gstreamer/gst/deinterlace2/tvtime/tomsmocomp/WierdBob.inc

190 lines
6.6 KiB
PHP
Raw Normal View History

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 <meiker at upb dot de> * 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.
2008-06-11 11:12:49 +00:00
// -*- c++ -*-
// First, get and save our possible Bob values
// Assume our pixels are layed out as follows with x the calc'd bob value
// and the other pixels are from the current field
//
// j a b c k current field
// x calculated line
// m d e f n current field
//
// we calc the bob value as:
// x2 = either avg(a,f), avg(c,d), avg(b,e), avg(j,n), or avg(k,m)
// selected for the smallest of abs(a,f), abs(c,d), or abs(b,e), etc.
// a,f
"movq -2(%%"XBX"), %%mm0\n\t" // value a from top left
"movq 2(%%"XBX", %%"XCX"), %%mm1\n\t" // value f from bottom right
"movq %%mm0, %%mm6\n\t"
// pavgb %%mm6, %%mm1 // avg(a,f), also best so far
V_PAVGB ("%%mm6", "%%mm1", "%%mm7", _ShiftMask) // avg(a,f), also best so far
"movq %%mm0, %%mm7\n\t"
"psubusb %%mm1, %%mm7\n\t"
"psubusb %%mm0, %%mm1\n\t"
"por %%mm1, %%mm7\n\t" // abs diff, also best so far
// c,d
"movq 2(%%"XBX"), %%mm0\n\t" // value a from top left
"movq -2(%%"XBX", %%"XCX"), %%mm1\n\t" // value f from bottom right
"movq %%mm0, %%mm2\n\t"
// pavgb %%mm2, %%mm1 // avg(c,d)
V_PAVGB ("%%mm2", "%%mm1", "%%mm3", _ShiftMask) // avg(c,d)
"movq %%mm0, %%mm3\n\t"
"psubusb %%mm1, %%mm3\n\t"
"psubusb %%mm0, %%mm1\n\t"
"por %%mm1, %%mm3\n\t" // abs(c,d)
"movq %%mm3, %%mm1\n\t" // keep copy
"psubusb %%mm7, %%mm3\n\t" // nonzero where new weights bigger, else 0
"pxor %%mm4, %%mm4\n\t"
"pcmpeqb %%mm4, %%mm3\n\t" // now ff where new better, else 00
"pcmpeqb %%mm3, %%mm4\n\t" // here ff where old better, else 00
"pand %%mm3, %%mm1\n\t" // keep only better new avg and abs
"pand %%mm3, %%mm2\n\t"
"pand %%mm4, %%mm6\n\t"
"pand %%mm4, %%mm7\n\t"
"por %%mm2, %%mm6\n\t" // and merge new & old vals keeping best
"por %%mm1, %%mm7\n\t"
"por "_UVMask", %%mm7\n\t" // but we know chroma is worthless so far
"pand "_YMask", %%mm5\n\t" // mask out chroma from here also
// j,n
"movq -4(%%"XBX"), %%mm0\n\t" // value j from top left
"movq 4(%%"XBX", %%"XCX"), %%mm1\n\t" // value n from bottom right
"movq %%mm0, %%mm2\n\t"
// pavgb %%mm2, %%mm1 // avg(j,n)
V_PAVGB ("%%mm2", "%%mm1", "%%mm3", _ShiftMask) // avg(j,n)
"movq %%mm0, %%mm3\n\t"
"psubusb %%mm1, %%mm3\n\t"
"psubusb %%mm0, %%mm1\n\t"
"por %%mm1, %%mm3\n\t" // abs(j-n)
"movq %%mm3, %%mm1\n\t" // keep copy
"psubusb %%mm7, %%mm3\n\t" // nonzero where new weights bigger, else 0
"pxor %%mm4, %%mm4\n\t"
"pcmpeqb %%mm4, %%mm3\n\t" // now ff where new better, else 00
"pcmpeqb %%mm3, %%mm4\n\t" // here ff where old better, else 00
"pand %%mm3, %%mm1\n\t" // keep only better new avg and abs
"pand %%mm2, %%mm3\n\t"
"pand %%mm4, %%mm6\n\t"
"pand %%mm4, %%mm7\n\t"
"por %%mm3, %%mm6\n\t" // and merge new & old vals keeping best
"por %%mm1, %%mm7\n\t" // "
// k, m
"movq 4(%%"XBX"), %%mm0\n\t" // value k from top right
"movq -4(%%"XBX", %%"XCX"), %%mm1\n\t" // value n from bottom left
"movq %%mm0, %%mm4\n\t"
// pavgb %%mm4, %%mm1 // avg(k,m)
V_PAVGB ("%%mm4", "%%mm1", "%%mm3", _ShiftMask) // avg(k,m)
