Merge branch experimental

This commit is contained in:
Dominic Szablewski 2021-12-11 15:46:13 +01:00
commit 6310d49ee8
2 changed files with 414 additions and 302 deletions

338
qoi.h
View File

@ -80,75 +80,66 @@ you can define QOI_MALLOC and QOI_FREE before including this library.
-- Data Format -- Data Format
A QOI file has a 14 byte header, followed by any number of data "chunks". A QOI file has a 14 byte header, followed by any number of data "chunks" and 8
zero-bytes to mark the end of the data stream.
struct qoi_header_t { struct qoi_header_t {
char magic[4]; // magic bytes "qoif" char magic[4]; // magic bytes "qoif"
uint32_t width; // image width in pixels (BE) uint32_t width; // image width in pixels (BE)
uint32_t height; // image height in pixels (BE) uint32_t height; // image height in pixels (BE)
uint8_t channels; // must be 3 (RGB) or 4 (RGBA) uint8_t channels; // 3 = RGB, 4 = RGBA
uint8_t colorspace; // a bitmap 0000rgba where uint8_t colorspace; // 0 = sRGB with linear alpha, 1 = all channels linear
// - a zero bit indicates sRGBA,
// - a one bit indicates linear (user interpreted)
// colorspace for each channel
}; };
The decoder and encoder start with {r: 0, g: 0, b: 0, a: 255} as the previous The decoder and encoder start with {r: 0, g: 0, b: 0, a: 0} as the previous
pixel value. Pixels are either encoded as pixel value. Pixels are either encoded as
- a run of the previous pixel - a run of the previous pixel
- an index into a previously seen pixel - an index into an array of previously seen pixels
- a difference to the previous pixel value in r,g,b,a - a difference to the previous pixel value in r,g,b
- full r,g,b,a values - full r,g,b or r,g,b,a values
A running array[64] of previously seen pixel values is maintained by the encoder The color channels are assumed to not be premultiplied with the alpha channel
and decoder. Each pixel that is seen by the encoder and decoder is put into this (un-premultiplied alpha).
array at the position (r^g^b^a) % 64. In the encoder, if the pixel value at this
index matches the current pixel, this index position is written to the stream. A running array[64] (zero-initialized) of previously seen pixel values is
maintained by the encoder and decoder. Each pixel that is seen by the encoder
and decoder is put into this array at the position formed by a hash function of
the color value. In the encoder, if the pixel value at the index matches the
current pixel, this index position is written to the stream as QOI_OP_INDEX.
The hash function for the index is:
index_position = (r * 3 + g * 5 + b * 7 + a * 11) % 64
Each chunk starts with a 2- or 8-bit tag, followed by a number of data bits. The
bit length of chunks is divisible by 8 - i.e. all chunks are byte aligned. All
values encoded in these data bits have the most significant bit on the left.
The 8-bit tags have precedence over the 2-bit tags. A decoder must check for the
presence of an 8-bit tag first.
The byte stream is padded with 8 zero-bytes at the end.
Each chunk starts with a 2, 3 or 4 bit tag, followed by a number of data bits.
The bit length of chunks is divisible by 8 - i.e. all chunks are byte aligned.
All values encoded in these data bits have the most significant bit (MSB) on the
left.
The possible chunks are: The possible chunks are:
- QOI_INDEX -------------
.- QOI_OP_INDEX ----------.
| Byte[0] | | Byte[0] |
| 7 6 5 4 3 2 1 0 | | 7 6 5 4 3 2 1 0 |
|-------+-----------------| |-------+-----------------|
| 0 0 | index | | 0 0 | index |
`-------------------------`
2-bit tag b00 2-bit tag b00
6-bit index into the color index array: 0..63 6-bit index into the color index array: 0..63
- QOI_RUN_8 ------------- .- QOI_OP_DIFF -----------.
| Byte[0] |
| 7 6 5 4 3 2 1 0 |
|----------+--------------|
| 0 1 0 | run |
3-bit tag b010
5-bit run-length repeating the previous pixel: 1..32
- QOI_RUN_16 --------------------------------------
| Byte[0] | Byte[1] |
| 7 6 5 4 3 2 1 0 | 7 6 5 4 3 2 1 0 |
|----------+----------------------------------------|
| 0 1 1 | run |
3-bit tag b011
13-bit run-length repeating the previous pixel: 33..8224
- QOI_DIFF_8 ------------
| Byte[0] | | Byte[0] |
| 7 6 5 4 3 2 1 0 | | 7 6 5 4 3 2 1 0 |
|-------+-----+-----+-----| |-------+-----+-----+-----|
| 1 0 | dr | dg | db | | 0 1 | dr | dg | db |
`-------------------------`
2-bit tag b10 2-bit tag b01
2-bit red channel difference from the previous pixel between -2..1 2-bit red channel difference from the previous pixel between -2..1
2-bit green channel difference from the previous pixel between -2..1 2-bit green channel difference from the previous pixel between -2..1
2-bit blue channel difference from the previous pixel between -2..1 2-bit blue channel difference from the previous pixel between -2..1
@ -156,58 +147,78 @@ The possible chunks are:
The difference to the current channel values are using a wraparound operation, The difference to the current channel values are using a wraparound operation,
so "1 - 2" will result in 255, while "255 + 1" will result in 0. so "1 - 2" will result in 255, while "255 + 1" will result in 0.
