The Curious Case of Subnormals in Audio Code
Floating-point subnormals have long presented an interesting edge case in digital signal processing that can cause severe performance hits, particularly when dealing with decaying infinite impulse response (IIR) filters. Most audio applications therefore opt to set flush-to-zero (FTZ) and denormals-are-zero (DAZ) CPU control register flags in their workflows.
The talk will chart an exploratory journey prompted by failing platform tests for web audio and an ancient ECMAScript forum thread, leading the author to investigate the current situation and impact on common devices.
Are subnormals still a problem for floating-point arithmetic in 2024? Is there a difference between native and web targets? What does the future hold for fast math and cross-platform audio code?
Attila Haraszti
Lead Engineer / Director
Haywirez
Attila Haraszti / h̷a̷y̷w̷i̷r̷e̷z̶ is a veteran dance music producer and independent software professional specializing in networked media applications. He is currently developing Songsling, a web-based creation and publishing suite for interactive music and audio.