Daniel Lemire
@lemire
Followers
21,465
Following
1,348
Media
1,910
Statuses
28,199
Computer science professor at @TELUQ (Université du Québec, @ReseauUQ ), programmer, blogger. @Nasorg 👨💻
Montreal, Quebec
Joined November 2007
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
#母の日
• 337969 Tweets
#นาฏราชครั้งที่15
• 309964 Tweets
WIN AT NATARAJA AWARDS
• 119280 Tweets
Feliz Dia
• 117976 Tweets
テンハッピーローズ
• 109104 Tweets
Mães
• 84346 Tweets
#光る君へ
• 71957 Tweets
カーネーション
• 54573 Tweets
MC NUNEW EP1
• 49618 Tweets
Cibeles
• 44500 Tweets
#やまラスト
• 37877 Tweets
新ビジュ
• 35582 Tweets
DONBELLE BOX OFFICE LEGACY
• 31905 Tweets
Ohm X Nataraja Awards
• 17094 Tweets
#光る君ヘ
• 16011 Tweets
GENCelebrate Music With BINI
• 12262 Tweets
清少納言
• 10366 Tweets
Pinned Tweet
You can sponsor my open source work on GitHub.
3
5
56
Quebec is lifting its universal mask mandate tonight at midnight. The mandate lasted 666 days.
965
3K
12K
Floating-point Number Parsing w/Perfect Accuracy at GB/sec
Talk by myself at Go Systems SF.
I don't know what happened to my hair that day.
8
50
4K
Your operating system, your browser, your database engine, your web server, git, your JavaScript engine, your Python interpreter… all of these are likely written in C/C++.
I repeatedly encourage people to learn many programming languages. I do not (at all) think that C or C++…
people here talking about how C/C++ is unsafe, no good, don't learn it
Your browser was written in C. Your OS was written in C. AAA games are in C. game engines are in C. "safe" languages are (or were) in C. Your art tools, written in C. Audio tools, C.
211
269
4K
104
249
3K
When Java became popular, people (me included) claimed that it was massively better than C/C++. This was highly controversial and people mocked me for using Java. I was hammered by the referees during my first grant application for picking Java as my language of choice.
In some…
217
258
3K
"Data Engineering at the Speed of Your Disk", my talk at the 3rd Performance Summit (supposedly in Seattle but I was in my bedroom with my cat)
3
113
2K
Knuth did it: The Art of Computer Programming, Volume 4, Fascicle 5 is out!!! (Was supposed to come out in 1960.)
28
743
2K
It is hard to overstate how strong the push for object-oriented programming was. It even bled out into other fields like education (look up "learning objects"). You had to organize your programming projects into hierarchical classes and you would be ridiculed if you did not. Java…
The older you get, the harder to resist saying "I told you so."
When OO programming came in, it made no sense to me, and I've never used it. Everyone said I was too old to understand.
Thirty years later, everyone's snapping out of it and wondering wtf they'd been thinking.
222
184
3K
128
231
2K
We released simdjson 0.3: the fastest JSON parser in the world is even better!
cc
@jkeiser2
@EgorBo
@markpapadakis
@_msw_
@PaulDreik
@heinz_gies
@geofflangdale
@worksonarm
@eddelbuettel
@sunnygleason
@emaxerrno
5
107
2K
Parsing JSON Really Quickly: Lessons Learned (on YouTube)
0
65
1K
People believe that if they could somehow walk into a research lab, they could then 'steal the knowledge'.
It does not work like that. Innovation is illegible.
You can go, right now, on many campuses all over the world, just walk in and talk to the professors, the students,…
68
193
1K
Intel is extending its instruction set. "Intel® APX doubles the number of general-purpose registers (GPRs) from 16 to 32. This allows the compiler to keep more values in registers; as a result, APX-compiled code contains 10% fewer loads and more than 20% fewer stores than the…
27
186
1K
Faster remainders when the divisor is a constant: beating compilers and libdivide
paper:
code:
6
91
1K
@AstraKernel
There is no need to mention Rust because 50 people will do it in the comments.
15
40
1K
Don’t EVER make the mistake that you can design something better than what you get from ruthless massively parallel trial-and-error with a feedback cycle.