Values are stored as unsigned integers with a bias of 2. E.g. -2 is stored as
0 (b00). 1 is stored as 3 (b11).
- QOI_DIFF_16 -------------------------------------
.- QOI_OP_LUMA -------------------------------------.
| Byte[0] | Byte[1] | | Byte[0] | Byte[1] |
| 7 6 5 4 3 2 1 0 | 7 6 5 4 3 2 1 0 | | 7 6 5 4 3 2 1 0 | 7 6 5 4 3 2 1 0 |
|----------+--------------|------------ +-----------| |-------+-----------------+-------------+-----------|
| 1 1 0 | red diff | green diff | blue diff | | 1 0 | green diff | dr - dg | db - dg |
`---------------------------------------------------`
2-bit tag b10
6-bit green channel difference from the previous pixel -32..31
4-bit red channel difference minus green channel difference -8..7
4-bit blue channel difference minus green channel difference -8..7
3-bit tag b110 The green channel is used to indicate the general direction of change and is
5-bit red channel difference from the previous pixel between -16..15 encoded in 6 bits. The red and green channels (dr and db) base their diffs off
4-bit green channel difference from the previous pixel between -8..7 of the green channel difference and are encoded in 4 bits. I.e.:
4-bit blue channel difference from the previous pixel between -8..7 dr_dg = (last_px.r - cur_px.r) - (last_px.g - cur_px.g)
db_dg = (last_px.b - cur_px.b) - (last_px.g - cur_px.g)
The difference to the current channel values are using a wraparound operation, The difference to the current channel values are using a wraparound operation,
so "10 - 13" will result in 253, while "250 + 7" will result in 1. so "10 - 13" will result in 253, while "250 + 7" will result in 1.
Values are stored as unsigned integers with a bias of 32 for the green channel
- QOI_DIFF_24 --------------------------------------------------------------- and a bias of 8 for the red and blue channel.
| Byte[0] | Byte[1] | Byte[2] |
| 7 6 5 4 3 2 1 0 | 7 6 5 4 3 2 1 0 | 7 6 5 4 3 2 1 0 |
|-------------+----------------+--------------+----------------+--------------|
| 1 1 1 0 | red diff | green diff | blue diff | alpha diff |
4-bit tag b1110
5-bit red channel difference from the previous pixel between -16..15
5-bit green channel difference from the previous pixel between -16..15
5-bit blue channel difference from the previous pixel between -16..15
5-bit alpha channel difference from the previous pixel between -16..15
The difference to the current channel values are using a wraparound operation,
so "10 - 13" will result in 253, while "250 + 7" will result in 1.
- QOI_COLOR ------------- .- QOI_OP_RUN ------------.
| Byte[0] | | Byte[0] |
| 7 6 5 4 3 2 1 0 | | 7 6 5 4 3 2 1 0 |
|-------------+--+--+--+--| |-------+-----------------|
| 1 1 1 1 |hr|hg|hb|ha| | 1 1 | run |
`-------------------------`
2-bit tag b11
6-bit run-length repeating the previous pixel: 1..62
4-bit tag b1111 The run-length is stored with a bias of 1. Note that the run-lengths 63 and 64
1-bit red byte follows (b111110 and b111111) are illegal as they are occupied by the QOI_OP_RGB and
1-bit green byte follows QOI_OP_RGBA tags.
1-bit blue byte follows
1-bit alpha byte follows
For each set bit hr, hg, hb and ha another byte follows in this order. If such a
byte follows, it will replace the current color channel value with the value of
this byte.
The byte stream is padded at the end with 4 zero bytes. Size the longest chunk .- QOI_OP_RGB ------------------------------------------.
we can encounter is 5 bytes (QOI_COLOR with RGBA set), with this padding we just | Byte[0] | Byte[1] | Byte[2] | Byte[3] |
have to check for an overrun once per decode loop iteration. | 7 6 5 4 3 2 1 0 | 7 .. 0 | 7 .. 0 | 7 .. 0 |
|-------------------------+---------+---------+---------|
| 1 1 1 1 1 1 1 0 | red | green | blue |
`-------------------------------------------------------`
8-bit tag b11111110
8-bit red channel value
8-bit green channel value
8-bit blue channel value
.- QOI_OP_RGBA ---------------------------------------------------.
| Byte[0] | Byte[1] | Byte[2] | Byte[3] | Byte[4] |
| 7 6 5 4 3 2 1 0 | 7 .. 0 | 7 .. 0 | 7 .. 0 | 7 .. 0 |
|-------------------------+---------+---------+---------+---------|
| 1 1 1 1 1 1 1 1 | red | green | blue | alpha |
`-----------------------------------------------------------------`
8-bit tag b11111111
8-bit red channel value
8-bit green channel value
8-bit blue channel value
8-bit alpha channel value
The byte stream is padded at the end with 8 zero bytes. Since the longest legal
chunk is 5 bytes (QOI_OP_RGBA), with this padding it is possible to check for an
overrun only once per decode loop iteration. These 0x00 bytes also mark the end
of the data stream, as an encoder should never produce 8 consecutive zero bytes
within the stream.
*/ */
@ -222,19 +233,20 @@ have to check for an overrun once per decode loop iteration.
extern "C" { extern "C" {
#endif #endif
// A pointer to qoi_desc struct has to be supplied to all of qoi's functions. It // A pointer to a qoi_desc struct has to be supplied to all of qoi's functions.
