Long ago I began my programming "career" (not really) with Pascal, and I still remember that I was frequently screaming at the screen that "IF YOU KNOW THERE SHOULD BE A SEMICOLON, WHY DON'T YOU JUST PUT IT THERE?" You don't know cyberbullying until you've been bullied by your compiler. Oh, and you could have counted all the extra development time you used in C and assembly to just run the program you wrote in Python. Writing it and letting it sit until the assembler programmer who is paid by the hour has finally finished juggling registers is just unfair.
In my company, we have a C program which takes three hours to load some data to a database. A modern Java program has been written to replace it. That takes six hours.
@@mamneo2 Even outside the public sector companies love being inefficient because they don't seem to realize that certain projects don't need 500 people or 10M lines
Don't forget that.. • Python is an interpreted programming language. • C and C++ are compiled languages which get optimized during compilation • Assembly is just sequence of instructions labeled with names typeable on a Latin keyboard. It isn't optimized or whatever. As long as it gets the job done 👍 C/C++ was the winner, but there are contexts where Python or Assembly would be a better tool for the job :)
In what case Python is better? Genuine question. Because in FE you use JS because that´s what the browser understands, and even then JS is just faster than Python. In BE its always a fight between performance (the faster a routine runs the less Besos steals your money) and easily of use (likeness). I think the only use case on top of my head is Django but at some amount of usage it is just a bad architecture decision to use that over Java, Rust, Go, etc. And machine learning I think there are libraries for other programming languages other than Python.
@@MarkVonBaldi As a general programming language, it's popular because it's an easy to learn, easy to use scripting language. What does it matter if your program runs in 1 msec or 100 msec if you don't need to run it that often? Especially if the quick python script only takes 10 minutes to make vs 1 hour for the C or C++ code; you're end up actually saving time overall. There are libraries written in other more efficient languages that optimize any intensive tasks you would want to do anyways, so Python's performance usually isn't a bottleneck Realistically though, nobody is using Python for its performance. Python is known as a 'glue' language and similar to other scripting languages, it's used to glue other languages and programs together. Scripting languages actually evolved from bash scripts, which were small scripts we used to glue programs together; take the input of a text file and replace every word with another word, take that output and pipe it into a java program, take that output and do something else to it, etc. The idea is that they're optimized so you can write code quickly and efficiently with almost no boilerplate, which speeds up development. They're just another level of abstraction up. With that in mind, as a programming language, you should be using Python to perform very high level coding that 'glues' other libraries together. You shouldn't be writing Python code that's performing hundreds of thousands of loops and manipulating integers. In a similar sense you're not going to write your base neural network class in Python, that would be incredibly slow, but if you're only using tensorflow functions? You can prototype your ideas in a fraction of the time it would take you in a language like C++. It's all about how you use the language.
I think, this is a little bit like comparing command line and GUI. The command line is in many situations more efficient, but only, if the commands are already known by the user. Still, with a GUI, even a noob will most likely succeed with simple tasks, simply by clicking and reading what is already on the screen. Less efficient, but still better than someone who doesn't know what to type at all. Python will be most tolerant to missing technical understanding, while ASM doesn't care about the users feelings at all. 🙂
Python users: Just use a package for that! C/C++ users: Who do you think made the package? C/C++ users: BTW you should compile with -O3, the compiler knows everything! ASM users: Who do you think made the compiler?
@@certidailyfacts ok but who makes the incredible asembly-level code optimizations? (i know you can write asembly within C files, but that requires you to know asembly)
@@certidailyfacts It's not important which language you wrote the compiler in, you could even write a C compiler in python. You still need to understand assembly language of whatever architecture you are targeting very well, to be able to compile efficient code into that.
One thing thats interesting about assembly vs high level languages is that back when high level languages were new, there was a similar sentiment about compiled languages as there is now with interpreted languages. The whole idea that compiled languages were so slow and unskilled. But now compiled has taken over since its way better optimized and allows for really complex programs to be way simpler. Maybe the same will happen for interpreted languages
> maybe the same will happen for interpreted Nah. They just compile the code at runtime lol. Chrome started JIT (just-in-time) compiling JavaScript back in 2008, and now everyone does it. Chrome's JIT is the fastest in the industry, and it's still not even comparable to even the "slow" compiled languages.
I don't think so because each better optimalization of c/c++ takes generally more time to compile. In interpreted languages you don't have this trade off because "compiling time" is also execusion time. But is very likely that languages will become more and more easy to use and compilators would do more and more work for programmers.
Python compiles directly to bytecode in the first stage of it's compiler, so it has practically no overhead due to it being interpreted, it's mostly slow just because it's single-threaded and dynamic.
I use python to generate lookup tables and the sort of preprocessing that is done once, because slapping something together in python is fast and intuitive. Everything else I do is C/C++ (I play around with embedded stuff).
Kind of surprised that nobody mentioned this, but when people talk about "Python saves development time", they don't mean the time that it takes you to type out an algorithm, that you already know ahead of time. Like, this is really obvious. "Development time" includes everything from "Let's start thinking about how I would solve this problem" and until "The program is written, documented, tested, packaged and ready to be shipped to production". Also, development time includes "Oh, I'll just use this library" vs "Oh, let's go implement a B-Tree in ASM".
"YoU cAn UsE pAcKaGeS!" Says everything python user. The lazy, most unintelligent response, especially when you have to remind them that "pAcKaGeS!" aren't exclusive to python, but that near every language out there, even javascript (lmao) has the ability to use third party libraries.
@@MechanicalMooCow strawmen are made of straw 1) The video compares Python to C and ASM. Name 5 C packages. Then explain, how to install and use them. Then do the same for ASM. I'll do it for Python: numpy, pandas, matplotlib, flask, requests (not even counting stdlib packages btw). pip install {package}, import {package} 2) My point was that the video attempts to sell "time spent typing out a program" as a useful metirc for "Programming time", which is an *obviously* flawed approach. By the videos logic, some esoteric golfing language would be the winner in the "Programming time" category.
@@ruroruro One difference I've personally noticed is that distribution is a lot nicer with c/c++, because it includes all the libraries without bundling in the entire thing, it only compiles the functions that you use. The vast majority of python software I've used required a lot of debugging before I could get it to run and often would bundle dozens of libraries in their entirety
@@ruroruro But it's pretty much unbeatable when it comes to writing scripts for use by other programmers and backend stuff or knocking out a quick program to process some data. It's just not great for consumer stuff imo
Everyone is forgetting that his code has the same form for all 3 languages. Yes there is better ways to do it in python but same goes for c and c++. Python is slower its just a fact doesn't mean its useless. All the man did was test code of O(n^2) in 3 languages to see witch runs faster. And how long it takes to develop it.
For me personally, the one of greatest strength of python is the concise code. It makes it easy to read and maintain. It would be nice to see the number of characters for each language...
C++ to me is cleaner than python can ever be maybe because I'm more experienced but throwing somebody who's freshly learned how to program into a C++ project isn't a good idea
@HackerExecute idk, assembly is definitely a step below the others when it comes to legibility. It's a speciality tool these days for pretty good reason. Bad programmers will write shite code regardless of the language.
@@h3xad3cimaldev61 I think it also depends on what you've started with. If you start with something statically typed C++ makes sense. Now I'm more experienced I prefer Rust for anything performant and python for the rest.
@HackerExecute dabbled with assembly in uni, haven't dealt with it since. It's interesting but definitely not something I could pick up and use right now! Can I ask what area these assembly programmers are working in? Is it embedded stuff? Drivers? My impression is that it's pretty niche these days compared to higher level languages. Re: legibility, I'm not sure we're talking about quite the same thing. I'm saying the abstractions available in higher level languages make them more legible than assembly for large tasks. For example, good luck writing a game or a machine learning model in assembly, you'll be there a long time. I think you're maybe talking about smaller low level tasks? Can't say I've run into the same issues you have with types in python. If I wanted to get specific about types I'd probs use rust/C++. What were you working on in python that was giving you hassles? I mainly use python for higher level tasks eg data science workflows, manipulating files. If I spend a morning working on a python script and it runs 2 minutes slower than another language I really don't care. The data science libraries available in python save me that many times over. As always, the right tool for the job.
Once upon a time, when the first extended instructions (MMX, etc.) first came out, and C/C++ compilers were still being optimized, it did make sense to go with hardcoded x86 assembly if you needed the bleeding edge performance. I've maintained embedded systems written entirely in x86 (technically 286) assembly because at the time that was the only way to get the necessary performance (at least, without going with a 68k instead :/). But nowadays? C/C++ compilers (yes, even GCC/MSVC) are going to extract more performance then anything you will ever write. Let them do their job.
Occasionally Assamby still can beat the optimizer. I know a library which PWM or bitbangs something which actually has AVR assamby within the c++ library. The optimizer ain't good enough. Obviously this is an rare exception and only applies to a very specific Achitecture which is predominantly used in embedded systems. I would even claim, that most of the time the optimizer beats hand written assamby.
@@IamTheHolypumpkin Yeah, but those are generally *Very* specific situations. You are right they do pop up more often in embedded systems though. Same with legit compiler bugs (of which I've encountered two in my lifetime).
@@IamTheHolypumpkin optimiser is better than the average c/c++ user's understanding of assembly code though, which is whats important. an experienced assembly programmer might be able to write faster code than the compiler, but anybody teaching themselves assembly to try to reach that point is going to struggle to write something faster.
one thing to take into account is that python can't inline func calls, so with such task where you have a lot of calls it's better to use nested loop instead of a separate function
Bahaha.... And it cares about whitespaces.. Idiotic, incapable and slow.... C however, where you can just typecast anything to anything, and use data as code and vice versa... Pointers rule!!!
@@MrZnarffy you're arguing that Fiskars axe is worse than Husqvarna chainsaw dude... Both languages have different purposes and are good at them, and your talk about some C features you like makes you look like one of those programmers that do only some code exercises online and no real job.