That's giving your intelligence much too much credit.
—Linus Torvalds
21
179
1K
You can learn C in a few weeks. Do it.
Most people do not become full-time C programmers, and that's a good thing... But C is still in the top 5 most used programming languages and it has been the case for the last 40 years. It is hard to beat as a track record.
Let us say you…
@techsavvytravvy
@lemire
I remember buying "Learn C in 21 days" when I was 17. Used a couple of more weeks, so yeah. But useful? Nah. Took a decade before I had any use for it (compilers), and that's 13 years ago. Being proficient enough to do something useful is a different thing than learning ofc
3
1
22
65
123
1K
C23: a slightly better C
One of the established and most popular programming languages is the C programming language. It is relatively easy to learn, and highly practical.
Maybe surprisingly, the C programming language keeps evolving, slowly and carefully. If you have GCC 23 or…
46
158
847
In C++, a function can return *3* values. Not just two.
😜
In Go, a function can return MULTIPLE results?! This is mindblowing 🤯
344
66
1K
74
26
740
In one of my classes, I always use 42 as a seed.
Today, a student asked what was special about 42.
Kids don’t know anything.
86
31
735
You cannot compress random data. In fact, that is what random means… if you can compress purely random data, then it is not random.
87
40
716
Learn C. You can learn it in weeks.
“You dont need to learn C, I didnt”
Anyone telling you that is limiting you in favor of their ego
Most of the worlds software is in C/C++
Operating systems, browsers, game engines, your favorite programming language
Is most likely written in C/C++
It certainly helps to…
135
145
1K
48
39
714
The Windows Dev Kit 2023 is great. You get this tiny box running a mobile ARM processor (8 cores) with 32 GB of RAM, and 512GB fast NVMe. You get a full Windows: you could not tell that it is not Intel. Microsoft is ready for 64-bit ARM systems.
I have…
48
49
557
I have been asked by some organization in France to send a signed letter in three copies. I asked whether I could send a scanned letter by email instead. They agreed. They insisted that it be in 3 copies. So I sent three emails and asked whether it was ok. They said I am good.
27
69
533
I wasted almost half a day of research yesterday because I trusted ChatGPT's answer too much. I asked ChatGPT how to do X using the programming language Y (it is not a secret, but details are irrelevant). The answer looked quite credible and there were references to human experts…
106
65
490
The man who invented the hash table never studied computer science. In fact, he never went to college.
Stop making excuses and build stuff.
The German-American inventor Hans Peter Luhn (IBM):
- invented hash tables
- invented document summarization
- invented modern knowledge management
- invented KWIC index
- invented modulo-10 checksums
- invented publish-subscribe alert services AKA "SDI";
all in the 1950s!
1
24
153
9
66
470
Apple's A12 reaches 3.7 instructions per cycle on a bitmap decoding test, that's 40% better than an Intel Skylake. Put another way, the 2.5 GHz A12 retires instructions like a 3.5 GHz Intel Skylake. cc
@vielmetti
@thecomp1ler
@geofflangdale
@trav_downs
@stephentyrone
14
123
417
The ARM-based servers (e.g., graviton 3) on AWS appear to also default on 4 kB page sizes. That's unfortunate because it makes memory allocation and page faults more expensive/likely.
With Linux, you can use transparent huge pages, however, and this can be tremendous benefits in…
people like to talk about the M1’s CPU performance but ignore the perf benefits from page size grow from 4096 -> 16 KB
Page faults are expensive and not much desktop software uses < 16KB anyway
13
11
237
13
46
405
When nearly the entire world required face coverings, Sweden did not. When nearly all countries closed schools, they did not. Many predicted that Sweden would be a catastrophe. Today we know that Sweden has one of the lowest excess mortality in the world, same as Japan and South…
The USA would have had 1.50 million fewer deaths if it had the performance of Sweden during the pandemic.
3
5
28
23
83
407
Safer than Rust, Newer than Zig, Prettier than Python. Unlike Cython, it has a corporate website.
Go Modular!
22
25
392
"But Daniel, how do I divide two integers in C++?"
I got you covered...
Apparently, it is how you compute the integer division of 455 by 10 in JavaScript.
