On 06/20/2018 10:33 AM, David Hildenbrand wrote:
> On 20.06.2018 21:33, Richard Henderson wrote:
>> On 06/20/2018 12:08 AM, David Hildenbrand wrote:
>>> +/* Converts ns to s390's clock format */
>>> +static inline uint64_t time2tod(uint64_t ns)
>>> +{
>>> + return (ns << 9) / 125;
>>> +}
>>> +
>>> +/* Converts s390's clock format to ns */
>>> +static inline uint64_t tod2time(uint64_t t)
>>> +{
>>> + return (t * 125) >> 9;
>>> +}
>>
>
> In this patch I'm only moving the code. If we find this is a problem,
> this should go into a separate patch.
Ah, right.
>> How many significant bits on input here?
>
> Basically all are significant, and as it is a clock, we will reach these
> bits at one point.
>
>> Do you in fact want to be using muldiv64?
>
> Looking at linux:
>
> arch/s390/include/asm/timex.h
>
> They have a lengthy documentation, resulting in (a spli to avoid overflows)
>
> return ((todval >> 9) * 125) + (((todval & 0x1ff) * 125) >> 9);
>
> Maybe we should do the same?
That would work too.
r~
