Which real use cases exist for arithmetic right bit shifting? - java

I stumbled upon a question that asks whether you ever had to use bit shifting in real projects. I have used bit shifts quite extensively in many projects, however, I never had to use arithmetic bit shifting, i.e., bit shifting where the left operand could be negative and the sign bit should be shifted in instead of zeros. For example, in Java, you would do arithmetic bit shifting with the >> operator (while >>> would perform a logical shift). After thinking a lot, I came to the conclusion that I have never used the >> with a possibly negative left operand.
As stated in this answer arithmetic shifting is even implementation defined in C++, so – in contrast to Java – there is not even a standardized operator in C++ for performing arithmetic shifting. The answer also states an interesting problem with shifting negative numbers that I was not even aware of:
+63 >> 1 = +31 (integral part of quotient E1/2E2)
00111111 >> 1 = 00011111
-63 >> 1 = -32
11000001 >> 1 = 11100000
So -63>>1 yields -32 which is obvious when looking at the bits, but maybe not what most programmers would anticipate on first sight. Even more surprising (but again obvious when looking at the bits) is that -1>>1 is -1, not 0.
So, what are concrete use cases for arithmetic right shifting of possibly negative values?

Perhaps the best known is the branchless absolute value:
int m = x >> 31;
int abs = x + m ^ m;
Which uses an arithmetic shift to copy the signbit to all bits. Most uses of arithmetic shift that I've encountered were of that form. Of course an arithmetic shift is not required for this, you could replace all occurrences of x >> 31 (where x is an int) by -(x >>> 31).
The value 31 comes from the size of int in bits, which is 32 by definition in Java. So shifting right by 31 shifts out all bits except the signbit, which (since it's an arithmetic shift) is copied to those 31 bits, leaving a copy of the signbit in every position.

It has come in handy for me before, in the creation of masks that were then used in '&' or '|' operators when manipulating bit fields, either for bitwise data packing or bitwise graphics.
I don't have a handy code sample, but I do recall using that technique many years ago in black-and-white graphics to zoom in (by extending a bit, either 1 or 0). For a 3x zoom, '0' would become '000' and '1' would become '111' without having to know the initial value of the bit. The bit to be expanded would be placed in the high order position, then an arithmetic right shift would extend it, regardless of whether it was 0 or 1. A logical shift, either left or right, always brings in zeros to fill vacated bit positions. In this case the sign bit was the key to the solution.

Here's an example of a function that will find the least power of two greater than or equal to the input. There are other solutions to this problem that are probably faster, namly any hardware oriented solution or just a series of right shifts and ORs. This solution uses arithmetic shift to perform a binary search.
unsigned ClosestPowerOfTwo(unsigned num) {
int mask = 0xFFFF0000;
mask = (num & mask) ? (mask << 8) : (mask >> 8);
mask = (num & mask) ? (mask << 4) : (mask >> 4);
mask = (num & mask) ? (mask << 2) : (mask >> 2);
mask = (num & mask) ? (mask << 1) : (mask >> 1);
mask = (num & mask) ? mask : (mask >> 1);
return (num & mask) ? -mask : -(mask << 1);
}

