Support for floating-point operations

The ARM processor core does not contain floating-point hardware. Floating-point arithmetic must be supported separately, either:

The differences between software and hardware support for floating-point arithmetic are illustrated with Example 4.1, which shows a function implementing floating-point arithmetic operations in C.

float foo(float num1, float num2)
{
    float temp, temp2;

    temp = num1 + num2;
    temp2 = num2 * num2;
    return temp2-temp;
}

When the C code of Example 4.1 is compiled with the command-line option --softvfp, the compiler produces machine code with the disassembly of Example 4.2. In this example, floating-point arithmetic is performed in software through calls to library routines such as __aeabi_fmul.

||foo|| PROC
    PUSH     {r4-r6, lr}
    MOV      r4, r1
    BL       __aeabi_fadd
    MOV      r5, r0
    MOV      r1, r4
    MOV      r0, r4
    BL       __aeabi_fmul
    MOV      r1, r5
    POP      {r4-r6, lr}
    B        __aeabi_fsub
    ENDP

When the C code of Example 4.1 is compiled with the command-line option --vfp, the compiler produces machine code with the disassembly of Example 4.3. In this example, floating-point arithmetic is performed in hardware through floating-point arithmetic instructions such as VMUL.F32.

||foo|| PROC
    VADD.F32 s2, s0, s1
    VMUL.F32 s0, s1, s1
    VSUB.F32 s0, s0, s2
    BX       lr
    ENDP

In practice, code that makes use of hardware support for floating-point arithmetic is more compact and offers better performance than code that performs floating-point arithmetic in software. However, hardware support for floating-point arithmetic requires a VFP coprocessor.

The default option is to support floating-point arithmetic using the software floating-point library fplib. fplib is available as part of the standard distribution of the RealView Development Suite C libraries.