Support uint8 quantized rsqrt
Two changes:
1) Add support for uint8 quantized rsqrt using lut
2) TFLite doesn't support uint8 quantized rsqrt natively, it converts
the op into to [quantize, int8 rsqrt, quantize]. Add an optimisar pass
to convert it back to uint8 rsqrt.
Signed-off-by:
Limin Tang <limintang@meta.com>
Change-Id: I30b0c0ef314d89695fc3489098ea876b4df6ea09
Loading
Please register or sign in to comment