From b7a7830dd41fa68c1b1143f4d94501f4648fa373 Mon Sep 17 00:00:00 2001 From: Stefano Sabatini Date: Sun, 18 Mar 2012 10:59:01 +0100 Subject: [PATCH] lavfi: port 2xsai filter from libmpcodecs Partially based on the port by Niel van der Westhuizen , done for GCI 2010. Same output as the original filter and as fast. See thread: Subject: [FFmpeg-devel] [PATCH] Port MPlayer 2xSaI filter to libavfilter Date: Thu, 25 Nov 2010 01:31:24 +1000 --- Changelog | 1 + configure | 1 + doc/filters.texi | 7 + libavfilter/Makefile | 1 + libavfilter/allfilters.c | 1 + libavfilter/version.h | 2 +- libavfilter/vf_super2xsai.c | 350 ++++++++++++++++++++++++++++++++++++ 7 files changed, 362 insertions(+), 1 deletion(-) create mode 100644 libavfilter/vf_super2xsai.c diff --git a/Changelog b/Changelog index dbd98bf48e..e0584b7a9b 100644 --- a/Changelog +++ b/Changelog @@ -25,6 +25,7 @@ version next: - drop support for ffmpeg without libavfilter - drawtext video filter: fontconfig support - ffmpeg -benchmark_all option +- super2xsai filter ported from libmpcodecs version 0.10: diff --git a/configure b/configure index 419d2526e1..3c0cd7b47c 100755 --- a/configure +++ b/configure @@ -1692,6 +1692,7 @@ ocv_filter_deps="libopencv" pan_filter_deps="swresample" removelogo_filter_deps="avcodec avformat swscale" scale_filter_deps="swscale" +super2xsai_filter_deps="gpl" tinterlace_filter_deps="gpl" yadif_filter_deps="gpl" diff --git a/doc/filters.texi b/doc/filters.texi index d16644582b..daf93e7210 100644 --- a/doc/filters.texi +++ b/doc/filters.texi @@ -2716,6 +2716,13 @@ For example: will create two separate outputs from the same input, one cropped and one padded. +@section super2xsai + +Scale the input by 2x and smooth using the Super2xSaI (Scale and +Interpolate) pixel art scaling algorithm. + +Useful for enlarging pixel art images without reducing sharpness. + @section swapuv Swap U & V plane. diff --git a/libavfilter/Makefile b/libavfilter/Makefile index 5f99fe65d1..ad65c7b279 100644 --- a/libavfilter/Makefile +++ b/libavfilter/Makefile @@ -94,6 +94,7 @@ OBJS-$(CONFIG_SETTB_FILTER) += vf_settb.o OBJS-$(CONFIG_SHOWINFO_FILTER) += vf_showinfo.o OBJS-$(CONFIG_SLICIFY_FILTER) += vf_slicify.o OBJS-$(CONFIG_SPLIT_FILTER) += vf_split.o +OBJS-$(CONFIG_SUPER2XSAI_FILTER) += vf_super2xsai.o OBJS-$(CONFIG_SWAPUV_FILTER) += vf_swapuv.o OBJS-$(CONFIG_THUMBNAIL_FILTER) += vf_thumbnail.o OBJS-$(CONFIG_TILE_FILTER) += vf_tile.o diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c index b2aef7bb22..5241fc09ad 100644 --- a/libavfilter/allfilters.c +++ b/libavfilter/allfilters.c @@ -99,6 +99,7 @@ void avfilter_register_all(void) REGISTER_FILTER (SHOWINFO, showinfo, vf); REGISTER_FILTER (SLICIFY, slicify, vf); REGISTER_FILTER (SPLIT, split, vf); + REGISTER_FILTER (SUPER2XSAI, super2xsai, vf); REGISTER_FILTER (SWAPUV, swapuv, vf); REGISTER_FILTER (THUMBNAIL, thumbnail, vf); REGISTER_FILTER (TILE, tile, vf); diff --git a/libavfilter/version.h b/libavfilter/version.h index 66d6552c76..1179f3fdf5 100644 --- a/libavfilter/version.h +++ b/libavfilter/version.h @@ -29,7 +29,7 @@ #include "libavutil/avutil.h" #define LIBAVFILTER_VERSION_MAJOR 2 -#define LIBAVFILTER_VERSION_MINOR 68 +#define LIBAVFILTER_VERSION_MINOR 69 #define LIBAVFILTER_VERSION_MICRO 100 #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \ diff --git a/libavfilter/vf_super2xsai.c b/libavfilter/vf_super2xsai.c new file mode 100644 index 0000000000..4406a4396e --- /dev/null +++ b/libavfilter/vf_super2xsai.c @@ -0,0 +1,350 @@ +/* + * Copyright (c) 2010 Niel van der Westhuizen + * Copyright (c) 2002 A'rpi + * Copyright (c) 1997-2001 ZSNES Team ( zsknight@zsnes.com / _demo_@zsnes.com ) + * + * This file is part of FFmpeg. + * + * FFmpeg 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. + * + * FFmpeg 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 FFmpeg; if not, write to the Free Software Foundation, Inc., + * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. + */ + +/** + * @file + * Super 2xSaI video filter + * Ported from MPlayer libmpcodecs/vf_2xsai.c. + */ + +#include "libavutil/pixdesc.h" +#include "libavutil/intreadwrite.h" +#include "avfilter.h" + +typedef struct { + /* masks used for two pixels interpolation */ + uint32_t hi_pixel_mask; + uint32_t lo_pixel_mask; + + /* masks used for four pixels interpolation */ + uint32_t q_hi_pixel_mask; + uint32_t q_lo_pixel_mask; + + int bpp; ///< bytes per pixel, pixel stride for each (packed) pixel +} Super2xSaIContext; + +#define GET_RESULT(A, B, C, D) ((A != C || A != D) - (B != C || B != D)) + +#define INTERPOLATE(A, B) (((A & hi_pixel_mask) >> 1) + ((B & hi_pixel_mask) >> 1) + (A & B & lo_pixel_mask)) + +#define Q_INTERPOLATE(A, B, C, D) ((A & q_hi_pixel_mask) >> 2) + ((B & q_hi_pixel_mask) >> 2) + ((C & q_hi_pixel_mask) >> 2) + ((D & q_hi_pixel_mask) >> 2) \ + + ((((A & q_lo_pixel_mask) + (B & q_lo_pixel_mask) + (C & q_lo_pixel_mask) + (D & q_lo_pixel_mask)) >> 2) & q_lo_pixel_mask) + +static void super2xsai(AVFilterContext *ctx, + uint8_t *src, int src_linesize, + uint8_t *dst, int dst_linesize, + int width, int height) +{ + Super2xSaIContext *sai = ctx->priv; + unsigned int x, y; + uint32_t color[4][4]; + unsigned char *src_line[4]; + const int bpp = sai->bpp; + const uint32_t hi_pixel_mask = sai->hi_pixel_mask; + const uint32_t lo_pixel_mask = sai->lo_pixel_mask; + const uint32_t q_hi_pixel_mask = sai->q_hi_pixel_mask; + const uint32_t q_lo_pixel_mask = sai->q_lo_pixel_mask; + + /* Point to the first 4 lines, first line is duplicated */ + src_line[0] = src; + src_line[1] = src; + src_line[2] = src + src_linesize*FFMIN(1, height-1); + src_line[3] = src + src_linesize*FFMIN(2, height-1); + +#define READ_COLOR4(dst, src_line, off) dst = *((const uint32_t *)src_line + off) +#define READ_COLOR3(dst, src_line, off) dst = AV_RL24 (src_line + 3*off) +#define READ_COLOR2(dst, src_line, off) dst = *((const uint16_t *)src_line + off) + + /* Initialise the color matrix for this row. */ + switch (bpp) { + case 4: + READ_COLOR4(color[0][0], src_line[0], 0); color[0][1] = color[0][0]; READ_COLOR4(color[0][2], src_line[0], 1); READ_COLOR4(color[0][3], src_line[0], 2); + READ_COLOR4(color[1][0], src_line[1], 0); color[1][1] = color[1][0]; READ_COLOR4(color[1][2], src_line[1], 1); READ_COLOR4(color[1][3], src_line[1], 2); + READ_COLOR4(color[2][0], src_line[2], 0); color[2][1] = color[2][0]; READ_COLOR4(color[2][2], src_line[2], 1); READ_COLOR4(color[2][3], src_line[2], 2); + READ_COLOR4(color[3][0], src_line[3], 0); color[3][1] = color[3][0]; READ_COLOR4(color[3][2], src_line[3], 1); READ_COLOR4(color[3][3], src_line[3], 2); + break; + case 3: + READ_COLOR3(color[0][0], src_line[0], 0); color[0][1] = color[0][0]; READ_COLOR3(color[0][2], src_line[0], 1); READ_COLOR3(color[0][3], src_line[0], 2); + READ_COLOR3(color[1][0], src_line[1], 0); color[1][1] = color[1][0]; READ_COLOR3(color[1][2], src_line[1], 1); READ_COLOR3(color[1][3], src_line[1], 2); + READ_COLOR3(color[2][0], src_line[2], 0); color[2][1] = color[2][0]; READ_COLOR3(color[2][2], src_line[2], 1); READ_COLOR3(color[2][3], src_line[2], 2); + READ_COLOR3(color[3][0], src_line[3], 0); color[3][1] = color[3][0]; READ_COLOR3(color[3][2], src_line[3], 1); READ_COLOR3(color[3][3], src_line[3], 2); + break; + default: + READ_COLOR2(color[0][0], src_line[0], 0); color[0][1] = color[0][0]; READ_COLOR2(color[0][2], src_line[0], 1); READ_COLOR2(color[0][3], src_line[0], 2); + READ_COLOR2(color[1][0], src_line[1], 0); color[1][1] = color[1][0]; READ_COLOR2(color[1][2], src_line[1], 1); READ_COLOR2(color[1][3], src_line[1], 2); + READ_COLOR2(color[2][0], src_line[2], 0); color[2][1] = color[2][0]; READ_COLOR2(color[2][2], src_line[2], 1); READ_COLOR2(color[2][3], src_line[2], 2); + READ_COLOR2(color[3][0], src_line[3], 0); color[3][1] = color[3][0]; READ_COLOR2(color[3][2], src_line[3], 1); READ_COLOR2(color[3][3], src_line[3], 2); + } + + for (y = 0; y < height; y++) { + uint8_t *dst_line[2]; + + dst_line[0] = dst + dst_linesize*2*y; + dst_line[1] = dst + dst_linesize*(2*y+1); + + for (x = 0; x < width; x++) { + uint32_t product1a, product1b, product2a, product2b; + +//--------------------------------------- B0 B1 B2 B3 0 1 2 3 +// 4 5* 6 S2 -> 4 5* 6 7 +// 1 2 3 S1 8 9 10 11 +// A0 A1 A2 A3 12 13 14 15 +//-------------------------------------- + if (color[2][1] == color[1][2] && color[1][1] != color[2][2]) { + product2b = color[2][1]; + product1b = product2b; + } else if (color[1][1] == color[2][2] && color[2][1] != color[1][2]) { + product2b = color[1][1]; + product1b = product2b; + } else if (color[1][1] == color[2][2] && color[2][1] == color[1][2]) { + int r = 0; + + r += GET_RESULT(color[1][2], color[1][1], color[1][0], color[3][1]); + r += GET_RESULT(color[1][2], color[1][1], color[2][0], color[0][1]); + r += GET_RESULT(color[1][2], color[1][1], color[3][2], color[2][3]); + r += GET_RESULT(color[1][2], color[1][1], color[0][2], color[1][3]); + + if (r > 0) + product1b = color[1][2]; + else if (r < 0) + product1b = color[1][1]; + else + product1b = INTERPOLATE(color[1][1], color[1][2]); + + product2b = product1b; + } else { + if (color[1][2] == color[2][2] && color[2][2] == color[3][1] && color[2][1] != color[3][2] && color[2][2] != color[3][0]) + product2b = Q_INTERPOLATE(color[2][2], color[2][2], color[2][2], color[2][1]); + else if (color[1][1] == color[2][1] && color[2][1] == color[3][2] && color[3][1] != color[2][2] && color[2][1] != color[3][3]) + product2b = Q_INTERPOLATE(color[2][1], color[2][1], color[2][1], color[2][2]); + else + product2b = INTERPOLATE(color[2][1], color[2][2]); + + if (color[1][2] == color[2][2] && color[1][2] == color[0][1] && color[1][1] != color[0][2] && color[1][2] != color[0][0]) + product1b = Q_INTERPOLATE(color[1][2], color[1][2], color[1][2], color[1][1]); + else if (color[1][1] == color[2][1] && color[1][1] == color[0][2] && color[0][1] != color[1][2] && color[1][1] != color[0][3]) + product1b = Q_INTERPOLATE(color[1][2], color[1][1], color[1][1], color[1][1]); + else + product1b = INTERPOLATE(color[1][1], color[1][2]); + } + + if (color[1][1] == color[2][2] && color[2][1] != color[1][2] && color[1][0] == color[1][1] && color[1][1] != color[3][2]) + product2a = INTERPOLATE(color[2][1], color[1][1]); + else if (color[1][1] == color[2][0] && color[1][2] == color[1][1] && color[1][0] != color[2][1] && color[1][1] != color[3][0]) + product2a = INTERPOLATE(color[2][1], color[1][1]); + else + product2a = color[2][1]; + + if (color[2][1] == color[1][2] && color[1][1] != color[2][2] && color[2][0] == color[2][1] && color[2][1] != color[0][2]) + product1a = INTERPOLATE(color[2][1], color[1][1]); + else if (color[1][0] == color[2][1] && color[2][2] == color[2][1] && color[2][0] != color[1][1] && color[2][1] != color[0][0]) + product1a = INTERPOLATE(color[2][1], color[1][1]); + else + product1a = color[1][1]; + + /* Set the calculated pixels */ + switch (bpp) { + case 4: + AV_WN32A(dst_line[0] + x * 8, product1a); + AV_WN32A(dst_line[0] + x * 8 + 4, product1b); + AV_WN32A(dst_line[1] + x * 8, product2a); + AV_WN32A(dst_line[1] + x * 8 + 4, product2b); + break; + case 3: + AV_WL24(dst_line[0] + x * 6, product1a); + AV_WL24(dst_line[0] + x * 6 + 3, product1b); + AV_WL24(dst_line[1] + x * 6, product2a); + AV_WL24(dst_line[1] + x * 6 + 3, product2b); + break; + default: // bpp = 2 + AV_WN32A(dst_line[0] + x * 4, product1a | (product1b << 16)); + AV_WN32A(dst_line[1] + x * 4, product2a | (product2b << 16)); + } + + /* Move color matrix forward */ + color[0][0] = color[0][1]; color[0][1] = color[0][2]; color[0][2] = color[0][3]; + color[1][0] = color[1][1]; color[1][1] = color[1][2]; color[1][2] = color[1][3]; + color[2][0] = color[2][1]; color[2][1] = color[2][2]; color[2][2] = color[2][3]; + color[3][0] = color[3][1]; color[3][1] = color[3][2]; color[3][2] = color[3][3]; + + if (x < width - 3) { + x += 3; + switch (bpp) { + case 4: + READ_COLOR4(color[0][3], src_line[0], x); + READ_COLOR4(color[1][3], src_line[1], x); + READ_COLOR4(color[2][3], src_line[2], x); + READ_COLOR4(color[3][3], src_line[3], x); + break; + case 3: + READ_COLOR3(color[0][3], src_line[0], x); + READ_COLOR3(color[1][3], src_line[1], x); + READ_COLOR3(color[2][3], src_line[2], x); + READ_COLOR3(color[3][3], src_line[3], x); + break; + default: /* case 2 */ + READ_COLOR2(color[0][3], src_line[0], x); + READ_COLOR2(color[1][3], src_line[1], x); + READ_COLOR2(color[2][3], src_line[2], x); + READ_COLOR2(color[3][3], src_line[3], x); + } + x -= 3; + } + } + + /* We're done with one line, so we shift the source lines up */ + src_line[0] = src_line[1]; + src_line[1] = src_line[2]; + src_line[2] = src_line[3]; + + /* Read next line */ + src_line[3] = src_line[2]; + if (y < height - 3) + src_line[3] += src_linesize; + + switch (bpp) { + case 4: + READ_COLOR4(color[0][0], src_line[0], 0); color[0][1] = color[0][0]; READ_COLOR4(color[0][2], src_line[0], 1); READ_COLOR4(color[0][3], src_line[0], 2); + READ_COLOR4(color[1][0], src_line[1], 0); color[1][1] = color[1][0]; READ_COLOR4(color[1][2], src_line[1], 1); READ_COLOR4(color[1][3], src_line[1], 2); + READ_COLOR4(color[2][0], src_line[2], 0); color[2][1] = color[2][1]; READ_COLOR4(color[2][2], src_line[2], 1); READ_COLOR4(color[2][3], src_line[2], 2); + READ_COLOR4(color[3][0], src_line[3], 0); color[3][1] = color[3][0]; READ_COLOR4(color[3][2], src_line[3], 1); READ_COLOR4(color[3][3], src_line[3], 2); + break; + case 3: + READ_COLOR3(color[0][0], src_line[0], 0); color[0][1] = color[0][0]; READ_COLOR3(color[0][2], src_line[0], 1); READ_COLOR3(color[0][3], src_line[0], 2); + READ_COLOR3(color[1][0], src_line[1], 0); color[1][1] = color[1][0]; READ_COLOR3(color[1][2], src_line[1], 1); READ_COLOR3(color[1][3], src_line[1], 2); + READ_COLOR3(color[2][0], src_line[2], 0); color[2][1] = color[2][1]; READ_COLOR3(color[2][2], src_line[2], 1); READ_COLOR3(color[2][3], src_line[2], 2); + READ_COLOR3(color[3][0], src_line[3], 0); color[3][1] = color[3][0]; READ_COLOR3(color[3][2], src_line[3], 1); READ_COLOR3(color[3][3], src_line[3], 2); + break; + default: + READ_COLOR2(color[0][0], src_line[0], 0); color[0][1] = color[0][0]; READ_COLOR2(color[0][2], src_line[0], 1); READ_COLOR2(color[0][3], src_line[0], 2); + READ_COLOR2(color[1][0], src_line[1], 0); color[1][1] = color[1][0]; READ_COLOR2(color[1][2], src_line[1], 1); READ_COLOR2(color[1][3], src_line[1], 2); + READ_COLOR2(color[2][0], src_line[2], 0); color[2][1] = color[2][1]; READ_COLOR2(color[2][2], src_line[2], 1); READ_COLOR2(color[2][3], src_line[2], 