@@mcroyal8000 ahahaha... Dude, if you can't use pointers, and function pointers, you shoulndn't even speak about C... I started with Assembler waaay back, done verilog, and lisp, and python, and anything between them. Python is wrtten by morons for morons.. I mean indentation sensitive???? Come on.. Oh.. And interpretative language.. LOL you don't do serious stuff in that.. It's like the C++ hype.. Why on earth have classes and inheritance? So you can be lazy and make bloated messes that is impossible to maintain.. I know COBOL too, but I wouldn't use it if I possibly could avoid it, and Python is just as crappy
@@MrZnarffy you don't do serious stuff in Python and yet it's the 2nd most used programming language after JavaScript. Keep living in your bubble, thinking that only C is good and for everything "bEcAusE p0iNTerS" lol
It just shows the power of the languages designed to be fast vs the one designed for ease of use/learning automation etc (plus it has numpy etc from the past languages, so there's really the best of both worlds)
@@nebulium6641 C compiler are very good optimized. But i like to use assembler to create some tiny executable. In assembly you see all instructions like the CPU.
I remember vividly taking this course called "Compilers"... and it was so much fun to know behind the scene how compilers work, how to save variable, how halt work... I am glad to take that course...
@@neeljoshi3923 unfortunately at my univ this is closed course... but there are many online course you can choose. the thing is the concept of compiler course is converting program from a to b.... once you get the idea the rest is EASY!!!
All python integers are bigNum structs that contain an array of integers along with a size in number of cells, and use non-native arithmetic operations (following the python mantra of "everything is an object"). So it makes sense it'd be this slow.
I remember reading a post by some retired game dev who started out programming with assembly and ended with C++. So with the first consoles they used to write everything exclusively in the native assembly because it was the only available option. When consoles started supporting compilation from C, they still used assembly for any loops because the compiler wouldn't optimize them. Then when the compilers finally started becoming feasible to write only in C, C++ started getting popular. Except then the C++ compilers weren't well optimized either lol. One of the devs even made their own C++ compiler better suited for the console but it still wasn't great. so they opted to continue writing in C until publicly available C++ compilers became very effective. Then everything was done exclusively in C++ and the guy retired several years later. I believe he worked mostly with Nintendo consoles (NES, SNES, N64, Gameboy, etc.) but also did a few projects on Sega and PlayStation.
Yeah, C++ is the standard these days for pretty much any game development that's not just automation scripts. There's not a lot of reason to learn ASM unless you plan to work with it a lot for a specific type of job.
@@sisyphus_strives5463 It's better to create a goal and master skills that are vital to the accomplishment of that goal. Learning random stuff for the sake of learning is usually idle behavior that doesn't accomplish anything. A programmer who knows a few languages really well is more useful than one who only knows the basics of 20.
good point - that would significantly decrease the number of iterations, but then I would have to massively increase the range (>>250,000) to get a similar disparity in C-vs-python performance :) . and at that point i*i would get even larger (i believe for this range, it already exceeds the maximum for unsigned int), or for even larger numbers maybe pull it out of the loop and do something fancy. or you can just use a sqrt, but I didn't want to include any math libraries (doing so in asm is a pain & a lot of typing), or have to implement that myself. obviously there are much better ways to compute & count primes than this - this is just something very quick to compare the relative speeds of execution when you have a lot of computations
@@xmdi0 i*i would get too large eh? no problem just compute an upper bound for sqrt(n) using a few iterations of newton's method - run n_ = n_ - 0.5n_ + n/(2*n_) a few times :)))), it's O(1) and should be negligible now you can iterate from 2 to approximately an upper bound for sqrt(n) now go make the range bigger :>
@@EmmanuelMayor that's odd, I just wrote a program using both the i*i method and the sqrt(n) using the math library and it still detect 97 as prime. (91, 93, 95 and 99 are all composite)
For all the people saying the Python could be faster, it could not, not in relations to other languages, at least. The WHOLE POINT of Python is to be flexible and easy do code, that's why it was the fastest to write in the video. But for that, CPyhton has to do a lot of work, like deciding all the types. That way, we can take advantage of dynamic typing. It's a tradeoff, and I think every python user should understand that.
If you need that level of performance in your app you wouldn't even think about starting coding in python, otherwise, python is the best choice, always
@@AbeDillon Sure! I was in a discuss these days where one of my workmates was saying he doesn't like python because it's slow... ok, sure it is... But you're not writing a powerful algorithm that really needs performance, it wouldn't make any difference in the end and you would got job done faster with python.
@@arthurbc6 well, try making a sudoku generator in python. I did. Made it in python, c# and c++. Guess what. The difference is really visible. Now don't blame my algorithm. Because any algorithmic optimisation would always reflect well in any language. The only real advantage of Python is the many libraries or packages. But that's more because it is the most used language by non-programmers. And python isn't very readable imo. It's also harder to debug. The reason python isn't for me is it's syntax. I like more structure in my programs, which I'm sure many others do as well.
@@arthurbc6 If you know you need performance... You'd still probably write the code in easiest language you have available to you, regardless of speed, and then only when you have the program ready and provably be too slow, you'd profile the performance and optimize most important parts. I don't think there exists a program where more than 10% of the code has its slowdown by 100x actually in any measurable way impact performance. So even if you know for sure 100% that the program as whole can't run fully on Python, that's 90% of the code you could write in a clearer, cleaner language with absolutely no performance impact, no downsides whatsoever. And then you have numpy and such, which just might actually be enough for the missing 10% as well. Premature optimization is the root of all evil.
Started out with Basic as a kid on the C64, learned C/C++ for Windows development, then later in my career it was JavaScript as applications went to the Web. Funny how after moving to the cloud I've been using Python for the last 3 years and loving it.
No Pascal. I was getting an Engineering degree and for some reason the only programming language we were required to take was Fortran 77. I also used a truly terrible language called ISETAL in an experimental Calculus class. Looking back I'm sure that my professor was just using us as research subjects to get published.
Also, you don't need to get up to n/2, sqrt(n) is enough, if there aren't any devisors up to that, there won't be any above that. But this would apply to all three of course. Plus, I don't know what the time cost of calculating the square root would be and if it would be worst in the end.
@@cg0wamp at the very least there's an old fast inverse square root algorithm out there, so just use it and use 1 over the result, that would be better than going up to n/2.
😆 Way too funny and certainly great. You should do one with the same languages and compile times for modestly sized source. Maybe add Delphi or Lazarus?
Theoretically, if you use the ASM optimized instructions generated from the compiled C code and you compare it to the C code, you should get the same results since it should generate the same object code (binary). The only way one would outweigh the other is if it uses different assemblers. In this case, OP seems to be using gcc for both C and ASM so it will be identical.
Note the C compiler won't do this but: since this code is entirely deterministic the most optimal compilation would be to compute the loop at compile time and then print the number.
Your semicolons as well as unnecessary parentheses and lack of spacing made my day. Anyway, I ran the exact same code, down to those semicolons, on my machine and it took less than 30 seconds. Also, when I ran the code on pypy, it took less than 4 seconds. And of course you get less than 0.1 second for both python and pypy if you use sieve.
First off, epic username. Way better than "The quick brown fox jumps over the lazy dog". Second off, I still have much to learn about python. This is ever more proof that the guy in the video is a dumbass.
I wrote a small program in C to long divide calculate primes like in the video. On my M1 MacBook Air, it took 1.6s to calculate the first 20,000 prime numbers. I tried writing the same thing in Python. It took just over 2 minutes to calculate the first 20,000 primes. Amazingly, my 10700k takes around 4 seconds for the C version. M1 single threading seems to be godlike.
Try using a built-in function isPrime() in the SymPy module to see how. It is a direct function and returns True if the number to be checked is prime and False if the number is not prime. Having many modules and built-in functions are the strength of Python.
Why not test for i < sqrt(n), or i*i < n in isPrime? A more expensive comparison, but many fewer iterations even for the comparatively smaller numbers you use.
I've developed a bit of C too. Mostly for small corner case projects. Few companies wants to pay for the time C takes to write though. Python, C#, Javascript, PHP etc. may be slower than C, but the coding most definitely is quicker and it'll pay your mortgage and food.
@@mattreigada3745 you are not wrong, but the amount of job postings requiring PHP or C# skills massively outnumber the amount of job postings requiring C.
@@dschledermann I'd be surprised if that still holds true, especially if we distinguish jobs and careers. JS and Java are considerably more popular than C, C#, or PHP but C# and PHP aren't really in their prime any more whereas C really only faces competition from C++ (and increasingly Rust). I generally find the comparison of job listings misleading since the industries these languages are used in have very different turnover rates.
I could argue that this algorithm doesn't make for a good example of the differences between C and assembly. Basically the whole inner loop will be completely bottlenecked on the integer division unit, and the whole rest of the processor will lie mostly idle just trying to feed it, so it doesn't showcase very well how a good assembly programmer may be able to outperform the compiler in various ways. When playing iwth microbenchmarks, I've often been able to improve register allocation or other minor details in the compiled output, but in this case, none of that will matter since it's all just waiting around for that one DIV instruction. That being said, if you really wanted to optimize this further is assembly, you'd probably convert the integers to floats and use the AVX division unit to increase parallelism.
@G E T R E K T Especially when you consider systems that may run the same small instructions millions and millions of times in rapid succession. The small small speed difference really adds up and it can become mission critical to maintain code in ASM instead of even C.
With proper optimization from the compiler I suspect this is a case where the compiler will result much faster machine code than the machine code generated from the assembly that was written. Two easy optimizations a compiler could reasonably implement are (1) calculating the division a single time outside the loop and (2) substituting this division with a right bitshift. Implementing either of those would give a performance edge over the assembly that was written.
About 99% of the time is taken by the DIV instruction, because of this it is useless to optimize this code, for example, the isPrimes inline function does not give anything. Tested the optimized version (i
@@BrightBlueJim The difference is you need the source code. If you know ASM any code can be disassembled and read as long as you know the platform it was compiled for (which is likely since if you ran the program you're on that platform). ASM -> Machine Code is 1:1, therefore Machine Code -> ASM is also 1:1. Of course python is likely going to be handed to you in python or bytecode at best which can also be decompiled. But things like C/C++ program can't easily be decompiled into C/C++. And this ubiquitousness of ASM is the crux of the "everything" part of the joke.
