question of double/ float

[Joakim Hove]

Hello,

consider a function:

double some_func(double arg) {
/*

*/
}

When this function is called with a float argument, and the result
assigned to a float variable like this:

float float_arg;
float float_result;

float_result = some_func(float_arg);

Are the necessary conversions guaranteed tok place automatically? And
if this is autoamtically taken care of, why does the standard (C99)
require functions like logf and sinf. Is it just to gain speed by
avoiding conversions, and possible using a 'cruder' algorithm?

Regards

Joakim

 

[Richard Tobin]

consider a function:

double some_func(double arg) {
/*

*/
}

When this function is called with a float argument, and the result
assigned to a float variable like this:

float float_arg;
float float_result;

float_result = some_func(float_arg);

Are the necessary conversions guaranteed tok place automatically?

Yes. The prototype tells the compiler the required types, and it
does the conversion. You could even use ints.

And
if this is autoamtically taken care of, why does the standard (C99)
require functions like logf and sinf. Is it just to gain speed by
avoiding conversions, and possible using a 'cruder' algorithm?

Yes. Many processors have built-in instructions for these operations
in a variety of sizes.

-- Richard

 

[Joakim Hove]

Thanks for answering

Yes. The prototype tells the compiler the required types, and it
does the conversion.

But, the compiler can only do this in terms of literals, otherwise it
must insert a call to a double -> float conversion function?

Yes. Many processors have built-in instructions for these operations
in a variety of sizes.

Good - then I will stick to one function implementation for both float
and double.

Joakim

 

[Richard Tobin]

Yes. The prototype tells the compiler the required types, and it
does the conversion.

[/QUOTE]

But, the compiler can only do this in terms of literals, otherwise it
must insert a call to a double -> float conversion function?

Yes, I didn't mean "do the conversion at compile time", I meant "arrange
for the conversion to be done".

-- Richard

 

[Chris Dollin]

Thanks for answering


But, the compiler can only do this in terms of literals,

No. The compiler can -- must -- arrange for the conversion of
any legal value to the required type.

otherwise it
must insert a call to a double -> float conversion function?

No.

There's no "must". If there's an instruction to hand, it can -- one
might say should -- use that.

Good - then I will stick to one function implementation for both float
and double.

Was your concern correctness or performance?

 

[Rg]

[...]

float float_arg;
float float_result;

float_result = some_func(float_arg);

Are the necessary conversions guaranteed tok place automatically? [...]

Yes, but keep in mind that conversions from "big" types to "smaller"
types (such as double to float) may incur in loss of precision or of
significance.

 

[Keith Thompson]

[...]

float float_arg;
float float_result;

float_result = some_func(float_arg);

Are the necessary conversions guaranteed tok place automatically? [...]

Yes, but keep in mind that conversions from "big" types to "smaller"
types (such as double to float) may incur in loss of precision or of
significance.

Such a conversion can lose precision, *or* it can overflow if the
double value is too big to be represented as a float. In that case,
the behavior is undefined.