Win64 compile using MinGW GCC 14.2.0 -O3 (static, UCRT, MCF)
attached is also a AVX2 compile using -march=native on i7-13700H

New: profile --extrahigh
AVX2 + AVX512 usage in some functions (e.g. dot-product)
NLMS: bipartite buffer, instead of a rolling one (slower write, much faster read, enables SIMD)
NLMS: larger search range
OLS: slightly faster
OLS: regularization param scales with system size
BIAS: improved mean,var estimator
Opt: new optimization method Differential Evolution (DE), select using --opt-cfg=de
Opt: multi-threading, use with e.g. --opt-cfg=dds,4 or --opt-cfg=de,20

Overall 30-100% faster encoding (using AVX2) and better compression

Win64 compile using MinGW GCC 14.2.0 -O3 (static, UCRT, MCF)

changes since v0.7.11

faster OLS stage
enabled LMS_N0 in the LMS cascade
enabled LMS_CLAMPW in the LMS cascade
added --opt-sigma to set the starting radius for optimization
replaced static sigma decay with adaptive method based on performance
increased flexibility of stereo decorrelation
increased search range for lms cascade
--best now uses cost-function: bpn

fixed bug: --reset-opt was not handled correctly on the cmd-line interface (thanks @Eddison007), also renamed it to --opt-reset
fixed cosmetic bug: overflow in calculating duration (thanks Porcus@HA for providing a sample)
fixed critical bug: adaptive block switching was using wrong blocks when more than 2 blocks per frame where present (thanks Porcus@HA)

Win64 compile using MinGW GCC 14.2.0 -O3 (static, UCRT, MCF)

changes since v0.7.9

use lms weights for target decay in lms cascade
increase default framelen to 20s
adaptive frame splitting for sparse-pcm model
specify the update interval for ols stage while optimizing
tweaked the bias model (less memory, faster)
fixed buffer overflow for WAVE_FORMAT_EXTENSIBLE
fixed reading WAVE_FORMAT_EXTENSIBLE when not using --verbose
added LICENSE model: MIT

Win64 compile using MinGW GCC 14.2.0 (static, UCRT, MCF)

encode/decode residuals in parallel for stereo-files (--mt-mode=1)
calc cost function in parallel for stereo-files (--mt-mode=2)
set manual cost-function (l1,rms,glb,ent,bpn)
use l1 instead of entropy for sparse model detection
tuned compiler options

Win64 compile using MinGW GCC 14.2.0 -O3 (static, UCRT)

increased dds sigma decay from [0.20,0.05] to [0.25,0.05] for all profiles
replaced linear target decay in lms cascade with exp decay
more flexible stereo model in ols stage

Win64 compile using MinGW GCC 13.2.0 -O3

Win64 compile using MinGW GCC 13.2.0 -O3

Win64 compile using MinGW GCC 13.2.0 -O2

Win64 compile using MinGW GCC 13.2.0

Windows 64-bit compile using GCC 7.3.0