[FFmpeg-devel] [PATCH] avcodec: add RPZA encoder

Paul B Mahol onemda at gmail.com
Thu Jul 16 17:54:31 EEST 2020


Signed-off-by: Paul B Mahol <onemda at gmail.com>
---
 libavcodec/Makefile    |   1 +
 libavcodec/allcodecs.c |   1 +
 libavcodec/rpzaenc.c   | 857 +++++++++++++++++++++++++++++++++++++++++
 3 files changed, 859 insertions(+)
 create mode 100644 libavcodec/rpzaenc.c

diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index 18353da549..a01c64e066 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -578,6 +578,7 @@ OBJS-$(CONFIG_ROQ_ENCODER)             += roqvideoenc.o roqvideo.o elbg.o
 OBJS-$(CONFIG_ROQ_DPCM_DECODER)        += dpcm.o
 OBJS-$(CONFIG_ROQ_DPCM_ENCODER)        += roqaudioenc.o
 OBJS-$(CONFIG_RPZA_DECODER)            += rpza.o
+OBJS-$(CONFIG_RPZA_ENCODER)            += rpzaenc.o
 OBJS-$(CONFIG_RSCC_DECODER)            += rscc.o
 OBJS-$(CONFIG_RV10_DECODER)            += rv10.o
 OBJS-$(CONFIG_RV10_ENCODER)            += rv10enc.o
diff --git a/libavcodec/allcodecs.c b/libavcodec/allcodecs.c
index a5048290f7..d9ce0e649e 100644
--- a/libavcodec/allcodecs.c
+++ b/libavcodec/allcodecs.c
@@ -266,6 +266,7 @@ extern AVCodec ff_rawvideo_decoder;
 extern AVCodec ff_rl2_decoder;
 extern AVCodec ff_roq_encoder;
 extern AVCodec ff_roq_decoder;
+extern AVCodec ff_rpza_encoder;
 extern AVCodec ff_rpza_decoder;
 extern AVCodec ff_rscc_decoder;
 extern AVCodec ff_rv10_encoder;
diff --git a/libavcodec/rpzaenc.c b/libavcodec/rpzaenc.c
new file mode 100644
index 0000000000..7270386c28
--- /dev/null
+++ b/libavcodec/rpzaenc.c
@@ -0,0 +1,857 @@
+/*
+ * Quicktime RPZA Video Encoder.
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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 2.1 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
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser 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 rpzaenc.c
+ * QT RPZA Video Encoder by Todd Kirby <doubleshot at pacbell.net> and David Adler
+ */
+
+#include "libavutil/avassert.h"
+#include "libavutil/common.h"
+#include "libavutil/opt.h"
+
+#include "avcodec.h"
+#include "internal.h"
+#include "put_bits.h"
+
+typedef struct RpzaContext {
+    AVClass *avclass;
+
+    int skip_frame_thresh;
+    int start_one_color_thresh;
+    int continue_one_color_thresh;
+    int sixteen_color_thresh;
+
+    AVFrame *prev_frame;    // buffer for previous source frame
+    PutBitContext pb;       // buffer for encoded frame data.
