Hacker News Reader

HNHacker News

Top New Best Ask Show Jobs

Comment by pella | Hacker News Reader

Full thread

pella·

Congrats !

If anyone is interested in the clearlinux optimized zstd build config ( imho it is useful for the 1.5.5 building )

https://github.com/clearlinux-pkgs/zstd/blob/main/zstd.spec

( CFLAGS, 4 patch )

View on HN

What's stopping those 4 patches from being upstreamed?

What advantages or disadvantages am I getting from them?

Looking at it it seems to require avx2. And use the "m" mode for fopen(), which I have no idea what it does, it's not in any documentation I can find.

From my fopen(2) docs:

m (since glibc 2.3)

Attempt to access the file using mmap(2), rather than I/O system calls (read(2), write(2)). Currently, use of mmap(2) is attempted only for a file opened for reading.

I'm very skeptical that this actually makes things faster. For one, zstd is definitely going to be CPU bound rather than I/O bound. But even if it wasn't, there shouldn't be a performance benefit to reading a file sequentially using mmap compared to reading it sequentially using read, which you can easily verify yourself by writing a small program that just reads files both ways.

I think for clear linux, it might be for memory saving in constraint devices rather than for speed optimization.

When a distro maintains patches, it's mostly down to either:

* Change default value for some parameters, that can't be upstream due to compatibility reasons.

* Change build flag / macro for optimization and/or workaround compiler bugs.

Reply on news.ycombinator.com