Every language has its own pros and cons, a real programmer uses a language depending on what he/she is going to do. Python is slow (than others I mean) but is easy and has lots of libraries. actually even its more used for machine learning and scientific calculations, but it can be used for scripting small projects, backend of Web applications, etc.
Python is easy for non programmers to learn, for example those who are interest in analysing data but don't want to learn something more complex and are willing to not learn a better language in exchange for much slower performances. Python is a scripting language, it can be good for that. Every language has libraries, C, C++, Java, Dart... literally every programming language does. Admittedly using a library with C++ might be harder than installing a package on Dart, but after you learn how the building process works, it's not hard at all, because you know what you're doing.
For those saying "python" is slow, a JIT will be added to CPython 3.12, you can compile python using mypyc (you need to use type annotations to get the maximum performance), and Cython now has pure python syntax.
@@diablominero Numba is very limited it doesn't work everywhere and you need to add it manually to every function. In the other hand CPython jit will be like PyPy's you don't need to change anything at all to gain speed.
Its all about the application build approach. Python is using interpreter approach, wich is make Python very easy to code, but very slow when the application is launching. C/C++ and x86 ASM are using Compiler So its not easy than python to code, but they are very fast when the application is running
Total Time : Python: 1min 50.4 sec C/C++: 1 min 24.3 sec x86 ASM: 2 min 25.3 sec So overall C/C++ takes less time for this specific program. Not talking about other big programs which has complexity and lots of debugging and errors takes in.
Quem escreveu o ASM não o escreveu tão bem como o compilador. Pense, você acha que o programa em C foi transcrito como? Não é uma questão de linguagem é uma questão de otimização.
When I saw this I wanted to test Lua as well and I was surprised. Writing: 61s Execution: Lua: 38s Luajit: 7.8s PS: Writing was calculated based on number of characters in python file compared to lua (+9.5%) since Everyone has different writing speed
I'm curious as to what for you guys code the most. I code mostly to do some matrix computation, interpolting images, extrapolating the price of the beer, and for my hobbies :) Some of these are even faster, using Numpy for example (at least faster than my implementation in C). Are you all developing the next Far Cry? Have you all HUGE databases to handle? Or are you just competing to find the most amount of prime numbers? Tell me, I'm curious 😁
Tbh I see python as just a scripting language. I use it only for prototyping small ideas or when I'm fed up with my current shell's scripting language.
Yeah I've had trouble finding something to do with it that I wouldn't prefer to use another language for. Network programming is something it's good at... it might be useful for writing some services or for gluing components together, but I haven't done any of that in a professional context.
For those who wants to see the prove: A = b × c (any number can be wrote as a product of another two. That's fact, because all numbers have a root. Example; pi = sqrt pi x sqrt pi). But if B and C are not the same (which is not always true, for example 1 = sqrt 1 x sqrt 1 and that's the only possibility), then B < sqrt a or C < sqrt A. Reason: if B and C are > than sqrt a, that's a contradiction, once root's definition implies "the product of two numbers, which ones are equal, that results at another number is the root of that number". The same logical can be apply if B and C < sqrt a. In that case, thejr product can't be A. Examples: 2 x 5 = 10 or sqrt 10 x sqrt 10 = 10. Why does all of that helps? Because a prime number can't be divide by any number that is not his root, once a prime is not a product of another two primes (which is the case of 10, for example). That means we don't need to check all numbers before the possibly prime number "N". We only have to check it before sqrt N, because if the number is not a prime number, then B < sqrt N or C < sqrt N. Once you catch a natural number that divides N, you prove that N is not prime. But if N is prime, then only his root can divide himself. Because we are checking everythig before sqrt N, after you check the first natural number to comes before sqrt N and it does not divide N, then you know that B == C, which implies B == C == sqrt N. Hope that helps.
C would actually be faster if he passed -O3 to the GCC compiler. The reason is that compilers are so intelligent in 2021 that they "know" what programmers tried to achieve with their code and so the compiler writes extremely efficient machine code compared to handwritten assembly which is almost always going to be slower as barely any human can match the 3 decades of tricks programmed into C compilers.
@@joey199412 Very true. This comment needs more likes. Hundreds of the best programmers optimize the compilers for max performance. An ordinary dudes asm code can never match all that. In such cases it really makes more sense to use the more abstract c code
@@joey199412when i searched about O3 found that it's known for being unstable and weird But even without that, as you can see in the video he uses the gcc compiler to compile the .o file so he Can use O3 for asm
@@ioneocla6577 Not really unstable, neither weird. If your code is proper won't have problems. When building a c program look at all those warning, they mean something.
For basic cases like this, C and C++ would likely produce identical code, assuming the compiler used the same level of optimization. C++ having many more features doesn't change how it compiles code not using any C++ features. Optimal generated assembly from C is likely the same, or similar in C++. Different compilers may have different methods of optimizing execution speed, but generally C code compiled in a C++ compiler will be the same or similar speed.
@@squee222 Backwards compatible with what? Any respectable C++ isn't backwards compatible with pretty much anything. Compiling any C(-style) code in C++ is a bloody bad idea.
you can check numbers up to sqrt(n) + 1 to check if it's prime. Don't know if it would make it run faster because less checking or slower because of sqrt call.
in general, it probably does get slightly better (a commenter below mentioned that -O1 gave him the best result), but I wouldn't expect it to be noticeably faster than the ASM. this ASM is literally the barebones procedural implementation of the algorithm described in the C code. the compiler might change the jump conditions or something fancy, but I personally don't see how you can get much "faster" for this algorithm unless you cheat and mess with the algorithm (and we can always cheat in the same way for the ASM code). I think if you have much more complex code, the difference between what ASM the compiler can generate and what ASM a human will casually generate will become larger. However, I personally believe in the limit humans will always be able to generate code as fast if not faster than the compiler (because we can literally hand-write the compiler's best guess), and there may sometimes be shortcuts that humans can take, using our knowledge of the problem. Of course, the compiler runs in a fraction of a second, whereas it would take you and me much much longer.
Absolutely does. I tried it and it ran 30% faster with optimization enabled. No noticeable difference between the different optimization levels but a VERY noticeable difference from no optimization at all.
Nice video Have you tried PyPy? Its a modified form of Python that runs a lot faster than standard Python, and is iirc completely compatible with normal Python
And this doesn't seem to include thinking time, only writing time. If you include the former, Python, C and C++ should be similar, but assembly would likely be even slower.
(EDIT : I was wrong about the interpreter overhead as python interpeter compiles to bytecode in first stage, but rather it’s mainly slow because of its single threaded, dynamic nature, NOT because of it being an interpreter) Python is an interpreted language, so the python interpreter has to parse and lex the code while executing at the same time, it doesn’t mean it’s a bad language though, its portability, simplicity and great ecosystem of tools that work nicely together makes it attractive for many applications, especially in science and engineering fields, moreover, libraries like numpy and tensorflow are native libraries written in c/c++ so they take care a lot of the gory details in things like matrix operations and other math ops. natively, but can be used from python through nice and clean APIs , without too much interpreter overhead, and that’s the beauty of Python.
Your point about interpreter is not completely correct. Let's take JavaScript for example, it is an interpreted language. Because of just in time compiling it runs very fast.
@@boem231 yeah you’re right, but it was a plain interpreter back in the old days, but with something like the V8 engine made JS much faster thanks to JIT (just in time) compilation and multithread utilisation (like in asynchronous programs to prevent IO blocking), which made chrome browser (which uses v8) stand out in performance a decade ago, and the v8 engine still powerful even in large-scale backend applications using nodejs, there are similar implementations for python as well. the vanilla out-of-the-box python interpreter has no fancy features like JIT. Python is just a language, so engineers are always finding creative ways to optimise and expand the functionality of it.
@@FakhryHTatanaki Yeah, sure, CPython doesn't have a JIT, but your statement about "the python interpreter has to parse and lex the code while executing at the same time" is still wrong (or at least very misleading). You make it sound like python is processing the text of the program as it is running it. This is obviously not the case. Python *is* an interpreted language, but it isn't interpreting the program text. It's interpreting the bytecode, that was compiled once, ahead of time. That's what all the __pycache__ folders are for. The reasons why Python is slow have almost nothing to do with it being an interpreted language. Python is slow because it has duck typing (no compile time type info -> no aggressive optimizations), it's slow because it's a dynamic language (everything is a dict -> every name lookup is a runtime string comparison), it's slow because of the GIL. Compared to these reasons, Python being an interpreted language makes almost no difference.
Back in the days beyond recall, when I was taking my first programming course it was Fortran on an IBM 1130. We had a lab with a graduate student TA and we had to program the sieve of Eratosthenes using 16 bit signed variables to print out the primes that would fit. Pretty basic stuff. Well programs would start out printing five lines per second on that chain printer... and get slower and slower. Up steps the TA and his program ran much longer before slowing down. He had written an interrupt controlled printer driver in ASM of which he was rather proud. I then ran my little Fortran program and it was faster. I thought that everyone knew that all composite numbers had a divisor equal to or less than the square root. The algorithm counts. Well, so does the language. Horses for courses as the British say.
Choose the language to be written according to the appropriate project, this is not much to discuss...According to the needs, the fastest way to run is of course machine language, then assembly, c/c++, the disadvantage is that the development efficiency is not as fast as high-level languages. If you don't pursue high running speed, you want to use high-level languages such as python and java for rapid development.
I used to write assembly (Z80) on paper with pencil, looked up mnemonics from a printed chart, typed in the bytes one by one into a Basic program’s DATA, POKEed it into the RAM with a FOR loop, run the code, watched the PC to freeze 95% of the times with no info about what might have gone wrong, and started again from scratch.
My 3 favorite programming languages. It depends on what you want to. Lately I use python most of the time. ASM was necessary back in the 80s and early90s before floating point became common. Or writing DSP code.