+
+    int frame_width;        // width in pixels of source frame
+    int frame_height;       // height in pixesl of source frame
+
+    int first_frame;        // flag set to one when the first frame is being processed
+                            // so that comparisons with previous frame data in not attempted
+} RpzaContext;
+
+typedef enum channel_offset {
+    RED = 2,
+    GREEN = 1,
+    BLUE = 0,
+} channel_offset;
+
+typedef struct rgb {
+    uint8_t r;
+    uint8_t g;
+    uint8_t b;
+} rgb;
+
+#define SQR(x) ((x) * (x))
+
+/* 15 bit components */
+#define GET_CHAN(color, chan) ((color) >> ((chan) * 5) & 0x1F)
+#define R(color) GET_CHAN(color, RED)
+#define G(color) GET_CHAN(color, GREEN)
+#define B(color) GET_CHAN(color, BLUE)
+
+typedef struct BlockInfo {
+    int row;
+    int col;
+    int block_width;
+    int block_height;
+    int image_width;
+    int image_height;
+    int block_index;
+    uint16_t start;
+    int rowstride;
+    int blocks_per_row;
+    int total_blocks;
+} BlockInfo;
+
+static void get_colors(uint8_t *min, uint8_t *max, uint8_t color4[4][3])
+{
+    uint8_t step;
+
+    color4[0][0] = min[0];
+    color4[0][1] = min[1];
+    color4[0][2] = min[2];
+
+    color4[3][0] = max[0];
+    color4[3][1] = max[1];
+    color4[3][2] = max[2];
+
+    // red components
+    step = (color4[3][0] - color4[0][0] + 1) / 3;
+    color4[1][0] = color4[0][0] + step;
+    color4[2][0] = color4[3][0] - step;
+
+    // green components
+    step = (color4[3][1] - color4[0][1] + 1) / 3;
+    color4[1][1] = color4[0][1] + step;
+    color4[2][1] = color4[3][1] - step;
+
+    // blue components
+    step = (color4[3][2] - color4[0][2] + 1) / 3;
+    color4[1][2] = color4[0][2] + step;
+    color4[2][2] = color4[3][2] - step;
+}
+
+/* Fill BlockInfo struct with information about a 4x4 block of the image */
+static int get_block_info(BlockInfo *bi, int block)
+{
+    bi->row = block / bi->blocks_per_row;
+    bi->col = block % bi->blocks_per_row;
+
+    // test for right edge block
+    if (bi->col == bi->blocks_per_row - 1 && (bi->image_width % 4) != 0) {
+        bi->block_width = bi->image_width % 4;
+    } else {
+        bi->block_width = 4;
+    }
+
+    // test for bottom edge block
+    if (bi->row == (bi->image_height / 4) && (bi->image_height % 4) != 0) {
+        bi->block_height = bi->image_height % 4;
+    } else {
+        bi->block_height = 4;
+    }
+
+    return block ? (bi->col * 4) + (bi->row * bi->rowstride * 4) : 0;
+}
+
+static uint16_t rgb24_to_rgb555(uint8_t *rgb24)
+{
+    uint16_t rgb555 = 0;
+    uint32_t r, g, b;
+
+    r = rgb24[0] >> 3;
+    g = rgb24[1] >> 3;
+    b = rgb24[2] >> 3;
+
+    rgb555 |= (r << 10);
+    rgb555 |= (g << 5);
+    rgb555 |= (b << 0);
+
+    return rgb555;
+}
+
+/*
+ * Returns the total difference between two 24 bit color values
+ */
+static int diff_colors(uint8_t *colorA, uint8_t *colorB)
+{
+    int tot;
+
+    tot  = SQR(colorA[0] - colorB[0]);
+    tot += SQR(colorA[1] - colorB[1]);
+    tot += SQR(colorA[2] - colorB[2]);
+
+    return tot;
+}
+
+/*
+ * Returns the maximum channel difference
+ */
+static int max_component_diff(uint16_t *colorA, uint16_t *colorB)
+{
+    int diff, max = 0;
+
+    diff = FFABS(R(colorA[0]) - R(colorB[0]));
+    if (diff > max) {
+        max = diff;
+    }
+    diff = FFABS(G(colorA[0]) - G(colorB[0]));
+    if (diff > max) {
+        max = diff;
+    }
+    diff = FFABS(B(colorA[0]) - B(colorB[0]));
+    if (diff > max) {
+        max = diff;
+    }
+    return max * 8;
+}
+
+/*
+ * Find the channel that has the largest difference between minimum and maximum
+ * color values. Put the minimum value in min, maximum in max and the channel
+ * in chan.
