diff options
Diffstat (limited to 'src/post')
| -rw-r--r-- | src/post/deinterlace/plugins/Makefile.am | 27 | ||||
| -rw-r--r-- | src/post/deinterlace/plugins/greedyh.asm | 309 | ||||
| -rw-r--r-- | src/post/deinterlace/plugins/greedyhmacros.h | 74 | ||||
| -rw-r--r-- | src/post/deinterlace/plugins/kdetv_greedyh.c | 129 | ||||
| -rw-r--r-- | src/post/deinterlace/plugins/x86-64_macros.inc | 81 | ||||
| -rw-r--r-- | src/post/deinterlace/xine_plugin.c | 4 |
6 files changed, 618 insertions, 6 deletions
diff --git a/src/post/deinterlace/plugins/Makefile.am b/src/post/deinterlace/plugins/Makefile.am index f08fa250a..817e206cb 100644 --- a/src/post/deinterlace/plugins/Makefile.am +++ b/src/post/deinterlace/plugins/Makefile.am @@ -1,8 +1,25 @@ include $(top_srcdir)/misc/Makefile.common -AM_CPPFLAGS = -I$(top_srcdir)/src/post/deinterlace +# plugins/Makefile.am distributes the plugins that come with tvtime. +# Copyright (C) 2002, 2003, 2004 Billy Biggs +# +# This program is free software; you can redistribute it and/or modify +# it under the terms of the GNU General Public License as published by +# the Free Software Foundation; either version 2 of the License, or (at +# your option) any later version. +# +# This program is distributed in the hope that it will be useful, but +# WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +# General Public License for more details. +# +# You should have received a copy of the GNU General Public License +# along with this program; if not, write to the Free Software +# Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + +EXTRA_DIST = greedy2frame_template.c greedyh.asm x86-64_macros.inc -EXTRA_DIST = greedy2frame_template.c +AM_CPPFLAGS = -I$(top_srcdir)/src/post/deinterlace libdir = $(XINE_PLUGINDIR)/post @@ -16,8 +33,10 @@ libdeinterlaceplugins_la_SOURCES = \ vfir.c \ weave.c \ greedy2frame.c \ - scalerbob.c + scalerbob.c \ + kdetv_greedyh.c +# kdetv_tomsmocomp.c libdeinterlaceplugins_la_LIBADD = $(XINE_LIB) libdeinterlaceplugins_la_LDFLAGS = -avoid-version -module @XINE_PLUGIN_MIN_SYMS@ -noinst_HEADERS = plugins.h +noinst_HEADERS = plugins.h greedyhmacros.h diff --git a/src/post/deinterlace/plugins/greedyh.asm b/src/post/deinterlace/plugins/greedyh.asm new file mode 100644 index 000000000..d2b3f3566 --- /dev/null +++ b/src/post/deinterlace/plugins/greedyh.asm @@ -0,0 +1,309 @@ +// -*- c++ -*- +///////////////////////////////////////////////////////////////////////////// +// Copyright (c) 2001 Tom Barry. All rights reserved. +///////////////////////////////////////////////////////////////////////////// +// +// This file is subject to the terms of the GNU General Public License as +// published by the Free Software Foundation. A copy of this license is +// included with this software distribution in the file COPYING. If you +// do not have a copy, you may obtain a copy by writing to the Free +// Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. +// +// This software is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details +// +///////////////////////////////////////////////////////////////////////////// + +#include "x86-64_macros.inc" + +static void FUNCT_NAME(uint8_t *output, int outstride, + deinterlace_frame_data_t *data, + int