Indeed logical right shift is much more commonly used. However there are many operations that require an arithmetic shift (or are solved much more elegantly with an arithmetic shift)
Sign extension:
Most of the time you only deal with the available types in C and the compiler will automatically sign extend when casting/promoting a narrower type to a wider one (like short to int) so you may not notice it, but under the hood a left-then-right shift is used if the architecture doesn't have an instruction for sign extension. For "odd" number of bits you'll have to do the sign extension manually so this would be much more common. For example if a 10-bit pixel or ADC value is read into the top bits of a 16-bit register: value >> 6 will move the bits to the lower 10 bit positions and sign extend to preserve the value. If they're read into the low 10 bits with the top 6 bits being zero you'll use value << 6 >> 6 to sign extend the value to work with it
You also need signed extension when working with signed bit fields
struct bitfield {
int x: 15;
int y: 12;
int z: 5;
};
int f(bitfield b) {
return (b.x/8 + b.y/5) * b.z;
}
Demo on Godbolt. The shifts are generated by the compiler but usually you don't use bitfields (as they're not portable) and operate on raw integer values instead so you'll need to do arithmetic shifts yourself to extract the fields
Another example: sign-extend a pointer to make a canonical address in x86-64. This is used to store additional data in the pointer: char* pointer = (char*)((intptr_t)address << 16 >> 16). You can think of this as a 48-bit bitfield at the bottom
V8 engine's SMI optimization stores the value in the top 31 bits so it needs a right shift to restore the signed integer
Round signed division properly when converting to a multiplication, for example x/12 will be optimized to x*43691 >> 19 with some additional rounding. Of course you'll never do this in normal scalar code because the compiler already does this for you but sometimes you may need to vectorize the code or make some related libraries then you'll need to calculate the rounding yourself with arithmetic shift. You can see how compilers round the division results in the output assembly for bitfield above
Saturated shift or shifts larger than bit width, i.e. the value becomes zero when the shift count >= bit width
uint32_t lsh_saturated(uint32_t x, int32_t n) // returns 0 if n == 32
{
return (x << (n & 0x1F)) & ((n-32) >> 5);
}
uint32_t lsh(uint32_t x, int32_t n) // returns 0 if n >= 32
{
return (x << (n & 0x1F)) & ((n-32) >> 31);
}
Bit mask, useful in various cases like branchless selection (i.e. muxer). You can see lots of ways to conditionally do something on the famous bithacks page. Most of them are done by generating a mask of all ones or all zeros. The mask is usually calculated by propagating the sign bit of a subtraction like this (x - y) >> 31 (for 32-bit ints). Of course it can be changed to -(unsigned(x - y) >> 31) but that requires 2's complement and needs more operations. Here's the way to get the min and max of two integers without branching:
min = y + ((x - y) & ((x - y) >> (sizeof(int) * CHAR_BIT - 1)));
max = x - ((x - y) & ((x - y) >> (sizeof(int) * CHAR_BIT - 1)));
Another example is m = m & -((signed)(m - d) >> s); in Compute modulus division by (1 << s) - 1 in parallel without a division operator

I am not too sure what you mean. BUt i'm going to speculate that you want to use the bit shift as an arithmetic function.
One interesting thing i have seen is this property of binary numbers.
int n = 4;
int k = 1;
n = n << k; // is the same as n = n * 2^k
//now n = (4 * 2) i.e. 8
n = n >> k; // is the same as n = n / 2^k
//now n = (8 / 2) i.e. 4
hope that helps.
But yes you want to be careful of negative numbers
i would mask and then turn it back accordingly

In C when writing device drivers, bit shift operators are used extensively since bits are used as switches that need to be turned on and off. Bit shift allow one to easily and correctly target the right switch.
Many hashing and cryptographic functions make use of bit shift. Take a look at Mercenne Twister.
Lastly, it is sometimes useful to use bitfields to contain state information. Bit manipulation functions including bit shift are useful for these things.