ASM is still necessary, any programmer worth their salt knows how to debug their code. I've been programing since I was a child, C64 era, and every time a new HLL (high level language) comes out I laugh and ask myself how many cpu cores does this one take to print hello world. Today programmers are lazy, their was a time (MSDOS-win98) we didn't even have protected memory, do you think todays programmers could deal with that? I don't, their lack of understanding is annoying to the point I catch myself forking their project and writing it in a competent language and optimizing it
i made a similar comparison and assembly was about 6-7% faster than c even when i turned the iterators and the counter into registers, however with a quick assembly tweaking into the c code the difference is seemless.
In a large, real-world program, C is usually better in every way. The expertise of people tweaking compilers for decades is beyond nearly everyone trying to code in assembly, especially for complex algorithms. Inline assembly is fine for certain critical portions, but often C with a good compiler and the correct optimization flags will beat it.
after reading some of the comments - there is a reason why python is preferred by people. You can argue all day but more people than not prefer the simplicity of python. Is it slower yes, but that is why lots of python modules are written in C to gain the best of both worlds. Simplicity of python syntax plus the speed of C in case you need it
ok, at first I'm not in python, so how fast does you code this is you use python and C means you make the function isPrime in C and the rest in Python. But i think mostly its quicker to code all in C then.
Yeah, and many programs don't need extreme speed. Unless things are within like 5 nested loops or are called very frequently and have to be fast, python is just fine.
Most of the libraries for ML in Python are written in C, for example, NumPy, imagine doing all the logic in Python and speed-requiring work (as matrix operations ) in C.
There are 2 type of python guy : 1) a newcomer who pick python as his 1st language and always ridicule other programming language by saying the *syntax is ugly* & *too verbose* 2) a professional who had written many apps in many languages and *having passion* in statistic, calculus & linear algebra
@ANoobyPlayer UwU this. I’m a C or V or Zig user but if I want to start a new project I’ll make as barebones as possible a basic thing in python before rewriting it in C or another low-level speedy language
I wonder if there are no algorithmic techniques , in ASM, to use X64 CPUs ability to execute more than one instruction per core. Writing ASM, especially when C compilers already optimize for that, does appear to be an inefficient way to spend programming time.
Out of order CPUs always try to execute more than 1 instruction per tick, up to ~5, but it is quite hard woth this code. Division operation is very slow by the way.
From sympy import isprime pcount = 0 for i in range(2,250001): pcount += isprime(i) print(pcount) running time .25 seconds; and if I really wanted to make it fast I would look for a library that will eliminate the loops remember, you use python for the libraries, if you are not using them, you are not really programming python also dead giveaway video maker doesn't program much python is the str conversion in the print statement
i mean, there are a lot badmouthing about python because they compare exact same code to a compiled language which doesn't make any sense. so comparing python loop and c loop doesn't prove anything because loop is not the best approach to take with python.
C is way lot cooler than python. And python looks and is better than c. To be honest indentation really makes code look so good without need for verifying if you missed a curly brace.
Sympy doesn't do this bruteforce remainder check, it knows to discard even numbers and divisible by 3 immediately among other tricks. On my pc the video version takes 58 seconds. When I do range(3, sqrt(n), 2) and check for n%2 first instead, it takes just 0.35 seconds. Sympy takes 0.5 seconds. Proper primesieve where you only divide by other primes becomes faster somewhere between 10mil and 20mil.
The thing is that python is very very very good for home tasks and quick algorithms that you won't use millions of times. But cpp and c are just way better for that. If you have an algorithm that takes 1 min to run a cpp/c but 2 in python. And your gonna be running it millions of times it's much more efficient to write it in cpp/c and not in python. It just depends on the circumstanses
I wish you tested the speed of the execution of each program and how it take each in terms of memory, that way everyone can see the pros and cons of these 3 programming languages
the diference between that's examples could be the weight of final binary compiled, in my experience when compile an hello world! example in asm the result is a ligthweight compiled file but in c++ the compiled file is heavier than asm file
Наука
stop cyberbullying me over python semicolons. it's a coping mechanism
Haha. Get cyberbullied
Long ago I began my programming "career" (not really) with Pascal, and I still remember that I was frequently screaming at the screen that "IF YOU KNOW THERE SHOULD BE A SEMICOLON, WHY DON'T YOU JUST PUT IT THERE?" You don't know cyberbullying until you've been bullied by your compiler.
Oh, and you could have counted all the extra development time you used in C and assembly to just run the program you wrote in Python. Writing it and letting it sit until the assembler programmer who is paid by the hour has finally finished juggling registers is just unfair.
Wait, I didn't even notice that, I'm just too used to semicolons
Pep8 or else
😂😂😂😂😂😂 everyone here is guilty of mixing up language syntax, the amount of times I have written python code into a c file is insane.
In my company, we have a C program which takes three hours to load some data to a database. A modern Java program has been written to replace it. That takes six hours.
I laughed so hard with this one XD!
well the next logical thing is to rewrite the whole code in asm!
It's evolving just backwards
IBM?
lmaoooo that made me laugh
"Use ASM if you are paid hourly"
That's such a genius trick
Becoming replaced in no time by a C programmer doing the same in less time! GENIUS
@@realedna Not if you work for the public sector, where nobody cares about efficiency xD
@@mamneo2 Even outside the public sector companies love being inefficient because they don't seem to realize that certain projects don't need 500 people or 10M lines
@@yasserarguelles6117 you sure they dont love shortcuts instead?
Nah, try coding in binary 😁
Don't forget that..
• Python is an interpreted programming language.
• C and C++ are compiled languages which get optimized during compilation
• Assembly is just sequence of instructions labeled with names typeable on a Latin keyboard. It isn't optimized or whatever.
As long as it gets the job done 👍
C/C++ was the winner, but there are contexts where Python or Assembly would be a better tool for the job :)
Overall, assembly language is still the faster language.
In what case Python is better? Genuine question. Because in FE you use JS because that´s what the browser understands, and even then JS is just faster than Python. In BE its always a fight between performance (the faster a routine runs the less Besos steals your money) and easily of use (likeness). I think the only use case on top of my head is Django but at some amount of usage it is just a bad architecture decision to use that over Java, Rust, Go, etc. And machine learning I think there are libraries for other programming languages other than Python.
@@MarkVonBaldi As a general programming language, it's popular because it's an easy to learn, easy to use scripting language. What does it matter if your program runs in 1 msec or 100 msec if you don't need to run it that often? Especially if the quick python script only takes 10 minutes to make vs 1 hour for the C or C++ code; you're end up actually saving time overall. There are libraries written in other more efficient languages that optimize any intensive tasks you would want to do anyways, so Python's performance usually isn't a bottleneck
Realistically though, nobody is using Python for its performance. Python is known as a 'glue' language and similar to other scripting languages, it's used to glue other languages and programs together.
Scripting languages actually evolved from bash scripts, which were small scripts we used to glue programs together; take the input of a text file and replace every word with another word, take that output and pipe it into a java program, take that output and do something else to it, etc. The idea is that they're optimized so you can write code quickly and efficiently with almost no boilerplate, which speeds up development. They're just another level of abstraction up.
With that in mind, as a programming language, you should be using Python to perform very high level coding that 'glues' other libraries together. You shouldn't be writing Python code that's performing hundreds of thousands of loops and manipulating integers. In a similar sense you're not going to write your base neural network class in Python, that would be incredibly slow, but if you're only using tensorflow functions? You can prototype your ideas in a fraction of the time it would take you in a language like C++. It's all about how you use the language.
fair enough❤
don't forget that interpreter and compiler are programs written by asm programmers.
I think, this is a little bit like comparing command line and GUI. The command line is in many situations more efficient, but only, if the commands are already known by the user. Still, with a GUI, even a noob will most likely succeed with simple tasks, simply by clicking and reading what is already on the screen. Less efficient, but still better than someone who doesn't know what to type at all.
Python will be most tolerant to missing technical understanding, while ASM doesn't care about the users feelings at all. 🙂
Python users: Just use a package for that!
C/C++ users: Who do you think made the package?
C/C++ users: BTW you should compile with -O3, the compiler knows everything!
ASM users: Who do you think made the compiler?
The C/C++ compilers are written in C but ok.
@@certidailyfacts ok but who makes the incredible asembly-level code optimizations?
(i know you can write asembly within C files, but that requires you to know asembly)
@@certidailyfacts It's not important which language you wrote the compiler in, you could even write a C compiler in python. You still need to understand assembly language of whatever architecture you are targeting very well, to be able to compile efficient code into that.
Hardware engineers: Who do you think made the instructions.
Hardware engineers: Use the VHDL optimizer, it knows everything.
Physicists: ...
Philosophers: !
One thing thats interesting about assembly vs high level languages is that back when high level languages were new, there was a similar sentiment about compiled languages as there is now with interpreted languages. The whole idea that compiled languages were so slow and unskilled. But now compiled has taken over since its way better optimized and allows for really complex programs to be way simpler. Maybe the same will happen for interpreted languages
Most things can still be done faster in assembly.
> maybe the same will happen for interpreted
Nah. They just compile the code at runtime lol. Chrome started JIT (just-in-time) compiling JavaScript back in 2008, and now everyone does it. Chrome's JIT is the fastest in the industry, and it's still not even comparable to even the "slow" compiled languages.
I don't think so because each better optimalization of c/c++ takes generally more time to compile. In interpreted languages you don't have this trade off because "compiling time" is also execusion time. But is very likely that languages will become more and more easy to use and compilators would do more and more work for programmers.
Python compiles directly to bytecode in the first stage of it's compiler, so it has practically no overhead due to it being interpreted, it's mostly slow just because it's single-threaded and dynamic.
Good luck waiting for this!!!
🤣🤣🤣🤣
I use python to generate lookup tables and the sort of preprocessing that is done once, because slapping something together in python is fast and intuitive. Everything else I do is C/C++ (I play around with embedded stuff).
"Use ASM if you are paid hourly" 😂
*goes to RUclips....'Assembly crash course for beginners'*
Still laughing!
Personally i love asm.