// describes either the input format (for qoi_write, qoi_encode), or is filled // It describes either the input format (for qoi_write and qoi_encode), or is
// with the description read from the file header (for qoi_read, qoi_decode). // filled with the description read from the file header (for qoi_read and
// qoi_decode).
// The colorspace in this qoi_desc is a bitmap with 0000rgba where a 0-bit // The colorspace in this qoi_desc is an enum where
// indicates sRGB and a 1-bit indicates linear colorspace for each channel. You // 0 = sRGB, i.e. gamma scaled RGB channels and a linear alpha channel
// may use one of the predefined constants: QOI_SRGB, QOI_SRGB_LINEAR_ALPHA or // 1 = all channels are linear
// QOI_LINEAR. The colorspace is purely informative. It will be saved to the // You may use the constants QOI_SRGB or QOI_LINEAR. The colorspace is purely
// file header, but does not affect en-/decoding in any way. // informative. It will be saved to the file header, but does not affect
// en-/decoding in any way.
#define QOI_SRGB 0x00 #define QOI_SRGB 0
#define QOI_SRGB_LINEAR_ALPHA 0x01 #define QOI_LINEAR 1
#define QOI_LINEAR 0x0f
typedef struct { typedef struct {
unsigned int width; unsigned int width;
@ -309,24 +321,21 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels);
#define QOI_FREE(p) free(p) #define QOI_FREE(p) free(p)
#endif #endif
#define QOI_INDEX 0x00 // 00xxxxxx #define QOI_OP_INDEX 0x00 // 00xxxxxx
#define QOI_RUN_8 0x40 // 010xxxxx #define QOI_OP_DIFF 0x40 // 01xxxxxx
#define QOI_RUN_16 0x60 // 011xxxxx #define QOI_OP_LUMA 0x80 // 10xxxxxx
#define QOI_DIFF_8 0x80 // 10xxxxxx #define QOI_OP_RUN 0xc0 // 11xxxxxx
#define QOI_DIFF_16 0xc0 // 110xxxxx #define QOI_OP_RGB 0xfe // 11111110
#define QOI_DIFF_24 0xe0 // 1110xxxx #define QOI_OP_RGBA 0xff // 11111111
#define QOI_COLOR 0xf0 // 1111xxxx
#define QOI_MASK_2 0xc0 // 11000000 #define QOI_MASK_2 0xc0 // 11000000
#define QOI_MASK_3 0xe0 // 11100000
#define QOI_MASK_4 0xf0 // 11110000
#define QOI_COLOR_HASH(C) (C.rgba.r ^ C.rgba.g ^ C.rgba.b ^ C.rgba.a) #define QOI_COLOR_HASH(C) (C.rgba.r*3 + C.rgba.g*5 + C.rgba.b*7 + C.rgba.a*11)
#define QOI_MAGIC \ #define QOI_MAGIC \
(((unsigned int)'q') << 24 | ((unsigned int)'o') << 16 | \ (((unsigned int)'q') << 24 | ((unsigned int)'o') << 16 | \
((unsigned int)'i') << 8 | ((unsigned int)'f')) ((unsigned int)'i') << 8 | ((unsigned int)'f'))
#define QOI_HEADER_SIZE 14 #define QOI_HEADER_SIZE 14
#define QOI_PADDING 4 #define QOI_PADDING 8
typedef union { typedef union {
struct { unsigned char r, g, b, a; } rgba; struct { unsigned char r, g, b, a; } rgba;
@ -353,7 +362,7 @@ void *qoi_encode(const void *data, const qoi_desc *desc, int *out_len) {
data == NULL || out_len == NULL || desc == NULL || data == NULL || out_len == NULL || desc == NULL ||
desc->width == 0 || desc->height == 0 || desc->width == 0 || desc->height == 0 ||
desc->channels < 3 || desc->channels > 4 || desc->channels < 3 || desc->channels > 4 ||
(desc->colorspace & 0xf0) != 0 desc->colorspace > 2
) { ) {
return NULL; return NULL;
} }
@ -380,7 +389,7 @@ void *qoi_encode(const void *data, const qoi_desc *desc, int *out_len) {
qoi_rgba_t index[64] = {0}; qoi_rgba_t index[64] = {0};
int run = 0; int run = 0;
qoi_rgba_t px_prev = {.rgba = {.r = 0, .g = 0, .b = 0, .a = 255}}; qoi_rgba_t px_prev = {.rgba = {.r = 0, .g = 0, .b = 0, .a = 0}};
qoi_rgba_t px = px_prev; qoi_rgba_t px = px_prev;
int px_len = desc->width * desc->height * desc->channels; int px_len = desc->width * desc->height * desc->channels;
@ -392,80 +401,68 @@ void *qoi_encode(const void *data, const qoi_desc *desc, int *out_len) {
px = *(qoi_rgba_t *)(pixels + px_pos); px = *(qoi_rgba_t *)(pixels + px_pos);
} }
else { else {
px.rgba.r = pixels[px_pos]; px.rgba.r = pixels[px_pos + 0];
px.rgba.g = pixels[px_pos + 1]; px.rgba.g = pixels[px_pos + 1];
px.rgba.b = pixels[px_pos + 2]; px.rgba.b = pixels[px_pos + 2];
} }
if (px.v == px_prev.v) { if (px.v == px_prev.v) {
run++; run++;