"movq %%mm0, %%mm3\n\t"
"psubusb %%mm1, %%mm3\n\t"
"psubusb %%mm0, %%mm1\n\t"
"por %%mm1, %%mm3\n\t" // abs(k,m)
"movq %%mm3, %%mm1\n\t" // keep copy
"movq %%mm4, %%mm2\n\t" // avg(k,m)
"psubusb %%mm7, %%mm3\n\t" // nonzero where new weights bigger, else 0
"pxor %%mm4, %%mm4\n\t"
"pcmpeqb %%mm4, %%mm3\n\t" // now ff where new better, else 00
"pcmpeqb %%mm3, %%mm4\n\t" // here ff where old better, else 00
"pand %%mm3, %%mm1\n\t" // keep only better new avg and abs
"pand %%mm2, %%mm3\n\t"
"pand %%mm4, %%mm6\n\t"
"pand %%mm4, %%mm7\n\t"
"por %%mm3, %%mm6\n\t" // and merge new & old vals keeping best
"por %%mm1, %%mm7\n\t" // "
// b,e
"movq (%%"XBX"), %%mm0\n\t" // value b from top
"movq (%%"XBX", %%"XCX"), %%mm1\n\t" // value e from bottom
// We will also calc here the max/min values to later limit comb
// so the max excursion will not exceed the Max_Comb constant
#ifdef SKIP_SEARCH
"movq %%mm0, %%mm2\n\t"
// pminub %%mm2, %%mm1
V_PMINUB ("%%mm2", "%%mm1", "%%mm4")
// pmaxub %%mm6, %%mm2 // clip our current results so far to be above this
V_PMAXUB ("%%mm6", "%%mm2")
"movq %%mm0, %%mm2\n\t"
V_PMAXUB ("%%mm2", "%%mm1")
// pminub %%mm6, %%mm2 // clip our current results so far to be below this
V_PMINUB ("%%mm6", "%%mm2", "%%mm4")
#else
"movq %%mm0, %%mm2\n\t"
"movq (%%"XAX"), %%mm4\n\t"
"psubusb %%mm4, %%mm2\n\t"
"psubusb %%mm0, %%mm4\n\t"
"por %%mm2, %%mm4\n\t" // abs diff
"movq %%mm1, %%mm2\n\t"
"movq (%%"XAX", %%"XCX"), %%mm3\n\t"
"psubusb %%mm3, %%mm2\n\t"
"psubusb %%mm1, %%mm3\n\t"
"por %%mm2, %%mm3\n\t" // abs diff
// pmaxub %%mm3, %%mm4 // top or bottom pixel moved most
V_PMAXUB ("%%mm3", "%%mm4") // top or bottom pixel moved most
"psubusb "_Max_Mov", %%mm3\n\t" // moved more than allowed? or goes to 0?
"pxor %%mm4, %%mm4\n\t"
"pcmpeqb %%mm4, %%mm3\n\t" // now ff where low motion, else high motion
"movq %%mm0, %%mm2\n\t"
// pminub %%mm2, %%mm1
V_PMINUB ("%%mm2", "%%mm1", "%%mm4")
// pmaxub %%mm6, %%mm2 // clip our current results so far to be above this
V_PMAXUB ("%%mm6", "%%mm2")
"psubusb %%mm3, %%mm2\n\t" // maybe decrease it to 0000.. if no surround motion
"movq %%mm2, "_Min_Vals"\n\t"
"movq %%mm0, %%mm2\n\t"
V_PMAXUB ("%%mm2", "%%mm1")
// pminub %%mm6, %%mm2 // clip our current results so far to be below this
V_PMINUB ("%%mm6", "%%mm2", "%%mm4")
"paddusb %%mm3, %%mm2\n\t" // maybe increase it to ffffff if no surround motion
"movq %%mm2, "_Max_Vals"\n\t"
#endif
"movq %%mm0, %%mm2\n\t"
// pavgb %%mm2, %%mm1 // avg(b,e)
V_PAVGB ("%%mm2", "%%mm1", "%%mm3", _ShiftMask) // avg(b,e)
"movq %%mm0, %%mm3\n\t"
"psubusb %%mm1, %%mm3\n\t"
"psubusb %%mm0, %%mm1\n\t"
"por %%mm1, %%mm3\n\t" // abs(c,d)
"movq %%mm3, %%mm1\n\t" // keep copy of diffs
"pxor %%mm4, %%mm4\n\t"
"psubusb %%mm7, %%mm3\n\t" // nonzero where new weights bigger, else 0
"pcmpeqb %%mm4, %%mm3\n\t" // now ff where new better, else 00
"pcmpeqb %%mm3, %%mm4\n\t" // here ff where old better, else 00
"pand %%mm3, %%mm1\n\t"
"pand %%mm3, %%mm2\n\t"
"pand %%mm4, %%mm6\n\t"
"pand %%mm4, %%mm7\n\t"
"por %%mm2, %%mm6\n\t" // our x2 value
"por %%mm1, %%mm7\n\t" // our x2 diffs
"movq %%mm7, %%mm4\n\t" // save as bob uncertainty indicator