@@akumaquik becouse youre paid hourly
That one had me crying 😂
Kind of surprised that nobody mentioned this, but when people talk about "Python saves development time", they don't mean the time that it takes you to type out an algorithm, that you already know ahead of time. Like, this is really obvious. "Development time" includes everything from "Let's start thinking about how I would solve this problem" and until "The program is written, documented, tested, packaged and ready to be shipped to production". Also, development time includes "Oh, I'll just use this library" vs "Oh, let's go implement a B-Tree in ASM".
"YoU cAn UsE pAcKaGeS!" Says everything python user. The lazy, most unintelligent response, especially when you have to remind them that "pAcKaGeS!" aren't exclusive to python, but that near every language out there, even javascript (lmao) has the ability to use third party libraries.
@@MechanicalMooCow your point?
@@MechanicalMooCow strawmen are made of straw
1) The video compares Python to C and ASM. Name 5 C packages. Then explain, how to install and use them. Then do the same for ASM.
I'll do it for Python: numpy, pandas, matplotlib, flask, requests (not even counting stdlib packages btw). pip install {package}, import {package}
2) My point was that the video attempts to sell "time spent typing out a program" as a useful metirc for "Programming time", which is an *obviously* flawed approach. By the videos logic, some esoteric golfing language would be the winner in the "Programming time" category.
@@ruroruro One difference I've personally noticed is that distribution is a lot nicer with c/c++, because it includes all the libraries without bundling in the entire thing, it only compiles the functions that you use. The vast majority of python software I've used required a lot of debugging before I could get it to run and often would bundle dozens of libraries in their entirety
@@ruroruro But it's pretty much unbeatable when it comes to writing scripts for use by other programmers and backend stuff or knocking out a quick program to process some data. It's just not great for consumer stuff imo
This comparison is for fun. Thanks for this. The important is what is your purpose.
It's interesting to see a similar review for languages Python, JavaScript, Delphi
Everyone is forgetting that his code has the same form for all 3 languages. Yes there is better ways to do it in python but same goes for c and c++. Python is slower its just a fact doesn't mean its useless. All the man did was test code of O(n^2) in 3 languages to see witch runs faster. And how long it takes to develop it.
"Never thought I'd die fighting side-by-side with an anime profile pic." "What about side-by-side with a friend?" "Aye"
exactly in which part he tested the code to see which runs faster? The code was not executed
@@S3Kglitches I think you need to re-watch the video as the programs clearly were run after being coded.
Yeah. Try using C for networks. You will regret that you were born.
@@Alexander-jm1oi and for machine learning and you will decide to destroy entire humanity
now do a Assembly vs Punching cards side-by-side comparasion
All languages can be done on punched cards, as long as you stay within 80 characters per line and don't drop the deck.
Would give essentially the same result
@@SwampKryakwa You can get some children from ozher countries to do that coding for you, it saves a lot of time
For me personally, the one of greatest strength of python is the concise code. It makes it easy to read and maintain. It would be nice to see the number of characters for each language...
I love python for that, there are plenty of instances where your time is more important than shaving seconds off your runtime.
C++ to me is cleaner than python can ever be maybe because I'm more experienced but throwing somebody who's freshly learned how to program into a C++ project isn't a good idea
@HackerExecute idk, assembly is definitely a step below the others when it comes to legibility. It's a speciality tool these days for pretty good reason. Bad programmers will write shite code regardless of the language.
@@h3xad3cimaldev61 I think it also depends on what you've started with. If you start with something statically typed C++ makes sense. Now I'm more experienced I prefer Rust for anything performant and python for the rest.
@HackerExecute dabbled with assembly in uni, haven't dealt with it since. It's interesting but definitely not something I could pick up and use right now!
Can I ask what area these assembly programmers are working in? Is it embedded stuff? Drivers? My impression is that it's pretty niche these days compared to higher level languages.
Re: legibility, I'm not sure we're talking about quite the same thing. I'm saying the abstractions available in higher level languages make them more legible than assembly for large tasks. For example, good luck writing a game or a machine learning model in assembly, you'll be there a long time. I think you're maybe talking about smaller low level tasks?
Can't say I've run into the same issues you have with types in python. If I wanted to get specific about types I'd probs use rust/C++. What were you working on in python that was giving you hassles?
I mainly use python for higher level tasks eg data science workflows, manipulating files. If I spend a morning working on a python script and it runs 2 minutes slower than another language I really don't care. The data science libraries available in python save me that many times over.
As always, the right tool for the job.
Once upon a time, when the first extended instructions (MMX, etc.) first came out, and C/C++ compilers were still being optimized, it did make sense to go with hardcoded x86 assembly if you needed the bleeding edge performance. I've maintained embedded systems written entirely in x86 (technically 286) assembly because at the time that was the only way to get the necessary performance (at least, without going with a 68k instead :/).
But nowadays? C/C++ compilers (yes, even GCC/MSVC) are going to extract more performance then anything you will ever write. Let them do their job.
Occasionally Assamby still can beat the optimizer.
I know a library which PWM or bitbangs something which actually has AVR assamby within the c++ library.
The optimizer ain't good enough.
Obviously this is an rare exception and only applies to a very specific Achitecture which is predominantly used in embedded systems.
I would even claim, that most of the time the optimizer beats hand written assamby.
They still can fail at few lines of trivial code
@@IamTheHolypumpkin Yeah, but those are generally *Very* specific situations. You are right they do pop up more often in embedded systems though. Same with legit compiler bugs (of which I've encountered two in my lifetime).
@@IamTheHolypumpkin optimiser is better than the average c/c++ user's understanding of assembly code though, which is whats important. an experienced assembly programmer might be able to write faster code than the compiler, but anybody teaching themselves assembly to try to reach that point is going to struggle to write something faster.
Wdym "even" gcc, it's the literal best compiler for C and C±+ that's around in terms of the speed of the resulting binary
python semicolons, you're on another level
one thing to take into account is that python can't inline func calls, so with such task where you have a lot of calls it's better to use nested loop instead of a separate function
Bahaha.... And it cares about whitespaces.. Idiotic, incapable and slow.... C however, where you can just typecast anything to anything, and use data as code and vice versa... Pointers rule!!!
@@MrZnarffy you're arguing that Fiskars axe is worse than Husqvarna chainsaw dude... Both languages have different purposes and are good at them, and your talk about some C features you like makes you look like one of those programmers that do only some code exercises online and no real job.
@@mcroyal8000 ahahaha... Dude, if you can't use pointers, and function pointers, you shoulndn't even speak about C...
I started with Assembler waaay back, done verilog, and lisp, and python, and anything between them. Python is wrtten by morons for morons.. I mean indentation sensitive???? Come on.. Oh.. And interpretative language.. LOL you don't do serious stuff in that.. It's like the C++ hype.. Why on earth have classes and inheritance? So you can be lazy and make bloated messes that is impossible to maintain..
I know COBOL too, but I wouldn't use it if I possibly could avoid it, and Python is just as crappy
@@MrZnarffy you don't do serious stuff in Python and yet it's the 2nd most used programming language after JavaScript. Keep living in your bubble, thinking that only C is good and for everything "bEcAusE p0iNTerS" lol
@@MrZnarffy omg THAT type of programmer...
It just shows the power of the languages designed to be fast vs the one designed for ease of use/learning automation etc (plus it has numpy etc from the past languages, so there's really the best of both worlds)
it's not like they're designed to be fast, assembly is literally just telling the cpu exactly what to do.
It really doesn't if you check how compiling python will produce very similar speeds. It only shows the uploader is a fool.
The winner is assembler.
Assembly*, the assembler turns the assembly into machine code. Like a compiler does with C.
@@nebulium6641 C compiler are very good optimized. But i like to use assembler to create some tiny executable. In assembly you see all instructions like the CPU.
Damn bro dont give out spoilers like that
@@nebulium6641 Python work over a C interpreter. C first translate the code to assembly. Assembly directly turn the file in machine code.
@@Alexis-hj6ci I know. I was saying that Max's terminology is wrong. The assembler is what converts assembly language source code to machine code.
I remember vividly taking this course called "Compilers"... and it was so much fun to know behind the scene how compilers work, how to save variable, how halt work... I am glad to take that course...
Please paste the course link 🙏
I'm taking a Compilers course right now. It's really interesting but the assignments are quite the challenge
@@neeljoshi3923 unfortunately at my univ this is closed course... but there are many online course you can choose. the thing is the concept of compiler course is converting program from a to b.... once you get the idea the rest is EASY!!!
@@banjoguy9000 yes, the mind set behind that is far complicated without good prior knowledge
I enjoyed this course so much more than any
All python integers are bigNum structs that contain an array of integers along with a size in number of cells, and use non-native arithmetic operations (following the python mantra of "everything is an object"). So it makes sense it'd be this slow.
I remember reading a post by some retired game dev who started out programming with assembly and ended with C++.
So with the first consoles they used to write everything exclusively in the native assembly because it was the only available option. When consoles started supporting compilation from C, they still used assembly for any loops because the compiler wouldn't optimize them. Then when the compilers finally started becoming feasible to write only in C, C++ started getting popular. Except then the C++ compilers weren't well optimized either lol. One of the devs even made their own C++ compiler better suited for the console but it still wasn't great. so they opted to continue writing in C until publicly available C++ compilers became very effective.
Then everything was done exclusively in C++ and the guy retired several years later.
I believe he worked mostly with Nintendo consoles (NES, SNES, N64, Gameboy, etc.) but also did a few projects on Sega and PlayStation.
i'd like to read the full post, i wonder where it is
Neat
Yeah, C++ is the standard these days for pretty much any game development that's not just automation scripts. There's not a lot of reason to learn ASM unless you plan to work with it a lot for a specific type of job.
@@coolbrotherf127 Well it is useful learning it for edification, which is never a waste of time.
