Hi
SNPE recently introduced 16-bit Fixed-Point quantization (in the description of snpe-dlc-quantize, it says it allows both 8 / 16 bit TF style fixed-point).
However, it gives the following error on Snapdragon 820 / Hexagon 680 DSP:
error_code=202; error_message=Invalid fixed point parameter. Tf spans can't be coerced from 0x210 to 0x208; error_component=Dl System; line_no=222
What is the 16-bit alternative for DSP_FIXED8_TF flag? Or if this is a work in progress as the SNPE runtime description does not mention it yet.