1
mirror of https://github.com/mpv-player/mpv synced 2024-10-30 04:46:41 +01:00
mpv/fli.c
pl a54006e1cf warning fix
git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@3738 b3059339-0415-0410-9bf9-f77b7e298cf2
2001-12-25 16:43:27 +00:00

307 lines
8.6 KiB
C

/*
FLI Decoder for MPlayer
(C) 2001 Mike Melanson
32bpp support (c) alex
*/
#include "config.h"
#include "bswap.h"
#include <stdio.h>
#define LE_16(x) (le2me_16(*(unsigned short *)(x)))
#define LE_32(x) (le2me_32(*(unsigned int *)(x)))
#define FLI_256_COLOR 4
#define FLI_DELTA 7
#define FLI_COLOR 11
#define FLI_LC 12
#define FLI_BLACK 13
#define FLI_BRUN 15
#define FLI_COPY 16
#define FLI_MINI 18
// 256 RGB entries; 25% of these bytes will be unused, but it's faster
// to index 4-byte entries
static unsigned char palette[256 * 4];
void Decode_Fli(
unsigned char *encoded,
int encoded_size,
unsigned char *decoded,
int width,
int height,
int bytes_per_pixel)
{
int stream_ptr = 0;
int pixel_ptr;
int palette_ptr1;
int palette_ptr2;
unsigned int frame_size;
int num_chunks;
unsigned int chunk_size;
int chunk_type;
int i, j;
int color_packets;
int color_changes;
int color_scale;
int lines;
int compressed_lines;
int starting_line;
signed short line_packets;
int y_ptr;
int line_inc;
signed char byte_run;
frame_size = LE_32(&encoded[stream_ptr]);
stream_ptr += 6; // skip the magic number
num_chunks = LE_16(&encoded[stream_ptr]);
stream_ptr += 10; // skip padding
// iterate through the chunks
frame_size -= 16;
while ((frame_size > 0) && (num_chunks > 0))
{
chunk_size = LE_32(&encoded[stream_ptr]);
stream_ptr += 4;
chunk_type = LE_16(&encoded[stream_ptr]);
stream_ptr += 2;
switch (chunk_type)
{
case FLI_256_COLOR:
case FLI_COLOR:
if (chunk_type == FLI_COLOR)
color_scale = 4;
else
color_scale = 1;
// set up the palette
color_packets = LE_16(&encoded[stream_ptr]);
stream_ptr += 2;
palette_ptr1 = 0;
for (i = 0; i < color_packets; i++)
{
// first byte is how many colors to skip
palette_ptr1 += (encoded[stream_ptr++] * 4);
// next byte indicates how many entries to change
color_changes = encoded[stream_ptr++];
// if there are 0 color changes, there are actually 256
if (color_changes == 0)
color_changes = 256;
for (j = 0; j < color_changes; j++)
{
palette[palette_ptr1++] = encoded[stream_ptr + 2] * color_scale;
palette[palette_ptr1++] = encoded[stream_ptr + 1] * color_scale;
palette[palette_ptr1++] = encoded[stream_ptr + 0] * color_scale;
palette_ptr1++;
stream_ptr += 3;
}
}
// it seems that a color packet has to be an even number of bytes
// so account for a pad byte
if (stream_ptr & 0x01)
stream_ptr++;
break;
case FLI_DELTA:
line_inc = width * bytes_per_pixel;
y_ptr = 0;
compressed_lines = LE_16(&encoded[stream_ptr]);
stream_ptr += 2;
while (compressed_lines > 0)
{
line_packets = LE_16(&encoded[stream_ptr]);
stream_ptr += 2;
if (line_packets < 0)
{
line_packets = -line_packets;
y_ptr += (line_packets * line_inc);
}
else
{
pixel_ptr = y_ptr;
for (i = 0; i < line_packets; i++)
{
// account for the skip bytes
pixel_ptr += encoded[stream_ptr++] * bytes_per_pixel;
byte_run = encoded[stream_ptr++];
if (byte_run < 0)
{
byte_run = -byte_run;
palette_ptr1 = encoded[stream_ptr++] * 4;
palette_ptr2 = encoded[stream_ptr++] * 4;
for (j = 0; j < byte_run; j++)
{
decoded[pixel_ptr++] = palette[palette_ptr1 + 0];
decoded[pixel_ptr++] = palette[palette_ptr1 + 1];
decoded[pixel_ptr++] = palette[palette_ptr1 + 2];
if (bytes_per_pixel == 4) /* 32bpp */
pixel_ptr++;
decoded[pixel_ptr++] = palette[palette_ptr2 + 0];
