I'm looking at an algorithm I'm trying to optimize, and it's basically a lot of bit twiddling, followed by some additions in a tight feedback. If I could use carry-save addition for the adders, it would really help me speed things up, but I'm not sure if I can distribute the operations over the addition.
Specifically if I represent:
a = sa+ca (state + carry)
b = sb+cb
can I represent (a >>> r) in terms of s and c? How about a | b and a & b?
AFAIK XOR is not distributive over addition modulo 2^32, so no you can't do that.
On most older microprocessors, bitwise operations are slightly faster than addition and subtraction operations and usually significantly faster than multiplication and division operations.
Addition and bit-wise or would be the same as bit-wise or would include any bits in either, and normal addition would do exactly the same given the mutually exclusive nature of your bits.
The bitwise operators have precedence and no special rules about avoid evaluation of subexpressions.
Think about it...
sa = 1 ca = 1
sb = 1 cb = 1
a = sa + ca = 2
b = sb + cb = 2
(a | b) = 2
(a & b) = 2
(sa | sb) + (ca | cb) = (1 | 1) + (1 | 1) = 1 + 1 = 2 # Coincidence?
(sa & sb) + (ca & cb) = (1 & 1) + (1 & 1) = 1 + 1 = 2 # Coincidence?
Let's try some other values:
sa = 1001 ca = 1 # Binary
sb = 0100 cb = 1
a = sa + ca = 1010
b = sb + cb = 0101
(a | b) = 1111
(a & b) = 0000
(sa | sb) + (ca | cb) = (1001 | 0101) + (1 | 1) = 1101 + 1 = 1110 # Oh dear!
(sa & sb) + (ca & cb) = (1001 & 0101) + (1 & 1) = 0001 + 1 = 2 # Oh dear!
So, proof by 4-bit counter example that you cannot distribute AND or OR over addition.
What about '>>>' (unsigned or logical right shift). Using the last example values, and r = 1:
sa = 1001
ca = 0001
sa >>> 1 = 0101
ca >>> 1 = 0000
(sa >>> 1) + (ca >>> 1) = 0101 + 0000 = 0101
(sa + ca) >>> 1 = (1001 + 0001) >>> 1 = 1010 >>> 1 = 0101 # Coincidence?
Let's see whether that is coincidence too:
sa = 1011
ca = 0001
sa >>> 1 = 0101
ca >>> 1 = 0000
(sa >>> 1) + (ca >>> 1) = 0101 + 0000 = 0101
(sa + ca) >>> 1 = (1011 + 0001) >>> 1 = 1100 >>> 1 = 0110 # Oh dear!
Proof by counter-example again.
So logical right shift is not distributive over addition either.
No, you cannot distribute AND or OR over binary operators.
Explanation
Let P be a proposition where P: (A+B)&C = A&C + B&C
let us take A=2,B=3 =>A+B=5.
We are to prove A&C + B&C != (A+B)&C
A=2= 010
B=3= 011
let 010&C = x, where x is some integer whose value is the resultant of bitwise AND of 010 and C
similarly 011&C = y, where y is some integer whose value is the resultant of bitwise AND of 011 and C
since we cannot say P holds for all C in the set of Natural numbers ( {0,1,...} ), consequently P is false.
In this case, take C=2=010
x=010 & 010 = 010 = 2
y=011 & 010 = 010 = 2
5&2=101 & 010 = 000 = 0
clearly, x+y!=0 , which means (A+B)&C != A&C + B&C.
Hence proved!
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With