bottom_field, int second_field, int width, int height ) +{ + int64_t i; + int stride = (width*2); + int InfoIsOdd = bottom_field; + + // in tight loop some vars are accessed faster in local storage + int64_t YMask = 0x00ff00ff00ff00ffull; // to keep only luma + int64_t UVMask = 0xff00ff00ff00ff00ull; // to keep only chroma + int64_t ShiftMask = 0xfefffefffefffeffull; // to avoid shifting chroma to luma + int64_t QW256 = 0x0100010001000100ull; // 4 256's + + int64_t MaxComb; + int64_t MotionThreshold; + int64_t MotionSense; + + int Line; + long LoopCtr; + long oldbx; + unsigned int Pitch = stride*2; + int FieldHeight = height / 2; + + unsigned char* L1; // ptr to Line1, of 3 + unsigned char* L2; // ptr to Line2, the weave line + unsigned char* L3; // ptr to Line3 + + unsigned char* L2P; // ptr to prev Line2 + unsigned char* Dest = output; + + int64_t QW256B; + int64_t LastAvg=0; //interp value from left qword + + // Set up our two parms that are actually evaluated for each pixel + i=GreedyMaxComb; + MaxComb = i << 56 | i << 48 | i << 40 | i << 32 | i << 24 | i << 16 | i << 8 | i; + + i = GreedyMotionThreshold; // scale to range of 0-257 + MotionThreshold = i << 48 | i << 32 | i << 16 | i | UVMask; + + i = GreedyMotionSense; // scale to range of 0-257 + MotionSense = i << 48 | i << 32 | i << 16 | i; + + + i = 0xffffffff - 256; + QW256B = i << 48 | i << 32 | i << 16 | i; // save a couple instr on PMINSW instruct. + + // copy first even line no matter what, and the first odd line if we're + // processing an EVEN field. (note diff from other deint rtns.) + if( second_field ) { + L1 = data->f0; + L2 = data->f0; + L2P = data->f1; + } else { + L1 = data->f1; + L2 = data->f0; + L2P = data->f1; + } + + if( InfoIsOdd ) { + L1 += 0; + L2 += stride; + L3 = L1 + Pitch; + L2P += stride; + + // copy first even line + xine_fast_memcpy(Dest, L1, stride); + Dest += outstride; + } else { + // copy first even line + xine_fast_memcpy(Dest, L2, stride); + Dest += outstride; + + L1 += stride; + L2 += Pitch; + L3 = L1 + Pitch; + L2P += Pitch; + + // then first odd line + xine_fast_memcpy(Dest, L1, stride); + Dest += outstride; + } + + for (Line = 0; Line < (FieldHeight - 1); ++Line) { + LoopCtr = stride / 8 - 1; // there are LineLength / 8 qwords per line but do 1 less, adj at end of loop + + // For ease of reading, the comments below assume that we're operating on an odd + // field (i.e., that InfoIsOdd is true). Assume the obvious for even lines.. + __asm__ __volatile__ + ( + // save ebx (-fPIC) + MOVX" %%"XBX", %[oldbx]\n\t" + + MOVX" %[L1], %%"XAX"\n\t" + LEAX" 8(%%"XAX"), %%"XBX"\n\t" // next qword needed by DJR + MOVX" %[L3], %%"XCX"\n\t" + SUBX" %%"XAX", %%"XCX"\n\t" // carry L3 addr as an offset + MOVX" %[L2P], %%"XDX"\n\t" + MOVX" %[L2], %%"XSI"\n\t" + MOVX" %[Dest], %%"XDI"\n\t" // DL1 if Odd or DL2 if Even + + ".align 8\n\t" + "1:\n\t" + + "movq (%%"XSI"), %%mm0\n\t" // L2 - the newest weave pixel value + "movq (%%"XAX"), %%mm1\n\t" // L1 - the top pixel + "movq (%%"XDX"), %%mm2\n\t" // L2P - the prev weave pixel + "movq (%%"XAX", %%"XCX"), %%mm3\n\t" // L3, next odd row + "movq %%mm1, %%mm6\n\t" // L1 - get simple single pixel interp + // pavgb mm6, mm3 // use macro below + V_PAVGB ("%%mm6", "%%mm3", "%%mm4", "%[ShiftMask]") + + // DJR - Diagonal Jaggie Reduction + // In the event that we are going to use an average (Bob) pixel we do not want a jagged + // stair step effect. To combat this we avg in the 2 horizontally adjacen pixels into the + // interpolated Bob mix. This will do horizontal smoothing for only the Bob'd pixels. + + "movq %[LastAvg], %%mm4\n\t" // the bob value from prev qword in row + "movq %%mm6, %[LastAvg]\n\t" // save for next pass + "psrlq $48, %%mm4\n\t" // right justify 1 pixel + "movq %%mm6, %%mm7\n\t" // copy of simple bob pixel + "psllq $16, %%mm7\n\t" // left justify 3 pixels + "por %%mm7, %%mm4\n\t" // and combine + + "movq (%%"XBX"), %%mm5\n\t" // next horiz qword from L1 + // pavgb mm5, qword ptr[ebx+ecx] // next horiz qword from L3, use macro below + V_PAVGB ("%%mm5", "(%%"XBX",%%"XCX")", "%%mm7", "%[ShiftMask]") + "psllq $48, %%mm5\n\t" // left just 1 pixel + "movq %%mm6, %%mm7\n\t" // another copy of simple bob pixel + "psrlq $16, %%mm7\n\t" // right just 3 pixels + "por %%mm7, %%mm5\n\t" // combine + // pavgb mm4, mm5 // avg of forward and prev by 1 pixel, use macro + V_PAVGB ("%%mm4", "%%mm5", "%%mm5", "%[ShiftMask]") // mm5 gets modified if MMX + // pavgb mm6, mm4 // avg of center and surround interp vals, use macro + V_PAVGB ("%%mm6", "%%mm4", "%%mm7", "%[ShiftMask]") + + // Don't do any more averaging than needed for mmx. It hurts performance and causes rounding errors. +#ifndef IS_MMX + // pavgb mm4, mm6 // 1/4 center, 3/4 adjacent + V_PAVGB ("%%mm4", "%%mm6", "%%mm7", "%[ShiftMask]") + // pavgb mm6, mm4 // 3/8 center, 5/8 adjacent + V_PAVGB ("%%mm6", "%%mm4", "%%mm7", "%[ShiftMask]") +#endif + + // get abs value of possible L2 comb + "movq %%mm6, %%mm4\n\t" // work copy of interp val + "movq %%mm2, %%mm7\n\t" // L2 + "psubusb %%mm4, %%mm7\n\t" // L2 - avg + "movq %%mm4, %%mm5\n\t" // avg + "psubusb %%mm2, %%mm5\n\t" // avg - L2 + "por %%mm7, %%mm5\n\t" // abs(avg-L2) + + // get abs value of possible L2P comb + "movq %%mm0, %%mm7\n\t" // L2P + "psubusb %%mm4, %%mm7\n\t" // L2P - avg + "psubusb %%mm0, %%mm4\n\t" // avg - L2P + "por %%mm7, %%mm4\n\t" // abs(avg-L2P) + + // use L2 or L2P depending upon which makes smaller comb + "psubusb %%mm5, %%mm4\n\t" // see if it goes to zero + "psubusb %%mm5, %%mm5\n\t" // 0 + "pcmpeqb %%mm5, %%mm4\n\t" // if (mm4=0) then FF else 0 + "pcmpeqb %%mm4, %%mm5\n\t" // opposite of mm4 + + // if Comb(L2P) <= Comb(L2) then mm4=ff, mm5=0 else mm4=0, mm5 = 55 + "pand %%mm2, %%mm5\n\t" // use L2 if mm5 == ff, else 0 + "pand %%mm0, %%mm4\n\t" // use L2P if mm4 = ff, else 0 + "por %%mm5, %%mm4\n\t" // may the best win + + // Inventory: at this point we have the following values: + // mm0 = L2P (or L2) + // mm1 = L1 + // mm2 = L2 (or L2P) + // mm3 = L3 + // mm4 = the best of L2,L2P weave pixel, base upon comb + // mm6 = the avg interpolated value, if we need to use it + + // Let's measure movement, as how much the weave pixel has changed + "movq %%mm2, %%mm7\n\t" + "psubusb %%mm0, %%mm2\n\t" + "psubusb %%mm7, %%mm0\n\t" + "por %%mm2, %%mm0\n\t" // abs value of change, used later + + // Now lets clip our chosen value to be not outside of the range + // of the high/low range L1-L3 by more than MaxComb. + // This