Page 212 of 299

Re: lekt player

Posted: Thu Oct 02, 2014 3:35 pm
by jesuscheung
lekt wrote:
jesuscheung wrote:...
c2sm and nortcc missing some emotion. maybe mid is weaker. don't feel sweet tune.
they have compiled without /RTCc, file size much reduced coz assembly code doesn't check smaller data type in loop.

difficult choose: /RTCc or not. will learn more about relationship of compiler's options, more than hundred options!!
c2sm all things technically better than noind. seems only issue is cold emotion.

maybe my OS is wrong something. better code usually identifies new issues

-----
c2sm is current best piano. weight/clarity/agility all things more correct.
c2sm MOD19 better skills. original more realism.

Re: lekt player

Posted: Thu Oct 02, 2014 6:00 pm
by lekt
comparing C code and assembly code from compiler's report file changed my mind. now C code of v2.100 xx much better. think need install vs2013, take ML64.exe and file set for compile MASM. learn more and will try MASM, don't know what i can make, hehe...
compiler optimization with /Ox makes assembly code very complex, ignore/reduce ~50% variables, how difficult read code. i see very smart use registers swap, stack pointers in rsp. smart!! too smart, but why sound not good as needed, maybe caused by imposition or misuse of registers. not sure.

c2sm code much better, need compile as noind to improve organic of sound.

Re: lekt player

Posted: Fri Oct 03, 2014 2:58 am
by jesuscheung
lekt wrote: ...compiler optimization with /Ox makes assembly code very complex, ignore/reduce ~50% variables, how difficult read code. i see very smart use registers swap, stack pointers in rsp. smart!! too smart, but why sound not good as needed, maybe caused by imposition or misuse of registers. not sure.
the trouble is that intel CPU tries to prefetch or predict your code. predict = overhead.
is optimized code more easily predict? maybe not.

the most analog piano sound is when i turn off "hardware prefetch" + "adjacent cache line" in bios
sharp, clear, accurate. digital sweetness/harshness/bright/3D will disappear/much reduce.

weight can easily go over or not enough when "prefetch=enabled"

listening to c2sm right now.
when prefetches are disabled, weight is wonderfully stable.
actually, c2sm is suprisingly sweet! wtf...
vocal is worse. micro detail has disappeared

the downside is vinyl noise/earache. vibration/quantity lesser.
the worse is OS, the more is earache/noise.

Re: lekt player

Posted: Fri Oct 03, 2014 3:06 am
by jesuscheung
c2sm is different sound again... yesterday morning. yersterday night. now morning.
3 different jitter responses. haha... has sweet sound today.
good news is that correctness in notes is consistent all the time.

Re: lekt player

Posted: Fri Oct 03, 2014 5:07 am
by lekt
notice that: from v2.7x to now i don't use /Ox
need learn more and use /O2 or /Ot. will test assembler code of these cases.

could you read about /fp:except- , testing now, very good. much removed 3D effect. seems kept echo in enough level.

learning how to remove .pdata, .realloc by compiler settings.

Re: lekt player

Posted: Fri Oct 03, 2014 5:59 am
by jesuscheung
lekt wrote:notice that: from v2.7x to now i don't use /Ox
need learn more and use /O2 or /Ot. will test assembler code of these cases.

could you read about /fp:except- , testing now, very good. much removed 3D effect. seems kept echo in enough level.

learning how to remove .pdata, .realloc by compiler settings.
firstly,
/02 = /Og /Oi /Ot /Oy /Ob2 /Gs /GF /Gy

apparently /Og is "Deprecated" for vs2013
http://msdn.microsoft.com/en-us/library/1yk3ydd7.aspx

so....
seems like we may or may not need all of these:
/Og /Oi /Ot /Oy /Ob2 /Gs /GF /Gy

now, to test all combinations, can take a million years

/GF is interesting... if you make all strings named "!", i think compiler will only use 1 copy of string "!".

---
fp seems easy to test
only 3-4 cases to test.

Re: lekt player

Posted: Fri Oct 03, 2014 9:35 am
by jesuscheung
lekt wrote:notice that: from v2.7x to now i don't use /Ox
...
do you remember using /Ox for 2.4.3??

---
relistening to 2.97.2
that treble is similar like MOD19 on c7/noind

Re: lekt player

Posted: Fri Oct 03, 2014 9:43 am
by jesuscheung
try 2.97.2 MOD21...
only deleted sections and changed stack heap

i don't have to give up any bass! all things better.


-----
finally understand c2sm jitter response.
when my computer is good mood, emotion is back!
c2sm mod19b is best sound here. weight seems a bit too manly? -> true drum, piano lesser a little. not a big issue.

Re: lekt player

Posted: Fri Oct 03, 2014 3:19 pm
by jesuscheung
jesuscheung wrote:uploading (40% right now) "Rolling in the Deep - Adele Acoustic Cover (Jorge and Alexa Narvaez)"
i got it from youtube

father and 5 year old daughter singing.
all kids have the brightest voice.

only MOD19 can make bright voice sounds transparent with air.

99.9% of software makes her voice very hard
...
i meant noiman MOD19 does it best

Re: lekt player

Posted: Sat Oct 04, 2014 5:10 am
by lekt
i was busy, JC. just only today can back to compile code writen after c2sm.

uploaded:
v2.100.1 160 ca1
v2.100.1 160 ca3
v2.100.1 160 casm

v2.100.2 160 vg1
v2.100.2 160 vg3
v2.100.2 160 vguid

good code after refference assembler of C compiler's report.
prefer vg3. these 6 versions have good emotion. try test ca1, i feel very good emotion, may have effect?