@@sisyphus_strives5463 It's better to create a goal and master skills that are vital to the accomplishment of that goal. Learning random stuff for the sake of learning is usually idle behavior that doesn't accomplish anything. A programmer who knows a few languages really well is more useful than one who only knows the basics of 20.
why check from 2 to n/2? you only need to check from 2 to sqrt(n) for prime numbers - you can implement this by doing "while i*i
good point - that would significantly decrease the number of iterations, but then I would have to massively increase the range (>>250,000) to get a similar disparity in C-vs-python performance :) . and at that point i*i would get even larger (i believe for this range, it already exceeds the maximum for unsigned int), or for even larger numbers maybe pull it out of the loop and do something fancy. or you can just use a sqrt, but I didn't want to include any math libraries (doing so in asm is a pain & a lot of typing), or have to implement that myself. obviously there are much better ways to compute & count primes than this - this is just something very quick to compare the relative speeds of execution when you have a lot of computations
@@xmdi0 i*i would get too large eh? no problem just compute an upper bound for sqrt(n) using a few iterations of newton's method - run n_ = n_ - 0.5n_ + n/(2*n_) a few times :)))), it's O(1) and should be negligible
now you can iterate from 2 to approximately an upper bound for sqrt(n)
now go make the range bigger :>
oh and just start with n_ = n>>1
I remember In C I had a case where prime numbers between 90 and 100 were not detected as primes if checking until sqrt(n), not the case with n/2
@@EmmanuelMayor that's odd, I just wrote a program using both the i*i method and the sqrt(n) using the math library and it still detect 97 as prime. (91, 93, 95 and 99 are all composite)
For all the people saying the Python could be faster, it could not, not in relations to other languages, at least. The WHOLE POINT of Python is to be flexible and easy do code, that's why it was the fastest to write in the video. But for that, CPyhton has to do a lot of work, like deciding all the types. That way, we can take advantage of dynamic typing. It's a tradeoff, and I think every python user should understand that.
How about Cython?
If you need that level of performance in your app you wouldn't even think about starting coding in python, otherwise, python is the best choice, always
@@AbeDillon Sure! I was in a discuss these days where one of my workmates was saying he doesn't like python because it's slow... ok, sure it is... But you're not writing a powerful algorithm that really needs performance, it wouldn't make any difference in the end and you would got job done faster with python.
@@arthurbc6 well, try making a sudoku generator in python. I did. Made it in python, c# and c++. Guess what. The difference is really visible. Now don't blame my algorithm. Because any algorithmic optimisation would always reflect well in any language. The only real advantage of Python is the many libraries or packages. But that's more because it is the most used language by non-programmers. And python isn't very readable imo. It's also harder to debug. The reason python isn't for me is it's syntax. I like more structure in my programs, which I'm sure many others do as well.
@@arthurbc6 If you know you need performance... You'd still probably write the code in easiest language you have available to you, regardless of speed, and then only when you have the program ready and provably be too slow, you'd profile the performance and optimize most important parts.
I don't think there exists a program where more than 10% of the code has its slowdown by 100x actually in any measurable way impact performance. So even if you know for sure 100% that the program as whole can't run fully on Python, that's 90% of the code you could write in a clearer, cleaner language with absolutely no performance impact, no downsides whatsoever.
And then you have numpy and such, which just might actually be enough for the missing 10% as well.
Premature optimization is the root of all evil.
A year ago when this got recommended to me, I was still learning python.
Now, i already know python and c++ and currently learning assembly :)
I learned these 3 languages in uni and I ended up being a Java programmer
Started out with Basic as a kid on the C64, learned C/C++ for Windows development, then later in my career it was JavaScript as applications went to the Web. Funny how after moving to the cloud I've been using Python for the last 3 years and loving it.
I had a class in Fortran 77. You may actually know what that is.
@@plumreid Same. Did you take a class in Pascal too? Folks are writing malware in it now (Lazarus IDE).
No Pascal. I was getting an Engineering degree and for some reason the only programming language we were required to take was Fortran 77. I also used a truly terrible language called ISETAL in an experimental Calculus class. Looking back I'm sure that my professor was just using us as research subjects to get published.
even better, if you pass -O3 to the C solution it's gonna be much more faster than the hand written assembly
in general, yes, but not here. -o3 has no benefit on this simple procedural algorithm. give it a try if you don't believe me
@@xmdi0 I'll try it, it might be auto vectorized with clang
ironically gcc seems to make it worse with -O3 for me
if I only pass -O1 it's better by a few tenths of a second
Also, you don't need to get up to n/2, sqrt(n) is enough, if there aren't any devisors up to that, there won't be any above that. But this would apply to all three of course. Plus, I don't know what the time cost of calculating the square root would be and if it would be worst in the end.
@@cg0wamp at the very least there's an old fast inverse square root algorithm out there, so just use it and use 1 over the result, that would be better than going up to n/2.
this is one of the coolest videos ever. Hope you keep making more of these.
😆 Way too funny and certainly great. You should do one with the same languages and compile times for modestly sized source. Maybe add Delphi or Lazarus?
Filesize (standalone executable):
Python: some Megabytes
C++: some Kilobytes
ASM: some Bytes
However no one uses standalone python
C++ runtime isn't all that?!
Would be very interested in a comparison of the asm code with the disassembly of the c code(also with optimizations)
Now this is interesting.
Theoretically, if you use the ASM optimized instructions generated from the compiled C code and you compare it to the C code, you should get the same results since it should generate the same object code (binary). The only way one would outweigh the other is if it uses different assemblers. In this case, OP seems to be using gcc for both C and ASM so it will be identical.
Note the C compiler won't do this but: since this code is entirely deterministic the most optimal compilation would be to compute the loop at compile time and then print the number.
if he is using x86-64 gcc 12.1 then his C code looks like this, without optimizations, just straight C translation:
isPrime:
push rbp
mov rbp, rsp
mov DWORD PTR [rbp-20], edi
mov DWORD PTR [rbp-4], 2
jmp .L2
.L5:
mov eax, DWORD PTR [rbp-20]
cdq
idiv DWORD PTR [rbp-4]
mov eax, edx
test eax, eax
jne .L3
mov eax, 0
jmp .L4
.L3:
add DWORD PTR [rbp-4], 1
.L2:
mov eax, DWORD PTR [rbp-20]
mov edx, eax
shr edx, 31
add eax, edx
sar eax
cmp DWORD PTR [rbp-4], eax
jle .L5
mov eax, 1
.L4:
pop rbp
ret
.LC0:
.string "%d
"
main:
push rbp
mov rbp, rsp
sub rsp, 16
mov DWORD PTR [rbp-4], 0
mov DWORD PTR [rbp-8], 2
jmp .L7
.L8:
mov eax, DWORD PTR [rbp-8]
mov edi, eax
call isPrime
add DWORD PTR [rbp-4], eax
add DWORD PTR [rbp-8], 1
.L7:
cmp DWORD PTR [rbp-8], 250000
jle .L8
mov eax, DWORD PTR [rbp-4]
mov esi, eax
mov edi, OFFSET FLAT:.LC0
mov eax, 0
call printf
nop
leave
ret
isPrime is the declaration of the isPrime function, and int i = 2;
.L2 is the condition i
@@dpeRUclips true. I believe Java does that at compiling (at least that's what I recall from college).
Your semicolons as well as unnecessary parentheses and lack of spacing made my day. Anyway, I ran the exact same code, down to those semicolons, on my machine and it took less than 30 seconds. Also, when I ran the code on pypy, it took less than 4 seconds. And of course you get less than 0.1 second for both python and pypy if you use sieve.
First off, epic username. Way better than "The quick brown fox jumps over the lazy dog".
Second off, I still have much to learn about python. This is ever more proof that the guy in the video is a dumbass.
I wrote a small program in C to long divide calculate primes like in the video. On my M1 MacBook Air, it took 1.6s to calculate the first 20,000 prime numbers. I tried writing the same thing in Python. It took just over 2 minutes to calculate the first 20,000 primes. Amazingly, my 10700k takes around 4 seconds for the C version. M1 single threading seems to be godlike.
Try using a built-in function isPrime() in the SymPy module to see how. It is a direct function and returns True if the number to be checked is prime and False if the number is not prime.
Having many modules and built-in functions are the strength of Python.
Assembly gets extra credit for being 1337.
Why not test for i < sqrt(n), or i*i < n in isPrime? A more expensive comparison, but many fewer iterations even for the comparatively smaller numbers you use.
"next i will compare fortran and 4chan" X'D you made my day
I've developed a bit of C too. Mostly for small corner case projects. Few companies wants to pay for the time C takes to write though. Python, C#, Javascript, PHP etc. may be slower than C, but the coding most definitely is quicker and it'll pay your mortgage and food.
That depends on the libraries, goals and person.
Really depends on your application space. I've used far more C than Python, C#, or PHP. Programming in C will definitely also pay your mortgage.
@@mattreigada3745 you are not wrong, but the amount of job postings requiring PHP or C# skills massively outnumber the amount of job postings requiring C.
@@dschledermann I'd be surprised if that still holds true, especially if we distinguish jobs and careers. JS and Java are considerably more popular than C, C#, or PHP but C# and PHP aren't really in their prime any more whereas C really only faces competition from C++ (and increasingly Rust). I generally find the comparison of job listings misleading since the industries these languages are used in have very different turnover rates.
I could argue that this algorithm doesn't make for a good example of the differences between C and assembly. Basically the whole inner loop will be completely bottlenecked on the integer division unit, and the whole rest of the processor will lie mostly idle just trying to feed it, so it doesn't showcase very well how a good assembly programmer may be able to outperform the compiler in various ways. When playing iwth microbenchmarks, I've often been able to improve register allocation or other minor details in the compiled output, but in this case, none of that will matter since it's all just waiting around for that one DIV instruction.
That being said, if you really wanted to optimize this further is assembly, you'd probably convert the integers to floats and use the AVX division unit to increase parallelism.
I think he made his point even without all those optimizations
@G E T R E K T Especially when you consider systems that may run the same small instructions millions and millions of times in rapid succession. The small small speed difference really adds up and it can become mission critical to maintain code in ASM instead of even C.
With proper optimization from the compiler I suspect this is a case where the compiler will result much faster machine code than the machine code generated from the assembly that was written. Two easy optimizations a compiler could reasonably implement are (1) calculating the division a single time outside the loop and (2) substituting this division with a right bitshift. Implementing either of those would give a performance edge over the assembly that was written.