allows some comb but limits the damages and also allows more + // detail than a boring oversmoothed clip. + "movq %%mm1, %%mm2\n\t" // copy L1 + // pmaxub mm2, mm3 // use macro + V_PMAXUB ("%%mm2", "%%mm3") // now = Max(L1,L3) + "movq %%mm1, %%mm5\n\t" // copy L1 + // pminub mm5, mm3 // now = Min(L1,L3), use macro + V_PMINUB ("%%mm5", "%%mm3", "%%mm7") + // allow the value to be above the high or below the low by amt of MaxComb + "psubusb %[MaxComb], %%mm5\n\t" // lower min by diff + "paddusb %[MaxComb], %%mm2\n\t" // increase max by diff + // pmaxub mm4, mm5 // now = Max(best,Min(L1,L3) use macro + V_PMAXUB ("%%mm4", "%%mm5") + // pminub mm4, mm2 // now = Min( Max(best, Min(L1,L3), L2 )=L2 clipped + V_PMINUB ("%%mm4", "%%mm2", "%%mm7") + + // Blend weave pixel with bob pixel, depending on motion val in mm0 + "psubusb %[MotionThreshold], %%mm0\n\t"// test Threshold, clear chroma change >>>?? + "pmullw %[MotionSense], %%mm0\n\t" // mul by user factor, keep low 16 bits + "movq %[QW256], %%mm7\n\t" +#ifdef IS_SSE + "pminsw %%mm7, %%mm0\n\t" // max = 256 +#else + "paddusw %[QW256B], %%mm0\n\t" // add, may sat at fff.. + "psubusw %[QW256B], %%mm0\n\t" // now = Min(L1,256) +#endif + "psubusw %%mm0, %%mm7\n\t" // so the 2 sum to 256, weighted avg + "movq %%mm4, %%mm2\n\t" // save weave chroma info before trashing + "pand %[YMask], %%mm4\n\t" // keep only luma from calc'd value + "pmullw %%mm7, %%mm4\n\t" // use more weave for less motion + "pand %[YMask], %%mm6\n\t" // keep only luma from calc'd value + "pmullw %%mm0, %%mm6\n\t" // use more bob for large motion + "paddusw %%mm6, %%mm4\n\t" // combine + "psrlw $8, %%mm4\n\t" // div by 256 to get weighted avg + + // chroma comes from weave pixel + "pand %[UVMask], %%mm2\n\t" // keep chroma + "por %%mm4, %%mm2\n\t" // and combine + + V_MOVNTQ ("(%%"XDI")", "%%mm2") // move in our clipped best, use macro + + // bump ptrs and loop + LEAX" 8(%%"XAX"), %%"XAX"\n\t" + LEAX" 8(%%"XBX"), %%"XBX"\n\t" + LEAX" 8(%%"XDX"), %%"XDX"\n\t" + LEAX" 8(%%"XDI"), %%"XDI"\n\t" + LEAX" 8(%%"XSI"), %%"XSI"\n\t" + DECX" %[LoopCtr]\n\t" + "jg 1b\n\t" // loop if not to last line + // note P-III default assumes backward branches taken + "jl 1f\n\t" // done + MOVX" %%"XAX", %%"XBX"\n\t" // sharpness lookahead 1 byte only, be wrong on 1 + "jmp 1b\n\t" + + "1:\n\t" + MOVX" %[oldbx], %%"XBX"\n\t" + + : /* no outputs */ + + : [LastAvg] "m"(LastAvg), + [L1] "m"(L1), + [L3] "m"(L3), + [L2P] "m"(L2P), + [L2] "m"(L2), + [Dest] "m"(Dest), + [ShiftMask] "m"(ShiftMask), + [MaxComb] "m"(MaxComb), + [MotionThreshold] "m"(MotionThreshold), + [MotionSense] "m"(MotionSense), + [QW256B] "m"(QW256B), + [YMask] "m"(YMask), + [UVMask] "m"(UVMask), + [LoopCtr] "m"(LoopCtr), + [QW256] "m"(QW256), + [oldbx] "m"(oldbx) + + : XAX, XCX, XDX, XSI, XDI, +#ifdef ARCH_X86 + "st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)", +#endif + "mm0", "mm1", "mm2", "mm3", "mm4", "mm5", "mm6", "mm7", + "memory", "cc" + ); + + Dest += outstride; + xine_fast_memcpy(Dest, L3, stride); + Dest += outstride; + + L1 += Pitch; + L2 += Pitch; + L3 += Pitch; + L2P += Pitch; + } + + if (InfoIsOdd) { + xine_fast_memcpy(Dest, L2, stride); + } + + // clear out the MMX registers ready for doing floating