Brianegge at 13:41, 30 January 2013

2013-01-30T13:41:21Z

← Older revision		Revision as of 13:41, 30 January 2013
Line 19:		Line 19:
	Using gcc 3.4.4, I compiled the above program. Here's the assembly it generated.		Using gcc 3.4.4, I compiled the above program. Here's the assembly it generated.

	<geshi>		<geshi lang="asm">
	cmpl $1, 8(%ebp)		cmpl $1, 8(%ebp)
	jne L2		jne L2
Line 33:		Line 33:

	Recompiling the program with [http://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html -O3] generates the following:		Recompiling the program with [http://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html -O3] generates the following:
	<geshi>		<geshi lang="asm">
	xorl %eax, %eax		xorl %eax, %eax
	cmpl $1, 8(%ebp)		cmpl $1, 8(%ebp)
Line 57:		Line 57:

	Without optimization, we create the same assembly as before:		Without optimization, we create the same assembly as before:
	<geshi>		<geshi lang="asm">
	cmpl $1, 8(%ebp)		cmpl $1, 8(%ebp)
	jne L2		jne L2
Line 71:		Line 71:

	Turning the optimization on results in the same optimization which was made when we had the ternary operator.		Turning the optimization on results in the same optimization which was made when we had the ternary operator.
	<geshi>		<geshi lang="asm">
	cmpl $1, 8(%ebp)		cmpl $1, 8(%ebp)
	leave		leave

Brianegge: Created page with 'In writing C, I've often liked the ternary operator. It allows expressions to be computed in line, and shortens the amount of code one has to write. I've also believed that any…'

2009-11-29T21:57:07Z

Created page with 'In writing C, I've often liked the ternary operator. It allows expressions to be computed in line, and shortens the amount of code one has to write. I've also believed that any…'

New page

In writing C, I've often liked the ternary operator. It allows expressions to be computed in line, and shortens the amount of code one has to write. I've also believed that any ternary expression could be turned into a formula without any branching instructions. This can be important in a tight loop, as deeper pipelining can occur. In other languages which don't support the ternary operator, I've often used boolean expressions to achieve the same result. The following two expressions should be equivalent:

<geshi lang="c">
i = expr ? value1 : value2;
// or
i = expr * value1 + (!expr) * value2;
</geshi>

This works for languages in which boolean expressions evaluate to 0 or 1. There in C is reasons that any ternary operator could be turned into a function.

Consider the following program:

<geshi lang="c">
int main(int argc, char *argv[]) {
return argc == 1 ? 2 : 3;
}
</geshi>

Using gcc 3.4.4, I compiled the above program. Here's the assembly it generated.

<geshi>
cmpl $1, 8(%ebp)
jne L2
movl $2, -4(%ebp)
jmp L3
L2:
movl $3, -4(%ebp)
L3:
movl -4(%ebp), %eax
leave
ret
</geshi>

Recompiling the program with [http://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html -O3] generates the following:
<geshi>
xorl %eax, %eax
cmpl $1, 8(%ebp)
leave
setne %al
addl $2, %eax
ret
</geshi>

So, in the first case, it does use branching instructions, but in the second case it inlines the result.

What would happen if we wrote the same code procedurally?

<geshi lang="c">
int main(int argc, char *argv[]) {
if (argc == 1) {
return 2;
} else {
return 3;
}
}
</geshi>

Without optimization, we create the same assembly as before:
<geshi>
cmpl $1, 8(%ebp)
jne L2
movl $2, -4(%ebp)
jmp L1
L2:
movl $3, -4(%ebp)
L1:
movl -4(%ebp), %eax
leave
ret
</geshi>

Turning the optimization on results in the same optimization which was made when we had the ternary operator.
<geshi>
cmpl $1, 8(%ebp)
leave
setne %al
addl $2, %eax
ret
</geshi>

In conclusion then, the ternary operator offers no performance benefit over using a control structure, and is simply a matter of style and preference.

[[Category:c]]
[[Category:Performance]]

Ternary Operator in C - Revision history

Brianegge at 13:41, 30 January 2013

Brianegge: Created page with 'In writing C, I've often liked the ternary operator. It allows expressions to be computed in line, and shortens the amount of code one has to write. I've also believed that any…'