if (run == 62 || px_pos == px_end) {
bytes[p++] = QOI_OP_RUN | (run - 1);
run = 0;
} }
if (
run > 0 &&
(run == 0x2020 || px.v != px_prev.v || px_pos == px_end)
) {
if (run < 33) {
run -= 1;
bytes[p++] = QOI_RUN_8 | run;
} }
else { else {
run -= 33; if (run > 0) {
bytes[p++] = QOI_RUN_16 | run >> 8; bytes[p++] = QOI_OP_RUN | (run - 1);
bytes[p++] = run;
}
run = 0; run = 0;
} }
if (px.v != px_prev.v) {
int index_pos = QOI_COLOR_HASH(px) % 64; int index_pos = QOI_COLOR_HASH(px) % 64;
if (index[index_pos].v == px.v) { if (index[index_pos].v == px.v) {
bytes[p++] = QOI_INDEX | index_pos; bytes[p++] = QOI_OP_INDEX | index_pos;
} }
else { else {
index[index_pos] = px; index[index_pos] = px;
int vr = px.rgba.r - px_prev.rgba.r; if (px.rgba.a == px_prev.rgba.a) {
int vg = px.rgba.g - px_prev.rgba.g; char vr = px.rgba.r - px_prev.rgba.r;
int vb = px.rgba.b - px_prev.rgba.b; char vg = px.rgba.g - px_prev.rgba.g;
int va = px.rgba.a - px_prev.rgba.a; char vb = px.rgba.b - px_prev.rgba.b;
char vg_r = vr - vg;
char vg_b = vb - vg;
if ( if (
vr > -17 && vr < 16 &&
vg > -17 && vg < 16 &&
vb > -17 && vb < 16 &&
va > -17 && va < 16
) {
if (
va == 0 &&
vr > -3 && vr < 2 && vr > -3 && vr < 2 &&
vg > -3 && vg < 2 && vg > -3 && vg < 2 &&
vb > -3 && vb < 2 vb > -3 && vb < 2
) { ) {
bytes[p++] = QOI_DIFF_8 | ((vr + 2) << 4) | (vg + 2) << 2 | (vb + 2); bytes[p++] = QOI_OP_DIFF | (vr + 2) << 4 | (vg + 2) << 2 | (vb + 2);
} }
else if ( else if (
va == 0 && vg_r > -9 && vg_r < 8 &&
vr > -17 && vr < 16 && vg > -33 && vg < 32 &&
vg > -9 && vg < 8 && vg_b > -9 && vg_b < 8
vb > -9 && vb < 8
) { ) {
bytes[p++] = QOI_DIFF_16 | (vr + 16); bytes[p++] = QOI_OP_LUMA | (vg + 32);
bytes[p++] = (vg + 8) << 4 | (vb + 8); bytes[p++] = (vg_r + 8) << 4 | (vg_b + 8);
} }
else { else {
bytes[p++] = QOI_DIFF_24 | (vr + 16) >> 1; bytes[p++] = QOI_OP_RGB;
bytes[p++] = (vr + 16) << 7 | (vg + 16) << 2 | (vb + 16) >> 3; bytes[p++] = px.rgba.r;
bytes[p++] = (vb + 16) << 5 | (va + 16); bytes[p++] = px.rgba.g;
bytes[p++] = px.rgba.b;
} }
} }
else { else {
bytes[p++] = QOI_COLOR | (vr ? 8 : 0) | (vg ? 4 : 0) | (vb ? 2 : 0) | (va ? 1 : 0); bytes[p++] = QOI_OP_RGBA;
if (vr) { bytes[p++] = px.rgba.r; } bytes[p++] = px.rgba.r;
if (vg) { bytes[p++] = px.rgba.g; } bytes[p++] = px.rgba.g;
if (vb) { bytes[p++] = px.rgba.b; } bytes[p++] = px.rgba.b;
if (va) { bytes[p++] = px.rgba.a; } bytes[p++] = px.rgba.a;
} }
} }
} }
@ -501,6 +498,7 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels) {
if ( if (
desc->width == 0 || desc->height == 0 || desc->width == 0 || desc->height == 0 ||
desc->channels < 3 || desc->channels > 4 || desc->channels < 3 || desc->channels > 4 ||
desc->colorspace > 2 ||
header_magic != QOI_MAGIC header_magic != QOI_MAGIC
) { ) {
return NULL; return NULL;
@ -516,7 +514,7 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels) {
return NULL; return NULL;
} }
qoi_rgba_t px = {.rgba = {.r = 0, .g = 0, .b = 0, .a = 255}}; qoi_rgba_t px = {.rgba = {.r = 0, .g = 0, .b = 0, .a = 0}};
qoi_rgba_t index[64] = {0}; qoi_rgba_t index[64] = {0};
int run = 0; int run = 0;
@ -528,40 +526,34 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels) {
else if (p < chunks_len) { else if (p < chunks_len) {
int b1 = bytes[p++]; int b1 = bytes[p++];
if ((b1 & QOI_MASK_2) == QOI_INDEX) { if (b1 == QOI_OP_RGB) {
px = index[b1 ^ QOI_INDEX]; px.rgba.r = bytes[p++];
px.rgba.g = bytes[p++];
px.rgba.b = bytes[p++];
} }
else if ((b1 & QOI_MASK_3) == QOI_RUN_8) { else if (b1 == QOI_OP_RGBA) {
run = (b1 & 0x1f); px.rgba.r = bytes[p++];
px.rgba.g = bytes[p++];
px.rgba.b = bytes[p++];
px.rgba.a = bytes[p++];
} }
else if ((b1 & QOI_MASK_3) == QOI_RUN_16) { else if ((b1 & QOI_MASK_2) == QOI_OP_INDEX) {
int b2 = bytes[p++]; px = index[b1];
run = (((b1 & 0x1f) << 8) | (b2)) + 32;
} }
else if ((b1 & QOI_MASK_2) == QOI_DIFF_8) { else if ((b1 & QOI_MASK_2) == QOI_OP_DIFF) {
px.rgba.r += ((b1 >> 4) & 0x03) - 2; px.rgba.r += ((b1 >> 4) & 0x03) - 2;
