Forum Discussion

Altera_Forum's avatar
Altera_Forum
Icon for Honored Contributor rankHonored Contributor
16 years ago

Optimizing code for II/e vs II/f

Hi,

I noticed when I compiled for II/e that it's putting a bunch of multiply instructions in the assembler, where I would not normally want them. Example: Subtracting two pointer ended up using a multiply. Obviously, I want not multiplies generated in the assembler for II/e.

Does the compiler not receive a hint as to whether it's compiling for II/e vs II/f? Can I tell to that I want to compile for II/e? Can I tell it not to use multiplies?

I know they're basically using GCC, which has a -b and -m option, but I'm not sure they're supported for the Nios II cross-compiled tools.

Thanks, Peter

1 Reply

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    16 years ago

    Subtracting 2 pointers is more likely to lead to a divide than a multiply!

    Adding an integral value to a pointer is likely to generate a multiply.

    If the compiler doesn't generate a multiply, it can only generate a shift instead, and fast shifts require the 'mul' instruction.

    Certainly a sequence of shifts and adds (which would likely be faster than a multiply on a system with hardware shift but not multiply) is likely to always be slower on a Nios2.

    Try -fno-hw-mul