Conversation
Rangi42
force-pushed
the
qoi
branch
8 times, most recently
from
7b0e23a to
80e87ff
Compare
Rangi42
force-pushed
the
qoi
branch
19 times, most recently
from
e74c280 to
60b51a0
Compare
Rangi42
force-pushed
the
qoi
branch
2 times, most recently
from
22b584e to
ae1e589
Compare
| @@ -0,0 +1,59 @@ | |||
| #include "proto.h" | |||
|
|
|||
| #define pixelcolor(px) (((uint32_t) (px).a << 24) | ((uint32_t) (px).b << 16) | ((uint32_t) (px).g << 8) | (uint32_t) (px).r) | |||
There was a problem hiding this comment.
I know this is basically PLUM_COLOR_VALUE_32, but (a) then I'd have to do PLUM_COLOR_VALUE_32(px.r, px.g, px.b, px.a), and (b) none of your src files actually use any of the convenience macros in header/color.h anyway.
| size_t datalength; | ||
| }; | ||
|
|
||
| struct QOI_pixel { |
There was a problem hiding this comment.
I did consider using four local uint8_t variables, or one packed uint32_t, but the overall complexity and line count would be higher, as values need to be packed into a lookup table and then unpacked for per-channel manipulation.
Rangi42
force-pushed
the
qoi
branch
2 times, most recently
from
9165397 to
8187f04
Compare
|
|
||
| #define pixelcolor(px) (((uint32_t) (px).a << 24) | ((uint32_t) (px).b << 16) | ((uint32_t) (px).g << 8) | (uint32_t) (px).r) | ||
|
|
||
| void load_QOI_data (struct context * context, unsigned flags, size_t limit) { |
| prev = px; | ||
| } | ||
| } | ||
| output += byteappend(output, 0, 0, 0, 0, 0, 0, 0, 1); |
There was a problem hiding this comment.
I considered using a memset for the 0s and then *output = 1;, but it would be more lines.
| } | ||
|
|
||
| static inline bool bit_depth_less_than (uint32_t depth, uint32_t target) { | ||
| // formally "less than or equal to", but that would be a very long name |
There was a problem hiding this comment.
"bit_depth_lte" would be shorter and not need a clarifying comment. Or "bit_depth_less_equal", or "bit_depth_not_greater", or "bit_depth_cap".
Rangi42
force-pushed
the
qoi
branch
2 times, most recently
from
cc8913f to
4834044
Compare
| return node -> data; | ||
| } | ||
|
|
||
| static inline void * resize_output_node (struct context * context, void * data, size_t size) { |
There was a problem hiding this comment.
The alternative I saw to "allocate one large-as-possible node and then shrink it after QOI compression is done" was "allocate lots of little nodes for every QOI chunk", and this seemed better.
| for (size_t cell = 0; cell < framesize; cell ++) { | ||
| struct QOI_pixel px = {.r = data[cell], .g = data[cell] >> 8, .b = data[cell] >> 16, .a = data[cell] >> 24}; |
There was a problem hiding this comment.
Other possibilities which would use an extra line:
uint32_t * dataend = data + framesize;
for (uint32_t * val = data; val < data; val ++) {
struct QOI_pixel px = {.r = *val, .g = *val >> 8, .b = *val >> 16, .a = *val >> 24};
...
if (run == 62 || val == dataend - 1) { for (size_t cell = 0; cell < framesize; cell ++) {
uint32_t val = data[cell];
struct QOI_pixel px = {.r = val, .g = val >> 8, .b = val >> 16, .a = val >> 24};
...
if (run == 62 || cell == framesize - 1) {1 participant
Spec: https://qoiformat.org/qoi-specification.pdf
Tested on the PNG and QOI images in qoi_test_images.zip, and on mandelbrot.png for a large 32768x32768 test case that the reference encoder can't handle (it's limited to 400-megapixel input, whereas this implementation is not).
After this is merged, fuzz.c should be updated to handle QOI files.
After the next version is released, the qoi repo should be PRed to list libplum in the supporting software.
This pull request is free and unencumbered software released into the public domain, under the same Unlicense as libplum.