A bitwise operation operates on one or more bit patterns or binary numerals at the level of their individual bits. It is a fast, primitive action directly supported by the processor, and is used to manipulate values for comparisons and calculations. On simple low-cost processors, typically, bitwise operations are substantially faster than division, several times faster than multiplication, and sometimes significantly faster than addition. While modern high-performance processors usually perform addition and multiplication as fast as bitwise operations,[1] the latter may still be optimal for overall power/performance due to lower resource use.
Contents |
Note that in the explanations below, any indication of a bit's position is counted from the right (least significant) side, advancing left. For example, the binary value 0001 (decimal 1) has zeroes at every position but the first one.
The bitwise NOT, or complement, is a unary operation that performs logical negation on each bit, forming the ones' complement of the given binary value. Digits which are 0 become 1, and those which are 1 become 0. For example:
NOT 0111 (decimal 7) = 1000 (decimal 8)
For two's complement signed integers, the bitwise complement is equal to the inverse of the value minus 1, i.e. NOT x = -x - 1. For unsigned integers, the bitwise complement of a number is the "mirror reflection" of the number across the half-way point of the unsigned integer's range. For example, for 8-bit unsigned integers, NOT x = 255 - x, which can be visualized on a graph as a downward line which effectively "flips" an increasing range from 0 to 255, to a decreasing range from 255 to 0. A simple but illustrative example use is to invert a grayscale image where each pixel is stored as an unsigned integer.
A bitwise AND takes two binary representations of equal length and performs the logical AND operation on each pair of corresponding bits. The result in each position is 1 if the first bit is 1 AND the second bit is 1; otherwise, the result is 0. In this, we perform the multiplication of two bits, i.e. 1 × 0 = 0 and 1 × 1 = 1. For example:
0101 (decimal 5)
AND 0011 (decimal 3)
= 0001 (decimal 1)
This may be used to determine whether a particular bit is set (1) or clear (0). For example, given a bit pattern 0011 (decimal 3), to determine whether the second bit is set we use a bitwise AND with a bit pattern containing 1 only in the second bit:
0011 (decimal 3)
AND 0010 (decimal 2)
= 0010 (decimal 2)
Because the result 0010 is non-zero, we know the second bit in the original pattern was set. This is often called bit masking. (By analogy, the use of masking tape covers, or masks, portions that should not be altered or portions that are not of interest. In this case, the 0 values mask the bits that are not of interest.)
If we store the result, this may be used to clear selected bits in a register. Given the example 0110 (decimal 6), the second bit may be cleared by using a bitwise AND with the pattern that has a zero only in the second bit:
0110 (decimal 6)
AND 1101 (decimal 13)
= 0100 (decimal 4)
A bitwise OR takes two bit patterns of equal length and performs the logical inclusive OR operation on each pair of corresponding bits. The result in each position is 1 if the first bit is 1 OR the second bit is 1 OR both bits are 1; otherwise, the result is 0. In this, we perform the addition of two bits, i.e. 0 + 1 = 1, however 1 + 1 = 1. For example:
0101 (decimal 5) OR 0011 (decimal 3) = 0111 (decimal 7)
The bitwise OR may be used to set selected bits, such as a specific bit (or flag) in a register where each bit represents an individual Boolean state. For example 0010 (decimal 2) can be considered a set of four flags, where the first, third, and fourth flags are clear (0) and the second flag is set (1). The fourth flag may be set by performing a bitwise OR between this value and a bit pattern containing 1 only in the fourth bit:
0010 (decimal 2) OR 1000 (decimal 8) = 1010 (decimal 10)
This technique is an efficient way to store a number of Boolean values using the minimum of memory.
A bitwise XOR takes two bit patterns of equal length and performs the logical exclusive OR operation on each pair of corresponding bits. The result in each position is 1 if only the first bit is 1 OR only the second bit is 1, but will be 0 if both are 0 or both are 1. This is equivalent to being 1 if the two bits are different, and 0 if they are the same. For example:
0101 (decimal 5)
XOR 0011 (decimal 3)
= 0110 (decimal 6)
The bitwise XOR may be used to invert selected bits in a register (also called toggle or flip). Given the bit pattern 0010 (decimal 2) the second and forth bits may be toggled by a bitwise XOR with a bit pattern containing 1 in the second and forth positions:
0010 (decimal 2)
XOR 1010 (decimal 10)
= 1000 (decimal 8)
This technique may be used to manipulate bit patterns representing sets of Boolean states.
Assembly language programmers sometimes use XOR as a short-cut to setting the value of a register to zero. Performing XOR on a value against itself always yields zero, and on many architectures this operation requires fewer clock cycles and/or memory than loading a zero value and saving it to the register.
The bit shifts are sometimes considered bitwise operations, because they operate on the binary representation of an integer instead of its numerical value; however, the bit shifts do not operate on pairs of corresponding bits, and therefore cannot properly be called bit-wise. In these operations the digits are moved, or shifted, to the left or right. Registers in a computer processor have a fixed width, so some bits will be "shifted out" of the register at one end, while the same number of bits are "shifted in" from the other end; the differences between bit shift operators lie in how they determine the values of the shifted-in bits.