Related

Negative of UInt8

I have this code in Objective-C and I want to write it in Java. However there is the statement (-(crc & 1)) that gives me problems. After some googeling it seems that the negative of an unsigned is generally not well defined. I didn't quite understand it though. Maybe somebody who knows CRC can tell me what the equivalent in Java would be.
I need exactly an equivalent method as a device with this Objective-C code checks my calculated crc against its self calculated.
+ (UInt32) crc32:(UInt32) crc ofByte: (UInt8) byte
{
int8_t i;
crc = crc ^ byte;
for (i = 7; i >= 0; i--)
{
crc = (crc >> 1) ^ (0xedb88320ul & (-(crc & 1)));
}
return(crc);
}
The solved code with help of comments. Seems like I searched the mistake in the wrong place.
private void addByteToCRC(byte oneByte) {
crc ^= oneByte & 0xFF;
for (int i = 7; i >= 0; i--)
{
crc = (crc >>> 1) ^ (0xedb88320 & (-(crc & 1)));
}
}
Both checksums start of with a crc value of 0xFFFFFFFF.
The -(crc & 1) works fine, and is not your problem. What it does is to copy the low bit of crc to all of the bits, since in two's complement representation, –1 is all ones.
There are two things you need to fix for Java, which does not have unsigned integers. First is that the right shift needs to be a logical right shift, >>>. That avoids copying the sign bit when shifting down, and instead shifts in a zero bit. Second, you need to guard against a character being signed-extended to an int, so byte (which will need a different name in Java) needs to be anded with 0xff. So:
crc ^= octet & 0xff;
and
crc = (crc >>> 1) ^ (0xedb88320ul & (-(crc & 1)));
There are some differences between (Objective-)C and Java, but at a guess you may have identified the wrong problem. Java does not have unsigned integer types. However it does have bitwise operators which treat integers as just strings of bits, and later versions of Java do have unsigned operators which when applied to integers operate on the underlying bit patterns as though they represented unsigned values.
The expression crc & 1 will yield a result of 0 or 1 depending on whether crc was even or odd.
The expression -(crc & 1) therefore evaluates to 0 or -1. The bit pattern of 0 is all zeros and for -1 it is all ones - the latter is true in (Objective-)C regardless of the underlying representing of integers as 1's or 2's complement, see Is it safe to use -1 to set all bits to true? for an explanation.
The expression 0xedb88320ul & (-(crc & 1)) therefore evaluates to 0 or 0xedb88320.
The expression crc >> 1 is the logical, or unsigned, shift right in (Objective-)C as src is defined as unsigned (right shifts for signed integers are "implementation defined" in (Objective-)C).
The expression x ^ 0 for any x evaluates to x.
Put that all together and your whole expression (crc >> 1) ^ (0xedb88320ul & (-(crc & 1))) conditionally xors crc >> 1 with 0xedb88320 base don whether crc is odd or even.
Now you just need to reproduce the same behaviour in Java.
Java uses 32-bit 2's complement integers for int (JLS: Primitive Types and Values) and provides bitwise operators which treat them as just bit-strings.
The hexadecimal constant comes just 0xedb88320 and has int (32-bit) type.
The expression -(crc & 1) therefore evaluates as in (Objective-)C.
However the expression crc >> 1 will not evaluate the same as in (Objective-)C. In Java the >> is an arithmetic shift right, which replicates the sign-bit during the shift. This choice stems from Java using signed types. Fortunately Java also provides the >>> operator to perform a logical, or unsigned, right-shift.
So my guess is you may have identified the wrong problem in your conversion and you need to use the >>> operator.
HTH

How to shift 32 bit int by 32 (yet again)

Ok, so I know that typically left- and right- shifts are well defined only for values 0..31. I was thinking how to best extend this to include 32, which simplifies some algorithms. I came up with:
int32 << n & (n-32) >> 5
Which seems to work. Question is, is it guaranteed to work on any architecture (C, C++, Java), and can it be done more effectively?
In Java it's guaranteed to work if those variables are of type int, since >> in Java does an arithmetic right shift and shifting more than 31 also has defined behavior. But beware of operator precedence
int lshift(int x, int n)
{
return (x << n) & ((n-32) >> 5);
}
This will work for shift count up to 32. But it can be modified to include any int values with shift counts larger than 31 return 0
return (x << n) & ((n-32) >> 31);
However in C and C++ the size of int type and the behavior of >> operator is implementation defined. Most (if not all) modern implementations implement it as arithmetic shift for signed types though. Besides, the behavior of shifting more than the variable width is undefined. Worse yet, signed overflow invokes UB so even a left shift by 31 is also UB (until C++14). Therefore to get a well defined output you need to
Use a unsigned fixed-width type like uint32_t (so x << 31 isn't UB)
Use a compiler that emits arithmetic right shift instruction for >> and use a signed type for n, or implement the arithmetic shift yourself
Mask the shift amount to limit it to 5 bits for int32_t
The result would be
uint32_t lshift(uint32_t x, int32_t n)
{
return (x << (n & 0x1F)) & ((n-32) >> 31);
}
If the architecture supports conditional instructions like x86 or ARM then the following way may be faster
return n < 32 ? x << n : 0;
On a 64-bit platform you can made this even simpler by shifting in a 64-bit type and then mask. Some 32-bit platforms like ARM does support shifting by 32 so this method is also efficient
return ((uint64_t)x << (n & 0x3F)) & 0xFFFFFFFFU;
You can see the output assembly here. I don't see how it can be improved further

Why does right shift perform better than divide in JAVA? [duplicate]

