dfttest v1.9.6 (20200324)

Changes:

2020.03.24 v1.9.6 - pinterf

• Enhanced: avx2 routines for mod-8 sbsize cases
• parameter opt allow 4 for avx2
• fix: a very broken filter_3_sse (ftype = 3)
• fix: filter_6_sse: number limited with 1e-15 instead of 0 (used in an ftype = 0 sub-case)
• filter_6_sse: replacing (rsqrt+rcp) with direct sqrt, more accurate and not slower

2020.03.23 v1.9.5 - pinterf

• Avisynth+ high bit depth support 10-32 bits (base version, needs more optimization)
  new formats: planar RGB
  Y (greyscale)
• Fix: minor image noise at stacked16 input (lsb_in = true)
  proc0_16_SSE2 contained a bug: every 2nd pixel contains traces of the lsb part of pixels 4 positions to the right
  Regression since 1.9.4
• source: no more external asm, YUY2 converters replaced with simd code

2020.03.23 v1.9.4.4 - pinterf

• fix: make fft3w plans thread safe
• MSVC project to VS2019, add v142, v141_xp and ClangCl config.
• refresh avisynth headers
• source to C++17 strict conformance
• AVX option is always available

2018.10.14 v1.9.4.3 - DJATOM

• Fixed one nasty bug, causing crash on non-AVX CPUs.

2017.09.04 - DJATOM

• Adaptive MT mode: MT_MULTI_INSTANCE for threads=1 and MT_SERIALIZED for > 1 internal
• Compilation: silence #693 for Intel Compiler

2017.08.14 - DJATOM

• x64 ready: ported almost all inline asm to intrinsics (dropped some SSE functions, we're in 2017 now).
• AddMean and RemoveMean got their AVX codepaths.
• PlanarFrame updated from JPSDR's NNEDI3 repo (x64 friendly, HBD ready).
• proc0_16 got SSE2 codepath (I can see noticeable speed-up against old version).
• opt parameter change: 2 - SSE/SSE2, 3 - SSE/SSE2/AVX codepath is used.