Expand description
Fixed-point math and support helper functions for embedded-nn.
Functionsยง
- clamp
- Clamps a scalar
xbetweenminandmax. - divide_
by_ power_ of_ two - Rounding divide by power of two (midpoint away from zero).
- doubling_
high_ mult_ no_ sat - Doubling high multiply without saturation.
- pack_
q15x2_ 32x1 - Packs two 16-bit signed integers into a single 32-bit integer.
- pack_
s8x4_ 32x1 - Packs four 8-bit signed integers into a single 32-bit integer.
- requantize
- Requantizes a 32-bit integer value given a multiplier and shift.
- requantize_
s64 - Requantizes a 64-bit accumulator value.