From 4f836b7732ccaf0421081cbe4198b97dabd60513 Mon Sep 17 00:00:00 2001 From: Martin Eikermann Date: Wed, 11 Jun 2008 11:12:49 +0000 Subject: gst/deinterlace2/: Add a deinterlacer plugin based on the tvtime/DScaler deinterlacer, which was relicensed to LGPL f... Original commit message from CVS: Based on a patch by: Martin Eikermann * gst/deinterlace2/Makefile.am: * gst/deinterlace2/gstdeinterlace2.c: (gst_deinterlace2_method_get_type), (gst_deinterlace2_fields_get_type), (gst_deinterlace2_field_layout_get_type), (gst_deinterlace2_base_init), (gst_deinterlace2_class_init), (gst_deinterlace2_init), (gst_deinterlace2_set_method), (gst_deinterlace2_set_property), (gst_deinterlace2_get_property), (gst_deinterlace2_finalize), (gst_deinterlace2_pop_history), (gst_deinterlace2_head_history), (gst_deinterlace2_push_history), (gst_deinterlace2_deinterlace_scanlines), (gst_deinterlace2_chain), (gst_deinterlace2_setcaps), (gst_deinterlace2_sink_event), (gst_deinterlace2_change_state), (gst_deinterlace2_src_event), (gst_deinterlace2_src_query), (gst_deinterlace2_src_query_types), (plugin_init): * gst/deinterlace2/gstdeinterlace2.h: * gst/deinterlace2/tvtime/greedy.c: (copy_scanline), (deinterlace_greedy_packed422_scanline_mmxext), (dscaler_greedyl_get_method): * gst/deinterlace2/tvtime/greedyh.asm: * gst/deinterlace2/tvtime/greedyh.c: (deinterlace_frame_di_greedyh), (dscaler_greedyh_get_method), (greedyh_init), (greedyh_filter_mmx), (greedyh_filter_3dnow), (greedyh_filter_sse): * gst/deinterlace2/tvtime/greedyh.h: * gst/deinterlace2/tvtime/greedyhmacros.h: * gst/deinterlace2/tvtime/mmx.h: * gst/deinterlace2/tvtime/plugins.h: * gst/deinterlace2/tvtime/speedtools.h: * gst/deinterlace2/tvtime/speedy.c: (multiply_alpha), (clip255), (comb_factor_packed422_scanline_mmx), (diff_factor_packed422_scanline_c), (diff_factor_packed422_scanline_mmx), (diff_packed422_block8x8_mmx), (diff_packed422_block8x8_c), (packed444_to_packed422_scanline_c), (packed422_to_packed444_scanline_c), (packed422_to_packed444_rec601_scanline_c), (vfilter_chroma_121_packed422_scanline_mmx), (vfilter_chroma_121_packed422_scanline_c), (vfilter_chroma_332_packed422_scanline_mmx), (vfilter_chroma_332_packed422_scanline_c), (kill_chroma_packed422_inplace_scanline_mmx), (kill_chroma_packed422_inplace_scanline_c), (invert_colour_packed422_inplace_scanline_mmx), (invert_colour_packed422_inplace_scanline_c), (mirror_packed422_inplace_scanline_c), (interpolate_packed422_scanline_c), (convert_uyvy_to_yuyv_scanline_mmx), (convert_uyvy_to_yuyv_scanline_c), (interpolate_packed422_scanline_mmx), (interpolate_packed422_scanline_mmxext), (blit_colour_packed422_scanline_c), (blit_colour_packed422_scanline_mmx), (blit_colour_packed422_scanline_mmxext), (blit_colour_packed4444_scanline_c), (blit_colour_packed4444_scanline_mmx), (blit_colour_packed4444_scanline_mmxext), (small_memcpy), (speedy_memcpy_c), (speedy_memcpy_mmx), (speedy_memcpy_mmxext), (blit_packed422_scanline_c), (blit_packed422_scanline_mmx), (blit_packed422_scanline_mmxext), (composite_colour4444_alpha_to_packed422_scanline_c), (composite_colour4444_alpha_to_packed422_scanline_mmxext), (composite_packed4444_alpha_to_packed422_scanline_c), (composite_packed4444_alpha_to_packed422_scanline_mmxext), (composite_packed4444_to_packed422_scanline_c), (composite_packed4444_to_packed422_scanline_mmxext), (composite_alphamask_to_packed4444_scanline_c), (composite_alphamask_to_packed4444_scanline_mmxext), (composite_alphamask_alpha_to_packed4444_scanline_c), (premultiply_packed4444_scanline_c), (premultiply_packed4444_scanline_mmxext), (blend_packed422_scanline_c), (blend_packed422_scanline_mmxext), (quarter_blit_vertical_packed422_scanline_mmxext), (quarter_blit_vertical_packed422_scanline_c), (subpix_blit_vertical_packed422_scanline_c), (a8_subpix_blit_scanline_c), (myround), (init_RGB_to_YCbCr_tables), (init_YCbCr_to_RGB_tables), (rgb24_to_packed444_rec601_scanline_c), (rgba32_to_packed4444_rec601_scanline_c), (packed444_to_rgb24_rec601_scanline_c), (packed444_to_nonpremultiplied_packed4444_scanline_c), (aspect_adjust_packed4444_scanline_c), (setup_speedy_calls), (speedy_get_accel): * gst/deinterlace2/tvtime/speedy.h: * gst/deinterlace2/tvtime/sse.h: * gst/deinterlace2/tvtime/tomsmocomp.c: (Fieldcopy), (deinterlace_frame_di_tomsmocomp), (dscaler_tomsmocomp_get_method), (tomsmocomp_init), (tomsmocomp_filter_mmx), (tomsmocomp_filter_3dnow), (tomsmocomp_filter_sse): * gst/deinterlace2/tvtime/tomsmocomp.h: * gst/deinterlace2/tvtime/tomsmocomp/SearchLoop0A.inc: * gst/deinterlace2/tvtime/tomsmocomp/SearchLoopBottom.inc: * gst/deinterlace2/tvtime/tomsmocomp/SearchLoopEdgeA.inc: * gst/deinterlace2/tvtime/tomsmocomp/SearchLoopEdgeA8.inc: * gst/deinterlace2/tvtime/tomsmocomp/SearchLoopOddA.inc: * gst/deinterlace2/tvtime/tomsmocomp/SearchLoopOddA2.inc: * gst/deinterlace2/tvtime/tomsmocomp/SearchLoopOddA6.inc: * gst/deinterlace2/tvtime/tomsmocomp/SearchLoopOddAH.inc: * gst/deinterlace2/tvtime/tomsmocomp/SearchLoopOddAH2.inc: * gst/deinterlace2/tvtime/tomsmocomp/SearchLoopTop.inc: * gst/deinterlace2/tvtime/tomsmocomp/SearchLoopVA.inc: * gst/deinterlace2/tvtime/tomsmocomp/SearchLoopVAH.inc: * gst/deinterlace2/tvtime/tomsmocomp/StrangeBob.inc: * gst/deinterlace2/tvtime/tomsmocomp/TomsMoCompAll.inc: * gst/deinterlace2/tvtime/tomsmocomp/TomsMoCompAll2.inc: * gst/deinterlace2/tvtime/tomsmocomp/WierdBob.inc: * gst/deinterlace2/tvtime/vfir.c: (deinterlace_line), (deinterlace_scanline_vfir), (copy_scanline), (dscaler_vfir_get_method): * gst/deinterlace2/tvtime/x86-64_macros.inc: Add a deinterlacer plugin based on the tvtime/DScaler deinterlacer, which was relicensed to LGPL for GStreamer and in theory provides better and faster results than the simple deinterlace element. Fixes bug #163578. Ported to GStreamer 0.10 but still not enabled or included in the build system by default because of bad artefacts caused by a bug somewhere and as it can be only build on x86/amd64 ATM and requires special CFLAGS. Will