General profiling approaches

"fermineutron" <>
4 Nov 2006 18:43:11 -0500

          From comp.compilers

Related articles
General profiling approaches (fermineutron) (2006-11-04)
Re: General profiling approaches (fermineutron) (2006-11-11)
Re: General profiling approaches (Chris F Clark) (2006-11-13)
Re: General profiling approaches (Wolfram Fenske) (2006-11-15)
| List of all articles for this month |

From: "fermineutron" <>
Newsgroups: comp.compilers
Date: 4 Nov 2006 18:43:11 -0500
Organization: Compilers Central
Keywords: performance, question, debug
Posted-Date: 04 Nov 2006 18:43:11 EST

What do you think is the best way to profile Low Level Language? LLL
hence forth. Now, in HLL each statement would normally be translated
into a series of simpler statements so adding the profiler code may not
alter the resulting assembly code significantly. It seems that using C
statements within C to profile the C code is the biggest challenge
since C is close enough to asm to have a single line of C translated
into few assembly statements, hence the fraction of profiler code to
the profiled code is large.

What do you all think about the challenge of profiling C or simmilar
code? Can someone argue that there can be a method of profiling in
which the profiler does not have to know what initial code would look
like in asm? What i mean is, one could write a profiler which will
insert the score keeping code into compiled asm hence not altering the
way that the initial code would have been compiled without the
profilers involvment.his would be acurate, but than the profiler would
also have to the the compiler.

I am kind of curious about fundamental limitation of profiling the
code. In physics the uncertanty principle states that the product of
the errors of measuring position and momentum of a particle is allways
equal or greater than h-bar divided by 2. Now, due to fundamentally
different limitations of profiling to limitations of measuring position
and velosity of a particle, the reasoning behing profiling problem
would be different, but it has to have a certain fundamental limitation
due to the ratio of profiler and profiled code.

I am curious to hear your thought on this matter.
[People have been profiling C code for 30 years quite successfully.
Perhaps this would be a good time to go find out how they've done so.

Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.