[9front] prof profiler changing program result

[9front] prof profiler changing program result

[sirjofri]

Hello,

today I did an implementation of mandelbrot and learned to use prof to profile my program. I noticed that the mandelbrot result is different when linked using -p. Instead of the nice mandelbrot image I get a circle (but still with lots of details). Funnily enough, julia seems to work better.

Since my installation of 9front isn't always up to date that issue might already be fixed. I also doubt that the issue appears very often, and I didn't find any notes about issues like that when reading prof(1).

For reproduction, find this small repository: https://shithub.us/sirjofri/fractals/HEAD/info.html .

You should be able to just run `mk` to build the program, and use `mk test` to build sample images. The mkfile contains a commented out "-p" ldflag which you can use to reproduce the issue.

In general, I don't expect anyone to go this far at this stage (or at all). I'll probably take a deeper look and try to find out what's happening. Did anyone encounter other (or similar) issues when using prof and the -p flag?

sirjofri

Re: [9front] prof profiler changing program result

[Steve Simon]

i have a very distant memory that there was a bug with the setting of tos (top of stack) which broke prof, this never got fixed in APE that i saw.

also there where various different attempts at tweaking the implementation of nsec() which may have effected prof.

-Steve

Re: [9front] prof profiler changing program result

[Amavect]

On Sat, Dec 9, 2023 at 12:30 PM sirjofri <sirjofri+ml-9front@sirjofri.de> wrote:
>today I did an implementation of mandelbrot and learned to use prof to profile my program.
>I noticed that the mandelbrot result is different when linked using -p.
>Instead of the nice mandelbrot image I get a circle (but still with lots of details).
>Funnily enough, julia seems to work better.
You have uninitialized variables.
I guess kencc doesn't warn if you don't fully initialize a struct.
Acid showed me that SetParams params was initialized differently by
the time it reaches generate().
The .julia field gets an uninitialized value of 0 without -p, and an
uninitialized value of 1 with -p.
After setting a zero struct initializer {0}, the mandelbrot generates correctly.
After setting a struct initializer {1}, the mandelbrot generates
incorrectly in the same way.

Thanks,
Amavect

Re: [9front] prof profiler changing program result

[sirjofri]

13.12.2023 04:30:51 Amavect <amavect@gmail.com>:

> On Sat, Dec 9, 2023 at 12:30 PM sirjofri <sirjofri+ml-9front@sirjofri.de> wrote:
>> today I did an implementation of mandelbrot and learned to use prof to profile my program.
>> I noticed that the mandelbrot result is different when linked using -p.
>> Instead of the nice mandelbrot image I get a circle (but still with lots of details).
>> Funnily enough, julia seems to work better.
> You have uninitialized variables.
> I guess kencc doesn't warn if you don't fully initialize a struct.
> Acid showed me that SetParams params was initialized differently by
> the time it reaches generate().
> The .julia field gets an uninitialized value of 0 without -p, and an
> uninitialized value of 1 with -p.
> After setting a zero struct initializer {0}, the mandelbrot generates correctly.
> After setting a struct initializer {1}, the mandelbrot generates
> incorrectly in the same way.

Good catch, I didn't think about that.

I fixed my program now and it works, thank you for pointing it out.

sirjofri