px.rgba.g += ((b1 >> 2) & 0x03) - 2; px.rgba.g += ((b1 >> 2) & 0x03) - 2;
px.rgba.b += ( b1 & 0x03) - 2; px.rgba.b += ( b1 & 0x03) - 2;
} }
else if ((b1 & QOI_MASK_3) == QOI_DIFF_16) { else if ((b1 & QOI_MASK_2) == QOI_OP_LUMA) {
int b2 = bytes[p++]; int b2 = bytes[p++];
px.rgba.r += (b1 & 0x1f) - 16; int vg = (b1 & 0x3f) - 32;
px.rgba.g += (b2 >> 4) - 8; px.rgba.r += vg - 8 + ((b2 >> 4) & 0x0f);
px.rgba.b += (b2 & 0x0f) - 8; px.rgba.g += vg;
px.rgba.b += vg - 8 + (b2 & 0x0f);
} }
else if ((b1 & QOI_MASK_4) == QOI_DIFF_24) { else if ((b1 & QOI_MASK_2) == QOI_OP_RUN) {
int b2 = bytes[p++]; run = (b1 & 0x3f);
int b3 = bytes[p++];
px.rgba.r += (((b1 & 0x0f) << 1) | (b2 >> 7)) - 16;
px.rgba.g += ((b2 & 0x7c) >> 2) - 16;
px.rgba.b += (((b2 & 0x03) << 3) | ((b3 & 0xe0) >> 5)) - 16;
px.rgba.a += (b3 & 0x1f) - 16;
}
else if ((b1 & QOI_MASK_4) == QOI_COLOR) {
if (b1 & 8) { px.rgba.r = bytes[p++]; }
if (b1 & 4) { px.rgba.g = bytes[p++]; }
if (b1 & 2) { px.rgba.b = bytes[p++]; }
if (b1 & 1) { px.rgba.a = bytes[p++]; }
} }
index[QOI_COLOR_HASH(px) % 64] = px; index[QOI_COLOR_HASH(px) % 64] = px;
@ -571,7 +563,7 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels) {
*(qoi_rgba_t*)(pixels + px_pos) = px; *(qoi_rgba_t*)(pixels + px_pos) = px;
} }
else { else {
pixels[px_pos] = px.rgba.r; pixels[px_pos + 0] = px.rgba.r;
pixels[px_pos + 1] = px.rgba.g; pixels[px_pos + 1] = px.rgba.g;
pixels[px_pos + 2] = px.rgba.b; pixels[px_pos + 2] = px.rgba.b;
} }

View File

@ -195,8 +195,12 @@ void png_decode_callback(png_structp png, png_bytep data, png_size_t length) {
read_data->pos += length; read_data->pos += length;
} }
void png_warning_callback(png_structp png_ptr, png_const_charp warning_msg) {
// Ingore warnings about sRGB profiles and such.
}
void *libpng_decode(void *data, int size, int *out_w, int *out_h) { void *libpng_decode(void *data, int size, int *out_w, int *out_h) {
png_structp png = png_create_read_struct(PNG_LIBPNG_VER_STRING, NULL, NULL, NULL); png_structp png = png_create_read_struct(PNG_LIBPNG_VER_STRING, NULL, NULL, png_warning_callback);
if (!png) { if (!png) {
ERROR("png_create_read_struct"); ERROR("png_create_read_struct");
} }
@ -308,6 +312,17 @@ void *fload(const char *path, int *out_size) {
// ----------------------------------------------------------------------------- // -----------------------------------------------------------------------------
// benchmark runner // benchmark runner
int opt_runs = 1;
int opt_nopng = 0;
int opt_nowarmup = 0;
int opt_noverify = 0;
int opt_nodecode = 0;
int opt_noencode = 0;
int opt_norecurse = 0;
int opt_onlytotals = 0;
typedef struct { typedef struct {
uint64_t size; uint64_t size;
uint64_t encode_time; uint64_t encode_time;
@ -315,6 +330,8 @@ typedef struct {
} benchmark_lib_result_t; } benchmark_lib_result_t;
typedef struct { typedef struct {
int count;
uint64_t raw_size;
uint64_t px; uint64_t px;
int w; int w;
int h; int h;
@ -324,12 +341,59 @@ typedef struct {
} benchmark_result_t; } benchmark_result_t;
void benchmark_print_result(benchmark_result_t res) {
res.px /= res.count;
res.raw_size /= res.count;
res.libpng.encode_time /= res.count;
res.libpng.decode_time /= res.count;
res.libpng.size /= res.count;
res.stbi.encode_time /= res.count;
res.stbi.decode_time /= res.count;
res.stbi.size /= res.count;
res.qoi.encode_time /= res.count;
res.qoi.decode_time /= res.count;
res.qoi.size /= res.count;
double px = res.px;
printf(" decode ms encode ms decode mpps encode mpps size kb rate\n");
if (!opt_nopng) {
printf(
"libpng: %8.1f %8.1f %8.2f %8.2f %8d %4.1f%%\n",
(double)res.libpng.decode_time/1000000.0,
(double)res.libpng.encode_time/1000000.0,
(res.libpng.decode_time > 0 ? px / ((double)res.libpng.decode_time/1000.0) : 0),
(res.libpng.encode_time > 0 ? px / ((double)res.libpng.encode_time/1000.0) : 0),
res.libpng.size/1024,
((double)res.libpng.size/(double)res.raw_size) * 100.0
);
printf(
"stbi: %8.1f %8.1f %8.2f %8.2f %8d %4.1f%%\n",
(double)res.stbi.decode_time/1000000.0,
