MQN

[sima66]

Gordon I hope you can get an avx one for hapless haswell less blokes?
This version is excellent and even if the project carries on this is a good benchmark point or end point. Slight understatement!

for the hapless and feckless have uploaded avx and sse2 24/96, you can explain to aleg why he is still waiting for 32 bit.

Not only Alex! :((

[sbgk]

uploaded 5.14 16/44 and 24/96 avx2

noticed in a 24/96 track classical quartet track the violin was a bit scratchy and the music just didn't make sense, so 5.14
is exactly the same as 5.07 just 2 instructions swapped position.

sounds very good

the instructions were

mov edx, 256
xor r8d, r8d
mov rcx, rdi

previously the xor was the first instruction, my thinking is that the xor is a faster instruction as the cpu sees that it is a special operation that is just zeroing r8d, so it's got some code set up to do that. the mov edx, 256 is slower because the immediate needs to be put into a temporary register before execution. there is an out of order instruction process so the cpu can pick up the xor while doing the mov edx. the other idea is that the 2 movs together would have had a bottleneck as they both use the same part of the cpu. anyway sounds much more civilised.

[nige2000]

uploaded 5.14 16/44 and 24/96 avx2

noticed in a 24/96 track classical quartet track the violin was a bit scratchy and the music just didn't make sense, so 5.14
is exactly the same as 5.07 just 2 instructions swapped position.

the previous 24 96 wasnt ideal

dunno if one formula will suit all

actually quite funny that it was the first 24 96 for 24 in 24, that won over many of the irish jplay crew

serious, serious potential in the hires

dunno if winning football matches will keep aleg going with out 24 in 32

will you have to pull out the old dac to test 24 in 32?

[sbgk]

uploaded 5.14 16/44 and 24/96 avx2

noticed in a 24/96 track classical quartet track the violin was a bit scratchy and the music just didn't make sense, so 5.14
is exactly the same as 5.07 just 2 instructions swapped position.

the previous 24 96 wasnt ideal

dunno if one formula will suit all

actually quite funny that it was the first 24 96 for 24 in 24, that won over many of the irish jplay crew

serious, serious potential in the hires

dunno if winning football matches will keep aleg going with out 24 in 32

will you have to pull out the old dac to test 24 in 32?

still got a mf vlink 192 which is 32 bit.

it's inertia with the 32 bit, need to find the code, understand it, modify it, test it etc

am going to try the minimum period size, to knock that one on the head.

[nige2000]

uploaded 5.14 16/44 and 24/96 avx2

noticed in a 24/96 track classical quartet track the violin was a bit scratchy and the music just didn't make sense, so 5.14
is exactly the same as 5.07 just 2 instructions swapped position.

the previous 24 96 wasnt ideal

dunno if one formula will suit all

actually quite funny that it was the first 24 96 for 24 in 24, that won over many of the irish jplay crew

serious, serious potential in the hires

dunno if winning football matches will keep aleg going with out 24 in 32

will you have to pull out the old dac to test 24 in 32?

still got a mf vlink 192 which is 32 bit.

it's inertia with the 32 bit, need to find the code, understand it, modify it, test it etc

am going to try the minimum period size, to knock that one on the head.

ya thought that the 32 bit was going to be more laboursome

whats the minimum period size?
might not be the min maybe just small

[sbgk]

I can get 3ms

[nige2000]

I can get 3ms

cool

hope it works

[Aleg]

uploaded 5.14 16/44 and 24/96 avx2

noticed in a 24/96 track classical quartet track the violin was a bit scratchy and the music just didn't make sense, so 5.14
is exactly the same as 5.07 just 2 instructions swapped position.

the previous 24 96 wasnt ideal

dunno if one formula will suit all

actually quite funny that it was the first 24 96 for 24 in 24, that won over many of the irish jplay crew

serious, serious potential in the hires

dunno if winning football matches will keep aleg going with out 24 in 32

will you have to pull out the old dac to test 24 in 32?

still got a mf vlink 192 which is 32 bit.

it's inertia with the 32 bit, need to find the code, understand it, modify it, test it etc

am going to try the minimum period size, to knock that one on the head.

Gordon

Thx.

I cannot comment on any of the current highres version as I have two DDC/DAC's in use and both are 24-in-32.

The RME Fireface UCX in my office system is 24-in-32 but I don't use that for judging sound quality as it is not as revealing.
In my main system the Sonicweld Diverter HR2 which is also 24-in-32.

The old code for 24-in-32 we used to have did just work fine on those devices, so could serve as a model.

Looking forward to a high res to try

In the end the target sample rates I would need to have are 44.1/24, 88.2/24 and 96/24.
Upwards of these would for me be a nice bonus but not essential.

Cheers

Aleg

[Ken Moreland]

5.14avx2 24/96 is much improved on 5.07 (even though I liked that one yesterday) . Just started listening to it but the improvement is clear, better bass, lots of detail and atmosphere.

[cvrle59]

I didn't try 5.14 24/96 yet, but I played an album with 5.14 16/44 last night. I like it better than 5.07 so far. Anyway, 5.07 was behind 4.82, in my ears.