21 commits

Author	SHA1	Message	Date
Pandora	599158e2ee	fix compiling on ARM ... thanks toshi	2017-12-08 22:26:20 -05:00
Pandora	4318dbe051	fix compiling for 32-bit machines / machine without SSE2	2017-12-07 16:09:13 -05:00
Pandora	2040285ce9	revert back to better blurring behaviour	2017-12-06 13:57:07 -05:00
Chris Guillott	9f8496441c	blurring stuff should work perfectly fine now	2017-12-05 22:07:38 -05:00
Chris Guillott	99842b3963	fix msse3	2017-12-05 17:47:21 -05:00
Chris Guillott	8946358b26	revert to former implementation with array size fix	2017-12-05 14:28:01 -05:00
Chris Guillott	5e0aeccbb3	first commit towards fixing this	2017-12-05 12:53:54 -05:00
Sebastian Frysztak	024dc2980e	Minor style changes	2017-02-15 11:27:43 +01:00
Sebastian Frysztak	3598cf19e8	Implement generic box blur	2017-02-15 11:22:06 +01:00
Sebastian Frysztak	6029c8e0b5	Clean up a bit.	2016-11-11 18:45:20 +01:00
Sebastian Frysztak	020af692e6	Remove AVX version.	2016-11-11 17:11:31 +01:00
Sebastian Frysztak	e5e6368926	Remove SSSE3 version.	2016-11-11 16:46:53 +01:00
Sebastian Frysztak	b47631d785	SSE2: resize filter to 7x7. clean up a little.	2016-11-11 13:11:11 +01:00
Sebastian Frysztak	ab41586b39	SSE2: switch from Gaussian to box blur	2016-11-05 16:01:40 +01:00
Sebastian Frysztak	252999f640	Slightly refactor border handling code.	2016-11-04 22:41:17 +01:00
Sebastian Frysztak	f06dc6cbc4	Add AVX version. ... It relies on some SSE2 instructions, so performance gain is not that huge (about 1.4x). I experimented with 256-bit loads, but they turned out to be slower (at least on Sandy Bridge).	2016-11-04 22:19:29 +01:00
Sebastian Frysztak	95c333cba5	SSSE3: use 16-bit weights. ... Overall, I'm very happy with performance of this code, but not so much with resulting image. It seems like integer approximations won't do. I might remove this code altogether, so I didn't update comments.	2016-11-03 20:16:06 +01:00
Sebastian Frysztak	72aec87047	Add SSSE3-based blur implementation. ... Calculations are done on integer, rather than floating point numbers, so this implementation is not as accurate (but when scale factor is reasonable enough, no artifacs are visible). It is, however, faster by a factor of ~3.	2016-10-29 14:32:49 +02:00
Sebastian Frysztak	3662b8e187	Improve border handling for larger kernels.	2016-10-28 17:36:43 +02:00
Sebastian Frysztak	afe41c5754	Extend kernel size to 15x15.	2016-10-28 17:35:33 +02:00
Sebastian Frysztak	fb5dbbe661	Add SSE2-optimized blur. ... About 4-6 times faster than naive implementation.	2016-10-22 15:30:27 +02:00