Atmel CAVR-4 Manual De Usuario

CAVR-4

Part1. Using the compiler

For details about representation of supported data types, pointers, and structures types, 
see the chapter Data representation.

Using floating-point types on a microprocessor without a math coprocessor is very 
inefficient, both in terms of code size and execution speed. The AVR IAR C/C++ 
Compiler supports two floating-point formats—32 and 64 bits. The 32-bit floating-point 
type 

float

 is more efficient in terms of code size and execution speed. However, the 

64-bit format 

double

 supports higher precision and larger numbers.

In the AVR IAR C/C++ Compiler, the floating-point type 

float

 always uses the 32-bit 

format. The format used by the 

double

 floating-point type depends on the compiler 

option 

--64bit_doubles

 (Use 64-bit doubles).

Unless the application requires the extra precision that 64-bit floating-point numbers 
give, we recommend using 32-bit floats instead. Also consider replacing code using 
floating-point operations with code using integers because these are more efficient.

Note that a floating-point constant in the source code is treated as being of the type 

double

. This can cause innocent-looking expressions to be evaluated in double 

precision. In the example below 

a

 is converted from a 

float

 to a 

double

, 

1

 is added 

and the result is converted back to a 

float

:

float test(float a)

{

   return a+1.0;

}

To treat a floating-point constant as a 

float

 rather than as a 

double

, add an 

f

 to it, for 

example:

float test(float a)

{

   return a+1.0f;

}