+ */
+static void get_max_component_diff(BlockInfo *bi, uint16_t *block_ptr,
+                                   uint8_t *min, uint8_t *max, channel_offset *chan)
+{
+    int x, y;
+    uint8_t min_r, max_r, min_g, max_g, min_b, max_b;
+    uint8_t r, g, b;
+
+    // fix warning about uninitialized vars
+    min_r = min_g = min_b = UINT8_MAX;
+    max_r = max_g = max_b = 0;
+
+    // loop thru and compare pixels
+    for (y = 0; y < bi->block_height; y++) {
+        for (x = 0; x < bi->block_width; x++){
+            // TODO:  optimize
+            min_r = FFMIN(R(block_ptr[x]) * 8, min_r);
+            min_g = FFMIN(G(block_ptr[x]) * 8, min_g);
+            min_b = FFMIN(B(block_ptr[x]) * 8, min_b);
+
+            max_r = FFMAX(R(block_ptr[x]) * 8, max_r);
+            max_g = FFMAX(G(block_ptr[x]) * 8, max_g);
+            max_b = FFMAX(B(block_ptr[x]) * 8, max_b);
+        }
+        block_ptr += bi->rowstride;
+    }
+
+    r = max_r - min_r;
+    g = max_g - min_g;
+    b = max_b - min_b;
+
+    if (r > g && r > b) {
+        *max = max_r;
+        *min = min_r;
+        *chan = RED;
+    } else if (g > b && g >= r) {
+        *max = max_g;
+        *min = min_g;
+        *chan = GREEN;
+    } else {
+        *max = max_b;
+        *min = min_b;
+        *chan = BLUE;
+    }
+}
+
+/*
+ * Compare two 4x4 blocks to determine if the total difference between the
+ * blocks is greater than the thresh parameter. Returns -1 if difference
+ * exceeds threshold or zero otherwise.
+ */
+static int compare_blocks(uint16_t *block1, uint16_t *block2, BlockInfo *bi, int thresh)
+{
+    int x, y, diff = 0;
+    for (y = 0; y < bi->block_height; y++) {
+        for (x = 0; x < bi->block_width; x++) {
+            diff = max_component_diff(&block1[x], &block2[x]);
+            if (diff >= thresh) {
+                return -1;
+            }
+        }
+        block1 += bi->rowstride;
+        block2 += bi->rowstride;
+    }
+    return 0;
+}
+
+/*
+ * Determine the fit of one channel to another within a 4x4 block. This
+ * is used to determine the best palette choices for 4-color encoding.
+ */
+static int leastsquares(uint16_t *block_ptr, BlockInfo *bi,
+                        channel_offset xchannel, channel_offset ychannel,
+                        double *slope, double *y_intercept, double *correlation_coef)
+{
+    double sumx = 0, sumy = 0, sumx2 = 0, sumy2 = 0, sumxy = 0,
+           sumx_sq = 0, sumy_sq = 0, tmp, tmp2;
+    int i, j, count;
+    uint8_t x, y;
+
+    count = bi->block_height * bi->block_width;
+
+    if (count < 2)
+        return -1;
+
+    for (i = 0; i < bi->block_height; i++) {
+        for (j = 0; j < bi->block_width; j++){
+            x = GET_CHAN(block_ptr[j], xchannel) * 8;
+            y = GET_CHAN(block_ptr[j], ychannel) * 8;
+            sumx += x;
+            sumy += y;
+            sumx2 += x * x;
+            sumy2 += y * y;
+            sumxy += x * y;
+        }
+        block_ptr += bi->rowstride;
+    }
+
+    sumx_sq = sumx * sumx;
+    tmp = (count * sumx2 - sumx_sq);
+
+    // guard against div/0
+    if (tmp == 0)
+        return -2;
+
+    sumy_sq = sumy * sumy;
+
+    *slope = (sumx * sumy - sumxy) / tmp;
+    *y_intercept = (sumy - (*slope) * sumx) / count;
+
+    tmp2 = count * sumy2 - sumy_sq;
+    if (tmp2 == 0) {
+        *correlation_coef = 0.0;
+    } else {
+        *correlation_coef = (count * sumxy - sumx * sumy) /
+            sqrt(tmp * tmp2);
+    }
+
+    return 0; // success
+}
+
+/*
+ * Determine the amount of error in the leastsquares fit.
