Sean Barrett
7c0ae19a42
documentation, release data
2014-09-17 07:10:21 -07:00
Sean Barrett
0fc13e997b
optimization attempts, no meaningful changes
2014-09-17 06:24:40 -07:00
Sean Barrett
8024673461
switch srgb convert to second ryg method, which is a bit faster (8% total speedup on upsampling test);
...
remove extra table in slowpath
2014-09-17 06:07:16 -07:00
Jorge Rodriguez
a12d3dedf0
Some todo notes.
2014-09-16 10:50:41 -07:00
Jorge Rodriguez
cbf5ebbd35
Install ryg's float -> uint8 sRGB conversion code, which is much faster.
2014-09-16 10:36:19 -07:00
Jorge Rodriguez
600d80387e
Make a guess at the value before doing the binary search, cuts the number of conditionals by half. Not as much gain as I hoped but something.
2014-09-15 14:01:11 -07:00
Jorge Rodriguez
7602c99e77
Quick formatting fix.
2014-09-15 11:29:40 -07:00
Sean Barrett
dd28033b34
don't encode alpha channel twice when alpha is different colorspace from other channels
2014-09-15 07:23:22 -07:00
Sean Barrett
bdbf1e0ef4
uint32 images need to round with uint32 cast, not int cast
2014-09-14 15:01:10 -07:00
Sean Barrett
9f66b441bd
new quickstart docs;
...
reformat docs;
new simple uint32 test
2014-09-14 14:55:23 -07:00
Jorge Rodriguez
3c261481a6
Add a very quick guide.
2014-09-13 14:33:13 -07:00
Sean Barrett
385d7a417c
delete some unused functions;
...
get rid of stbir__inline for some functions that don't need it
2014-09-13 08:50:10 -07:00
Sean Barrett
8849501a3f
add documentation explaining why even a 64K table for srgb_to_uint8 would be inaccurate
2014-09-13 08:21:14 -07:00
Sean Barrett
9a6af9a8d3
replace int(floor()) with int(), since that produces same results for non-negative values and is much faster on 32-bit x86
2014-09-13 08:12:37 -07:00
Sean Barrett
adbbe8dabd
move x loop inside switch in horizontal_downsample
2014-09-13 07:46:09 -07:00
Sean Barrett
7a8c9196d1
unroll vertical_upsample inner loop, and reverse loop nesting to minimize work
2014-09-13 07:41:39 -07:00
Sean Barrett
9c2ae9bdb8
cache filter-width derived computations that use floor/ceil because those were showing up like crazy on VC6 profile despite only being per-scanline-ish?!?
...
unroll inner loop in horizontal_resample
2014-09-13 07:29:04 -07:00
Sean Barrett
c9600c012f
unroll channel loops in downsampler
2014-09-13 06:23:07 -07:00
Sean Barrett
3ee97c221f
compile as C
2014-09-13 06:12:40 -07:00
Sean Barrett
59898db411
tweak grammar
2014-09-13 06:05:00 -07:00
Sean Barrett
a2f1cadde8
convert tabs to spaces so it's consistent with other stb libs
2014-09-13 06:03:48 -07:00
Sean Barrett
0fa5b5c66f
rename cubic to cubicbspline;
...
tweak filter type documentation
2014-09-13 05:52:19 -07:00
Sean Barrett
29b36b3dea
rename filters, document them
2014-09-12 05:28:24 -07:00
Sean Barrett
a9778b8dbd
tweak docs; fix compile error in UNUSED_PARAM stuff
2014-09-12 04:46:41 -07:00
Sean Barrett
99f3b78bbd
Merge branch 'resample' of http://github.com/bsvino/stb into resample
2014-09-12 03:54:17 -07:00
Sean Barrett
df128b7995
update STBIR__UNUSED_PARAM to match stb_image.h version
2014-09-12 03:53:36 -07:00
Jorge Rodriguez
ca88b6fba6
Calculate and store vertical contributors first so that they can be normalized and optimized.
2014-09-11 12:45:18 -07:00
Jorge Rodriguez
a25c1d2bbd
Merge remote-tracking branch 'remotes/nothings/resample' into resample
2014-09-11 10:00:57 -07:00
Jorge Rodriguez
36b417f6b4
Optimizations: Skip zero-coefficient contributions. Reduce the size of the coefficients array when downsampling.