decoded[pixel_ptr++] = palette[palette_ptr2 + 1];
decoded[pixel_ptr++] = palette[palette_ptr2 + 2];
if (bytes_per_pixel == 4) /* 32bpp */
pixel_ptr++;
}
}
else
{
for (j = 0; j < byte_run * 2; j++)
{
palette_ptr1 = encoded[stream_ptr++] * 4;
decoded[pixel_ptr++] = palette[palette_ptr1 + 0];
decoded[pixel_ptr++] = palette[palette_ptr1 + 1];
decoded[pixel_ptr++] = palette[palette_ptr1 + 2];
if (bytes_per_pixel == 4) /* 32bpp */
pixel_ptr++;
}
}
}
y_ptr += line_inc;
compressed_lines--;
}
}
break;
case FLI_LC:
// line compressed
line_inc = width * bytes_per_pixel;
starting_line = LE_16(&encoded[stream_ptr]);
stream_ptr += 2;
y_ptr = starting_line * line_inc;
compressed_lines = LE_16(&encoded[stream_ptr]);
stream_ptr += 2;
while (compressed_lines > 0)
{
pixel_ptr = y_ptr;
line_packets = encoded[stream_ptr++];
if (line_packets > 0)
{
for (i = 0; i < line_packets; i++)
{
// account for the skip bytes
pixel_ptr += encoded[stream_ptr++] * bytes_per_pixel;
byte_run = encoded[stream_ptr++];
if (byte_run > 0)
{
for (j = 0; j < byte_run; j++)
{
palette_ptr1 = encoded[stream_ptr++] * 4;
decoded[pixel_ptr++] = palette[palette_ptr1 + 0];
decoded[pixel_ptr++] = palette[palette_ptr1 + 1];
decoded[pixel_ptr++] = palette[palette_ptr1 + 2];
if (bytes_per_pixel == 4) /* 32bpp */
pixel_ptr++;
}
}
else
{
byte_run = -byte_run;
palette_ptr1 = encoded[stream_ptr++] * 4;
for (j = 0; j < byte_run; j++)
{
decoded[pixel_ptr++] = palette[palette_ptr1 + 0];
decoded[pixel_ptr++] = palette[palette_ptr1 + 1];
decoded[pixel_ptr++] = palette[palette_ptr1 + 2];
if (bytes_per_pixel == 4) /* 32bpp */
pixel_ptr++;
}
}
}
}
y_ptr += line_inc;
compressed_lines--;
}
break;
case FLI_BLACK:
// set the whole frame to color 0 (which is usually black)
for (pixel_ptr = 0; pixel_ptr < (width * height * bytes_per_pixel); pixel_ptr++)
{
decoded[pixel_ptr++] = palette[0];
decoded[pixel_ptr++] = palette[1];
decoded[pixel_ptr++] = palette[2];
if (bytes_per_pixel == 4) /* 32bpp */
pixel_ptr++;
}
break;
case FLI_BRUN:
// byte run compression
line_inc = width * bytes_per_pixel;
y_ptr = 0;
for (lines = 0; lines < height; lines++)
{
pixel_ptr = y_ptr;
line_packets = encoded[stream_ptr++];
for (i = 0; i < line_packets; i++)
{
byte_run = encoded[stream_ptr++];
if (byte_run > 0)
{
palette_ptr1 = encoded[stream_ptr++] * 4;
for (j = 0; j < byte_run; j++)
{
decoded[pixel_ptr++] = palette[palette_ptr1 + 0];
decoded[pixel_ptr++] = palette[palette_ptr1 + 1];
decoded[pixel_ptr++] = palette[palette_ptr1 + 2];
if (bytes_per_pixel == 4) /* 32bpp */
pixel_ptr++;
}
}
else // copy bytes if byte_run < 0
{
byte_run = -byte_run;
for (j = 0; j < byte_run; j++)
{
palette_ptr1 = encoded[stream_ptr++] * 4;
decoded[pixel_ptr++] = palette[palette_ptr1 + 0];
decoded[pixel_ptr++] = palette[palette_ptr1 + 1];
decoded[pixel_ptr++] = palette[palette_ptr1 + 2];
if (bytes_per_pixel == 4) /* 32bpp */
pixel_ptr++;
}
}
}
y_ptr += line_inc;
}
break;
case FLI_COPY:
// copy the chunk (uncompressed frame)
for (pixel_ptr = 0; pixel_ptr < chunk_size - 6; pixel_ptr++)
{
palette_ptr1 = encoded[stream_ptr++] * 4;
decoded[pixel_ptr++] = palette[palette_ptr1 + 0];
decoded[pixel_ptr++] = palette[palette_ptr1 + 1];
decoded[pixel_ptr++] = palette[palette_ptr1 + 2];
if (bytes_per_pixel == 4) /* 32bpp */
pixel_ptr++;
}
break;
case FLI_MINI:
// sort of a thumbnail? disregard this chunk...
stream_ptr += chunk_size - 6;
break;
default:
printf ("FLI: Unrecognized chunk type: %d\n", chunk_type);
break;
}
frame_size -= chunk_size;
num_chunks--;
}
}