"Hacker's Delight" book was really useful for me even though I already had a good grasp on bitwise operators. I sat down with a pencil and went through the bit matrix transpose example and it really opened my eyes to how much interesting stuff can be done this way. I think the key to understanding is to visualize the bitwise operations.
Most of it is pattern recognition and learning common idioms, like multiplying by 0x01010101..., extracting substrings of bits ((x >> y) & z), thinking of values as sets of integers and so on.
Also bitwise manipulations lend themselves well to function composition, so try to keep an eye out for that. For example, if you have a fast function that checks if any byte is zero, you can generalize it to any value by applying xor beforehand. Similarly, the fast bit matrix transpose (which absolutely terrified me) is simply applying the least number of "shift a subset of bits" operations such that each bit travels the required distance.
Most of it is pattern recognition and learning common idioms, like multiplying by 0x01010101..., extracting substrings of bits ((x >> y) & z), thinking of values as sets of integers and so on.
Also bitwise manipulations lend themselves well to function composition, so try to keep an eye out for that. For example, if you have a fast function that checks if any byte is zero, you can generalize it to any value by applying xor beforehand. Similarly, the fast bit matrix transpose (which absolutely terrified me) is simply applying the least number of "shift a subset of bits" operations such that each bit travels the required distance.