Jump to content
IGNORED

Visual studio 2012 c++ and wasapi minimalist player


sbgk

Recommended Posts

last 2 version 2.31 & first 2 version 2.32: real/true/clearly sounds, but emotion not enough as jesuscheung said. Why is that?

.....

i stand corrected- as i said i had a misconfiguration in bios that is bad for 2.32. now 2.32 is as true/real as nopgomemcpy in the mids and trebles. not sure about the bass coz 2.32 is colder than the warm nopgomemcpy. yea, 2.32 needs some warmth to get the emotion going.

Link to comment
i stand corrected- as i said i had a misconfiguration in bios that is bad for 2.32. now 2.32 is as true/real as nopgomemcpy in the mids and trebles. not sure about the bass coz 2.32 is colder than the warm nopgomemcpy. yea, 2.32 needs some warmth to get the emotion going.

 

right where were we, 2.32 had a bug which caused a cache miss at the start,oops, that's why it sounded a bit rough. Have uploaded 2.33 sse4 intel rel which is the fixed version, 2.33 sse4 intel buf buf which is another way of doing things, 2.34 sse4 intel buf buf no xmm which is a warmer version, 2.34 sse4 intel buf buf no xmm nopgo which doesn't have profile guided optimisation and finally the famous 2.34 nopgomemcpy which is an updated version of the original. I like the 2.34 sse4 intel buf buf no xmm version, the nopgomemcpy has good bass, but digital sounding on my system - it uses the original memcpy function

There is no harm in doubt and skepticism, for it is through these that new discoveries are made. Richard P Feynman

 

http://mqnplayer.blogspot.co.uk/

Link to comment
...Vocal need to nearer, bass need so far (little) at listeners, drum need to more power and pressure (but don't turn up volume).

...nopgomemcpy version give me ...pressure really enough...

 

tested 3 last versions:

(1) mqnplay.exe 2.34 nopgomemcpy,

(2) mqnplay.exe 2.34 sse4 intel buf buf no xmm,

(3) mqnplay.exe 2.35 sse4 intel

 

drum need to more power and pressure (but don't turn up volume).

in (1) bass OK;

in (2) and (3) bass need so far (little) at listeners, i feel 3D wide effect is turning up and not natural in stereo 2 channel presentation, bass is near and raise earache, vocal intense not enough (little).

 

I think reason is 3D wide effect. After these corrections maybe musical emotion automaticaly appear.

Link to comment
...I have tried to do was reduce noise as much as possible ...

 

...in 64 bit mode the default data size is 32 bits, so moving 64 into a 64 bit register is less optimum than moving it into a 32 bit one, ...move of 64 into a 32 bit register and then a register to register move for each loop which is faster than moving 64 into a register for each loop)...

 

I think in analog sound, noise is natural. we can not deny it and maybe not required reduce noise to ~zero (too clearly), because the singer's breath will disappear. Breath is emotion, intense, passion,... We need to hear the breath to feel emotion.

Link to comment
I think in analog sound, noise is natural. we can not deny it and maybe not required reduce noise to ~zero (too clearly), because the singer's breath will disappear.

 

But shouldn't the noise from the singer's breath come from the original recording and not be artificially generated in our gear?

Link to comment
I think in analog sound, noise is natural. we can not deny it and maybe not required reduce noise to ~zero (too clearly), because the singer's breath will disappear. Breath is emotion, intense, passion,... We need to hear the breath to feel emotion.
have uploaded 2 new versions of 2.35, I like the mqnplay.exe 2.35 ntdqa ntdq xmm x 2 no wc version, hopefully you will hear breath.

There is no harm in doubt and skepticism, for it is through these that new discoveries are made. Richard P Feynman

 

http://mqnplayer.blogspot.co.uk/

Link to comment

i think i have a tweak to improve SQ A LOT for MQn. i need you guys to experiment this. i will explain it if it actually works for others not just me.

 

if your ram's timing is 9-9-9-X-2, go to bios set it to 9-9-9-17-2.

for 8-8-8-X-2, set to 8-8-8-15-2

for 7-7-7-X-2, set to 7-7-7-13-2

for 6-6-6-X-2, set to 6-6-6-11-2

for 5-5-5-X-2, set to 5-5-5-9-2

etc

Link to comment
tested 3 last versions:

(1) mqnplay.exe 2.34 nopgomemcpy,

(2) mqnplay.exe 2.34 sse4 intel buf buf no xmm,

(3) mqnplay.exe 2.35 sse4 intel

 

drum need to more power and pressure (but don't turn up volume).

in (1) bass OK;

in (2) and (3) bass need so far (little) at listeners, i feel 3D wide effect is turning up and not natural in stereo 2 channel presentation, bass is near and raise earache, vocal intense not enough (little).

