Linux/TMP422: Linux driver temperature calculation

Taketo-san,

See Table 2 of datasheet. Notice that temperature result is binary weighted. Bits 0-3 are always zero, so they can be ignored or discarded. Bit 7 has a weight of 0.5°C, bit 6 has a weight of 0.25°C and so on. For this reason, the 16 bit value can be converted to floating point temperature by performing a right shift of 8.

2^1 = 2 and x << 1 is equivalent to x * 2
2^-1 = 0.5 and x >> 1 is equivalent to x * 0.5 and x / 2

If we have the raw 16 bit result from the temperature register stored correctly in a signed 16 bit integer datatype, then negative numbers are naturally handled correctly under 2's complement rules. If this same value were stored in a signed 32 bit datatype, it may not correctly resolve as negative, but I digress.

If we wanted to take that 16 bit signed integer and convert it to °C using only integer math operations, we could emulate a right shift of 8 using divide by 256. However, this would cause the fractional bits to fall off the right side and be lost if we don't store the result in a float datatype. (Additionally, float datatypes won't allow you to >> or <<.) Instead, if we scaled the value by 1000 before dividing by 256, we would be left with temperature in milliCelsius. If the temperature sensor were configured for RANGE=1, then you would need to subtract 64°C afterwards. I can't explain why they are performing +128. In summary, these are correct functions:

if (RANGE){
signed short x = 0x5950
mC = (x*1000)/256-64000;
}
else {
signed short x = 0x1950
mC = (x*1000)/256;
}

mC = 25312 or 25.312C

Thanks,
Ren