Daniel Lemire

BTW here is the current code in simdjson (the basis for simdcsv): ```C++ if (bits == 0) return; int cnt = static_cast(count_ones(bits)); // Do the first 8 all together for...

> i don't have access to legacy compiler tech. If you follow... https://github.com/lemire/Code-used-on-Daniel-Lemire-s-blog/tree/master/2019/05/03 You will notice that I offer Docker configurations and scripts allowing you to run it with the...

Here my results with LLVM clang 13 on Zen 2: ``` just scanning: matches = 74 words = 10 1-bit density 11.562 % 1-bit per word 7.400 bytes per index...

>>You can use Dockerfile.llvmlatest > do you mean this Dockerfile.llvmlatest Yes.

@abalkin Most compilers at the highest level of optimization will unroll short simple loops... short is about 16 iterations. Note that this is sensitive to the optimization flags. If you...

@kelas > this is a bold statement, Daniel I disagree. I carefully provided multiple reproducible tests, using multiple compilers and more than one processor, using CPU performance counters, showing no...

@kelas > there is one now. Yes. You are making this claim. I find the evidence lacking. It does not mean that your claim is not interesting. I'd love to...

@abalkin The simdcsv library will not build on anything but NEON or AVX2 targets. Please compile either for NEON or AVX2 targets (e.g., add `-march=haswell` or use an ARM 64...

@kelas @abalkin What happens if you change SIMDCSV_BUFFERSIZE ? Note how it was doubled in this PR. I remember setting SIMDCSV_BUFFERSIZE "approximately". --- Regarding the LEA, we can sum it...

I am querying twitter to see if someone has an idea https://twitter.com/lemire/status/1383210384321036289