The answer was upvoted 27 times.
I don't know what to think at this point.
28
19
288
12
19
368
Reminder: open-source projects are not a source of free technical support. If there is no documentation, offer to write some. If some function is missing, offer to write it. Do not demand that functionality be added, that documentation be provided… unless you are willing to pay.
10
73
361
Measuring energy usage: regular code vs. SIMD code
Modern processor have fancy instructions that can do many operations at one using wide registers: SIMD instructions. Intel and AMD have 512-bit registers and associated instructions under AVX-512.
You expect these instructions…
11
31
359
No, GitHub. This is not "commented-out code". This comment tells the reader that k is equal to b2->arraysize.
I am really starting to hate this trend of overbearing static analysis.
25
13
350
You can write C++/Rust/Go/Java code that is much slower than Python. So please use Python if it is fast enough for your needs. Do not assume that "written in Rust/C/C++" means "fast and efficient". It does not.
39
38
339
Do not update your Visual Studio to the latest version.
C/C++ code may wrongly compile to AVX-512 instructions under Visual Studio. Since not all hardware has AVX-512 support, this is causing crashes all over.
Bug:
14
49
309
This talk by Stonebraker (the guy behind PostgreSQL) should be a classic.
5
51
304
And we finally learn the horrible truth: C++ conferences do not run on C++ software. They use Python !!!
27
29
299
Currently, programmers who build new systems have higher status than people who maintain existing systems. In some companies, to quickest path to promotion is to build new software.
More advanced automated code generation might change the prestige hierarchy. AI reveals that…
I expect LLMs to eventually increase the status of the maintenance programmer. Partial or complete automation of code creation will lead to regulatory pressure for *someone* to understand systems in certain critical domains- a skill set not unlike maintenance programming.
3
5
47
21
36
300
Apparently, it is how you compute the integer division of 455 by 10 in JavaScript.
The answer was upvoted 27 times.
I don't know what to think at this point.
28
19
288
@jordanebelanger
Java wasted a lot of time playing with user interfaces... and getting it wrong time and time again.
5
1
291
@jeffreyatucker
Taleb and
@tylercowen
were major disappointments. Also
@reason
for their approval of vax passports both in print and in their podcasts.
23
24
279
Shamefully, I was stuck for a long time, unable to write a small web app in C++.
Turns out that it is easy enough.
19
18
279
Kids in 2050... "Why are we using 4kB pages on our computers when we have 12 Terabytes of RAM ?"
Old folks like me...
"Well, see, back in 1980, we thought that 4kB was good... and it is too late today to change it. Sorry. It is now a constant of the universe."
@lemire
4 KB as page size was usual around 1980 on mainframes with 16 MB memory. I wrote an asynchronous transaction system with own memory pool at this time. 10 years later the page size was increased.
0
0
25
11
18
279
Parsing IP addresses crazily fast
Most of us are familiar with IP addresses: they are strings of the form “ddd.ddd.ddd.ddd” where ddd is a decimal number of up to three digits in the range 0 to 255. For example, 127.0.0.1 or 192.168.0.2.
Each of the four number is a byte value,…
11
44
276
My blog post was badly wrong. I should have written:
1. The Apple M1 is a match for Intel/AMD AVX2.
2. Apple’s x64 emulation layer is so good that I mistakenly benchmarked under it without realizing it.
Lessons learned.
Cc
@thecomp1ler
@trav_downs
@stephentyrone
@Gok
11
20
263
At my school, overall revenues are up 35% over 7 years. Spending on teaching and research went down 0.5% while management's budget is up by 64%. We have more reports and more meetings than ever. It appears to be the same trend everywhere in higher education?
19
54
254
Parsing JSON faster with Intel AVX-512
Release 2.0.0.
Minification at 12 GB/s, find_tweet benchmark at 7.4 GB/s.
cc
@jkeiser2
@geofflangdale
8
47
254
It looks like mold will make our builds faster.
It is a super fast linker.
mold: A Modern Linker
"mold is a faster drop-in replacement for existing Unix linkers. It is several times quicker than the LLVM lld linker, the second-fastest open-source linker..."
2
24
160
8
14
257
Programming languages such as JavaScript, C++ and C are standards with multiple competing implementations.