 

I think reason is 3D wide effect. After these corrections maybe musical emotion automaticaly appear.

the stereo effect and bass effect are side effects of what the cpu is doing, I can't really correct for them, just try different instructions to see if they sound better. I like the mqnplay.exe 2.35 ntdqa ntdq xmm x 2 no wc version for it's detail.

There is no harm in doubt and skepticism, for it is through these that new discoveries are made. Richard P Feynman

 

http://mqnplayer.blogspot.co.uk/

Link to comment
have uploaded 2 new versions of 2.35, I like the mqnplay.exe 2.35 ntdqa ntdq xmm x 2 no wc version, hopefully you will hear breath.

 

nopgomemcpy>2.34. 2.34 has added harshness in some female vocal.

 

2.35, 2.35 no wc, 2.35 wc x2 also have added harshness in the treble. female vocal can be a little harsh, which feels more like intenser/harder than richer. in comparison, nopgomemcpy is soft but dense/rich in female vocal. in my opinion, the best vocal comes from nopgomemcpy, 3.82 and 2.10, depending on the situation.

 

i think in recent versions, there is a tendency of harder sound. i think softer+denser will be more lovable than harder/intenser.

 

does anyone else have similar feeling?

Link to comment
hey lekt, can you try 2.82? i think it is better than nopgomemcpy.

 

I have tested. OK, even better in stabilization. But sound is near at us, earache. nopgomemcpy is better. But I think need to consider this version, jesuscheung.

2.35xxx drum+bass not improve, even wrong volume. OK, we are waiting...

Link to comment
sbgk, I've just started trying out the two 2.35 versions. I've come across a Loreena McKennitt track (WAV) they both won't play. It plays ok via WMA. Does this just happen sometimes or is their an exception not being handled?

 

not as far as I know. does the track play on other versions ? what bit rate/frequency is the track ?

There is no harm in doubt and skepticism, for it is through these that new discoveries are made. Richard P Feynman

 

http://mqnplayer.blogspot.co.uk/

Link to comment
not as far as I know. does the track play on other versions ? what bit rate/frequency is the track ?

It's straight 16/44.1....I have found what the problem is. The track name has an á in it. I remove the letter with the accent and it plays. It seems like MQn can only handle normal letters or is this a laptop setup issue I have?

Link to comment
It's straight 16/44.1....I have found what the problem is. The track name has an á in it. I remove the letter with the accent and it plays. It seems like MQn can only handle normal letters or is this a laptop setup issue I have?

 

it's a feature of mqn, shall fix it one day

There is no harm in doubt and skepticism, for it is through these that new discoveries are made. Richard P Feynman

 

http://mqnplayer.blogspot.co.uk/

Link to comment
have uploaded 2 new versions of 2.35, I like the mqnplay.exe 2.35 ntdqa ntdq xmm x 2 no wc version, hopefully you will hear breath.

 

So not really getting anywhere with 2 loops and a buffer, lets try it without a buffer and just one loop. Have uploaded 4 files, 1 x 2.36 and 3 x 2.37, they demonstrate the different ways of copying the data, dqa dqa uses movdqa instruction and is the slowest/most noisy; dqa ntdq uses movdqa and movntdq - movntdq uses streaming write to bypass the cache and is the second fastest/noisy; wc and no wc versions use movntdqa and movntdq for streaming loads and writes, wc means the source file uses write combining and no wc means it doesn't have this, movntdqa requires it to have wc. I like the mqnplay.exe 2.37 wc no xmm sub -128 version - have tested it for bass with Eminem and Motorhead and have no complaints there, tested it for piano with Oscar Peterson and sounds ok, BB King and John Lee Hooker also sounded good, maybe verging on too soft a sound. The reason for pairs of instructions is that it loads from source to a register and from register to device, every bit you hear has to pass through the cpu registers, the optimum method is to prefetch the data so it is in a l1 cache line to prevent cache misses transfer it using 128 bit registers and then write it out. movntdqa uses streaming loads which loads a complete cache line (64 bytes) when it sees the first 16 bytes, so in theory it doesn't have to go back to the source for the next 48 bytes, it also bypasses the l2 and l3 cache. movntdq uses non temporal writes ie for data that is only read once and bypasses the l2 and l3 cache. the dqa versions can be played by cpus with sse2, the wc versions require sse4.1

There is no harm in doubt and skepticism, for it is through these that new discoveries are made. Richard P Feynman