About 99% of the time is taken by the DIV instruction, because of this it is useless to optimize this code, for example, the isPrimes inline function does not give anything. Tested the optimized version (i
A good reson to learn assembly? everything becomes open source. Complete dominance.
Makes no difference. If you have source code for Python, every Python program becomes open source.
@@BrightBlueJim it's called a joke
@@voxelrifts And your point is?
@@BrightBlueJim The difference is you need the source code.
If you know ASM any code can be disassembled and read as long as you know the platform it was compiled for (which is likely since if you ran the program you're on that platform). ASM -> Machine Code is 1:1, therefore Machine Code -> ASM is also 1:1.
Of course python is likely going to be handed to you in python or bytecode at best which can also be decompiled. But things like C/C++ program can't easily be decompiled into C/C++. And this ubiquitousness of ASM is the crux of the "everything" part of the joke.
@@lordofduct Yes, good point.
Every language has its own pros and cons, a real programmer uses a language depending on what he/she is going to do. Python is slow (than others I mean) but is easy and has lots of libraries. actually even its more used for machine learning and scientific calculations, but it can be used for scripting small projects, backend of Web applications, etc.
A real programmer is someone who programs for a living. Gate keeping is cringe, even if you are right.
@@Katt1n how is that not also gatekeeping? Bisqwit is an excellent programmer, but he drives a bus for a living
@@wrong1029 I was referring to "programmer" as a job title. Obviously you can be a brilliant programmer without doing it for a living.
Python libs are literally in C
Python is easy for non programmers to learn, for example those who are interest in analysing data but don't want to learn something more complex and are willing to not learn a better language in exchange for much slower performances.
Python is a scripting language, it can be good for that.
Every language has libraries, C, C++, Java, Dart... literally every programming language does. Admittedly using a library with C++ might be harder than installing a package on Dart, but after you learn how the building process works, it's not hard at all, because you know what you're doing.
It would be interesting to do it with mathematical graphs, plot etc.
For those saying "python" is slow, a JIT will be added to CPython 3.12, you can compile python using mypyc (you need to use type annotations to get the maximum performance), and Cython now has pure python syntax.
There's already numba if you really want JIT for your python.
@@diablominero Numba is very limited it doesn't work everywhere and you need to add it manually to every function. In the other hand CPython jit will be like PyPy's you don't need to change anything at all to gain speed.
@@diablomineroi tested numba with the video code and the time goes down to 4 seconds
Now do one in straight machine code with just a hex editor.
Pf. Hex editor. Do one where the programmer is toggling the front panel switches on a PDP-11/45.
@@BrightBlueJim YES!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Its all about the application build approach. Python is using interpreter approach, wich is make Python very easy to code, but very slow when the application is launching. C/C++ and x86 ASM are using Compiler So its not easy than python to code, but they are very fast when the application is running
Thanks for nothing
Total Time :
Python: 1min 50.4 sec
C/C++: 1 min 24.3 sec
x86 ASM: 2 min 25.3 sec
So overall C/C++ takes less time for this specific program. Not talking about other big programs which has complexity and lots of debugging and errors takes in.
Quem escreveu o ASM não o escreveu tão bem como o compilador. Pense, você acha que o programa em C foi transcrito como? Não é uma questão de linguagem é uma questão de otimização.
When I saw this I wanted to test Lua as well and I was surprised.
Writing:
61s
Execution:
Lua: 38s
Luajit: 7.8s
PS: Writing was calculated based on number of characters in python file compared to lua (+9.5%) since Everyone has different writing speed
In the case of python with list comprehension we can speed up the execution, but surely, even slower than the rest.
screw youtube for hiding dislikes i wanna see how many python users got pissed
currently 95.5% like to dislike, so they are outnumbered lol
@@xmdi0 cope
@@gr1mmd 1900 likes 153 dislikes. The api is still public
I'm curious as to what for you guys code the most. I code mostly to do some matrix computation, interpolting images, extrapolating the price of the beer, and for my hobbies :) Some of these are even faster, using Numpy for example (at least faster than my implementation in C). Are you all developing the next Far Cry? Have you all HUGE databases to handle? Or are you just competing to find the most amount of prime numbers? Tell me, I'm curious 😁
Potato vodka and a good assembler goes a long way
numpy is written in C
@@coolxy2 that's the point
Speed is just one of the many reasons I love C. Moreover, I get to SEE and UNDERSTAND what my code is doing.
I wonder what would you get if you constexpr the isPrime() function in the C++ solution (use g++ to build, and at least c++11)
Tbh I see python as just a scripting language. I use it only for prototyping small ideas or when I'm fed up with my current shell's scripting language.
Yeah I've had trouble finding something to do with it that I wouldn't prefer to use another language for. Network programming is something it's good at... it might be useful for writing some services or for gluing components together, but I haven't done any of that in a professional context.
The comparison we all deserve.
To check if a number is prime you only have to check up to sqrt(n). In python I usually do int(math.sqrt(n)) + 1 and it works 100% of the time
For those who wants to see the prove:
A = b × c (any number can be wrote as a product of another two. That's fact, because all numbers have a root. Example; pi = sqrt pi x sqrt pi).
But if B and C are not the same (which is not always true, for example 1 = sqrt 1 x sqrt 1 and that's the only possibility), then B < sqrt a or C < sqrt A.
Reason: if B and C are > than sqrt a, that's a contradiction, once root's definition implies "the product of two numbers, which ones are equal, that results at another number is the root of that number". The same logical can be apply if B and C < sqrt a. In that case, thejr product can't be A.
Examples:
2 x 5 = 10 or sqrt 10 x sqrt 10 = 10.
Why does all of that helps? Because a prime number can't be divide by any number that is not his root, once a prime is not a product of another two primes (which is the case of 10, for example).
That means we don't need to check all numbers before the possibly prime number "N". We only have to check it before sqrt N, because if the number is not a prime number, then B < sqrt N or C < sqrt N. Once you catch a natural number that divides N, you prove that N is not prime.
But if N is prime, then only his root can divide himself. Because we are checking everythig before sqrt N, after you check the first natural number to comes before sqrt N and it does not divide N, then you know that B == C, which implies B == C == sqrt N.
Hope that helps.
it's interesting how execution time of C matches that of ASM
C would actually be faster if he passed -O3 to the GCC compiler. The reason is that compilers are so intelligent in 2021 that they "know" what programmers tried to achieve with their code and so the compiler writes extremely efficient machine code compared to handwritten assembly which is almost always going to be slower as barely any human can match the 3 decades of tricks programmed into C compilers.
@@v0xl It's an abstraction on raw machine code. It's not much removed, but it is indeed a programing language.
@@joey199412 Very true. This comment needs more likes. Hundreds of the best programmers optimize the compilers for max performance. An ordinary dudes asm code can never match all that. In such cases it really makes more sense to use the more abstract c code
@@joey199412when i searched about O3 found that it's known for being unstable and weird
But even without that, as you can see in the video he uses the gcc compiler to compile the .o file so he Can use O3 for asm
@@ioneocla6577 Not really unstable, neither weird. If your code is proper won't have problems. When building a c program look at all those warning, they mean something.
I would suggest to separate c with c++. After c++11 c++ has started to diverge from c. After c++20, it is basically a different language all together
For basic cases like this, C and C++ would likely produce identical code, assuming the compiler used the same level of optimization. C++ having many more features doesn't change how it compiles code not using any C++ features. Optimal generated assembly from C is likely the same, or similar in C++. Different compilers may have different methods of optimizing execution speed, but generally C code compiled in a C++ compiler will be the same or similar speed.
@@xeridea the problem isn't the code performance, it's the association of C and C++.
@@au-lit I don't think it's a problem. C++ is more or less backwards compatible and C++ compilers will compile C-like code.
If you don't use the new features (which are often slow and/or buggy) it's still exactly the same as C.
@@squee222 Backwards compatible with what? Any respectable C++ isn't backwards compatible with pretty much anything. Compiling any C(-style) code in C++ is a bloody bad idea.
The world needs more videos like this.
you can check numbers up to sqrt(n) + 1 to check if it's prime. Don't know if it would make it run faster because less checking or slower because of sqrt call.
Of course much faster. You can go up to sqrt(n) and check only numbers of the form 6k+1 and 6k-1.
Thank you RUclips algorithms you did it well.
Does the performance of C get better with -Ofast in this case?
in general, it probably does get slightly better (a commenter below mentioned that -O1 gave him the best result), but I wouldn't expect it to be noticeably faster than the ASM. this ASM is literally the barebones procedural implementation of the algorithm described in the C code. the compiler might change the jump conditions or something fancy, but I personally don't see how you can get much "faster" for this algorithm unless you cheat and mess with the algorithm (and we can always cheat in the same way for the ASM code). I think if you have much more complex code, the difference between what ASM the compiler can generate and what ASM a human will casually generate will become larger. However, I personally believe in the limit humans will always be able to generate code as fast if not faster than the compiler (because we can literally hand-write the compiler's best guess), and there may sometimes be shortcuts that humans can take, using our knowledge of the problem. Of course, the compiler runs in a fraction of a second, whereas it would take you and me much much longer.
Absolutely does. I tried it and it ran 30% faster with optimization enabled. No noticeable difference between the different optimization levels but a VERY noticeable difference from no optimization at all.
Nice video
Have you tried PyPy? Its a modified form of Python that runs a lot faster than standard Python, and is iirc completely compatible with normal Python
Hmm good to know
That takeaway was amazing
And this doesn't seem to include thinking time, only writing time. If you include the former, Python, C and C++ should be similar, but assembly would likely be even slower.
That would depend how experienced you are with writing assembly
@@agfd5659 I am experienced enough with assembly to avoid it where possible. Other people have to be able to read the code, too.
@@davidwuhrer6704 not other people, but other programmers
@@agfd5659 Fair enough, but they still have to be able to read it.
