This decades old Linux function is now 4 times faster

Key Takeaways

an old function in linux called memchar has been improved recently with the latest patch of linux by four times but this is not really the most instructive thing because this function was introduced back in what 1991 but the most instructive thing here is how it was discovered and the way uh the the code is written back in the days versus taking advantage of modern systems and cpus and this is what what is fascinating to me this comes from phoronx i'm going to read it to you and then a little bit i'm going to discuss a little bit so optimized memchar implementations for linux kernel up to four times faster a set of proposed patches promises to make the linux kernel's memchar which is a function basically you guys that you give it a block of memory the start of the block of the memory you give it what character you want to find and it will give you where exactly in that memory that character exists so it's almost like a search you know contains if you will in tests carried out by the developer the new implementation can be nearly four times faster in large searches only in large searches so it's not really a big deal if you're searching in small binaries small strengths but in a large when you're searching large things yeah it can be beneficial but this is not really what i'm interested in i'm interested in the explanation here and how we can learn something from this from experts so this comes from one of the developers of linux kernel eugene chang who sent other patches let's read this and discuss a little bit the original version of mem char is implemented with a buy twice comparing technique which does not fully use the 64 or 32-bit registers in the cpu we that means them we use word wide comparing so that eight characters so eight bytes assuming uh one byte is one character that might not always be true for utf-8 but you get the point can be compared at the same time on the cpu this code is based on david lates implementation so this is and they they go on and explain the actual performance because you see if you compare a single byte that means you have to take it from memory copy to the cpu register and then compare it with the actual value that you will look for so if you're comparing byte by byte you're going to go back and forth from memory to the cpu what they those guys did it's like okay since we have plenty of room in the cpu register which is the closest cache to the cpu let's go p8 let's copy the whole 64-bit let's understand how much the register is using and then maybe change that they don't mention if if the cpus register is 32 they shift to a four board or four bytes or four characters i don't know but it's still comparing eight together you know it's much faster close to the cpu compared to one byte by one and uh this is the basically the the actual function this is the code right and as you can see it was written back in 1992 by linus trevor and the the the code is really simple you can read through it through it and really understand what's going on you know it it's this is the trick here the p here is the basically the storage where you're looping through each bite right so it's an inside character and you're looping through all the n size that's the memory size right and this is what you're looking for if you found it then exit right and this return the result otherwise return now that means you didn't find it didn't find it so we're searching bite by bite what those guys did is they changed this to award which is i believe it's a it's an eight eight byte so this is what fascinates me when i read this article i was like is it worth it to make a video on it for the benefits we might not get anything out of it as a as you as a user we might not feel it at all maybe if you use this function if it's exposed to you in the string.c which is the the c library that this is exposed to yeah you might use it if you're using really large string then you're going to start fitting it because you're going to start reading more and more you're going to utilize the cache of the cpu registers right but it doesn't really matter what this i think teaches us is how to look beyond the abstractions you know because if i'm look if i'm writing a code and it's looping through a character we don't really think in terms of software architecture and hardware architecture we really at least i don't personally you know but this kind of changes the game i believe those c the kernel guys think of this in this time all the time right but i personally don't you know because i i mostly write in high-level languages so i rarely think like this but this kind of this kind of news changes completely how i think you know and i think we can learn something of them there look at the code you're writing today and see can it really improve i believe every code can be improved but the goal here is we don't want to also go and improve everything right because it's like okay how much the gains are really worth it or not you have to think about it this way but just the fact that understanding what's going on gives you internal i don't know comfort of some sort at least to me that okay this is exactly how that code works instead of just i don't know it's just looping but what is it actually doing this trip between the memory and the cpu and it's less significant this trip between the cpu and memory when you're in a noma architecture where the cpu or or the the the new architecture with m1 or m2 the apple one where the cpu literally lives within the memory so there is no trap to go to a bus of some sort to read the memory there is no trap so this i believe will be even less significant right in in this kind of architecture where the things are so close to each other but i i find it just fascinating to as as you go through the code and understand what is exactly doing and by all means i i'm not really an expert it just makes me really think about this uh in details i don't know what do you guys think do do you feel the same or do you say yeah this is really just really nothing burger let me know in the comment section below gonna see in the next one quick news today