+ */
+static int calc_lsq_max_fit_error(uint16_t *block_ptr, BlockInfo *bi,
+                                  int min, int max, int tmp_min, int tmp_max,
+                                  channel_offset xchannel, channel_offset ychannel)
+{
+    int i, j, x, y;
+    int err;
+    int max_err = 0;
+
+    for (i = 0; i < bi->block_height; i++) {
+        for (j = 0; j < bi->block_width; j++){
+            int x_inc, lin_y, lin_x;
+            x = GET_CHAN(block_ptr[j], xchannel) * 8;
+            y = GET_CHAN(block_ptr[j], ychannel) * 8;
+
+            /* calculate x_inc as the 4-color index (0..3) */
+            x_inc = floor( (x - min) * 3.0 / (max - min) + 0.5);
+            x_inc = FFMAX(FFMIN(3, x_inc), 0);
+
+            /* calculate lin_y corresponding to x_inc */
+            lin_y = (int)(tmp_min + (tmp_max - tmp_min) * x_inc / 3.0 + 0.5);
+
+            err = FFABS(lin_y - y);
+            if (err > max_err)
+                max_err = err;
+
+            /* calculate lin_x corresponding to x_inc */
+            lin_x = (int)(min + (max - min) * x_inc / 3.0 + 0.5);
+
+            err = FFABS(lin_x - x);
+            if (err > max_err)
+                max_err += err;
+        }
+        block_ptr += bi->rowstride;
+    }
+
+    return max_err;
+}
+
+/*
+ * Find the closest match to a color within the 4-color palette
+ */
+static int match_color(uint16_t *color, uint8_t colors[4][3])
+{
+    int ret = 0;
+    int smallest_variance = INT_MAX;
+    uint8_t dithered_color[3];
+
+    for (int channel = 0; channel < 3; channel++) {
+        dithered_color[channel] = GET_CHAN(color[0], channel) * 8;
+    }
+
+    for (int palette_entry = 0; palette_entry < 4; palette_entry++) {
+        int variance = diff_colors(dithered_color, colors[palette_entry]);
+
+        if (variance < smallest_variance) {
+            smallest_variance = variance;
+            ret = palette_entry;
+        }
+    }
+
+    return ret;
+}
+
+/*
+ * Encode a block using the 4-color opcode and palette. return number of
+ * blocks encoded (until we implement multi-block 4 color runs this will
+ * always be 1)
+ */
+static int encode_four_color_block(uint8_t *min_color, uint8_t *max_color,
+                                   PutBitContext *pb, uint16_t *block_ptr, BlockInfo *bi)
+{
+    int x, y, idx;
+    uint8_t color4[4][3];
+    uint16_t rounded_max, rounded_min;
+
+    // round min and max wider
+    rounded_min = rgb24_to_rgb555(min_color);
+    rounded_max = rgb24_to_rgb555(max_color);
+
+    // put a and b colors
+    // encode 4 colors = first 16 bit color with MSB zeroed and...
+    put_bits(pb, 16, rounded_max & ~0x8000);
+    // ...second 16 bit color with MSB on.
+    put_bits(pb, 16, rounded_min | 0x8000);
+
+    get_colors(min_color, max_color, color4);
+
+    for (y = 0; y < 4; y++) {
+        for (x = 0; x < 4; x++) {
+            idx = match_color(&block_ptr[x], color4);
+            put_bits(pb, 2, idx);
+        }
+        block_ptr += bi->rowstride;
+    }
+    return 1; // num blocks encoded
+}
+
+/*
+ * Copy a 4x4 block from the current frame buffer to the previous frame buffer.
+ */
+static void update_block_in_prev_frame(const uint16_t *src_pixels,
+                                       uint16_t *dest_pixels,
+                                       const BlockInfo *bi, int block_counter)
+{
+    for (int y = 0; y < 4; y++) {
+        memcpy(dest_pixels, src_pixels, 8);
+        dest_pixels += bi->rowstride;
+        src_pixels += bi->rowstride;
+    }
+}
+
+/*
+ * update statistics for the specified block. If first_block,
+ * it initializes the statistics.  Otherwise it updates the statistics IF THIS
+ * BLOCK IS SUITABLE TO CONTINUE A 1-COLOR RUN. That is, it checks whether
+ * the range of colors (since the routine was called first_block != 0) are
+ * all close enough intensities to be represented by a single color.
