#
#asahi · 2026-09-14 20:29
waltmckI'm working on some post-processing tuning right now, I'll probably write a nix flake for people to test it after I get something shippable
#asahi · 2026-09-14 20:26
waltmckand here is a render using the current upstream of the same recording: http://waltmckelvie.com/files/upstream.wav
#asahi · 2026-09-14 20:25
waltmckHere is a render of a recording using my fork: http://waltmckelvie.com/files/post.wav
#asahi · 2026-09-14 20:20
waltmckhttps://github.com/waltmck/triforce/commits/stft-binned-mvdr/
#asahi · 2026-09-14 20:20
waltmckunfortunately it can never be upstreamed because it is AI assisted (I've only done DSP stuff in C++ so it was very helpful writing Rust). I'm using it and it sounds pretty good, feel free to look at it as a reference if you'd like
#asahi · 2026-09-14 20:19
waltmckBTW, I went ahead and implemented a binned MVDR beamformed based on STFT. Despite adding binning it also uses around 50% less CPU than the upstream IIR implementation
#asahi · 2026-09-14 20:14
nicolas17"-funsafe is neither fun nor safe" :3
#asahi · 2026-09-14 20:07
kitlithchaos_princess: I think I hadn't internalized the existence of -funsafe-math-optimizations (which -ffast-math includes), so I didn't catch your allusion to it.
#asahi · 2026-09-14 19:56
kitlith(e-core didn't have slowdown on my machine, this is probably not news)
#asahi · 2026-09-14 19:26
kitlithhm. rust-denormals-bench mentions a run on m2 max having no slowdown, but i don't know if that's p-core or e-core. (probably p-core)
#asahi · 2026-09-14 19:22
chaos_princessit was a legitimate question somewhat, as i misremembered what ffast-math does (i thought it only did fassociative-math, while ffunsafe-math-optimizatsions is what you needed for ftz/daz/nans-do-not-exist)
#asahi · 2026-09-14 19:20
kitlith(if that was tongue in cheek, my apologies)
#asahi · 2026-09-14 19:17
kitliththe way -ffast-math sets it is not particularly safe <https://moyix.blogspot.com/2022/09/someones-been-messing-with-my-subnormals.html>, i think rust considers having those set to be UB since compiler optimizations assume that they're not set.
#asahi · 2026-09-14 19:15
chaos_princessisn't that a part of fun and safe math optimizations?
#asahi · 2026-09-14 19:13
kitlithlinkmauve: almost, -ffast-math also tends to set DTZ/DAZ ("denormals to zero"/"denormals as zero")
#asahi · 2026-09-14 18:52
linkmauveAh, I missed the 1.98.0 release blog post due to being at a festival, that’s why I had never heard about it.
#asahi · 2026-09-14 18:50
linkmauveTIL, this is -ffast-math but localized to a single operation, nice!
#asahi · 2026-09-14 18:49
chaos_princessalgebraic_{add,sub,mul}. yes
#asahi · 2026-09-14 18:48
linkmauveThis code will exclusively run on an ecore right?
#asahi · 2026-09-14 18:48
linkmauveNeat, which new Rust feature did you need?
#asahi · 2026-09-14 18:36
chaos_princesslinkmauve: got it to autovectorize at the cost of raising msrv by a lot - 112 ecore, 40 pcore
#asahi · 2026-09-14 18:02
chaos_princessopt-level 2 - 106 ecore, 41 pcore
#asahi · 2026-09-14 18:01
linkmauveThat’s great!
#asahi · 2026-09-14 18:01
linkmauveWow, so about twice as fast as before!
#asahi · 2026-09-14 17:59
chaos_princessbefore optimizations - 192 ecore 71 pcore
#asahi · 2026-09-14 17:59
chaos_princesslinkmauve: 108ms ecore 34 ms pcore. Unfortunately includes time needed to taskset, but taskset-less is also 34
#asahi · 2026-09-14 17:48
linkmauveI just pushed a new version, the only change is to use vfmaq_n_f32() instead of vfmaq_f32() for the last step, which takes a f32 instead of a float32x4 which I had to vdupq_n_f32() at first.
#asahi · 2026-09-14 17:47
linkmauveFor the example no, it shouldn’t.
#asahi · 2026-09-14 17:44
chaos_princessi guess it shouldn't matter?
#asahi · 2026-09-14 17:44
chaos_princessmy m1pro has broken mics, somehow
#asahi · 2026-09-14 17:43
linkmauveWouldn’t it be more interesting to do it on an actual Apple core though?
#asahi · 2026-09-14 17:41
linkmauveOh…
#asahi · 2026-09-14 17:41
chaos_princessthis file only contains program counter values
#asahi · 2026-09-14 17:41
linkmauvechaos_princess, you mean line numbers from the Rust source code? In the hot loop, it’s basically 1:1 Rust-asm.
#asahi · 2026-09-14 17:39
chaos_princesslinkmauve: i don't think that is sufficient, since i can't map to line numbers
#asahi · 2026-09-14 17:36
calderweis[m]what are peopls thoughts on this ;https://gravitylinux.org/about/
#asahi · 2026-09-14 17:26
linkmauveI wanted to bump its syn dependency, to reduce compilation times (Cortex-A53 is slow, turns out), but there might be better use of my time.
#asahi · 2026-09-14 17:25
linkmauveOw, the lv2 crate hasn’t seen a single commit in the past five years…
#asahi · 2026-09-14 17:22
linkmauveperf report might give something sensible still.
#asahi · 2026-09-14 17:22
linkmauvehttps://linkmauve.fr/files/performance_test.O2.perf.data
#asahi · 2026-09-14 17:21
chaos_princessprobably not
#asahi · 2026-09-14 17:21
chaos_princessporque no los dos
#asahi · 2026-09-14 17:21
linkmauveWould you be able to make use of a perf.data?
#asahi · 2026-09-14 17:20
linkmauveAt O2 or O3?
#asahi · 2026-09-14 17:20
chaos_princesswould be interesting to see
#asahi · 2026-09-14 17:20
linkmauveI do.
#asahi · 2026-09-14 17:20
chaos_princessdo you have perf counters?
#asahi · 2026-09-14 17:19
linkmauveopt-level = 2 is much slower, at 715ms mean out of ten runs.
#asahi · 2026-09-14 17:18
chaos_princessneat.
#asahi · 2026-09-14 17:17
linkmauveOn my PinePhone.

plus ›