Make use of multiple CPU cores

[sa666666]

The docs state that scanmem is slow on large programs on the first pass. I've experienced this too. Would it be possible to partition the area to search, and divide it amongst multiple cores? I have an 8-core, 16-thread machine, where the rest is going to waste, as scanmem is only using one core. How feasible would this be?

[12345ieee]

How feasible would this be?

It's clearly doable, but needs a huge amount of work.

I'll leave this open, maybe someone young and brave wants to try.

[shenada]

Hi Andrea, I'm very interested on this topic, too. Any data flow or work flow documents for libscanmem ? Or even a hint ? Which will helps me quicker start and prevent misunderstanding on current design. Because old school classic signal based cancelable worker model somehow kind a little bit hard to trace/read. Respectfully, D.S

…

On Sun, Jun 7, 2020, 19:32 Andrea Stacchiotti ***@***.***> wrote: How feasible would this be? It's clearly doable, but needs a huge amount of work. I'll leave this open, maybe someone young and brave wants to try. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#384 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAHCDRWZLWDNNIANFMF4Q4DRVN3DZANCNFSM4NWFHTQA> .

[12345ieee]

While I could write down a nice document on how the pieces fit, it'd take me ages and I really don't want to do that kind of work.

We can discuss this on a chat, sm has a (mostly desert) slack room here:
Or if you like Discord more, I've been crashing at PINCE's place lately, it's a bit more lively: https://discord.gg/KCNDp9m

[shenada]

Hi Andrea,

On Sun, Jun 7, 2020 at 8:58 PM Andrea Stacchiotti ***@***.***> wrote: While I could write down a nice document on how the pieces fit, it'd take me ages and I really don't want to do that kind of work.

I hate to write documents too. :smile No need to be a real document. There could be few hints for keywords or symbols in the source to guide me to quickly understand why and how the scanmem's design.

We can discuss this on a chat, sm has a (mostly desert) slack room here: [image: Chat on Slack] <https://join.slack.com/t/scanmem/shared_invite/enQtMjU0OTE5NjczNjk4LWI1ZjEyNzMzZTdiMTZhOTUyNDRmMWFmNDQ3YTJhOWE5MDZmMjUzZmM1MzgzN2ViM2MzYmMzM2QzMWIxNDdiMjk> Or if you like Discord more, I've been crashing at PINCE's place lately, it's a bit more lively: https://discord.gg/KCNDp9m

Both are new things for me. I used to use IRC or maillist(We draw with ascii art. :LOL). Anyway, I tried 'Chat on Slack' which shows me an error:

This link is no longer active To join this workspace, you'll need to ask the person who originally

invited you for a new link.

[12345ieee]

I'll just close it, nobody uses it anyway.

IRC is fine by me, but I'd rather you try discord first, because it keeps history.

[shenada]

Discord is fine for me too. Just create a new account with DTag 'D.S#2209'.

[bkazemi]

Well, not v0.18 for sure, but the easiest place to start would be to pthread region searching, as those are logically divided anyways. Then you'll have to synchronize access to the matches array which will be kind of a pain because we want the addresses to be ordered properly. Then there's the issue of progress output, it'd get jumbled without proper care (or ncurses). Last thing is handling the user stopping the search prematurely, not difficult because we handle that in the main thread. Maybe ptrace() issues. Probably other things too ;-)