Re: [9fans] prof

Re: [9fans] prof

[Russ Cox]

the problem is likely that rio uses 4-space
tabs, while prof assumes 8-space.  try

prof | pr -t -e8

and the output will probably make more sense.

the extra . on the lock line as compared with
the vprintf line shows that they're not actually
lined up as they seem to be.

Re: [9fans] prof

[D. Brownlee]

forsyth@vitanuova.com wrote:
> 
> >>Another question: has anyone done anything with the
> >>compiler to generate a static flow graph -- something
> >>like 'cflow'?
> 
> i have changes to one of the linkers somewhere.
> it also checked stack depth.

I got curious and came up with this after a minor change:

%term diff /sys/src/cmd/8l/pass.c pass.c
306a307
>		mydiag("%s calls %s", TNAME, s->name);
308d308
<			diag("undefined: %s in %s\n", s->name, TNAME);

void mydiag(char *fmt, ...); -- after diag() in list.c


This yields alot of lines of the form:

	caller calls callee

It looks like a fairly simple job for 'awk', maybe 'rc', to
process those lines and produce the flow graph. Thanks for
the clue about using the linker instead of the compiler; I'll
even get things that were written in assembler.

Later,

D. Brownlee

Re: [9fans] prof

[forsyth]

[-- Attachment #1: Type: text/plain, Size: 90 bytes --]

the linker knows all.
i'll try to unearth the code, or reproduce it.
it wasn't hard.


[-- Attachment #2: Type: message/rfc822, Size: 2245 bytes --]

From: "D. Brownlee" <ancipites@earthlink.net>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] prof
Date: Fri, 11 Aug 2000 11:22:17 -0700
Message-ID: <39944459.C86EAAFE@earthlink.net>

It would be nice to see the linker
change. I was thinking of modifying
the compiler's 'codgen' to create a
bunch of temp files that identify
function definitions and the functions
called within them, then process those
to get a static flow graph.

forsyth@vitanuova.com wrote:
> 
> >>Another question: has anyone done anything with the
> >>compiler to generate a static flow graph -- something
> >>like 'cflow'?
> 
> i have changes to one of the linkers somewhere.
> it also checked stack depth.

Re: [9fans] prof

[D. Brownlee]

It would be nice to see the linker
change. I was thinking of modifying
the compiler's 'codgen' to create a
bunch of temp files that identify
function definitions and the functions
called within them, then process those
to get a static flow graph.

forsyth@vitanuova.com wrote:
> 
> >>Another question: has anyone done anything with the
> >>compiler to generate a static flow graph -- something
> >>like 'cflow'?
> 
> i have changes to one of the linkers somewhere.
> it also checked stack depth.

Re: [9fans] prof

[forsyth]

>>Another question: has anyone done anything with the
>>compiler to generate a static flow graph -- something
>>like 'cflow'?

i have changes to one of the linkers somewhere.
it also checked stack depth.

Re: [9fans] prof

[D. Brownlee]

Whoops! That should be:

main:0
.  f:0
.    printf:0
.      vprintf:0
.    qlock:0
.    .  lock:0
etc.



D. Brownlee wrote:
> 
deleted
> 
> produces, with 'prof -d' the following output:
> 
> main:0
> .  printf:0
> .    vfprintf:0
> .  qlock:0
> .  .  lock:0
> etc.
> 
> This indicates that 'main' calls 'qlock', but I
> think that 'qlock' is actually called by 'vfprintf'. (?)

[9fans] prof

[D. Brownlee]

Hello,

This program:

	main()
	{
		void f(char *);

		f("hello\n");
	}

	f(char *s)
	{
		printf("%s", s);
	}

produces, with 'prof -d' the following output:


main:0
.  printf:0
.    vfprintf:0
.  qlock:0
.  .  lock:0
etc.

This indicates that 'main' calls 'qlock', but I
think that 'qlock' is actually called by 'vfprintf'. (?)

Another question: has anyone done anything with the
compiler to generate a static flow graph -- something
like 'cflow'?

D. Brownlee