(double)res.stbi.encode_time/1000000.0,
(res.stbi.decode_time > 0 ? px / ((double)res.stbi.decode_time/1000.0) : 0),
(res.stbi.encode_time > 0 ? px / ((double)res.stbi.encode_time/1000.0) : 0),
res.stbi.size/1024,
((double)res.stbi.size/(double)res.raw_size) * 100.0
);
}
printf(
"qoi: %8.1f %8.1f %8.2f %8.2f %8d %4.1f%%\n",
(double)res.qoi.decode_time/1000000.0,
(double)res.qoi.encode_time/1000000.0,
(res.qoi.decode_time > 0 ? px / ((double)res.qoi.decode_time/1000.0) : 0),
(res.qoi.encode_time > 0 ? px / ((double)res.qoi.encode_time/1000.0) : 0),
res.qoi.size/1024,
((double)res.qoi.size/(double)res.raw_size) * 100.0
);
printf("\n");
}
// Run __VA_ARGS__ a number of times and meassure the time taken. The first // Run __VA_ARGS__ a number of times and meassure the time taken. The first
// run is ignored. // run is ignored.
#define BENCHMARK_FN(RUNS, AVG_TIME, ...) \ #define BENCHMARK_FN(NOWARMUP, RUNS, AVG_TIME, ...) \
do { \ do { \
uint64_t time = 0; \ uint64_t time = 0; \
for (int i = 0; i <= RUNS; i++) { \ for (int i = NOWARMUP; i <= RUNS; i++) { \
uint64_t time_start = ns(); \ uint64_t time_start = ns(); \
__VA_ARGS__ \ __VA_ARGS__ \
uint64_t time_end = ns(); \ uint64_t time_end = ns(); \
@ -341,13 +405,14 @@ typedef struct {
} while (0) } while (0)
benchmark_result_t benchmark_image(const char *path, int runs) { benchmark_result_t benchmark_image(const char *path) {
int encoded_png_size; int encoded_png_size;
int encoded_qoi_size; int encoded_qoi_size;
int w; int w;
int h; int h;
// Load the encoded PNG, encoded QOI and raw pixels into memory // Load the encoded PNG, encoded QOI and raw pixels into memory
void *pixels = (void *)stbi_load(path, &w, &h, NULL, 4); void *pixels = (void *)stbi_load(path, &w, &h, NULL, 4);
void *encoded_png = fload(path, &encoded_png_size); void *encoded_png = fload(path, &encoded_png_size);
void *encoded_qoi = qoi_encode(pixels, &(qoi_desc){ void *encoded_qoi = qoi_encode(pixels, &(qoi_desc){
@ -358,10 +423,25 @@ benchmark_result_t benchmark_image(const char *path, int runs) {
}, &encoded_qoi_size); }, &encoded_qoi_size);
if (!pixels || !encoded_qoi || !encoded_png) { if (!pixels || !encoded_qoi || !encoded_png) {
ERROR("Error decoding %s\n", path); ERROR("Error decoding %s", path);
} }
// Verify QOI Output
if (!opt_noverify) {
qoi_desc dc;
void *pixels_qoi = qoi_decode(encoded_qoi, encoded_qoi_size, &dc, 4);
if (memcmp(pixels, pixels_qoi, w * h * 4) != 0) {
ERROR("QOI roundtrip pixel missmatch for %s", path);
}
free(pixels_qoi);
}
benchmark_result_t res = {0}; benchmark_result_t res = {0};
res.count = 1;
res.raw_size = w * h * 4;
res.px = w * h; res.px = w * h;
res.w = w; res.w = w;
res.h = h; res.h = h;
@ -369,41 +449,48 @@ benchmark_result_t benchmark_image(const char *path, int runs) {
// Decoding // Decoding
BENCHMARK_FN(runs, res.libpng.decode_time, { if (!opt_nodecode) {
if (!opt_nopng) {
BENCHMARK_FN(opt_nowarmup, opt_runs, res.libpng.decode_time, {
int dec_w, dec_h; int dec_w, dec_h;
void *dec_p = libpng_decode(encoded_png, encoded_png_size, &dec_w, &dec_h); void *dec_p = libpng_decode(encoded_png, encoded_png_size, &dec_w, &dec_h);
free(dec_p); free(dec_p);
}); });
BENCHMARK_FN(runs, res.stbi.decode_time, { BENCHMARK_FN(opt_nowarmup, opt_runs, res.stbi.decode_time, {
int dec_w, dec_h, dec_channels; int dec_w, dec_h, dec_channels;
void *dec_p = stbi_load_from_memory(encoded_png, encoded_png_size, &dec_w, &dec_h, &dec_channels, 4); void *dec_p = stbi_load_from_memory(encoded_png, encoded_png_size, &dec_w, &dec_h, &dec_channels, 4);
free(dec_p); free(dec_p);
}); });
}
BENCHMARK_FN(runs, res.qoi.decode_time, { BENCHMARK_FN(opt_nowarmup, opt_runs, res.qoi.decode_time, {
qoi_desc desc; qoi_desc desc;
void *dec_p = qoi_decode(encoded_qoi, encoded_qoi_size, &desc, 4); void *dec_p = qoi_decode(encoded_qoi, encoded_qoi_size, &desc, 4);
free(dec_p); free(dec_p);
}); });
}
// Encoding // Encoding
BENCHMARK_FN(runs, res.libpng.encode_time, { if (!opt_noencode) {
if (!opt_nopng) {
