0.30000000000000004 VS glibc

This is due to the way floating point numbers are represented in memory, certain numbers like 0.3 are stored as a number very close to the original value (0.30000000000000004), but not the exact same. This problem also exists in other languages like JavaScript and C++, but is much more likely to cause problems in common applications of Python like machine learning and data science.

https://0.30000000000000004.com/

Although it would be good to move in the direction of using a BigDecimal equivalent by default when ingesting unknown data.

There's no such thing as precision for floats. Floating-point calculations are always inaccurate: read this: https://0.30000000000000004.com/

Edit: This specific example even has its own website: https://0.30000000000000004.com/

I'll take existence proofs [1] over personal insults but YMMV.

[1] https://github.com/bminor/glibc/commit/7a61e7f557a97ab597d6f...

Depending on how you count, the ratio might not be that small. A lot of hot code are written in hand-coded inline assembly, so in terms of CPU cycles run it's probably non-negligible.

i.e. take a look at the glibc implementation of 'strcmp` [0]

[0] https://github.com/bminor/glibc/blob/master/sysdeps/x86_64/m...

I wonder if you’re using a different definition of ‘vectorized’ from the one I would use. For example glibc provides a vectorized strlen. Here is the sse version: https://github.com/bminor/glibc/blob/master/sysdeps/x86_64/m...

It’s pretty simple to imagine how to write an unoptimized version: read a vector from the start of the string, compare it to 0, convert that to a bitvector, test for equal to zero, then loop or clz and finish.

I would call this vectorized because it operates on 16 bytes (sse) at a time.

There are a few issues:

1. You’re still spending a lot of time in the scalar code checking loop conditions.

2. You’re doing unaligned reads which are slower on old processors

3. You may read across a cache line forcing you to pull a second line into cache even if the string ends before then.

4. You may read across a page boundary which could cause a segfault if the next page is not accessible

So the fixes are to do 64-byte (ie cache line) aligned accesses which also means page-aligned (so you won’t read from a page until you know the string doesn’t end in the previous page). That deals with alignment problems. You read four vector registers at a time but this doesn’t really cost much more if the string is shorter as it all comes from one cache line. Another trick in the linked code is that it first finds the cache line by reading the first 16 bytes then merging in the next 3 groups with unsigned-min, so it only requires one test against a zero vector instead of 4. Then it finds the zero in the cache line. You need to do a bit of work in the first iteration to become aligned. With AVX, you can use mask registers on reads to handle that first step instead.

That was also my thought. To my knowledge `/etc/localtime` is the creation of Arthur David Olson, the founder of the tz database (now maintained by IANA), but his code never read `/etc/localtime` multiple times unless `TZ` environment variable was changed. Tzcode made into glibc but Ulrich Drepper changed it to not cache `/etc/localtime` when `TZ` is unset [1]; I wasn't able to locate the exact rationale, given that the commit was very ancient (1996-12) and no mailing list archive is available for this time period.

[1] https://github.com/bminor/glibc/commit/68dbb3a69e78e24a778c6...

That's not really what the problem is. The actual code is fine.

The issue is that the definition of `fd_set` has a constant size [1]. If you allocate the memory yourself, the select() system call will work with as many file descriptors as you care to pass to it. You can see that both glibc [2] and the kernel [3] support arbitrarily large arrays.

[1] https://github.com/bminor/glibc/blob/master/misc/sys/select....

[2] https://github.com/bminor/glibc/blob/master/sysdeps/unix/sys...

[3] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/lin...

The source code for that is here.

Expanding macro gives three GCC function attributes [2]: `__attribute__ ((malloc))`, `__attribute__ ((alloc_size(1)))` and `__attribute__ ((warn_unused_result))`. They are required for GCC (and others recognizing them) to actually ensure that they behave as the standard dictates. Your own malloc-like functions won't be treated same unless you give similar attributes.

[1] https://github.com/bminor/glibc/blob/807690610916df8aef17cd1...

[2] https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attribute...

IFunc relocations is what enables glibc to dynamically choose the best memcpy routine to use at runtime based on the CPU.

see https://github.com/bminor/glibc/blob/glibc-2.31/sysdeps/x86_...

memmove can be very well implemented in pure C, libc implementations usually have a "generic" (meaning, architecture independent) fallback. Here is musl generic implementation and its x86-64 assembly implementation. For glibc, implementation is a bit more complex, having multiple architectures implemented, but you could find a generic implementation with these two files: memmove.c and generic/memcopy.h.