Performance counters and LARGE

Performance counters and LARGE_INTEGER values

Thread view:

Enable EMail Alerts

Start New Thread

Thread rating:

Bob Altman - 20 Mar 2005 17:51 GMT

Hi all,

QueryPerformanceCounter returns a LARGE_INTEGER, which the documentation
states is equivalent to a signed 64-bit integer. My question is, can
QueryPerformanceCounter return negative values, or is its output restricted
to positive values? TIA!

Reply to this Message

Jochen Kalmbach - 20 Mar 2005 18:09 GMT

Hi Bob Altman,

> QueryPerformanceCounter returns a LARGE_INTEGER, which the
> documentation states is equivalent to a signed 64-bit integer. My
> question is, can QueryPerformanceCounter return negative values, or is
> its output restricted to positive values? TIA!

QueryperformanceCounter internaly sometimes uses the RDTSC instruction. And
this value is an unsigned 64-bit value. Therefor if you interpret this
value as signed, it might become negative...

BUT: On current processor speeds (3GHz) it will take about 95 years of a
continuous running CPU to reach the "negative" value.

Signature

Greetings
Jochen

My blog about Win32 and .NET
http://blog.kalmbachnet.de/

Reply to this Message

Tim Robinson - 21 Mar 2005 01:23 GMT

>> QueryPerformanceCounter returns a LARGE_INTEGER, which the
>> documentation states is equivalent to a signed 64-bit integer. My
[quoted text clipped - 5 lines]
> this value is an unsigned 64-bit value. Therefor if you interpret this
> value as signed, it might become negative...

To be exact, QPC/QPF use different counters under different circumstances.
On my system, the clock used for these functions is much slower than the CPU
clock; I believe it's using some clock on the motherboard. If you think
about the multi-CPU case, RDTSC stops making sense.

Signature

Tim Robinson
MVP, Windows SDK

Reply to this Message

Jochen Kalmbach - 21 Mar 2005 07:35 GMT

Hi Tim Robinson,

>>> QueryPerformanceCounter returns a LARGE_INTEGER, which the
>>> documentation states is equivalent to a signed 64-bit integer. My
[quoted text clipped - 11 lines]
> motherboard. If you think about the multi-CPU case, RDTSC stops making
> sense.

I just wanted to show that in the fastes possible implementation (RDTSC),
it will take 95 years to become negative!

Signature

Greetings
Jochen

My blog about Win32 and .NET
http://blog.kalmbachnet.de/

Reply to this Message

Tim Robinson - 21 Mar 2005 21:12 GMT

[...]

>> To be exact, QPC/QPF use different counters under different
>> circumstances. On my system, the clock used for these functions is
[quoted text clipped - 4 lines]
> I just wanted to show that in the fastes possible implementation (RDTSC),
> it will take 95 years to become negative!

A fair point :).

/me goes and tests Jochen's theory

Signature

Tim Robinson
MVP, Windows SDK

Reply to this Message

- 22 Mar 2005 03:45 GMT

But the TSC can be user-set to any value, at
any time. Not that you'd want to, but anyone
(typically) could.

Signature

40th Floor - Software @ http://40th.com/
iPlay : the ultimate audio player for PPCs
mp3,mp4,m4a,aac,ogg,flac,wav,play & record
parametric eq, xfeed, reverb: all on a ppc

Reply to this Message

Jochen Kalmbach - 22 Mar 2005 07:28 GMT

Hi ,

> But the TSC can be user-set to any value, at
> any time. Not that you'd want to, but anyone
> (typically) could.

Not 100% true...

<quote>
When the WRMSR instruction is used to write to the time-stamp
counter, only the low order 32-bits of the time-stamp counter can be
written to, and the highorder 32 bits cleared to all 0s.
</quote>

Signature

Greetings
Jochen

My blog about Win32 and .NET
http://blog.kalmbachnet.de/

Reply to this Message

- 22 Mar 2005 10:00 GMT

JK- [Mon, 21 Mar 2005 22:28:33 -0800]:

>Not 100% true...

Right you are ... not the full 64 bits since the P6,
apparently all 64 in the P5. Do you have a P5 to
try? I've got some...on the wall. Smart move in
any case.

Signature

40th Floor - Software @ http://40th.com/
iPlay : the ultimate audio player for PPCs
mp3,mp4,m4a,aac,ogg,flac,wav,play & record
parametric eq, xfeed, reverb: all on a ppc

Reply to this Message

Performance counters and LARGE_INTEGER values

Free Magazines