Intel SVT-AV1 0.6 Released With AV1 Decoding, SIMD Optimizations

[phoronix]

Phoronix: Intel SVT-AV1 0.6 Released With AV1 Decoding, SIMD Optimizations

Intel's open-source developers working on their Scalable Video Technology video encoders (and decoders) on Tuesday released SVT-AV1 0.6 as their latest work on high-performance AV1 support using CPU-based encoding/decoding...

http://www.phoronix.com/scan.php?page=news_item&px=Intel-SVT-AV1-0.6-Released

[tildearrow]

What about dav1d?

[Azrael5]

Which instructions of the cpus are involved in decoding of this codec?

[jacob]

Which instructions of the cpus are involved in decoding of this codec?

I hope it doesn't require AVX512.

[quikee]

What about dav1d?

They wanted to make a decoder to be complete (they had the pieces already as an encoder requires most of the things a decoder needs). It is not going to replace dav1d and that's not their goal, but it gives us another alternative. I think they want to propose svt-av1 as the reference codec for AV2 (instead ob libaom) so having a decoder is important from that aspect.

[FireBurn]

I hope it doesn't require AVX512.

It says Gen 5 Core is the minimum - which I think is Broadwell (the one before Skylake)

It would be great if someone could try compiling this on something older and see how well it works

The source suggests there's a pure C implementation with folders for SSE2 SSE3 SSE4_1 & AVX2 and AVX2 was added in the 4th Gen Haswell

[Azrael5]

I hope it doesn't require AVX512.

I fear that it involves just that kind of codec.

[ms178]

I hope it doesn't require AVX512.

They would shoot themselves into their own foot if they did that as practically no desktop user owns a AVX-512 capable processor. On the other hand, I'd like to see them pushing AVX-512 further as a performance booster as it is the next thing on x86 for SIMD. And I'd love to see AMD adopt it at least with Zen 3 as there are some instructions which should make SIMD more useful in more usage scenarios (at least according to the original programmer of the ISPC compiler). But Intel's first hardware implementation was hampered by its frequency penalty, hence I can understand AMD's position to wait until it is practically feasible to spend that amount of transistors on it. Also AVX-512 is scattered around different subfeatures and software developers had no incentive to implement something as fancy with no users of it.

[NateHubbard]

What about dav1d?

This is far more interesting to me, because it's actually an encoder, not just a decoder.
If you've ever tried encoding AV1, you'd know why a faster encoder is needed.

[tildearrow]

If you've ever tried encoding AV1, you'd know why a faster encoder is needed.

I have. However, reports say SVT-AV1 produces poor quality results...