be fixed soon. --- .../tvtime/tomsmocomp/SearchLoopTop.inc | 193 +++++++++++++++++++++ 1 file changed, 193 insertions(+) create mode 100644 gst/deinterlace2/tvtime/tomsmocomp/SearchLoopTop.inc (limited to 'gst/deinterlace2/tvtime/tomsmocomp/SearchLoopTop.inc') diff --git a/gst/deinterlace2/tvtime/tomsmocomp/SearchLoopTop.inc b/gst/deinterlace2/tvtime/tomsmocomp/SearchLoopTop.inc new file mode 100644 index 00000000..7560f404 --- /dev/null +++ b/gst/deinterlace2/tvtime/tomsmocomp/SearchLoopTop.inc @@ -0,0 +1,193 @@ +// -*- c++ -*- + +unsigned char* pDest; +const unsigned char* pSrcP; +const unsigned char* pSrc; +const unsigned char* pBob; +const unsigned char* pBobP; + +int64_t Max_Mov = 0x0404040404040404ull; +int64_t DiffThres = 0x0f0f0f0f0f0f0f0full; +int64_t YMask = 0x00ff00ff00ff00ffull; // keeps only luma +int64_t UVMask = 0xff00ff00ff00ff00ull; // keeps only chroma +int64_t TENS = 0x0a0a0a0a0a0a0a0aull; +int64_t FOURS = 0x0404040404040404ull; +int64_t ONES = 0x0101010101010101ull; +int64_t Min_Vals = 0x0000000000000000ull; +int64_t Max_Vals = 0x0000000000000000ull; +int64_t ShiftMask = 0xfefffefffefffeffull; + +// long is int32 on ARCH_368, int64 on ARCH_AMD64. Declaring it this way +// saves a lot of xor's to delete 64bit garbage. + +#if defined(DBL_RESIZE) || defined(USE_FOR_DSCALER) +long src_pitch2 = src_pitch; // even & odd lines are not interleaved in DScaler +#else +long src_pitch2 = 2 * src_pitch; // even & odd lines are interleaved in Avisynth +#endif + + +long dst_pitch2 = 2 * dst_pitch; +long y; + +#ifdef IS_SSE2 +long Last8 = (rowsize-16); // ofs to last 16 bytes in row for SSE2 +#else +long Last8 = (rowsize-8); // ofs to last 8 bytes in row +#endif + +long dst_pitchw = dst_pitch; // local stor so asm can ref + pSrc = pWeaveSrc; // points 1 weave line above + pSrcP = pWeaveSrcP; // " + +#ifdef DBL_RESIZE + +#ifdef USE_VERTICAL_FILTER + pDest = pWeaveDest + dst_pitch2; +#else + pDest = pWeaveDest + 3*dst_pitch; +#endif + +#else + +#ifdef USE_VERTICAL_FILTER + pDest = pWeaveDest + dst_pitch; +#else + pDest = pWeaveDest + dst_pitch2; +#endif + +#endif + + if (TopFirst) + { + pBob = pCopySrc + src_pitch2; // remember one weave line just copied previously + pBobP = pCopySrcP + src_pitch2; + } + else + { + pBob = pCopySrc; + pBobP = pCopySrcP; + } + +#ifndef _pBob +#define _pBob "%0" +#define _src_pitch2 "%1" +#define _ShiftMask "%2" +#define _pDest "%3" +#define _dst_pitchw "%4" +#define _Last8 "%5" +#define _pSrc "%6" +#define _pSrcP "%7" +#define _pBobP "%8" +#define _DiffThres "%9" +#define _Min_Vals "%10" +#define _Max_Vals "%11" +#define _FOURS "%12" +#define _TENS "%13" +#define _ONES "%14" +#define _UVMask "%15" +#define _Max_Mov "%16" +#define _YMask "%17" +#define _oldbx "%18" +#endif + + long oldbx; + + for (y=1; y < FldHeight-1; y++) + { + // pretend it's indented -->> + __asm__ __volatile__ + ( + // Loop general reg usage + // + // XAX - pBobP, then pDest + // XBX - pBob + // XCX - src_pitch2 + // XDX - current