Programming languages such as C#, Scala, Go, Swift, Rust, PHP are mostly single-vendor.
I recommend learning JavaScript, C and C++ among other programming languages.
46
27
246
Parsing time stamps faster with SIMD instructions
In software, it is common to represent time as a time-stamp string. It is usually specified by a time format string. Some standards use the format
%Y%m%d%H%M%S meaning that we print the year, the month, the day, the hours, the…
7
32
247
Expected performance of a Bloom filter
A hash function is a function that maps a value (such as a string) to an integer value. Typically, we want random-looking values.
A Bloom filter is a standard data structure in computer science to approximate a set. Basically, you start…
3
39
245
Book review: Theft of Fire by
@Devon_Eriksen_
When I was young, science fiction was the genre of choice for many engineers and scientists. But the genre declined significantly in recent years. Part of the problem is the rise dystopian fiction. In the imagined future, we are no…
12
32
243
Identifying a useless delay of hundreds of milliseconds saved the world.
Benchmarking saves lives.
@yagiznizipli
I'm still in shock the xz backdoor happened.
But even more surprising is that it got caught because a dev noticed login in to his machine via ssh was taking 0.8s instead of the usual 0.3s and decided to look into it.
And he happened to be familiar with the Valgrind situation
26
410
6K
1
24
243
Transcoding Unicode strings at crazy speeds with AVX-512
In software, we store strings of text as arrays of bytes in memory using one of the Unicode Transformation Formats (UTF), the most popular being UTF-8 and UTF-16. We must constantly convert our strings from UTF-8 to UTF-16…
7
28
235
Vinay is correct.
Science is fundamentally honour-based. Most researchers, me included, spot dubious work all the time. Calling it out publicly is a massive time sink.
So science has no error correction per se. Bad ideas linger and they are only displaced, over long periods, by…
Agree with Nate. Also think many academics do not read other people's work much. They are so focused on their own grants. And the culture of being a sycophant is common. See also med Twitter
5
13
134
12
50
229
There are many ways in JavaScript to read a file. I wrote a small benchmark where I repeated read a file from disk. It is a simple loop where the same file is accessed each time. I report the number of milliseconds needed to read the file 50,000 times. The file is relatively…
14
27
221
If you think that performance does not matter, watch this video by
@jarredsumner
comparing the performance of his fast JavaScript runtime with the standard Node.js runtime.
You see why we care?
cc
@yagiznizipli
Calling fetch(url) 15,000 times, in batches of 50
left: Bun v1.1.7
right: Node v22
51
116
2K
32
25
407
Recognizing string prefixes with SIMD instructions
Suppose that I give you a long list of strings (e.g., “A”, “A6”, “AAAA”, “AFSDB”, “APL”, “CAA”, “CDS”, “CDNSKEY”, “CERT”, “CH”, “CNAME”, “CS”, “CSYNC”, “DHC”, etc.). I give you a pointer inside a much larger string and I ask you…
14
36
219
Despite having 20 years of experience with C++, when I compile a non trivial chunk of code for the first time without any error or warning, I am suspicious. It is not, usually, a good sign.
8
24
217
Careful with Pair-of-Registers instructions on Apple Silicon
@EgorBo
is an engineer working on C# compiler technology at Microsoft. He had an intriguing remark about a performance regression on Apple hardware following what appears to be an optimization. The .NET 9.0 runtime…
5
26
219
Programming with iterators sounds like a good idea on paper... but in practice, it has limited usability. This is true irrespective of the programming language.
I hate it when languages try to be "smart" and return some weird type when I'm expecting an array/list. It's super awkward to have to deal with some LazyWeirdVoodooShitIterator or FunkyWeirdIterable that's not interchangeable with a normal f'ing list.
24
27
380
39
10
216
Measuring the size of the cache line empirically
Our computers do not read or write memory in units of bits or even bytes. Rather memory is accessed in small blocks of memory called “cache line”. For a given system, the cache line size is usually fixed and small (e.g., 16 to…
10
34
212
New release of the simdjson library: version 1.0. Parse JSON multiple times faster. Help fight climate change by using several times less energy.
cc
@jkeiser2
,
@geofflangdale
5
39
212
Hotspot performance engineering fails:
Developers often believe that software performance follows a Pareto distribution: 80% of the running time is spent in 20% of the code. Using this model, you can write most of your code without any care for performance and focus on the…
“Is performance really the bottle neck of X?” - This question will eventually make me stop contributing to open-source software. I can’t and will not understand why people are against fast code, and accepting slowness.