+
+ * The routine returns 0 if this block is too different to be part of
+ * the same run of 1-color blocks. The routine returns 1 if this
+ * block can be part of the same 1-color block run.
+
+ * If the routine returns 1, it also updates its arguments to include
+ * the statistics of this block. Otherwise, the stats are unchanged
+ * and don't include the current block.
+ */
+static int update_block_stats(RpzaContext *s, BlockInfo *bi, uint16_t *block,
+                              uint8_t min_color[3], uint8_t max_color[3],
+                              int *total_rgb, int *total_pixels,
+                              uint8_t avg_color[3], int first_block)
+{
+    int x, y;
+    int is_in_range;
+    int total_pixels_blk;
+    int threshold;
+
+    uint8_t min_color_blk[3], max_color_blk[3];
+    int total_rgb_blk[3];
+    uint8_t avg_color_blk[3];
+
+    if (first_block) {
+        min_color[0] = UINT8_MAX;
+        min_color[1] = UINT8_MAX;
+        min_color[2] = UINT8_MAX;
+        max_color[0] = 0;
+        max_color[1] = 0;
+        max_color[2] = 0;
+        total_rgb[0] = 0;
+        total_rgb[1] = 0;
+        total_rgb[2] = 0;
+        *total_pixels = 0;
+        threshold = s->start_one_color_thresh;
+    } else {
+        threshold = s->continue_one_color_thresh;
+    }
+
+    /*
+       The *_blk variables will include the current block.
+       Initialize them based on the blocks so far.
+     */
+    min_color_blk[0] = min_color[0];
+    min_color_blk[1] = min_color[1];
+    min_color_blk[2] = min_color[2];
+    max_color_blk[0] = max_color[0];
+    max_color_blk[1] = max_color[1];
+    max_color_blk[2] = max_color[2];
+    total_rgb_blk[0] = total_rgb[0];
+    total_rgb_blk[1] = total_rgb[1];
+    total_rgb_blk[2] = total_rgb[2];
+    total_pixels_blk = *total_pixels + bi->block_height * bi->block_width;
+
+    /*
+       Update stats for this block's pixels
+     */
+    for (y = 0; y < bi->block_height; y++) {
+        for (x = 0; x < bi->block_width; x++) {
+            total_rgb_blk[0] += R(block[x]) * 8;
+            total_rgb_blk[1] += G(block[x]) * 8;
+            total_rgb_blk[2] += B(block[x]) * 8;
+
+            min_color_blk[0] = FFMIN(R(block[x]) * 8, min_color_blk[0]);
+            min_color_blk[1] = FFMIN(G(block[x]) * 8, min_color_blk[1]);
+            min_color_blk[2] = FFMIN(B(block[x]) * 8, min_color_blk[2]);
+
+            max_color_blk[0] = FFMAX(R(block[x]) * 8, max_color_blk[0]);
+            max_color_blk[1] = FFMAX(G(block[x]) * 8, max_color_blk[1]);
+            max_color_blk[2] = FFMAX(B(block[x]) * 8, max_color_blk[2]);
+        }
+        block += bi->rowstride;
+    }
+
+    /*
+       Calculate average color including current block.
+     */
+    avg_color_blk[0] = total_rgb_blk[0] / total_pixels_blk;
+    avg_color_blk[1] = total_rgb_blk[1] / total_pixels_blk;
+    avg_color_blk[2] = total_rgb_blk[2] / total_pixels_blk;
+
+    /*
+       Are all the pixels within threshold of the average color?
+     */
+    is_in_range = (max_color_blk[0] - avg_color_blk[0] <= threshold &&
+                   max_color_blk[1] - avg_color_blk[1] <= threshold &&
+                   max_color_blk[2] - avg_color_blk[2] <= threshold &&
+                   avg_color_blk[0] - min_color_blk[0] <= threshold &&
+                   avg_color_blk[1] - min_color_blk[1] <= threshold &&
+                   avg_color_blk[2] - min_color_blk[2] <= threshold);
+
+    if (is_in_range) {
+        /*
+           Set the output variables to include this block.
