Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: FLAC v1.4.x Performance Tests (Read 80587 times) previous topic - next topic
0 Members and 2 Guests are viewing this topic.

Re: FLAC v1.4.x Performance Tests

Reply #175 –
Quote from: bennetng on
Upper: fast-math-noasm-manyflags-haswell-git
Lower: Case GCC 12.2.0
1579150352 bytes PCM size
-8 -b16384 -A "subdivide_tukey(3/2e-1);welch;hann;flattop"
Total encoding time: 0:41.172, 38.18x realtime
769068252 bytes
Total encoding time: 0:36.953, 42.54x realtime
769068236 bytes

Combining cheap windows sometimes can produce good results.

Re: FLAC v1.4.x Performance Tests

Reply #176 –
hann is tukey(1) - the tukey windowing is a rectangle with cosine tapering, and when the "rectangle" hits zero width there is only the cosine left: https://en.wikipedia.org/wiki/List_of_window_functions#Tukey_window
So again you got "two very differently tapered tukeys".
flattop is a weirdo, it is even negative somewhere, but at HA fifteen years ago, it would do well in combination with tukey, so it is an obvious "try this if you want another".

For high resolution - and I got a bit of my testing material from 2L too! - I recall that a gauss window sometimes did "surprisingly" well. That is, surprise compared to it being of very little value for CDDA.
Note, welch = parabola and gauss = exp(parabola).

Re: FLAC v1.4.x Performance Tests

Reply #177 –
Tested a loud 24-96 album and the performance hit is pretty hefty indeed. -8 -p single file.

Case 1.4.1
26.71x realtime

1.4.1 manyflags -Ofast
27.63x realtime

fast-math-noasm-manyflags-haswell-git
21.36x realtime
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: FLAC v1.4.x Performance Tests

Reply #178 –
Oh, 96/24 ... Attached: a few seconds of 96/24, squeezed down to < 2 MB.

Not at all randomly selected: 1.4.1's double precision makes savings like 17 percent at -7, that is YUGE. (But, it is not so that I cherry-picked the best-looking few seconds in the track. Though I did avoid the most dense part.)

* Compared to CDDA, it is easier to beat -p by stacking up with -A [functions]. For the full track, I could beat -p at half the encoding time
* -b [something] can often make a difference on 96/24, but default looks good on this clip. Also the gains from -r are not jaw-dropping either.

At https://hydrogenaud.io/index.php/topic,120158.msg1003288.html#msg1003288 I used the entire EP for testing, but of course I cannot share more than a clip. Buy it :-)

Music: "Temptation", 2017 remake, by Canadian band The Tea Party (who, name-wise, suffered halfway the same fate as ISIS ...)
Known for "Moroccan roll" style hard rock, however this track is largely industrial synth. I bought the EP from https://teaparty.com/tx20 , (2.99 Canadian $ - and 1.99 in MP3 for those of you who happily let others do the lossless testing). You can listen there or on Spotify, https://open.spotify.com/album/6Q3GV4HsGwPzQ9a2TA8cg0

Re: FLAC v1.4.x Performance Tests

Reply #179 –
Quote from: Wombat on
Tested a loud 24-96 album and the performance hit is pretty hefty indeed.
If it is unavoidable then I can still use 1.4.1 for hi-res and 1.4.2 for CDDA.

Re: FLAC v1.4.x Performance Tests

Reply #180 –
Quote from: Porcus on
flattop is a weirdo, it is even negative somewhere, but at HA fifteen years ago, it would do well in combination with tukey, so it is an obvious "try this if you want another".
Flattop tends to work better when the upper spectrum is more empty. An example of DSD to flac conversion with different filtering.
X
-8 -A subdivide_tukey(3/2e-1) with and without flattop

818465962 25kHz flattop.flac
818511667 25kHz.flac (.0055842% bigger)

886963408 Multistage flattop.flac
886996117 Multistage.flac (.0036878% bigger)

====================================

Another example, the "raw DXD" file:
http://www.2l.no/hires/DXD-DSD/index.html
-8 -A subdivide_tukey(3/2e-1) with and without flattop

209196492 JGH flattop.flac
209215084 JGH no flattop.flac (.0088873% bigger)

With optimal -b the effect is even bigger.
-8 -b16384 -A subdivide_tukey(3/2e-1) with and without flattop

207277309 JGH flattop.flac
207300539 JGH no flattop.flac (.0112072% bigger)

Re: FLAC v1.4.x Performance Tests

Reply #181 –
The weakness of git version is 24-bit.
Upper: Case GCC 12.2.0
Lower: fast-math-noasm-manyflags-haswell-git
-8p

16/96
Total encoding time: 0:05.204, 59.04x realtime
Total encoding time: 0:04.922, 62.43x realtime

16/192
Total encoding time: 0:12.406, 24.76x realtime
Total encoding time: 0:11.859, 25.91x realtime

16/352
Total encoding time: 0:26.329, 11.67x realtime
Total encoding time: 0:25.657, 11.97x realtime

24/96
Total encoding time: 0:12.828, 23.95x realtime
Total encoding time: 0:15.469, 19.86x realtime

24/192
Total encoding time: 0:27.688, 11.09x realtime
Total encoding time: 0:34.485, 8.91x realtime

24/352
Total encoding time: 0:53.562, 5.73x realtime
Total encoding time: 1:26.813, 3.53x realtime

Re: FLAC v1.4.x Performance Tests

Reply #182 –
flac git-92928f28 20221017

Clang 16 vs GCC 12.  Both compiled with the same cflags/cxxflags.