(EDIT : I was wrong about the interpreter overhead as python interpeter compiles to bytecode in first stage, but rather it’s mainly slow because of its single threaded, dynamic nature, NOT because of it being an interpreter) Python is an interpreted language, so the python interpreter has to parse and lex the code while executing at the same time, it doesn’t mean it’s a bad language though, its portability, simplicity and great ecosystem of tools that work nicely together makes it attractive for many applications, especially in science and engineering fields, moreover, libraries like numpy and tensorflow are native libraries written in c/c++ so they take care a lot of the gory details in things like matrix operations and other math ops. natively, but can be used from python through nice and clean APIs , without too much interpreter overhead, and that’s the beauty of Python.
Your point about interpreter is not completely correct. Let's take JavaScript for example, it is an interpreted language. Because of just in time compiling it runs very fast.
@@boem231 yeah you’re right, but it was a plain interpreter back in the old days, but with something like the V8 engine made JS much faster thanks to JIT (just in time) compilation and multithread utilisation (like in asynchronous programs to prevent IO blocking), which made chrome browser (which uses v8) stand out in performance a decade ago, and the v8 engine still powerful even in large-scale backend applications using nodejs, there are similar implementations for python as well. the vanilla out-of-the-box python interpreter has no fancy features like JIT. Python is just a language, so engineers are always finding creative ways to optimise and expand the functionality of it.
@@FakhryHTatanaki my bad then, didn't know there were better interpreters than the default one. Have to look in to that
Yeah, this video doesn't mean that python is a bad language, but python is a bad language
@@FakhryHTatanaki Yeah, sure, CPython doesn't have a JIT, but your statement about "the python interpreter has to parse and lex the code while executing at the same time" is still wrong (or at least very misleading). You make it sound like python is processing the text of the program as it is running it. This is obviously not the case.
Python *is* an interpreted language, but it isn't interpreting the program text. It's interpreting the bytecode, that was compiled once, ahead of time. That's what all the __pycache__ folders are for.
The reasons why Python is slow have almost nothing to do with it being an interpreted language. Python is slow because it has duck typing (no compile time type info -> no aggressive optimizations), it's slow because it's a dynamic language (everything is a dict -> every name lookup is a runtime string comparison), it's slow because of the GIL. Compared to these reasons, Python being an interpreted language makes almost no difference.
Back in the days beyond recall, when I was taking my first programming course it was Fortran on an IBM 1130. We had a lab with a graduate student TA and we had to program the sieve of Eratosthenes using 16 bit signed variables to print out the primes that would fit. Pretty basic stuff. Well programs would start out printing five lines per second on that chain printer... and get slower and slower. Up steps the TA and his program ran much longer before slowing down. He had written an interrupt controlled printer driver in ASM of which he was rather proud. I then ran my little Fortran program and it was faster. I thought that everyone knew that all composite numbers had a divisor equal to or less than the square root. The algorithm counts. Well, so does the language. Horses for courses as the British say.
Please tell me everybody clapped and panties were soiled?
One word ( CHAD )
Amazing! And the punchline genius!!!
In Python this is just
print(sum(all(n % i for i in range(2, n // 2 + 1)) for n in range(2, 250001)))
respect
I mean, the whole point of Python is to be portable and readable, this is neither lol.
@@TuxraGamer just using what's given, also respect the // man
@@guiAI You can do the same on C, it's just unnecessary and makes it harder to read.
@@TuxraGamer man i can't convince you
Choose the language to be written according to the appropriate project, this is not much to discuss...According to the needs, the fastest way to run is of course machine language, then assembly, c/c++, the disadvantage is that the development efficiency is not as fast as high-level languages. If you don't pursue high running speed, you want to use high-level languages such as python and java for rapid development.
There are pretty good tools for converting python to C++ if you have a performance heavy app you run a lot.
I used to write assembly (Z80) on paper with pencil, looked up mnemonics from a printed chart, typed in the bytes one by one into a Basic program’s DATA, POKEed it into the RAM with a FOR loop, run the code, watched the PC to freeze 95% of the times with no info about what might have gone wrong, and started again from scratch.
My 3 favorite programming languages. It depends on what you want to. Lately I use python most of the time. ASM was necessary back in the 80s and early90s before floating point became common. Or writing DSP code.
ASM is still necessary, any programmer worth their salt knows how to debug their code.
I've been programing since I was a child, C64 era, and every time a new HLL (high level language) comes out I laugh and ask myself how many cpu cores does this one take to print hello world.
Today programmers are lazy, their was a time (MSDOS-win98) we didn't even have protected memory, do you think todays programmers could deal with that? I don't, their lack of understanding is annoying to the point I catch myself forking their project and writing it in a competent language and optimizing it
wrote the same code in JS and was suprised how fast node runs it.
how many seconds ?
@@mohamedelsayed8428 43seg
@@gonzalolog i don't think so
@@mohamedelsayed8428
Just open the console in the browser and write this:
function isPrime(n) {
for (let i = 2; i
@@1Naif i want to see the performance on the same pc and the same processor
Whats the main differences between LC3 assmebly and x86 assembly?
i made a similar comparison and assembly was about 6-7% faster than c even when i turned the iterators and the counter into registers, however with a quick assembly tweaking into the c code the difference is seemless.
In a large, real-world program, C is usually better in every way. The expertise of people tweaking compilers for decades is beyond nearly everyone trying to code in assembly, especially for complex algorithms. Inline assembly is fine for certain critical portions, but often C with a good compiler and the correct optimization flags will beat it.
after reading some of the comments - there is a reason why python is preferred by people. You can argue all day but more people than not prefer the simplicity of python. Is it slower yes, but that is why lots of python modules are written in C to gain the best of both worlds. Simplicity of python syntax plus the speed of C in case you need it
ok, at first I'm not in python, so how fast does you code this is you use python and C means you make the function isPrime in C and the rest in Python. But i think mostly its quicker to code all in C then.
Well, if I need to quickly write a script to do something, I'd go with python, because you don't have to have a lot of setup
Yeah, and many programs don't need extreme speed. Unless things are within like 5 nested loops or are called very frequently and have to be fast, python is just fine.
Most of the libraries for ML in Python are written in C, for example, NumPy, imagine doing all the logic in Python and speed-requiring work (as matrix operations ) in C.
@@JKLKJ Ok, I'm not in this topic. But I thought all big frameworks are mostly written in C or C++
Wow python so fast, can't wait to write my OS on it.
its actually pretty cool to see that C/C++ and ASM are really close if not exactly the same
Até vc resolver a lógica do problema o Assembler já fez tudo faz tempo.. 😆😆👍👏👏👏
I liked this better when it was on Dave's Garage last year :-)
Wait a second...
Ahahahah
Hello there
Oh lord.
Agree. Too bad I couldn't understand a bit from that video.
There are 2 type of python guy :
1) a newcomer who pick python as his 1st language and always ridicule other programming language by saying the *syntax is ugly* & *too verbose*
2) a professional who had written many apps in many languages and *having passion* in statistic, calculus & linear algebra
My first language was C++, which was pain in the butt to learn. It however was worth it!
@ANoobyPlayer UwU this. I’m a C or V or Zig user but if I want to start a new project I’ll make as barebones as possible a basic thing in python before rewriting it in C or another low-level speedy language
Pyton: programming time ~ running time
Can you do a comparison with Zig, Rust, Carbon Google, and Bun?
I wonder if there are no algorithmic techniques , in ASM, to use X64 CPUs ability to execute more than one instruction per core.
Writing ASM, especially when C compilers already optimize for that, does appear to be an inefficient way to spend programming time.
Out of order CPUs always try to execute more than 1 instruction per tick, up to ~5, but it is quite hard woth this code. Division operation is very slow by the way.
Art.
Recursive functions will be slower in python, javascript etc. I did the same mistake comparing rust-wasm and js.
最近プログラミング始めました。
すみませんがこれは何に注目すればいいのでしょうか。
From sympy import isprime
pcount = 0
for i in range(2,250001):
pcount += isprime(i)
print(pcount)
running time .25 seconds; and if I really wanted to make it fast I would look for a library that will eliminate the loops
remember, you use python for the libraries, if you are not using them, you are not really programming python
also dead giveaway video maker doesn't program much python is the str conversion in the print statement
Pretty sure the whole point was to write basically the same code in all 3
i mean, there are a lot badmouthing about python because they compare exact same code to a compiled language which doesn't make any sense.
so comparing python loop and c loop doesn't prove anything because loop is not the best approach to take with python.
C is way lot cooler than python. And python looks and is better than c. To be honest indentation really makes code look so good without need for verifying if you missed a curly brace.
Sympy doesn't do this bruteforce remainder check, it knows to discard even numbers and divisible by 3 immediately among other tricks.
On my pc the video version takes 58 seconds. When I do range(3, sqrt(n), 2) and check for n%2 first instead, it takes just 0.35 seconds. Sympy takes 0.5 seconds.
Proper primesieve where you only divide by other primes becomes faster somewhere between 10mil and 20mil.
"Use ASM if you are paid hourly" LOLL
And that's why you would do this in python with a library implemented in C
Use ASM when you want Total Control right down to the machine
use python and type slow if you're paid hourly
The thing is that python is very very very good for home tasks and quick algorithms that you won't use millions of times. But cpp and c are just way better for that. If you have an algorithm that takes 1 min to run a cpp/c but 2 in python. And your gonna be running it millions of times it's much more efficient to write it in cpp/c and not in python. It just depends on the circumstanses
the thing is, if it takes a minute in C/C++ it takes half an hour in python
@@atiedebee1020 The thing is
@@SportSync_official the thing is
@@Avatarjasper2003 the thing is
@@atiedebee1020 the thing is, the thing
A question:
Which of the three generates a more robust and faster executable.
Up to which number were you calculating primes? I can't see it on the screen even if I enter full screen. Is it 25 million?
I wish you tested the speed of the execution of each program and how it take each in terms of memory, that way everyone can see the pros and cons of these 3 programming languages
It does.
Read the output. The program isn't run on the shell, it's passed into time, which shows the runtimes 'real', 'system', and 'user'.
Hello, could you please make the same comparison between Julia, C++ and Fortran? It would be very interesting for me to see the time difference.
the diference between that's examples could be the weight of final binary compiled, in my experience when compile an hello world! example in asm the result is a ligthweight compiled file but in c++ the compiled file is heavier than asm file
Please, how did you manage to time all 3 programs?