL_SHIFT);\r
+ In[8*6] = (int16_t) (mm6 >> COL_SHIFT);\r
+ BUTF(mm2, mm5, mm0);\r
+ In[8*2] = (int16_t) (mm2 >> COL_SHIFT);\r
+ In[8*5] = (int16_t) (mm5 >> COL_SHIFT);\r
+}\r
+\r
+static void Idct_Col_3(short * const In)\r
+{\r
+ int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, Spill;\r
+\r
+ // odd\r
+\r
+ mm7 = (int)In[1*8];\r
+ mm4 = MULT(Tan1, mm7, 16);\r
+\r
+ mm6 = mm7 + mm4;\r
+ mm5 = mm7 - mm4;\r
+ mm6 = 2*MULT(Sqrt2, mm6, 16); // 2*sqrt2\r
+ mm5 = 2*MULT(Sqrt2, mm5, 16);\r
+\r
+ // even\r
+\r
+ mm0 = mm1 = (int)In[0*8];\r
+ mm3 = (int)In[2*8];\r
+ mm2 = MULT(Tan2,mm3, 16);\r
+\r
+ BUTF(mm0, mm3, Spill);\r
+ BUTF(mm0, mm7, Spill);\r
+ In[8*0] = (int16_t) (mm0 >> COL_SHIFT);\r
+ In[8*7] = (int16_t) (mm7 >> COL_SHIFT);\r
+ BUTF(mm3, mm4, mm0);\r
+ In[8*3] = (int16_t) (mm3 >> COL_SHIFT);\r
+ In[8*4] = (int16_t) (mm4 >> COL_SHIFT);\r
+\r
+ BUTF(mm1, mm2, mm0);\r
+ BUTF(mm1, mm6, mm0);\r
+ In[8*1] = (int16_t) (mm1 >> COL_SHIFT);\r
+ In[8*6] = (int16_t) (mm6 >> COL_SHIFT);\r
+ BUTF(mm2, mm5, mm0);\r
+ In[8*2] = (int16_t) (mm2 >> COL_SHIFT);\r
+ In[8*5] = (int16_t) (mm5 >> COL_SHIFT);\r
+}\r
+\r
+#undef Tan1\r
+#undef Tan2\r
+#undef Tan3\r
+#undef Sqrt2\r
+\r
+#undef ROW_SHIFT\r
+#undef COL_SHIFT\r
+\r
+//////////////////////////////////////////////////////////\r
+\r
+void idct_int32(short *const In)\r
+{\r
+ int i, Rows = 0x07;\r
+\r
+ Idct_Row(In + 0*8, Tab04, Rnd0);\r
+ Idct_Row(In + 1*8, Tab17, Rnd1);\r
+ Idct_Row(In + 2*8, Tab26, Rnd2);\r
+ if (Idct_Row(In + 3*8, Tab35, Rnd3)) Rows |= 0x08;\r
+ if (Idct_Row(In + 4*8, Tab04, Rnd4)) Rows |= 0x10;\r
+ if (Idct_Row(In + 5*8, Tab35, Rnd5)) Rows |= 0x20;\r
+ if (Idct_Row(In + 6*8, Tab26, Rnd6)) Rows |= 0x40;\r
+ if (Idct_Row(In + 7*8, Tab17, Rnd7)) Rows |= 0x80;\r
+\r
+ if (Rows&0xf0) {\r
+ for(i=0; i<8; i++)\r
+ Idct_Col_8(In + i);\r
+ }\r
+ else if (Rows&0x08) {\r
+ for(i=0; i<8; i++)\r
+ Idct_Col_4(In + i);\r
+ }\r
+ else {\r
+ for(i=0; i<8; i++)\r
+ Idct_Col_3(In + i);\r
+ }\r
+}\r
projects / narabu / commitdiff