[Groucho2004]

Originally Posted by asarian

Isn't it time you upgraded to ICC16 too?! Just kidding, really; if I recall correctly, those Intel compilers are wickedly expensive.

I chose ICL10/11 because they are faster than newer versions in most cases.

Take for example jpdsr's nnedi3 which apparently has numerous (assembly) optimizations and uses Agner Fog's asmlib memcpy functions (which is weird because Intel's internal memcpy is at least as fast):

Script (source is a short PAL SD clip):

MPEG2Source("F:\Test\test.d2v", idct = 3).trim(0,2999)
fturnleft()
NNEDI3(dh = true, nsize = 0, nns = 4, qual = 2)
fturnright()
NNEDI3(dh = true, nsize = 0, nns = 4, qual = 2)

My nnedi3 0.9.4 ICL10:

[Runtime info]
Frames processed:               3000 (0 - 2999)
FPS (min | max | average):      22.74 | 179.7 | 75.49
Memory usage (phys | virt):     529 | 551 MiB
Thread count:                   14
CPU usage (average):            78%
Time (elapsed):                 00:00:39.739

jpdsr's NNEDI3_v0_9_4_21_x86
(Release_Intel_XP_Core2_SSE4.2):

[Runtime info]
Frames processed:               3000 (0 - 2999)
FPS (min | max | average):      22.64 | 177.0 | 74.76
Memory usage (phys | virt):     529 | 551 MiB
Thread count:                   14
CPU usage (average):            78%
Time (elapsed):                 00:00:40.130