Mail Archives: djgpp/1997/11/11/16:28:55
Message-Id: | <3.0.5.32.19971111162720.00a0dd30@modempool.com>
|
Date: | Tue, 11 Nov 1997 16:27:20 -0500
|
To: | djgpp AT delorie DOT com
|
From: | bpmann AT modempool DOT com (Brian Mann)
|
Subject: | Re: Curious: How does DJGPP profile __dpmi_int?
|
In-Reply-To: | <199711112016.MAA18277@adit.ap.net>
|
Mime-Version: | 1.0
|
At 12:16 PM 11/11/97 -0800, Nate Eldredge wrote:
>At 10:21 11/10/1997 +0200, Eli Zaretskii wrote:
>>
>>On Sun, 9 Nov 1997, Nate Eldredge wrote:
>>
>>> That's my question. How is it possible for __dpmi_int to show up on
>>> profiling outputs of DJGPP programs? I looked at its sources, expecting to
>>> find a call to _mcount (the function that's called on entry to a profiled
>>> function and sets up its profiling), but I didn't.
>>
>>I think it's for the same reason other library functions appear in
>>the profile, although the library wasn't compiled with -pg.
>>
>>The reason is that the profile is also gathered on timer tick
>>interrupts. The interrupt handler records the EIP and later
>>gprof translates it into a symbol by looking at the debug info
>>in the executable (much like the debugger does).
>Oh, okay. It just tends to show up more often since it usually takes longer.
>I get it. Thanks.
>>
>>> It's a nice feature, but how the heck is it done? I think it's magic.
>>
>>"Any sufficiently high technology is indistinguishable from magic."
>>(A. Einstein)
How about Arthur C. Clarke?<grin>
Brian
>Right. :)
>
>Nate Eldredge
>eldredge AT ap DOT net
>
>
>
>
- Raw text -