+         */
+        min_color[0] = min_color_blk[0];
+        min_color[1] = min_color_blk[1];
+        min_color[2] = min_color_blk[2];
+        max_color[0] = max_color_blk[0];
+        max_color[1] = max_color_blk[1];
+        max_color[2] = max_color_blk[2];
+        total_rgb[0] = total_rgb_blk[0];
+        total_rgb[1] = total_rgb_blk[1];
+        total_rgb[2] = total_rgb_blk[2];
+        *total_pixels = total_pixels_blk;
+        avg_color[0] = avg_color_blk[0];
+        avg_color[1] = avg_color_blk[1];
+        avg_color[2] = avg_color_blk[2];
+    }
+
+    return is_in_range;
+}
+
+static void rpza_encode_stream(RpzaContext *s, const AVFrame *pict)
+{
+    BlockInfo bi;
+    int block_counter = 0;
+    int n_blocks;
+    int total_blocks;
+    int prev_block_offset;
+    int block_offset = 0;
+    uint8_t min = 0, max = 0;
+    channel_offset chan;
+    int i;
+    int tmp_min, tmp_max;
+    int total_rgb[3];
+    uint8_t avg_color[3];
+    int pixel_count;
+    uint8_t min_color[3], max_color[3];
+    double slope, y_intercept, correlation_coef;
+    uint16_t *src_pixels = (uint16_t *)pict->data[0];
+    uint16_t *prev_pixels = (uint16_t *)s->prev_frame->data[0];
+
+    /* Number of 4x4 blocks in frame. */
+    total_blocks = ((s->frame_width + 3) / 4) * ((s->frame_height + 3) / 4);
+
+    bi.image_width = s->frame_width;
+    bi.image_height = s->frame_height;
+    bi.rowstride = pict->linesize[0] / 2;
+
+    bi.blocks_per_row = (s->frame_width + 3) / 4;
+
+    while (block_counter < total_blocks) {
+        // SKIP CHECK
+        // make sure we have a valid previous frame and we're not writing
+        // a key frame
+        if (!s->first_frame) {
+            n_blocks = 0;
+            prev_block_offset = 0;
+
+            while (n_blocks < 32 && block_counter + n_blocks < total_blocks) {
+
+                block_offset = get_block_info(&bi, block_counter + n_blocks);
+
+                // multi-block opcodes cannot span multiple rows.
+                // If we're starting a new row, break out and write the opcode
+                /* TODO: Should eventually use bi.row here to determine when a
+                   row break occurs, but that is currently breaking the
+                   quicktime player. This is probably due to a bug in the
+                   way I'm calculating the current row.
+                 */
+                if (prev_block_offset && block_offset - prev_block_offset > 12) {
+                    break;
+                }
+
+                prev_block_offset = block_offset;
+
+                if (compare_blocks(&prev_pixels[block_offset],
+                                   &src_pixels[block_offset], &bi, s->skip_frame_thresh) != 0) {
+                    // write out skipable blocks
+                    if (n_blocks) {
+
+                        // write skip opcode
+                        put_bits(&s->pb, 8, 0x80 | (n_blocks - 1));
+                        block_counter += n_blocks;
+
+                        goto post_skip;
+                    }
+                    break;
+                }
+
+                /*
+                 * NOTE: we don't update skipped blocks in the previous frame buffer
+                 * since skipped needs always to be compared against the first skipped
+                 * block to avoid artifacts during gradual fade in/outs.
+                 */
+
+                // update_block_in_prev_frame(&src_pixels[block_offset],
+                //   &prev_pixels[block_offset], &bi, block_counter + n_blocks);
+
+                n_blocks++;
+            }
+
+            // we're either at the end of the frame or we've reached the maximum
+            // of 32 blocks in a run. Write out the run.