BENCHMARK_FN(opt_nowarmup, opt_runs, res.libpng.encode_time, {
int enc_size; int enc_size;
void *enc_p = libpng_encode(pixels, w, h, &enc_size); void *enc_p = libpng_encode(pixels, w, h, &enc_size);
res.libpng.size = enc_size; res.libpng.size = enc_size;
free(enc_p); free(enc_p);
}); });
BENCHMARK_FN(runs, res.stbi.encode_time, { BENCHMARK_FN(opt_nowarmup, opt_runs, res.stbi.encode_time, {
int enc_size = 0; int enc_size = 0;
stbi_write_png_to_func(stbi_write_callback, &enc_size, w, h, 4, pixels, 0); stbi_write_png_to_func(stbi_write_callback, &enc_size, w, h, 4, pixels, 0);
res.stbi.size = enc_size; res.stbi.size = enc_size;
}); });
}
BENCHMARK_FN(runs, res.qoi.encode_time, { BENCHMARK_FN(opt_nowarmup, opt_runs, res.qoi.encode_time, {
int enc_size; int enc_size;
void *enc_p = qoi_encode(pixels, &(qoi_desc){ void *enc_p = qoi_encode(pixels, &(qoi_desc){
.width = w, .width = w,
@ -414,6 +501,7 @@ benchmark_result_t benchmark_image(const char *path, int runs) {
res.qoi.size = enc_size; res.qoi.size = enc_size;
free(enc_p); free(enc_p);
}); });
}
free(pixels); free(pixels);
free(encoded_png); free(encoded_png);
@ -422,102 +510,134 @@ benchmark_result_t benchmark_image(const char *path, int runs) {
return res; return res;
} }
void benchmark_print_result(const char *head, benchmark_result_t res) { void benchmark_directory(const char *path, benchmark_result_t *grand_total) {
double px = res.px; DIR *dir = opendir(path);
printf("## %s size: %dx%d\n", head, res.w, res.h); if (!dir) {
printf(" decode ms encode ms decode mpps encode mpps size kb\n"); ERROR("Couldn't open directory %s", path);
printf(
"libpng: %8.1f %8.1f %8.2f %8.2f %8d\n",
(double)res.libpng.decode_time/1000000.0,
(double)res.libpng.encode_time/1000000.0,
(res.libpng.decode_time > 0 ? px / ((double)res.libpng.decode_time/1000.0) : 0),
(res.libpng.encode_time > 0 ? px / ((double)res.libpng.encode_time/1000.0) : 0),
res.libpng.size/1024
);
printf(
"stbi: %8.1f %8.1f %8.2f %8.2f %8d\n",
(double)res.stbi.decode_time/1000000.0,
(double)res.stbi.encode_time/1000000.0,
(res.stbi.decode_time > 0 ? px / ((double)res.stbi.decode_time/1000.0) : 0),
(res.stbi.encode_time > 0 ? px / ((double)res.stbi.encode_time/1000.0) : 0),
res.stbi.size/1024
);
printf(
"qoi: %8.1f %8.1f %8.2f %8.2f %8d\n",
(double)res.qoi.decode_time/1000000.0,
(double)res.qoi.encode_time/1000000.0,
(res.qoi.decode_time > 0 ? px / ((double)res.qoi.decode_time/1000.0) : 0),
(res.qoi.encode_time > 0 ? px / ((double)res.qoi.encode_time/1000.0) : 0),
res.qoi.size/1024
);
printf("\n");
} }
int main(int argc, char **argv) { struct dirent *file;
if (argc < 3) {
printf("Usage: qoibench <iterations> <directory>\n"); if (!opt_norecurse) {
printf("Example: qoibench 10 images/textures/\n"); for (int i = 0; (file = readdir(dir)) != NULL; i++) {
exit(1); if (
file->d_type & DT_DIR &&
strcmp(file->d_name, ".") != 0 &&
strcmp(file->d_name, "..") != 0
) {
char subpath[1024];
snprintf(subpath, 1024, "%s/%s", path, file->d_name);
benchmark_directory(subpath, grand_total);
}
}
rewinddir(dir);
} }
float total_percentage = 0; float total_percentage = 0;
int total_size = 0; int total_size = 0;
benchmark_result_t totals = {0}; benchmark_result_t dir_total = {0};
int runs = atoi(argv[1]); int has_shown_heaad = 0;
DIR *dir = opendir(argv[2]); for (int i = 0; (file = readdir(dir)) != NULL; i++) {
if (runs <=0) {
runs = 1;
}
if (!dir) {
ERROR("Couldn't open directory %s", argv[2]);
}
printf("## Benchmarking %s/*.png -- %d runs\n\n", argv[2], runs);
struct dirent *file;
int count = 0;
for (int i = 0; dir && (file = readdir(dir)) != NULL; i++) {
if (strcmp(file->d_name + strlen(file->d_name) - 4, ".png") != 0) { if (strcmp(file->d_name + strlen(file->d_name) - 4, ".png") != 0) {
continue; continue;
} }
count++;
char *file_path = malloc(strlen(file->d_name) + strlen(argv[2])+8); if (!has_shown_heaad) {
sprintf(file_path, "%s/%s", argv[2], file->d_name); has_shown_heaad = 1;
printf("## Benchmarking %s/*.png -- %d runs\n\n", path, opt_runs);