offset + // XDI - prev weave pixels, 1 line up + // XSI - next weave pixels, 1 line up + + // Save "XBX" (-fPIC) + MOVX" %%"XBX", "_oldbx"\n\t" + +#ifdef IS_SSE2 + + // sse2 code deleted for now + +#else + // simple bob first 8 bytes + MOVX" "_pBob", %%"XBX"\n\t" + MOVX" "_src_pitch2", %%"XCX"\n\t" + +#ifdef USE_VERTICAL_FILTER + "movq (%%"XBX"), %%mm0\n\t" + "movq (%%"XBX", %%"XCX"), %%mm1\n\t" //, qword ptr["XBX"+"XCX"] + "movq %%mm0, %%mm2\n\t" + V_PAVGB ("%%mm2", "%%mm1", "%%mm3", _ShiftMask) // halfway between + V_PAVGB ("%%mm0", "%%mm2", "%%mm3", _ShiftMask) // 1/4 way + V_PAVGB ("%%mm1", "%%mm2", "%%mm3", _ShiftMask) // 3/4 way + MOVX" "_pDest", %%"XDI"\n\t" + MOVX" "_dst_pitchw", %%"XAX"\n\t" + V_MOVNTQ ("(%%"XDI")", "%%mm0") + V_MOVNTQ ("(%%"XDI", %%"XAX")", "%%mm1") // qword ptr["XDI"+"XAX"], mm1 + + // simple bob last 8 bytes + MOVX" "_Last8", %%"XDX"\n\t" + LEAX" (%%"XBX", %%"XDX"), %%"XSI"\n\t" // ["XBX"+"XDX"] + "movq (%%"XSI"), %%mm0\n\t" + "movq (%%"XSI", %%"XCX"), %%mm1\n\t" // qword ptr["XSI"+"XCX"] + "movq %%mm0, %%mm2\n\t" + V_PAVGB ("%%mm2", "%%mm1", "%%mm3", _ShiftMask) // halfway between + V_PAVGB ("%%mm0", "%%mm2", "%%mm3", _ShiftMask) // 1/4 way + V_PAVGB ("%%mm1", "%%mm2", "%%mm3", _ShiftMask) // 3/4 way + ADDX" %%"XDX", %%"XDI"\n\t" // last 8 bytes of dest + V_MOVNTQ ("%%"XDI"", "%%mm0") + V_MOVNTQ ("(%%"XDI", %%"XAX")", "%%mm1") // qword ptr["XDI"+"XAX"], mm1) + +#else + "movq (%%"XBX"), %%mm0\n\t" + // pavgb mm0, qword ptr["XBX"+"XCX"] + V_PAVGB ("%%mm0", "(%%"XBX", %%"XCX")", "%%mm2", _ShiftMask) // qword ptr["XBX"+"XCX"], mm2, ShiftMask) + MOVX" "_pDest", %%"XDI"\n\t" + V_MOVNTQ ("(%%"XDI")", "%%mm0") + + // simple bob last 8 bytes + MOVX" "_Last8", %%"XDX"\n\t" + LEAX" (%%"XBX", %%"XDX"), %%"XSI"\n\t" //"XSI", ["XBX"+"XDX"] + "movq (%%"XSI"), %%mm0\n\t" + // pavgb mm0, qword ptr["XSI"+"XCX"] + V_PAVGB ("%%mm0", "(%%"XSI", %%"XCX")", "%%mm2", _ShiftMask) // qword ptr["XSI"+"XCX"], mm2, ShiftMask) + V_MOVNTQ ("(%%"XDI", %%"XDX")", "%%mm0") // qword ptr["XDI"+"XDX"], mm0) +#endif + // now loop and get the middle qwords + MOVX" "_pSrc", %%"XSI"\n\t" + MOVX" "_pSrcP", %%"XDI"\n\t" + MOVX" $8, %%"XDX"\n\t" // curr offset longo all lines + + "1:\n\t" + MOVX" "_pBobP", %%"XAX"\n\t" + ADDX" $8, %%"XDI"\n\t" + ADDX" $8, %%"XSI"\n\t" + ADDX" $8, %%"XBX"\n\t" + ADDX" %%"XDX", %%"XAX"\n\t" + +#ifdef USE_STRANGE_BOB +#include "StrangeBob.inc" +#else +#include "WierdBob.inc" +#endif + + // For non-SSE2: + // through out most of the rest of this loop we will maintain + // mm4 our min bob value + // mm5 best weave pixels so far + // mm6 our max Bob value + // mm7 best weighted pixel ratings so far + + // We will keep a slight bias to using the weave pixels + // from the current location, by rating them by the min distance + // from the Bob value instead of the avg distance from that value. + // our best and only rating so far + "pcmpeqb %%mm7, %%mm7\n\t" // ffff, say we didn't find anything good yet + +#endif -- cgit v1.2.1