2014-09-11 09:53:14 -07:00
Sean Barrett
27f26f8337
Documentation;
...
rename STBIR_FLAG_PREMULTIPLED_ALPHA to STBIR_FLAG_ALPHA_PREMULTIPLIED so that both flags have "ALPHA" first
2014-09-11 03:02:20 -07:00
Sean Barrett
16d68d14f8
fix stbir__linear_to_srgb_uchar:
...
1. table stored threshhold of transition from i to i+1, but wants to be i-1 to i
2. table was computed by dividing uchar by 256.0 instead of 255.0, causing it to be 100% wrong
2014-09-11 02:05:53 -07:00
Sean Barrett
30c7a981ec
compile as C;
...
fix some unsigned/signed comparisons;
avoid round() since it's not in pre-C99 C;
remove MAX_CHANNELS since I never ended up needing it;
rename STBIR_EPSILON to STBIR_ALPHA_EPSILON;
don't use STBIR_ALPHA_EPSILON on float input (can't remove it properly due to numeric precision, and they can do it themselves);
optimize subtraction of STBIR_ALPHA_EPSILON;
sorry i forgot to commit these separately!;
2014-09-11 01:47:50 -07:00
Jorge Rodriguez
793818d1af
Skip zero-coefficient contributors, a fairly decent speedup.
2014-09-10 20:26:14 -07:00
Jorge Rodriguez
46dc8f84fb
Use vertical pixel width for the ring buffer.
2014-09-10 20:12:38 -07:00
Jorge Rodriguez
953a637841
Use a carefully shaped trapezoid instead of a box filter to avoid jumps between pixel values.
2014-09-10 19:08:11 -07:00
Jorge Rodriguez
6922628106
Use an even smaller epsilon. Allow the user to avoid the epsilon business if they want to preserve their alpha values.
2014-09-10 15:03:55 -07:00
Jorge Rodriguez
c9d67446d2
Reverse allocation routines parameter order so that the context is at the back.
2014-09-09 22:57:15 -07:00
Jorge Rodriguez
17b931047c
Add the epsilon value into the alpha instead of using an if statement. It's a tiny bit faster and it can be removed afterwards.
2014-09-09 22:47:53 -07:00
Jorge Rodriguez
1d5c902e2e
These minor optimizations were probably already done by the compiler but they can't hurt.
2014-09-09 22:07:40 -07:00
Jorge Rodriguez
41e6aad693
Avoid repeated adds and ensure correct round-trip result.
2014-09-09 21:57:46 -07:00
Jorge Rodriguez
f2102d906f
Use the info structure to keep track of our memory block sizes to reduce duplicated code and cut down on errors.
2014-09-09 20:58:19 -07:00
Jorge Rodriguez
969ff7c850
Make sure calculate_memory matches with the allocations. Fix some errant asserts.
2014-09-09 20:51:29 -07:00
Jorge Rodriguez
8355ea1184
Use height, not width, for correct progress report.
2014-09-09 18:41:28 -07:00
Jorge Rodriguez
4b69c0637e
If alpha is zero then sneakily replace it with a very small nonzero so that the color data is preserved.
...
Pre-process n1 so that we don't have to do it later on. Can't do n0 since it's used to find the coefficient index.
2014-09-09 18:38:05 -07:00
Jorge Rodriguez
d510d70b6a
Report downsample progress by when we empty the ring buffer, it's more accurate that way.
2014-09-09 16:35:00 -07:00
Jorge Rodriguez
497eab8339
Normalize downsample coefficients.
2014-09-09 15:22:29 -07:00
Jorge Rodriguez
43fbc1d5e3
Rearrange the algebra on two of the filters to avoid repeating decimals which cause precision loss. Use int32 to test filters because it's more accurate.
2014-09-09 14:10:14 -07:00
Jorge Rodriguez
ca241daefa
Use rounding to try to preserve the original value. Fix test case.
2014-09-09 12:16:23 -07:00
Jorge Rodriguez
145690788c
Use input_h + pixel_margin * 2 to prevent progress report from ever going above 1. Always report a 0 before and a 1 at the end of each resize.
2014-09-09 10:58:01 -07:00
Jorge Rodriguez
e6c47ec657
Fix kernel lookup for downsampling.
2014-09-07 12:57:53 -07:00
