projectNe10 / Ne10

An open optimized software library project for the ARM® Architecture
Other
1.46k stars 408 forks source link

Setting project compile flag to -Ofast breaks Ne10 FFT #241

Closed phaetrick closed 4 years ago

phaetrick commented 4 years ago

Hello. In our Android project (armv7, arm64, ndk20, llvm, clang) when compiled with -Ofast NE10 FFT (r2c, c2r) does not work correctly any more. No output at fftsizes >= 4096. Lower sizes work. With -O3 everything works, but this results in a huge speed decrease in most parts of the app. Is there any workaround or any suggestions?