point again +#ifdef ARCH_X86 + __asm__ __volatile__ ("emms\n\t"); +#endif +} diff --git a/src/post/deinterlace/plugins/greedyhmacros.h b/src/post/deinterlace/plugins/greedyhmacros.h new file mode 100644 index 000000000..5f65959c3 --- /dev/null +++ b/src/post/deinterlace/plugins/greedyhmacros.h @@ -0,0 +1,74 @@ +///////////////////////////////////////////////////////////////////////////// +// Copyright (c) 2001 Tom Barry. All rights reserved. +///////////////////////////////////////////////////////////////////////////// +// +// This file is subject to the terms of the GNU General Public License as +// published by the Free Software Foundation. A copy of this license is +// included with this software distribution in the file COPYING. If you +// do not have a copy, you may obtain a copy by writing to the Free +// Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. +// +// This software is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details +// +///////////////////////////////////////////////////////////////////////////// + +// Define a few macros for CPU dependent instructions. +// I suspect I don't really understand how the C macro preprocessor works but +// this seems to get the job done. // TRB 7/01 + +// BEFORE USING THESE YOU MUST SET: + +// #define SSE_TYPE SSE (or MMX or 3DNOW) + +// some macros for pavgb instruction +// V_PAVGB(mmr1, mmr2, mmr work register, smask) mmr2 may = mmrw if you can trash it + +#define V_PAVGB_MMX(mmr1, mmr2, mmrw, smask) \ + "movq "mmr2", "mmrw"\n\t" \ + "pand "smask", "mmrw"\n\t" \ + "psrlw $1, "mmrw"\n\t" \ + "pand "smask", "mmr1"\n\t" \ + "psrlw $1, "mmr1"\n\t" \ + "paddusb "mmrw", "mmr1"\n\t" +#define V_PAVGB_SSE(mmr1, mmr2, mmrw, smask) "pavgb "mmr2", "mmr1"\n\t" +#define V_PAVGB_3DNOW(mmr1, mmr2, mmrw, smask) "pavgusb "mmr2", "mmr1"\n\t" +#define V_PAVGB(mmr1, mmr2, mmrw, smask) V_PAVGB2(mmr1, mmr2, mmrw, smask, SSE_TYPE) +#define V_PAVGB2(mmr1, mmr2, mmrw, smask, ssetyp) V_PAVGB3(mmr1, mmr2, mmrw, smask, ssetyp) +#define V_PAVGB3(mmr1, mmr2, mmrw, smask, ssetyp) V_PAVGB_##ssetyp(mmr1, mmr2, mmrw, smask) + +// some macros for pmaxub instruction +#define V_PMAXUB_MMX(mmr1, mmr2) \ + "psubusb "mmr2", "mmr1"\n\t" \ + "paddusb "mmr2", "mmr1"\n\t" +#define V_PMAXUB_SSE(mmr1, mmr2) "pmaxub "mmr2", "mmr1"\n\t" +#define V_PMAXUB_3DNOW(mmr1, mmr2) V_PMAXUB_MMX(mmr1, mmr2) // use MMX version +#define V_PMAXUB(mmr1, mmr2) V_PMAXUB2(mmr1, mmr2, SSE_TYPE) +#define V_PMAXUB2(mmr1, mmr2, ssetyp) V_PMAXUB3(mmr1, mmr2, ssetyp) +#define V_PMAXUB3(mmr1, mmr2, ssetyp) V_PMAXUB_##ssetyp(mmr1, mmr2) + +// some macros for pminub instruction +// V_PMINUB(mmr1, mmr2, mmr work register) mmr2 may NOT = mmrw +#define V_PMINUB_MMX(mmr1, mmr2, mmrw) \ + "pcmpeqb "mmrw", "mmrw"\n\t" \ + "psubusb "mmr2", "mmrw"\n\t" \ + "paddusb "mmrw", "mmr1"\n\t" \ + "psubusb "mmrw", "mmr1"\n\t" +#define V_PMINUB_SSE(mmr1, mmr2, mmrw) "pminub "mmr2", "mmr1"\n\t" +#define V_PMINUB_3DNOW(mmr1, mmr2, mmrw) V_PMINUB_MMX(mmr1, mmr2, mmrw) // use MMX version +#define V_PMINUB(mmr1, mmr2, mmrw) V_PMINUB2(mmr1, mmr2, mmrw, SSE_TYPE) +#define V_PMINUB2(mmr1, mmr2, mmrw, ssetyp) V_PMINUB3(mmr1, mmr2, mmrw, ssetyp) +#define V_PMINUB3(mmr1, mmr2, mmrw, ssetyp) V_PMINUB_##ssetyp(mmr1, mmr2, mmrw) + +// some