mainrs (Sep 17 2025 at 08:14): mainrs (Sep 17 2025 at 08:14):Are there cases where AoT consistently runs worse (in terms of execution time) than interpreted or fast-interpreted?F
My Wasm binary basically allocates a struct on the heap, sets some values to random values using rand(), and returns the memory address of the struct.
The struct gets freed later on the host side. I see consistent peaks that tank performance at around every 10 invocations at a frequency of 500Hz.
The boxplot looks wild compared to the interpreted version, with even the means being the same, and the 25% quantiles are also larger than interpreted.
I was wondering if the memory allocation is the issue. It's probably hard to tell in general, but maybe someone has similar experiences. Thanks!
lum1n0us (Sep 18 2025 at 00:40):That's interesting. Would you mind sharing more details for reproduction?
Last updated: Dec 06 2025 at 06:05 UTC