sun qingyao - 6 months ago 44

C Question

The website in which I found this code

`int v, sign;`

// or, to avoid branching on CPUs with flag registers (IA32):

sign = -(int)((unsigned int)((int)v) >> (sizeof(int) * CHAR_BIT - 1)); // if v < 0 then -1, else 0.

This statement assigns variable sign with the sign of variable v (either -1 or 0). I wonder why

`(int)((unsigned int)((int)v)`

Answer

Quoting the C Standard 6.5.7p5:

The result of E1 >> E2 is E1 right-shifted E2 bit positions. If E1 has an unsigned type or if E1 has a signed type and a nonnegative value, the value of the result is the integral part of the quotient of E1 / 2E2. If E1 has a signed type and a negative value, the resulting value is implementation-defined.

The author is writing on how to implement a function `sign(int v)`

that returns `-1`

for negative numbers and `0`

for 0 and positive numbers efficiently. A naive approach is this:

```
int sign(int v) {
if (v < 0)
return -1;
else
return 0;
}
```

But this solution may compile to code that performs a comparison and branch on CPU flags set by the comparison. This is inefficient. He proposes a simpler and more direct solution:

```
sign = -(v > 0);
```

But this computation still requires compare and branch on CPUs that do not produce comparison results directly as boolean values. CPUs with flag registers typically set various flags upon comparison instructions or even most arithmetic instruction. So he proposes another solution based on shifting the sign bit, but as the Standard specifies above, he cannot rely on the result of right shifting a negative value.

Casting `v`

as `unsigned`

removes this problem because right shifting unsigned values is well specified. Assuming the sign bit is in the highest position, which is true for all modern processors but not mandated by the C standard, right shifting `(unsigned)v`

by the one less than the number of bits in its type produces a value of `1`

for negative values and `0`

otherwise. Negating the result should produce the expected values `-1`

for negative `v`

and `0`

for positive and zero `v`

. But the expression is unsigned, so plain negation vill produce `UINT_MAX`

or `0`

, which in turn causes arithmetic overflow when stored to an `int`

or even just cast as an `(int)`

. Casting this result back to `int`

before negating it correctly computes the desired result, `-1`

for negative `v`

and `0`

for positive or zero `v`

.

Arithmetic overflow is usually benign and widely ignored by most programmers, but modern compilers tend to take advantage of its undefinedness to perform aggressive optimizations, so it is unwise to rely on *expected* but unwarranted behavior and best to avoid arithmetic overflow in all cases.

The expression could be simplified as:

```
sign = -(int)((unsigned)v >> (sizeof(int) * CHAR_BIT - 1));
```

Note that if right shifting is defined as replicating the bit for your platform (an almost universal behavior with current CPUs), the expression would be much simpler (assuming `int v`

):

```
sign = v >> (sizeof(v) * CHAR_BIT - 1)); // works on x86 CPUs
```

The bithacks page https://graphics.stanford.edu/~seander/bithacks.html , very instructive indeed, contains a detailed explanation:

```
int v; // we want to find the sign of v
int sign; // the result goes here
// CHAR_BIT is the number of bits per byte (normally 8).
sign = -(v < 0); // if v < 0 then -1, else 0.
// or, to avoid branching on CPUs with flag registers (IA32):
sign = -(int)((unsigned int)((int)v) >> (sizeof(int) * CHAR_BIT - 1));
// or, for one less instruction (but not portable):
sign = v >> (sizeof(int) * CHAR_BIT - 1);
```

*The last expression above evaluates to sign = v >> 31 for 32-bit integers. This is one operation faster than the obvious way, sign = -(v < 0). This trick works because when signed integers are shifted right, the value of the far left bit is copied to the other bits. The far left bit is 1 when the value is negative and 0 otherwise; all 1 bits gives -1. Unfortunately, this behavior is architecture-specific.*

As an epilogue, I would recommend to use the *most readable* version and rely on the compiler to produce the most efficient code:

```
sign = -(v < 0);
```

As can be verified on this enlightening page: http://gcc.godbolt.org/# compiling the above code with `gcc -O3 -std=c99 -m64`

indeed produces the code below for all solutions above, even the most naive `if`

/`else`

statement:

```
sign(int):
movl %edi, %eax
sarl $31, %eax
ret
```