Does the Java compiler or the JIT compiler optimize divisions or multiplications by a constant power of two down to bitshifting?
For example, are the following two statements optimized to be the same?
int median = start + (end - start) >>> 1;
int median = start + (end - start) / 2;
(basically this question but for Java)
While the accepted answer is right in the sense that the division can't be simply replaced by a right shift, the benchmark is terribly wrong. Any Java benchmark running for less than one second is probably measuring the interpreter's performance - not something you usually care about.
I couldn't resist and wrote an own benchmark which mainly shows that it's all more complicated. I'm not trying to fully explain the results, but I can say that
a general division is a damn slow operation
it gets avoided is much as possible
division by a constant gets AFAIK always somehow optimized
division by a power of two gets replaced by a right shift and an adjustment for negative numbers
a manually optimized expression might be better
No, the Java compiler doesn't do that, because it can't be sure on what the sign of (end - start) will be. Why does this matter? Bit shifts on negative integers yield a different result than an ordinary division. Here you can see a demo: this simple test:
System.out.println((-10) >> 1); // prints -5
System.out.println((-11) >> 1); // prints -6
System.out.println((-11) / 2); // prints -5
Also note that I used >> instead of >>>. A >>> is an unsigned bitshift, while >> is signed.
System.out.println((-10) >>> 1); // prints 2147483643
#Mystical: I wrote a benchmark, that shows that the compiler / JVM doesn't do that optimization: https://ideone.com/aKDShA
If the JVM does not do it, you can easily do it yourself.
As noted, right shifts on negative numbers do not behave the same as division because the result is rounded in the wrong direction. If you know that the dividend is non-negative, you can safely replace the division by a shift. If it might be negative, you can use the following technique.
If you can express your original code in this form:
int result = x / (1 << shift);
You can replace it with this optimized code:
int result = (x + (x >> 31 >>> (32 - shift))) >> shift;
Or, alternatively:
int result = (x + ((x >> 31) & ((1 << shift) - 1))) >> shift;
These formulae compensate for the incorrect rounding by adding a small number computed from the sign bit of the dividend. This works for any x with all shift values from 1 to 30.
If the shift is 1 (i.e., you are dividing by 2) then the >> 31 can be removed in the first formula to give this very tidy snippet:
int result = (x + (x >>> 31)) >> 1;
I have found these techniques to be faster even when the shift is non-constant, but obviously they benefit the most if the shift is constant. Note: For long x instead of int, change 31 and 32 respectively to 63 and 64.
Examining the generated machine code shows that (unsurprisingly) the HotSpot Server VM can do this optimization automatically when the shift is constant, but (also unsurprisingly) the HotSpot Client VM is too stupid to.

Rotating a 4 bit integer

I have an int 00000000000000000000000000001101 which represents 13 in base ten. I am trying to circular rotate the the bits by treating the 32 bit integer as a 4 bit integer because if I rotate the integer the value becomes very large. My desired answer after a right rotation of 2 for the above example would be 00000000000000000000000000000111 which is 7 in base 10.
Any help on doing this is greatly appreciated.
Try this:
x = (x >> 2) | ((x & 3) << 2);
This is just simulating the rotation via shifts and masks. I don't think there is anything better you could do, short of maybe making a lookup table (which may not actually be better); the CPU doesn't have opcodes for natively dealing with nybbles.
To rotate the lower 4 bits to the right by n, where n is 1, 2, or 3:
((x >> n) | (x << (4-n))) & 0xF;
The first part shifts the leftmost 4-n bits to the right; the second part shifts the rightmost n bits to the left. Then you or them together, and use & 0xF to zero extra bits that may have been set by the left shift.

What is the purpose of the unsigned right shift operator ">>>" in Java?

