RE: patch for hp-timing.h

["Tang, Jun" <juntangc@amazon.com>]

Wilco,

Thanks for the suggestion, 

There are 7 zero bits at the end of freq (1050000000).  So the time can be extended to 2048 second before losing resolution.  Below is the updated patch for 1024 sec -

-   (Diff) = ((End) - (Start)) * (UINT64_C(1000000000) / freq); \
+   (Diff) = (((End) - (Start)) * UINT64_C(1000000000>>6)) / (freq>>6); \
 
Regards,

Jun
 
-----Original Message-----
From: Wilco Dijkstra <Wilco.Dijkstra@arm.com> 
Sent: Monday, February 6, 2023 10:54 AM
To: Tang, Jun <juntangc@amazon.com>
Cc: 'GNU C Library' <libc-alpha@sourceware.org>
Subject: RE: [EXTERNAL]patch for hp-timing.h

CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.



Hi Jun,

> Removing zero bits from multiply and making shifts afterwards can 
> extend the time before overflow.  Does it also reduce the resolution of the timer?

It wouldn't if we shift out zero bits. For example 1000000 has 6 zero bits at the bottom, so assuming every implementation uses a multiple of 1MHz then there is no loss of accuracy. And 6 extra bits would mean 1024 seconds before overflow at 1GHz which seems long enough.

Cheers,
Wilco