macros for movntq instruction +// V_MOVNTQ(mmr1, mmr2) +#define V_MOVNTQ_MMX(mmr1, mmr2) "movq "mmr2", "mmr1"\n\t" +#define V_MOVNTQ_3DNOW(mmr1, mmr2) "movq "mmr2", "mmr1"\n\t" +#define V_MOVNTQ_SSE(mmr1, mmr2) "movntq "mmr2", "mmr1"\n\t" +#define V_MOVNTQ(mmr1, mmr2) V_MOVNTQ2(mmr1, mmr2, SSE_TYPE) +#define V_MOVNTQ2(mmr1, mmr2, ssetyp) V_MOVNTQ3(mmr1, mmr2, ssetyp) +#define V_MOVNTQ3(mmr1, mmr2, ssetyp) V_MOVNTQ_##ssetyp(mmr1, mmr2) + +// end of macros diff --git a/src/post/deinterlace/plugins/kdetv_greedyh.c b/src/post/deinterlace/plugins/kdetv_greedyh.c new file mode 100644 index 000000000..04fbbb82e --- /dev/null +++ b/src/post/deinterlace/plugins/kdetv_greedyh.c @@ -0,0 +1,129 @@ +/** + * Copyright (C) 2004 Billy Biggs <vektor@dumbterm.net> + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2 of the License, or (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the + * Free Software Foundation, Inc., 59 Temple Place - Suite 330, + * Boston, MA 02111-1307, USA. + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +#if HAVE_INTTYPES_H +#include <inttypes.h> +#else +#include <stdint.h> +#endif + +#include "attributes.h" +#include "xineutils.h" +#include "deinterlace.h" +#include "speedtools.h" +#include "speedy.h" + +#include "greedyhmacros.h" + +#define MAXCOMB_DEFAULT 5 +#define MOTIONTHRESHOLD_DEFAULT 25 +#define MOTIONSENSE_DEFAULT 30 + +static unsigned int GreedyMaxComb = MAXCOMB_DEFAULT; +static unsigned int GreedyMotionThreshold = MOTIONTHRESHOLD_DEFAULT; +static unsigned int GreedyMotionSense = MOTIONSENSE_DEFAULT; + + +#define IS_SSE +#define SSE_TYPE SSE +#define FUNCT_NAME greedyh_filter_sse +#include "greedyh.asm" +#undef SSE_TYPE +#undef IS_SSE +#undef FUNCT_NAME + +#define IS_3DNOW +#define FUNCT_NAME greedyh_filter_3dnow +#define SSE_TYPE 3DNOW +#include "greedyh.asm" +#undef SSE_TYPE +#undef IS_3DNOW +#undef FUNCT_NAME + +#define IS_MMX +#define SSE_TYPE MMX +#define FUNCT_NAME greedyh_filter_mmx +#include "greedyh.asm" +#undef SSE_TYPE +#undef IS_MMX +#undef FUNCT_NAME + + +static void deinterlace_frame_di_greedyh( uint8_t *output, int outstride, + deinterlace_frame_data_t *data, + int bottom_field, int second_field, + int width, int height ) +{ + if( xine_mm_accel() & MM_ACCEL_X86_MMXEXT ) { + greedyh_filter_sse( output, outstride, data, + bottom_field, second_field, + width, height ); + } else if( xine_mm_accel() & MM_ACCEL_X86_3DNOW ) { + greedyh_filter_3dnow( output, outstride, data, + bottom_field, second_field, + width, height ); + } else { + greedyh_filter_mmx( output, outstride, data, + bottom_field, second_field, + width, height ); + } +} + + +static deinterlace_method_t greedymethod = +{ + "Greedy - High Motion (DScaler)", + "GreedyH", + /* + "Motion Adaptive: Advanced Detection", + "AdaptiveAdvanced", + */ + 4, + MM_ACCEL_X86_MMX, + 0, + 0, + 0, + 0, + deinterlace_frame_di_greedyh, + 1, + { "Uses heuristics to detect motion in the input", + "frames and reconstruct image detail where", + "possible. Use this for high quality output", + "even on monitors set to an arbitrary refresh", + "rate.", + "", + "Advanced detection uses linear interpolation", + "where motion is detected, using a four-field", + "buffer. This is