+            if (n_blocks) {
+                // write skip opcode
+                put_bits(&s->pb, 8, 0x80 | (n_blocks - 1));
+                block_counter += n_blocks;
+
+                continue;
+            }
+
+        } else {
+            block_offset = get_block_info(&bi, block_counter);
+        }
+post_skip :
+
+        // ONE COLOR CHECK
+        if (update_block_stats(s, &bi, &src_pixels[block_offset],
+                               min_color, max_color,
+                               total_rgb, &pixel_count, avg_color, 1)) {
+            int first_block_offset;
+            first_block_offset = prev_block_offset = block_offset;
+
+            n_blocks = 1;
+
+            /* update this block in the previous frame buffer */
+            update_block_in_prev_frame(&src_pixels[block_offset],
+                                       &prev_pixels[block_offset], &bi, block_counter + n_blocks);
+
+            // check for subsequent blocks with the same color
+            while (n_blocks < 32 && block_counter + n_blocks < total_blocks) {
+                block_offset = get_block_info(&bi, block_counter + n_blocks);
+
+                // multi-block opcodes cannot span multiple rows.
+                // If we've hit end of a row, break out and write the opcode
+                if (block_offset - prev_block_offset > 12) {
+                    break;
+                }
+
+                if (!update_block_stats(s, &bi, &src_pixels[block_offset],
+                                        min_color, max_color,
+                                        total_rgb, &pixel_count, avg_color, 0)) {
+                    break;
+                }
+
+                prev_block_offset = block_offset;
+
+                /* update this block in the previous frame buffer */
+                update_block_in_prev_frame(&src_pixels[block_offset],
+                                           &prev_pixels[block_offset], &bi, block_counter + n_blocks);
+
+                n_blocks++;
+            }
+
+            // write one color opcode.
+            put_bits(&s->pb, 8, 0xa0 | (n_blocks - 1));
+            // write color to encode.
+            put_bits(&s->pb, 16, rgb24_to_rgb555(avg_color));
+            // skip past the blocks we've just encoded.
+            block_counter += n_blocks;
+        } else { // FOUR COLOR CHECK
+            int err = 0;
+
+            // get max component diff for block
+            get_max_component_diff(&bi, &src_pixels[block_offset], &min, &max, &chan);
+
+            min_color[0] = 0;
+            max_color[0] = 0;
+            min_color[1] = 0;
+            max_color[1] = 0;
+            min_color[2] = 0;
+            max_color[2] = 0;
+
+            // run least squares against other two components
+            for (i = 0; i < 3; i++) {
+                if (i == chan) {
+                    min_color[i] = min;
+                    max_color[i] = max;
+                    continue;
+                }
+
+                slope = y_intercept = correlation_coef = 0;
+
+                if (leastsquares(&src_pixels[block_offset], &bi, chan, i,
+                                 &slope, &y_intercept, &correlation_coef)) {
+                    min_color[i] = GET_CHAN(src_pixels[block_offset], i) * 8;
+                    max_color[i] = GET_CHAN(src_pixels[block_offset], i) * 8;
+                } else {
+                    tmp_min = (int)(0.5 + min * slope + y_intercept);
+                    tmp_max = (int)(0.5 + max * slope + y_intercept);
+
+                    av_assert0(tmp_min <= tmp_max);
+                    // clamp min and max color values
+                    tmp_min = av_clip_uint8(tmp_min);
+                    tmp_max = av_clip_uint8(tmp_max);
+
+                    err = FFMAX(calc_lsq_max_fit_error(&src_pixels[block_offset], &bi,
+                                                       min, max, tmp_min, tmp_max, chan, i), err);
+
+                    min_color[i] = tmp_min;
+                    max_color[i] = tmp_max;
+                }
+            }
+
