0cac224431
Replace use of "Duff's device" with for-loop to allow compiler to optimise.
...
Reduces CPU instruction count of conv_gen function by ~30%.
Improves performance of convolution operation by 20-25%.
2016-01-25 21:17:41 +00:00
73bcc8b5d3
improve the using-C docs
...
add some stuff on giving options to operations
see https://github.com/jcupitt/libvips/issues/373
2016-01-25 09:35:06 +00:00
07a1903390
use g_assert_not_reached()
...
we were using g_assert(0) ... g_assert_not_reached() makes the
intention clearer
2016-01-24 13:21:53 +00:00
31e4168718
clip bicubic output range from int
...
The output range of the bicubic interpolator is outside the input range,
we must clip from a larger type.
See https://github.com/jcupitt/libvips/pull/372
2016-01-23 10:09:11 +00:00
0dd7f14ada
put interpolate param back
...
Although no current interpolators use the interpolate param, having it
there makes it possible for other (perhaps user-defined) interpolators
to be parameterised.
For example, you could write:
vips affine in.v out.v "2 0 0 2" --interpolate sharp[2]
and `sharp` could do a bicubic and sharpen by 2 in a single operation.
See https://github.com/jcupitt/libvips/pull/372
2016-01-23 09:57:27 +00:00
a7d889df06
use builtin floor / ceil
...
get rid of FAST_PSEUDO_FLOOR, add VIPS_FLOOR and VIPS_CEIL, use them
everywhere
see https://github.com/jcupitt/libvips/pull/372
2016-01-23 09:50:57 +00:00
e025a04d68
Merge pull request #372 from lovell/resample-builtin-floor
...
Small perf improvements to vips_affine + bicubic
2016-01-23 09:28:34 +00:00
25eeb82de6
Consolidate FAST_PSEUDO_FLOOR macros, prefer __builtin_floor if available
...
Use of vips_interpolate_get_method avoids need to pass interpolate instance
Remove an unnecessary temp variable to help prevent register starvation
5% less time in vips_interpolate_bicubic_interpolate
10% less time in vips_affine_gen
2016-01-22 22:26:29 +00:00
f3e3514b03
test embed profile in tiff
...
test embed from file
2016-01-22 22:06:20 +00:00
66e0676a40
fix compiler warning
2016-01-22 17:44:43 +00:00
94db738773
fix double free on tiff write
...
if you attached a profile to a tiff from a file during write, you'd get
a double-free
see https://github.com/jcupitt/libvips/issues/371
2016-01-22 17:43:42 +00:00
923fb43ae9
clear up tiny docs issue
...
not about SEQ behaviour
2016-01-22 15:02:18 +00:00
c7eb88b87f
fix subsample at the command-line
...
it had the SEQ flag set, causing confusion with vips_sequential() when
it skipped scanlines on startup
see https://github.com/jcupitt/libvips/issues/370
2016-01-22 14:37:42 +00:00
9b968491d3
unroll shrinkh inner loop
...
use VIPS_UNROLL for the inner loop, another 5% or so
2016-01-22 10:21:38 +00:00
dce70c82a5
vectorise vips_shrinkh()
...
add special paths for 1, 3 and 4 bands so we can vectorise
see https://github.com/jcupitt/libvips/pull/369
2016-01-22 09:49:20 +00:00
2e5880b69c
speed up for vips_shrinkh()
...
remove the sum buffer, we can just use a local, about 5% faster
see https://github.com/jcupitt/libvips/pull/369
2016-01-22 09:15:38 +00:00
d29710db66
Merge pull request #369 from lovell/shrinkh-loop-vector
...
Allow compiler to use vector unit with vips_shrinkh for ~25% boost
2016-01-22 08:50:02 +00:00
572354b793
Allow compiler to vectorise vips_shrinkh
...
Reduce for-loop count in shrinkh macros
Don't access loop variable outside loop
Remove unnecessary use of ceil from integer
2016-01-21 20:18:56 +00:00
16b76b2cfd
Merge branch 'master' of github.com:jcupitt/libvips
2016-01-21 07:47:14 +00:00
89d1127de6
Merge pull request #368 from lovell/builtin-isnan-isinf
...
Use gcc/clang isnan and isinf builtin functions, where available
2016-01-21 07:22:35 +00:00
90c1fb9a94
Use gcc/clang isnan and isinf builtin functions where available.
...
These are used heavily in the building of look-up tables at
start-up time and therefore improve vipsthumbnail perf by ~4%.
