Closed same-denik closed 4 months ago
Large coefficients can cause an int16_t overflow in FDTC calculation on NEON. The fix replaces the last unsafe butterfly operation with vector multiplication add/sub instructions with int32_t outputs.
Fixes #128
Thanks for the patch.
Large coefficients can cause an int16_t overflow in FDTC calculation on NEON. The fix replaces the last unsafe butterfly operation with vector multiplication add/sub instructions with int32_t outputs.
Fixes #128