Adam Stylinski

> @KungFuJesus I thought ARM was dual-endian... it's both little- and big-endian at the same time... I think technically the architecture doesn't explicitly define an endianness one way or the...

I guess the one thing left I might want to implement / want implemented is for someone to reproduce what minigzip and minideflate are doing as callable functions with compression...

I was thinking for benchmark we could just have a canonical path string for the benchmark data and allow the user to drop in whatever file they want to benchmark...

> I'm not sure if it is needed but I don't mind either. Yeah the overhead should be fairly minimal since it's a fixed one time cost, anyway. I would...

Here is a patch that fixes it: ``` python diff --git a/python/Chapter 2/Question2_4.py b/python/Chapter 2/Question2_4.py index e393ece..3ea0516 100644 --- a/python/Chapter 2/Question2_4.py +++ b/python/Chapter 2/Question2_4.py @@ -2,6 +2,7 @@ from classes.LinkedList...

So I did measure quite a bit of difference, now that things are inlined, in removing this specific fast path from the AVX version: https://github.com/zlib-ng/zlib-ng/blob/cd8b6eb8de425c1608153cb6399d4839a547257f/chunkset_tpl.h#L111 Now that we're not having...

> As far as code size goes, I am not seeing any notable differences between HEAD and this PR. _Somewhat_ surprising, we are wholesale duplicating the inffast decode loop. Which,...

So I'm not seeing significant differences with and without the inlining happening on the chunk functions. However, the compiler _is_ inlining chunkmemset for me now with that commit where it...

> `static inline` means the function can't be called from other compilation units, so compiler is more inclined to inline the function without creating non-inlined copy. Even without `inline` keyword,...

> Any code can be simplified, but there is always the risk that it becomes slower than the longer, more complex, version. We also need to keep the code readable...