correctly truncating bits for fpga arithmetic

[sweethomela8]

To avoid overflow, when you do a multiplication or large sum, you have enough bits to hold the largest value possible from the arithmetic. If you need to scale down in bits, do you take the number of bits from the MSB downto the number of bits you can use? say you have a 20 bit sum and need to truncate it down to 10 bits...do you take the MSB down to the 10th bit?

 

[randyest]

I'm not sure I fully understand your question, but if you have a multiplication result and you need to truncate it, of course you would take bits from MSB down rather than LSB up. Take a few examples and convert to decimal to see why -- it should be very obvious.

If I misunderstood, please forgive me and explain more.

 

[rakko]

Always discard the least significant bits. Also you should look at the most significant bit of the 10 bits you are dropping and make a decision like this; If the MSBit of the 10 bit you are dropping and the LSBit of the 10 bit you are keeping are both 1's, then add 1 to the 10 bits you are keeping (the MSB). This improves your precision by 1/2 point.