25
15
242
26
49
341
My oldest boy (18) is interested in learning more about cybersecurity, cryptography and stuff. What is a good book to get started?
84
30
206
@FrankGrimes_Jr
@atRachelGilmore
Major measles epidemic in the region of Quebec despite a 99% vaccine coverage
"The vaccination coverage among cases was at least 84.5%. Vaccination coverage for the total population was 99.0%. Incomplete vaccination coverage is not a valid explanation for the Quebec City measles…
12
12
206
« C (along w/ C++) is far more ubiquitous than COBOL ever was, and it has been the bedrock of most important software technology for pretty close to 50 years. You might think that’s hyperbole, but it absolutely is not. Currently, Python has been the world’s most popular…
18
26
212
Irrespective of your programming language, watching dynamic memory allocation... is a good practice.
As much as possible, allocate all memory at the start and stop there.
Programming languages like C and Zig that force you to make memory allocation explicit are helpful.
In…
think of malloc/free as so expensive, they might as well be filesystem APIs. use sparingly, like fopen/fclose
3
12
176
6
13
191
The following is likely to compile to only a handful of instructions and it is hard to beat in terms of performance.
33
1
192
Red Hat Enterprise Linux 9 and Rocky Linux 9 require x86-64-v2 on x64 platforms which means that they assume
- SSSE3,
- POPCNT,
- SSE4_1,
- SSE4_2,
- CMPXCHG16B.
Software built on these system will not run on older CPUs.
This is a good thing: we should not be continuously…
18
25
199
Parsing integers quickly with AVX-512
Gist: you can be at least 2x faster than conventional functions.
9
17
185
Parsing 8-bit integers quickly
Suppose that you want to parse quickly 8-bit integers (0, 1, 2, …, 254, 255) from an ASCII/UTF-8 string. You are given a string and its length: e.g., ’22’ and length is 2.
How fast can you go?
16
19
184
« Code coverage » is utterly useless as a software-quality metric. The only reason I sometimes report it is because some asked to see the number and it takes too long to explain why it is useless.
Please stop advertising your high code coverage. It's meaningless for your users.
high code coverage ≠ well-tested
15
7
71
37
15
179
We do not know that sunscreen prevents skin cancer. It has been claimed repeatedly by health agencies and the WHO, but there is no strong evidence behind it.
Don't believe me?
Here is the conclusion of a recent metastudy (1):
"While the current evidence suggests no increased…
I don’t wear sunscreen, and I never will.
We blame the sun for cancer when we should be blaming our diets.
But if we cleaned up our diets, how would Big Food and Big Pharma make their money?
Sunscreen and a poor diet will make you sick.
Sick people = $$$
Don’t be a pawn.
1K
308
2K
40
19
183
We released a new version of the fastest JSON parsers in the world. Version 0.5 has better performance under Visual Studio, a new UTF-8 validator (lookup4), better support for C++20, we dropped computed GOTOs without reducing the performance, and more
6
38
180
The AMD Rome processor learns to predict branches almost perfectly within 10 trials (much better than Intel Skylake)
credit:
@erwanaliasr1
6
61
176
C can be memory-safe by
@ErrataRob
Forking is a foolish idea. The core principle of computer-science is that we need to live with legacy, not abandon it.
And there's no need. Modern C compilers already have the ability to be memory-safe, we just need to make minor -- and…
20
29
182
People who are predicting the end of software programming have no clue.
All coding projects have two parts:
1. The fun part: where you get to "create"
2. The pain part: where you have to debug
Code LLMs are "automating" the fun parts while introducing bugs and not helping much with debugging. As a developer, you’re left with more pain to deal with.
147
249
2K
9
21
180
@Craig_A_Spencer
People who went to the hospital were told to come back when their symptoms were critical. No treatment attempted.
16
5
171