Yusuf Redžić

[REQUEST]: Option to filter vertical whitespace within same function

2

### Is your feature request related to a problem? Please describe The use of inline asm, at least in Rust, seems to cause extra vertical whitespace to be printed in...

Remove extra quotes and spaces from custom args

1

Currently, some unexpected behavior can happen if the user is not careful to properly pass the arguments to av1an. This allows everything to work properly even if extra quotes are...

Zero-copy scene detection

13

Scene detection, especially at higher resolutions, is currently bottlenecked by many allocations and copying of uncompressed frames. This PR uses the ffmpeg API directly to decode frames and removes copies...

Confusing error message for pixel format in some cases

3

Currently, if the input video has a pixel format that is considered unsupported by `--encoder`, when we try to get the pixel format for scene detection, a confusing error message...

Add initial aarch64 neon support

3

This PR adds a NEON implementation of all `mem{r}chr` functions. This PR does not add NEON support for `memmem`, as deeper changes in the code are needed for the `Vector`...

Poor performance on Zen 1/Threadripper due to loop unrolling

3

I have noticed that the SSE2 implementation of memchr in this crate unrolls the loop 4x. Unfortunately, this seems to lead to a significant performance drop on processors on the...

Use of `AtomicPtr` in `unsafe_ifunc` prevents memchr from being inlined when compiled with avx enabled

2

`is_x86_feature_detected!` resolves to just a `return true` when `-C target-cpu` or `-C target-feature` is set to a value that enables the feature. When using just a simple `is_x86_feature_detected!` and `-C...

Use fixed-sized array in luma/chroma pred mode counts

2

According to #2236, this was converted to an ArrayVec since at that time Rust did not support deriving `Debug` for fixed-sized arrays above 32 elements, but this is no longer...

In x86inc, `pxor m0, m0` is translated inefficiently

4

It seems like in x86inc.asm, `pxor m0, m0` gets translated into `vpxor ymm0, ymm0` when `mmsize == 32`. This is a bit inefficient because `vpxor xmm0, xmm0` does the same...

Add AVX2 implementation of `first_max_element`

11

This also now requires BMI1 and BMI2 for AVX2 in `CpuFeatureLevel` (for tzcnt).