the Greedy: High Motion", + "deinterlacer from DScaler." } +}; + +deinterlace_method_t *dscaler_greedyh_get_method( void ) +{ + return &greedymethod; +} + diff --git a/src/post/deinterlace/plugins/x86-64_macros.inc b/src/post/deinterlace/plugins/x86-64_macros.inc new file mode 100644 index 000000000..3dfd9b63a --- /dev/null +++ b/src/post/deinterlace/plugins/x86-64_macros.inc @@ -0,0 +1,81 @@ +/*************************************************************************** + KdetvDScalerFilter_x86-64.inc + ----------------------------- + begin : Thu Sep 26 2004 + copyright : (C) 2004 by Dirk Ziegelmeier + email : dziegel@gmx.de + ***************************************************************************/ + +/* + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Library General Public + * License as published by the Free Software Foundation; either + * version 2 of the License, or (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Library General Public License for more details. + * + * You should have received a copy of the GNU Library General Public License + * along with this library; see the file COPYING.LIB. If not, write to + * the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + * Boston, MA 02111-1307, USA. + */ + +/* + * This file is copied from TVTIME's sources. + * Original author: Achim Schneider <batchall@mordor.ch> + */ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#ifndef XAX + +#if defined (ARCH_X86) + +#define XAX "eax" +#define XBX "ebx" +#define XCX "ecx" +#define XDX "edx" +#define XSI "esi" +#define XDI "edi" +#define XSP "esp" +#define MOVX "movl" +#define LEAX "leal" +#define DECX "decl" +#define PUSHX "pushl" +#define POPX "popl" +#define CMPX "cmpl" +#define ADDX "addl" +#define SHLX "shll" +#define SHRX "shrl" +#define SUBX "subl" + +#elif defined (ARCH_X86_64) + +#define XAX "rax" +#define XBX "rbx" +#define XCX "rcx" +#define XDX "rdx" +#define XSI "rsi" +#define XDI "rdi" +#define XSP "rsp" +#define MOVX "movq" +#define LEAX "leaq" +#define DECX "decq" +#define PUSHX "pushq" +#define POPX "popq" +#define CMPX "cmpq" +#define ADDX "addq" +#define SHLX "shlq" +#define SHRX "shrq" +#define SUBX "subq" + +#else +#error Undefined architecture. Define either ARCH_X86 or ARCH_X86_64. +#endif + +#endif diff --git a/src/post/deinterlace/xine_plugin.c b/src/post/deinterlace/xine_plugin.c index a7999585a..4fea21798 100644 --- a/src/post/deinterlace/xine_plugin.c +++ b/src/post/deinterlace/xine_plugin.c @@ -17,7 +17,7 @@ * along with this program; if not, write to the Free Software * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA * - * $Id: xine_plugin.c,v 1.39 2005/05/16 01:39:02 miguelfreitas Exp $ + * $Id: xine_plugin.c,v 1.40 2005/05/27 15:22:52 miguelfreitas Exp $ * * advanced video deinterlacer plugin * Jun/2003 by Miguel Freitas @@ -310,7 +310,7 @@ static void *deinterlace_init_plugin(xine_t *xine, void *data) register_deinterlace_method( double_get_method() ); register_deinterlace_method( vfir_get_method() ); register_deinterlace_method( scalerbob_get_method() ); - + register_deinterlace_method( dscaler_greedyh_get_method() ); filter_deinterlace_methods( config_flags, 5 /*fieldsavailable*/ ); if( !get_num_deinterlace_methods() ) { |