}
benchmark_result_t res = benchmark_image(file_path, runs); char *file_path = malloc(strlen(file->d_name) + strlen(path)+8);
benchmark_print_result(file_path, res); sprintf(file_path, "%s/%s", path, file->d_name);
benchmark_result_t res = benchmark_image(file_path);
if (!opt_onlytotals) {
printf("## %s size: %dx%d\n", file_path, res.w, res.h);
benchmark_print_result(res);
}
free(file_path); free(file_path);
dir_total.count++;
dir_total.raw_size += res.raw_size;
dir_total.px += res.px;
dir_total.libpng.encode_time += res.libpng.encode_time;
dir_total.libpng.decode_time += res.libpng.decode_time;
dir_total.libpng.size += res.libpng.size;
dir_total.stbi.encode_time += res.stbi.encode_time;
dir_total.stbi.decode_time += res.stbi.decode_time;
dir_total.stbi.size += res.stbi.size;
dir_total.qoi.encode_time += res.qoi.encode_time;
dir_total.qoi.decode_time += res.qoi.decode_time;
dir_total.qoi.size += res.qoi.size;
totals.px += res.px; grand_total->count++;
totals.libpng.encode_time += res.libpng.encode_time; grand_total->raw_size += res.raw_size;
totals.libpng.decode_time += res.libpng.decode_time; grand_total->px += res.px;
totals.libpng.size += res.libpng.size; grand_total->libpng.encode_time += res.libpng.encode_time;
totals.stbi.encode_time += res.stbi.encode_time; grand_total->libpng.decode_time += res.libpng.decode_time;
totals.stbi.decode_time += res.stbi.decode_time; grand_total->libpng.size += res.libpng.size;
totals.stbi.size += res.stbi.size; grand_total->stbi.encode_time += res.stbi.encode_time;
totals.qoi.encode_time += res.qoi.encode_time; grand_total->stbi.decode_time += res.stbi.decode_time;
totals.qoi.decode_time += res.qoi.decode_time; grand_total->stbi.size += res.stbi.size;
totals.qoi.size += res.qoi.size; grand_total->qoi.encode_time += res.qoi.encode_time;
grand_total->qoi.decode_time += res.qoi.decode_time;
grand_total->qoi.size += res.qoi.size;
} }
closedir(dir); closedir(dir);
totals.px /= count; if (dir_total.count > 0) {
totals.libpng.encode_time /= count; printf("## Total for %s\n", path);
totals.libpng.decode_time /= count; benchmark_print_result(dir_total);
totals.libpng.size /= count; }
totals.stbi.encode_time /= count; }
totals.stbi.decode_time /= count;
totals.stbi.size /= count;
totals.qoi.encode_time /= count;
totals.qoi.decode_time /= count;
totals.qoi.size /= count;
benchmark_print_result("Totals (AVG)", totals); int main(int argc, char **argv) {
if (argc < 3) {
printf("Usage: qoibench <iterations> <directory> [options]\n");
printf("Options:\n");
printf(" --nowarmup ... don't perform a warmup run\n");
printf(" --nopng ...... don't run png encode/decode\n");
printf(" --noverify ... don't verify qoi roundtrip\n");
printf(" --noencode ... don't run encoders\n");
printf(" --nodecode ... don't run decoders\n");
printf(" --norecurse .. don't descend into directories\n");
printf(" --onlytotals . don't print individual image results\n");
printf("Examples\n");
printf(" qoibench 10 images/textures/\n");
printf(" qoibench 1 images/textures/ --nopng --nowarmup\n");
exit(1);
}
for (int i = 3; i < argc; i++) {
if (strcmp(argv[i], "--nowarmup") == 0) { opt_nowarmup = 1; }
else if (strcmp(argv[i], "--nopng") == 0) { opt_nopng = 1; }
else if (strcmp(argv[i], "--noverify") == 0) { opt_noverify = 1; }
else if (strcmp(argv[i], "--noencode") == 0) { opt_noencode = 1; }
else if (strcmp(argv[i], "--nodecode") == 0) { opt_nodecode = 1; }
else if (strcmp(argv[i], "--norecurse") == 0) { opt_norecurse = 1; }
else if (strcmp(argv[i], "--onlytotals") == 0) { opt_onlytotals = 1; }
else { ERROR("Unknown option %s", argv[i]); }
}
opt_runs = atoi(argv[1]);
if (opt_runs <=0) {
ERROR("Invalid number of runs %d", opt_runs);
}
benchmark_result_t grand_total = {0};
benchmark_directory(argv[2], &grand_total);
if (grand_total.count > 0) {
printf("# Grand total for %s\n", argv[2]);
benchmark_print_result(grand_total);
}
else {
printf("No images found in %s\n", argv[2]);
}
return 0; return 0;
} }