On 4 July 2014 18:13, Edward Nevill edward.nevill@linaro.org wrote:

Hi,

Does anyone know of a nice way to detect 64 x 64 -> 64 multiply overflow on aarch64?

On x86 you can just use the V flag, but aarch64 seems to have no option to set the flags after a MUL instruction which I find very odd.

The only solution I can think of is to do the equivalent of

r = a * b; if (r/a != b) overflow();

There is some short circuiting you can do to avoid the div if you know there is no possibility it could have overflowed.

Eg.

if (clz(labs(a)) + clz(labs(b)) < 32) { r = a * b; if (r/a != b) overflow(); }

Then there is the special case of a = 0x8000000000000000 and b == -1.

All in all its a mess just to detect overflow.

If you are saying AArch64, I am assuming you can use inline asm and umulh, no?