2016-01-20 19:02:14 +00:00
d3f87c5f44
notes
2016-01-20 11:58:59 +00:00
af8b43b16d
fix to make check in non-C locales
...
"make check" uses bc, which needs '.' as a decimal sep.
https://github.com/jcupitt/libvips/issues/367
2016-01-17 09:29:25 +00:00
4512400a3c
better rounding in new bilinear interp
...
was off by 0.5
2016-01-14 13:13:08 +00:00
6ca9a907b8
reorder load tests
...
and a slight memory saving on ifthenelse
2016-01-13 20:32:23 +00:00
52384606dd
bump version for new C++ patch
2016-01-12 17:12:46 +00:00
eeb0ef7ae4
Merge pull request #363 from lovell/vips-cpp-win
...
Enable compilation of vips8 C++ bindings with MSVC
2016-01-11 18:44:52 +00:00
20ae80397d
Enable compilation of vips8 C++ bindings with MSVC
...
Add VIPS_CPLUSPLUS_API macro to allow DLL creation
2016-01-11 17:56:02 +00:00
50db131865
VImage::get_typeof() returns GType
...
it was int, causing some compiler warnings on some platforms.
see https://github.com/jcupitt/libvips/issues/362
2016-01-09 20:27:14 +00:00
bf64fde872
regen C++ interface
2016-01-09 15:08:17 +00:00
66373e9b41
remove exception specs from vips8 C++ interface
...
We had exception specs on the C++ interface, but C++11 does not support
them and some compilers have never honoured them. Remove all specs.
Thanks Lovell.
See https://github.com/jcupitt/libvips/issues/362
2016-01-09 14:52:05 +00:00
3be5af9779
silence a clang warning
2016-01-09 14:51:35 +00:00
9419e07a6b
remove a lock from VImage.cc
...
it wasn't really needed, and windows will barf with pointers across
DLLs
2016-01-08 13:40:58 +00:00
b85231baba
small doc improvements
2016-01-07 11:16:12 +00:00
98f4a64a86
better Vips.py class names builder
2016-01-05 08:00:40 +00:00
e27aa4488a
clearer changelog entry
2016-01-04 17:24:00 +00:00
e81900b187
bandjoin is just an instance function
...
we had both a class member bandjoin, and an instance member
Vips.Image.bandjoin([i1, i2, i3..])
i1.ibandjoin([i2, i3..])
this was confusing and annoying ... get rid of the class one and just
use bandjoin everywhere, so this is now the way to do it:
i1.bandjoin([i2, i3..])
2016-01-04 14:18:10 +00:00
e6cbdbb516
add a stub for compat
...
see https://github.com/jcupitt/libvips/issues/361
2016-01-01 10:58:53 +00:00
69d10ea4b2
add C decls for bandjoin_const
...
oops, forgot them
2015-12-30 16:30:56 +00:00
8ce10058eb
oop, missed a vips-8.1
2015-12-30 11:49:32 +00:00
813e91c682
rename vips wrapper script
2015-12-30 11:31:59 +00:00
dbf9a5ed1c
fix dzsave tile-width default
...
was wrong for google / zoomify layouts
revise tests too
2015-12-30 11:23:33 +00:00
aa635f7816
Merge pull request #360 from ericcorriel/master
...
fixing typo, should be Ifdef
2015-12-28 11:03:35 +00:00
630f8bb02b
fixing typo, should be IFDEF
2015-12-27 20:39:55 -05:00
8302e2ffcd
notes on dzsave change
...
see https://github.com/jcupitt/libvips/pull/359
2015-12-23 09:16:58 +00:00
2efb5c1718
Merge pull request #359 from gasi/patch-1
...
Use 254 pixels as default tile size
2015-12-22 21:15:48 +00:00
58183088ed
another try at no-gsf
2015-12-21 15:21:46 +00:00
bb545542d7
oop, fix build without libgsf
2015-12-21 15:05:46 +00:00
73436ddfc2
add @properties flag to tiffsave
...
writes all vips metadata to imagedescription tag
2015-12-21 14:43:59 +00:00
0deb640bc4
load and save TIFFTAG_IMAGEDESCRIPTION
...
this often has useful metadata in, for example the OME spec has all the
metadata in there as an XML document
see https://github.com/jcupitt/libvips/issues/358
2015-12-21 13:53:33 +00:00
Lovell Fuller
John Cupitt
Eric Corriel