I understand what the unsigned right shift operator ">>>" in Java does, but why do we need it, and why do we not need a corresponding unsigned left shift operator?
The >>> operator lets you treat int and long as 32- and 64-bit unsigned integral types, which are missing from the Java language.
This is useful when you shift something that does not represent a numeric value. For example, you could represent a black and white bit map image using 32-bit ints, where each int encodes 32 pixels on the screen. If you need to scroll the image to the right, you would prefer the bits on the left of an int to become zeros, so that you could easily put the bits from the adjacent ints:
int shiftBy = 3;
int[] imageRow = ...
int shiftCarry = 0;
// The last shiftBy bits are set to 1, the remaining ones are zero
int mask = (1 << shiftBy)-1;
for (int i = 0 ; i != imageRow.length ; i++) {
// Cut out the shiftBits bits on the right
int nextCarry = imageRow & mask;
// Do the shift, and move in the carry into the freed upper bits
imageRow[i] = (imageRow[i] >>> shiftBy) | (carry << (32-shiftBy));
// Prepare the carry for the next iteration of the loop
carry = nextCarry;
}
The code above does not pay attention to the content of the upper three bits, because >>> operator makes them
There is no corresponding << operator because left-shift operations on signed and unsigned data types are identical.
>>> is also the safe and efficient way of finding the rounded mean of two (large) integers:
int mid = (low + high) >>> 1;
If integers high and low are close to the the largest machine integer, the above will be correct but
int mid = (low + high) / 2;
can get a wrong result because of overflow.
Here's an example use, fixing a bug in a naive binary search.
Basically this has to do with sign (numberic shifts) or unsigned shifts (normally pixel related stuff).
Since the left shift, doesn't deal with the sign bit anyhow, it's the same thing (<<< and <<)...
Either way I have yet to meet anyone that needed to use the >>>, but I'm sure they are out there doing amazing things.
As you have just seen, the >> operator automatically fills the
high-order bit with its previous contents each time a shift occurs.
This preserves the sign of the value. However, sometimes this is
undesirable. For example, if you are shifting something that does not
represent a numeric value, you may not want sign extension to take
place. This situation is common when you are working with pixel-based
values and graphics. In these cases you will generally want to shift a
zero into the high-order bit no matter what its initial value was.
This is known as an unsigned shift. To accomplish this, you will use
java’s unsigned, shift-right operator,>>>, which always shifts zeros
into the high-order bit.
Further reading:
http://henkelmann.eu/2011/02/01/java_the_unsigned_right_shift_operator
http://www.java-samples.com/showtutorial.php?tutorialid=60
The signed right-shift operator is useful if one has an int that represents a number and one wishes to divide it by a power of two, rounding toward negative infinity. This can be nice when doing things like scaling coordinates for display; not only is it faster than division, but coordinates which differ by the scale factor before scaling will differ by one pixel afterward. If instead of using shifting one uses division, that won't work. When scaling by a factor of two, for example, -1 and +1 differ by two, and should thus differ by one afterward, but -1/2=0 and 1/2=0. If instead one uses signed right-shift, things work out nicely: -1>>1=-1 and 1>>1=0, properly yielding values one pixel apart.
The unsigned operator is useful either in cases where either the input is expected to have exactly one bit set and one will want the result to do so as well, or in cases where one will be using a loop to output all the bits in a word and wants it to terminate cleanly. For example:
void processBitsLsbFirst(int n, BitProcessor whatever)
{
while(n != 0)
{
whatever.processBit(n & 1);
n >>>= 1;
}
}
If the code were to use a signed right-shift operation and were passed a negative value, it would output 1's indefinitely. With the unsigned-right-shift operator, however, the most significant bit ends up being interpreted just like any other.
The unsigned right-shift operator may also be useful when a computation would, arithmetically, yield a positive number between 0 and 4,294,967,295 and one wishes to divide that number by a power of two. For example, when computing the sum of two int values which are known to be positive, one may use (n1+n2)>>>1 without having to promote the operands to long. Also, if one wishes to divide a positive int value by something like pi without using floating-point math, one may compute ((value*5468522205L) >>> 34) [(1L<<34)/pi is 5468522204.61, which rounded up yields 5468522205]. For dividends over 1686629712, the computation of value*5468522205L would yield a "negative" value, but since the arithmetically-correct value is known to be positive, using the unsigned right-shift would allow the correct positive number to be used.
A normal right shift >> of a negative number will keep it negative. I.e. the sign bit will be retained.
An unsigned right shift >>> will shift the sign bit too, replacing it with a zero bit.
There is no need to have the equivalent left shift because there is only one sign bit and it is the leftmost bit so it only interferes when shifting right.
Essentially, the difference is that one preserves the sign bit, the other shifts in zeros to replace the sign bit.
For positive numbers they act identically.
For an example of using both >> and >>> see BigInteger shiftRight.
In the Java domain most typical applications the way to avoid overflows is to use casting or Big Integer, such as int to long in the previous examples.
int hiint = 2147483647;
System.out.println("mean hiint+hiint/2 = " + ( (((long)hiint+(long)hiint)))/2);
System.out.println("mean hiint*2/2 = " + ( (((long)hiint*(long)2)))/2);
BigInteger bhiint = BigInteger.valueOf(2147483647);
System.out.println("mean bhiint+bhiint/2 = " + (bhiint.add(bhiint).divide(BigInteger.valueOf(2))));

Categories