 

http://mqnplayer.blogspot.co.uk/

Link to comment
So not really getting anywhere with 2 loops and a buffer, lets try it without a buffer and just one loop. Have uploaded 4 files, 1 x 2.36 and 3 x 2.37, they demonstrate the different ways of copying the data, dqa dqa uses movdqa instruction and is the slowest/most noisy; dqa ntdq uses movdqa and movntdq - movntdq uses streaming write to bypass the cache and is the second fastest/noisy; wc and no wc versions use movntdqa and movntdq for streaming loads and writes, wc means the source file uses write combining and no wc means it doesn't have this, movntdqa requires it to have wc. I like the mqnplay.exe 2.37 wc no xmm sub -128 version - have tested it for bass with Eminem and Motorhead and have no complaints there, tested it for piano with Oscar Peterson and sounds ok, BB King and John Lee Hooker also sounded good, maybe verging on too soft a sound. The reason for pairs of instructions is that it loads from source to a register and from register to device, every bit you hear has to pass through the cpu registers, the optimum method is to prefetch the data so it is in a l1 cache line to prevent cache misses transfer it using 128 bit registers and then write it out. movntdqa uses streaming loads which loads a complete cache line (64 bytes) when it sees the first 16 bytes, so in theory it doesn't have to go back to the source for the next 48 bytes, it also bypasses the l2 and l3 cache. movntdq uses non temporal writes ie for data that is only read once and bypasses the l2 and l3 cache. the dqa versions can be played by cpus with sse2, the wc versions require sse4.1

 

tried 2.37 all three of them. i am not sure about this one...

 

i think that 2.29 is a product of 2.15 to 2.28. it is the state of the art after SSE4 version is introduced. this is a great version. from 2.30, mods/changes here and there, some losses and some gains.

prior to 2.29, i like 2.10 to 2.13 for vocal with emotion.

prior to 2.10, i love 2.82 for piano sounds, i am telling you, this version is god like for piano. it makes my CD player/speakers way overpriced. it is open/airy/crisp/real. the better the tuning of OS/BIOS, the better this 2.82 sounds. bass is good too, extremely good vibration, but like lekt said, bass is near and earache comes.

finally, the nopgomemcpy. it is the most real/intense. bass is best, has great vibration and correct positioning. the vocal has breath.

Link to comment
tried 2.37 all three of them. i am not sure about this one...

 

i think that 2.29 is a product of 2.15 to 2.28. it is the state of the art after SSE4 version is introduced. this is a great version. from 2.30, mods/changes here and there, some losses and some gains.

prior to 2.29, i like 2.10 to 2.13 for vocal with emotion.

prior to 2.10, i love 2.82 for piano sounds, i am telling you, this version is god like for piano. it makes my CD player/speakers way overpriced. it is open/airy/crisp/real. the better the tuning of OS/BIOS, the better this 2.82 sounds. bass is good too, extremely good vibration, but like lekt said, bass is near and earache comes.

finally, the nopgomemcpy. it is the most real/intense. bass is best, has great vibration and correct positioning. the vocal has breath.

 

last versions 2.3x sounds is analog, bass in some versions is OK, some is still near and raised earache. But why drum and drum+bass effect in all versions are not good, insufficient pressure, not enough bouncing (must be as like as bouncing ball when drum+bass plays). MQn.exenopgomemcpy give this, very good, really better than JPlay5.1, JEP2.4.1x64 (but not AVX, JEP2.4.1x64 is good, analog, excluding drum+bass effect).

MQn.exenopgomemcpy is reasonable in all tasks. Ax, difficult for you. Don't give up.

I recommend test drum beat of Phil Collins - ...But Seriously/03.Do You Remember (first seconds). This is rare sound effect for testing.

Link to comment
last versions 2.3x sounds is analog, bass in some versions is OK, some is still near and raised earache. But why drum and drum+bass effect in all versions are not good, insufficient pressure, not enough bouncing (must be as like as bouncing ball when drum+bass plays). MQn.exenopgomemcpy give this, very good, really better than JPlay5.1, JEP2.4.1x64 (but not AVX, JEP2.4.1x64 is good, analog, excluding drum+bass effect).

MQn.exenopgomemcpy is reasonable in all tasks. Ax, difficult for you. Don't give up.

I recommend test drum beat of Phil Collins - ...But Seriously/03.Do You Remember (first seconds). This is rare sound effect for testing.

 

my tuned win7 has much better bass than my tuned win2012. it could also be an issue of win2012.

Link to comment
my tuned win7 has much better bass than my tuned win2012. it could also be an issue of win2012.

 

on my laptop (Win8pro) when Pro Audio ClockRate set to 23220 or 46440 sound like as bundle, feeling vocal has more emotion. But it's false effect, not true. I set to default 10000 and true sound appear.

Link to comment
on my laptop (Win8pro) when Pro Audio ClockRate set to 23220 or 46440 sound like as bundle, feeling vocal has more emotion. But it's false effect, not true. I set to default 10000 and true sound appear.

 

since you like exenopgomemcpy, i assume you are used to the intense/control/rich/dense sound. 10000 will make sound intenser and more control than using a optimal clock rate. maybe why you like it. i mean true sound includes the airy feeling. it is missing in exenopgomemcpy. 23220 should make it more airy/less earache for MQn. and coz exenopgomemcpy sounds control, it sounds better in higher jitter/more digital noise environment, i think you are too used to control/intense/rich sound.

Link to comment

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now



×
×
  • Create New...