On 08/22/2013 07:48 PM, Paolo Bonzini wrote:
> Il 22/08/2013 11:47, Peter Maydell ha scritto:
>> On 22 August 2013 10:09, Paolo Bonzini <[email protected]> wrote:
>>> Il 22/08/2013 10:20, Alexey Kardashevskiy ha scritto:
>>>> +static inline Int128 int128_exts64(int64_t a)
>>>> +{
>>>> + return (Int128) { .lo = a, .hi = (a >> 63) ? -1 : 0 };
>>>> +}
>>>
>>> The "? -1 : 0" is not necessary, but the compiler will remove it at -O1
>>> or more (interestingly, or -O0 it will remove the shift and leave the
>>> conditional!).
>>
>> We can avoid relying on implementation defined
>> behaviour here by using
>> .hi = (a < 0) ? -1 : 0;
>>
>> (I know we allow ourselves to assume right-shift of signed
>> ints is arithmetic shift, but I think it's nicer to avoid it unless
>> it really makes the code better.)
>
> This is what Alexey proposed. I suggested (a >> 63) without the ?: but
> he misunderstood my (probably not clear enough) suggestion.
Yes, I misunderstood. It was not obvious to me that (signed long
long)-1>>63 will be still -1. I really (really) envy people who can easily
read stuff like but I cannot :(
1) return (Int128) { .lo = a, .hi = (a < 0) ? -1 : 0 };
2) return (Int128) { .lo = a, .hi = (a < 0) };
3) return (Int128) { .lo = a, .hi = a >> 63 };
So with which one should I repost the patch?
--
Alexey