2); + READ_COLOR2(color[3][0], src_line[3], 0); color[3][1] = color[3][0]; READ_COLOR2(color[3][2], src_line[3], 1); READ_COLOR2(color[3][3], src_line[3], 2); + } + } // y loop +} + +static int query_formats(AVFilterContext *ctx) +{ + static const enum PixelFormat pix_fmts[] = { + PIX_FMT_RGBA, PIX_FMT_BGRA, PIX_FMT_ARGB, PIX_FMT_ABGR, + PIX_FMT_RGB24, PIX_FMT_BGR24, + PIX_FMT_RGB565, PIX_FMT_BGR565, PIX_FMT_RGB555, PIX_FMT_BGR555, + PIX_FMT_NONE + }; + + avfilter_set_common_pixel_formats(ctx, avfilter_make_format_list(pix_fmts)); + return 0; +} + +static int config_input(AVFilterLink *inlink) +{ + Super2xSaIContext *sai = inlink->dst->priv; + + sai->hi_pixel_mask = 0xFEFEFEFE; + sai->lo_pixel_mask = 0x01010101; + sai->q_hi_pixel_mask = 0xFCFCFCFC; + sai->q_lo_pixel_mask = 0x03030303; + sai->bpp = 4; + + switch (inlink->format) { + case PIX_FMT_RGB24: + case PIX_FMT_BGR24: + sai->bpp = 3; + break; + + case PIX_FMT_RGB565: + case PIX_FMT_BGR565: + sai->hi_pixel_mask = 0xF7DEF7DE; + sai->lo_pixel_mask = 0x08210821; + sai->q_hi_pixel_mask = 0xE79CE79C; + sai->q_lo_pixel_mask = 0x18631863; + sai->bpp = 2; + break; + + case PIX_FMT_BGR555: + case PIX_FMT_RGB555: + sai->hi_pixel_mask = 0x7BDE7BDE; + sai->lo_pixel_mask = 0x04210421; + sai->q_hi_pixel_mask = 0x739C739C; + sai->q_lo_pixel_mask = 0x0C630C63; + sai->bpp = 2; + break; + } + + return 0; +} + +static int config_output(AVFilterLink *outlink) +{ + AVFilterLink *inlink = outlink->src->inputs[0]; + + outlink->w = inlink->w*2; + outlink->h = inlink->h*2; + + av_log(inlink->dst, AV_LOG_INFO, "fmt:%s size:%dx%d -> size:%dx%d\n", + av_get_pix_fmt_name(inlink->format), + inlink->w, inlink->h, outlink->w, outlink->h); + + return 0; +} + +static void null_draw_slice(AVFilterLink *inlink, int y, int h, int slice_dir) { } + +static void end_frame(AVFilterLink *inlink) +{ + AVFilterLink *outlink = inlink->dst->outputs[0]; + AVFilterBufferRef *inpicref = inlink->cur_buf; + AVFilterBufferRef *outpicref = outlink->out_buf; + + super2xsai(inlink->dst, inpicref->data[0], inpicref->linesize[0], + outpicref->data[0], outpicref->linesize[0], + inlink->w, inlink->h); + + avfilter_unref_buffer(inpicref); + avfilter_draw_slice(outlink, 0, outlink->h, 1); + avfilter_end_frame(outlink); + avfilter_unref_buffer(outpicref); +} + +AVFilter avfilter_vf_super2xsai = { + .name = "super2xsai", + .description = NULL_IF_CONFIG_SMALL("Scale the input by 2x using the Super2xSaI pixel art algorithm."), + .priv_size = sizeof(Super2xSaIContext), + .query_formats = query_formats, + + .inputs = (const AVFilterPad[]) { + { .name = "default", + .type = AVMEDIA_TYPE_VIDEO, + .config_props = config_input, + .draw_slice = null_draw_slice, + .end_frame = end_frame, + .min_perms = AV_PERM_READ }, + { .name = NULL } + }, + .outputs = (const AVFilterPad[]) { + { .name = "default", + .type = AVMEDIA_TYPE_VIDEO, + .config_props = config_output }, + { .name = NULL } + }, +};