In an arithmetic shift, the bits that are shifted out of either end are discarded. In a left arithmetic shift, zeros are shifted in on the right; in a right arithmetic shift, the sign bit is shifted in on the left, thus preserving the sign of the operand. Further on, while shifting right, the empty spaces will be filled up with a copy of the most significant bit (MSB). Meaning by shifting the second arithmetic shift register (ASR#2), with a MSB=1, you fill up with 1.
This example uses an 8-bit register:
00010111 LEFT-SHIFT = 00101110
00010111 RIGHT-SHIFT = 00001011
In the first case, the leftmost digit was shifted past the end of the register, and a new 0 was shifted into the rightmost position. In the second case, the rightmost 1 was shifted out (perhaps into the carry flag), and a new 0 was copied into the leftmost position, preserving the sign of the number. Multiple shifts are sometimes shortened to a single shift by some number of digits. For example:
00010111 LEFT-SHIFT-BY-TWO = 01011100
A left arithmetic shift by n is equivalent to multiplying by 2n (provided the value does not overflow), while a right arithmetic shift by n of a two's complement value is equivalent to dividing by 2n and rounding toward negative infinity. If the binary number is treated as ones' complement, then the same right-shift operation results in division by 2n and rounding toward zero.
In a logical shift, the bits that are shifted out are discarded, and zeros are shifted in (on either end). Therefore, the logical and arithmetic left-shifts are exactly the same operation. However, the logical right-shift inserts bits with value 0 instead of copying in the sign bit. Hence the logical shift is suitable for unsigned binary numbers, while the arithmetic shift is suitable for signed two's complement binary numbers.
Another form of shift is the circular shift or bit rotation. In this operation, the bits are "rotated" as if the left and right ends of the register were joined. The value that is shifted in on the right during a left-shift is whatever value was shifted out on the left, and vice versa. This operation is useful if it is necessary to retain all the existing bits, and is frequently used in digital cryptography.
Rotate through carry is similar to the rotate no carry operation, but the two ends of the register are considered to be separated by the carry flag. The bit that is shifted in (on either end) is the old value of the carry flag, and the bit that is shifted out (on the other end) becomes the new value of the carry flag.
A single rotate through carry can simulate a logical or arithmetic shift of one position by setting up the carry flag beforehand. For example, if the carry flag contains 0, then x RIGHT-ROTATE-THROUGH-CARRY-BY-ONE is a logical right-shift, and if the carry flag contains a copy of the sign bit, then x RIGHT-ROTATE-THROUGH-CARRY-BY-ONE is an arithmetic right-shift. For this reason, some microcontrollers such as PICs just have rotate and rotate through carry, and don't bother with arithmetic or logical shift instructions.
Rotate through carry is especially useful when performing shifts on numbers larger than the processor's native word size, because if a large number is stored in two registers, the bit that is shifted off the end of the first register must come in at the other end of the second. With rotate-through-carry, that bit is "saved" in the carry flag during the first shift, ready to shift in during the second shift without any extra preparation.
In C-inspired languages, the left and right shift operators are "<<" and ">>", respectively. The number of places to shift is given as the second argument to the shift operators. For example,
x = y << 2;
assigns x the result of shifting y to the left by two bits.
In C and C++, computations with the left operand as an unsigned integer use logical shifts. In C#, the right-shift is an arithmetic shift when the first operand is an int or long. If the first operand is of type uint or ulong, the right-shift is a logical shift.[2] In C, the results with the left operand as a signed integer are:[3] In general case:
x = a << b then x = a*2^b;
x = a >> b then x = a/2^b;
<<": y×2right (undefined if an overflow occurs);>>": implementation-defined (most often the result of the arithmetic shift: y/2right).There are also compiler-specific intrinsics implementing circular shifts, like _rotl8, _rotl16, _rotr8, _rotr16 in Microsoft Visual C++.
In Java, all integer types are signed, and the "<<" and ">>" operators perform arithmetic shifts. Java adds the operator ">>>" to perform logical right shifts, but because the logical and arithmetic left-shift operations are identical, there is no "<<<" operator in Java. These general rules are affected in several ways by the default type promotions; for example, because the eight-bit type byte is promoted to int in shift-expressions,[4] the expression "b >>> 2" effectively performs an arithmetic shift of the byte value b instead of a logical shift. Such effects can be mitigated by judicious use of casts or bitmasks; for example, "(b & 0xFF) >>> 2" effectively results in a logical shift.
In Pascal, as well as in all its dialects, for example Object Pascal and Standard Pascal, the left and right shift operators are "shl" and "shr", respectively. The number of places to shift is given as the second argument to the shift operators. For example,
x := y shl 2;
assigns x the result of shifting y to the left by two bits.
Bitwise operations are necessary for much low-level programming, such as writing device drivers, low-level graphics, communications protocol packet assembly, and decoding.
Although machines often have efficient built-in instructions for performing arithmetic and logical operations, in fact, all these operations can be performed by combining the bitwise operators and zero-testing in various ways.
For example, here is a pseudocode example showing how to multiply two arbitrary integers a and b (a greater than b) using only bitshifts and addition:
c := 0
while b ≠ 0
if (b and 1) ≠ 0
c := c + a
shift a left by one
shift b right by one
return c
This implementation of ancient Egyptian multiplication, like most multiplication algorithms, involves bitshifts. In turn, even addition can be written using just bitshifts and zero-testing:
c := b and a
while a ≠ 0
c := b and a
b := b xor a
shift c left by one
a := c
return b