+            if (err > s->sixteen_color_thresh) { // DO SIXTEEN COLOR BLOCK
+                uint16_t *row_ptr;
+                int rgb555;
+
+                block_offset = get_block_info(&bi, block_counter);
+
+                row_ptr = &src_pixels[block_offset];
+
+                for (int y = 0; y < 4; y++) {
+                    for (int x = 0; x < 4; x++){
+                        rgb555 = row_ptr[x] & ~0x8000;
+
+                        put_bits(&s->pb, 16, rgb555);
+                    }
+                    row_ptr += bi.rowstride;
+                }
+
+                block_counter++;
+            } else { // FOUR COLOR BLOCK
+                block_counter += encode_four_color_block(min_color, max_color,
+                                                         &s->pb, &src_pixels[block_offset], &bi);
+            }
+
+            /* update this block in the previous frame buffer */
+            update_block_in_prev_frame(&src_pixels[block_offset],
+                                       &prev_pixels[block_offset], &bi, block_counter);
+        }
+    }
+}
+
+static int rpza_encode_init(AVCodecContext *avctx)
+{
+    RpzaContext *s = avctx->priv_data;
+
+    s->frame_width = avctx->width;
+    s->frame_height = avctx->height;
+
+    s->prev_frame = av_frame_alloc();
+    if (!s->prev_frame)
+        return AVERROR(ENOMEM);
+
+    return 0;
+}
+
+static int rpza_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
+                                const AVFrame *frame, int *got_packet)
+{
+    RpzaContext *s = avctx->priv_data;
+    const AVFrame *pict = frame;
+    uint8_t *buf;
+    int ret;
+
+    if ((ret = ff_alloc_packet2(avctx, pkt, avctx->height * avctx->width * 6, 0)) < 0)
+        return ret;
+
+    init_put_bits(&s->pb, pkt->data, pkt->size);
+
+    // skip 4 byte header, write it later once the size of the chunk is known
+    put_bits32(&s->pb, 0x00);
+
+    if (!s->prev_frame->data[0]) {
+        s->first_frame = 1;
+        ret = ff_get_buffer(avctx, s->prev_frame, 0);
+        if (ret < 0)
+            return ret;
+    } else {
+        s->first_frame = 0;
+    }
+
+    rpza_encode_stream(s, pict);
+
+    flush_put_bits(&s->pb);
+
+    pkt->size = put_bits_count(&s->pb) >> 3;
+    buf = pkt->data;
+
+    // write header opcode
+    buf[0] = 0xe1; // chunk opcode
+
+    // write chunk length
+    buf[1] = (pkt->size >> 16) & 0xFF;
+    buf[2] = (pkt->size >> 8) & 0xFF;
+    buf[3] =  pkt->size & 0xFF;
+
+    *got_packet = 1;
+
+    return 0;
+}
+
+static int rpza_encode_end(AVCodecContext *avctx)
+{
+    RpzaContext *s = (RpzaContext *)avctx->priv_data;
+
+    av_frame_free(&s->prev_frame);
+
+    return 0;
+}
+
+#define OFFSET(x) offsetof(RpzaContext, x)
+#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
+static const AVOption options[] = {
+    { "skip_frame_thresh", NULL, OFFSET(skip_frame_thresh), AV_OPT_TYPE_INT, {.i64=1}, 0, 24, VE},
+    { "start_one_color_thresh", NULL, OFFSET(start_one_color_thresh), AV_OPT_TYPE_INT, {.i64=8}, 0, 24, VE},
+    { "continue_one_color_thresh", NULL, OFFSET(continue_one_color_thresh), AV_OPT_TYPE_INT, {.i64=0}, 0, 24, VE},
+    { "sixteen_color_thresh", NULL, OFFSET(sixteen_color_thresh), AV_OPT_TYPE_INT, {.i64=1}, 0, 24, VE},
+    { NULL },
+};
+
+static const AVClass rpza_class = {
+    .class_name = "rpza encoder",
+    .item_name  = av_default_item_name,
+    .option     = options,
+    .version    = LIBAVUTIL_VERSION_INT,
+};
+
+AVCodec ff_rpza_encoder = {
+    .name           = "rpza",
+    .long_name      = NULL_IF_CONFIG_SMALL("QuickTime video (RPZA)"),
+    .type           = AVMEDIA_TYPE_VIDEO,
+    .id             = AV_CODEC_ID_RPZA,
+    .priv_data_size = sizeof(RpzaContext),
+    .priv_class     = &rpza_class,
+    .init           = rpza_encode_init,
+    .encode2        = rpza_encode_frame,
+    .close          = rpza_encode_end,
+    .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_RGB555,
+                                                     AV_PIX_FMT_NONE},
+};
-- 
2.17.1



More information about the ffmpeg-devel mailing list