Compiled with GCC 12.2.0
Code: [Select]
flac git-92928f28 20221017
Copyright (C) 2000-2009  Josh Coalson, 2011-2022  Xiph.Org Foundation
flac comes with ABSOLUTELY NO WARRANTY.  This is free software, and you are
welcome to redistribute it under certain conditions.  Type `flac' for details.

the_fragile_album.wav: Verify OK, wrote 658916991 bytes, ratio=0.600

/usr/local/bin/flac -V -8 the_fragile_album.wav
Encode Time: 1:34.67



flac git-92928f28 20221017
Copyright (C) 2000-2009  Josh Coalson, 2011-2022  Xiph.Org Foundation
flac comes with ABSOLUTELY NO WARRANTY.  This is free software, and you are
welcome to redistribute it under certain conditions.  Type `flac' for details.

the_fragile_album.wav: Verify OK, wrote 658690092 bytes, ratio=0.600

/usr/local/bin/flac -V -8 -e the_fragile_album.wav
Encode Time: 1:56.36



flac git-92928f28 20221017
Copyright (C) 2000-2009  Josh Coalson, 2011-2022  Xiph.Org Foundation
flac comes with ABSOLUTELY NO WARRANTY.  This is free software, and you are
welcome to redistribute it under certain conditions.  Type `flac' for details.

the_fragile_album.wav: Verify OK, wrote 658391508 bytes, ratio=0.599

/usr/local/bin/flac -V -8 -p the_fragile_album.wav
Encode Time: 1:55.56



flac git-92928f28 20221017
Copyright (C) 2000-2009  Josh Coalson, 2011-2022  Xiph.Org Foundation
flac comes with ABSOLUTELY NO WARRANTY.  This is free software, and you are
welcome to redistribute it under certain conditions.  Type `flac' for details.

the_fragile_album.wav: Verify OK, wrote 658115494 bytes, ratio=0.599

/usr/local/bin/flac -V -8 -e -p the_fragile_album.wav
Encode Time: 7:12.51



flac git-92928f28 20221017
Copyright (C) 2000-2009  Josh Coalson, 2011-2022  Xiph.Org Foundation
flac comes with ABSOLUTELY NO WARRANTY.  This is free software, and you are
welcome to redistribute it under certain conditions.  Type `flac' for details.

the_fragile_album.wav: Verify OK, wrote 659488938 bytes, ratio=0.600

/usr/local/bin/flac -V -8 -b 2304 the_fragile_album.wav
Encode Time: 1:22.54



flac git-92928f28 20221017
Copyright (C) 2000-2009  Josh Coalson, 2011-2022  Xiph.Org Foundation
flac comes with ABSOLUTELY NO WARRANTY.  This is free software, and you are
welcome to redistribute it under certain conditions.  Type `flac' for details.

the_fragile_album.wav: Verify OK, wrote 659082364 bytes, ratio=0.600

/usr/local/bin/flac -V -8 -A tukey(5e-1);partial_tukey(2);punchout_tukey(3) the_fragile_album.wav
Encode Time: 1:25.99

Clang 16.0.0
Code: [Select]
flac git-92928f28 20221017
Copyright (C) 2000-2009  Josh Coalson, 2011-2022  Xiph.Org Foundation
flac comes with ABSOLUTELY NO WARRANTY.  This is free software, and you are
welcome to redistribute it under certain conditions.  Type `flac' for details.

the_fragile_album.wav: Verify OK, wrote 658916991 bytes, ratio=0.600

/usr/local/bin/flac -V -8 the_fragile_album.wav
Encode Time: 1:06.14



flac git-92928f28 20221017
Copyright (C) 2000-2009  Josh Coalson, 2011-2022  Xiph.Org Foundation
flac comes with ABSOLUTELY NO WARRANTY.  This is free software, and you are
welcome to redistribute it under certain conditions.  Type `flac' for details.

the_fragile_album.wav: Verify OK, wrote 658690092 bytes, ratio=0.600

/usr/local/bin/flac -V -8 -e the_fragile_album.wav
Encode Time: 1:56.12



flac git-92928f28 20221017
Copyright (C) 2000-2009  Josh Coalson, 2011-2022  Xiph.Org Foundation
flac comes with ABSOLUTELY NO WARRANTY.  This is free software, and you are
welcome to redistribute it under certain conditions.  Type `flac' for details.

the_fragile_album.wav: Verify OK, wrote 658391508 bytes, ratio=0.599

/usr/local/bin/flac -V -8 -p the_fragile_album.wav
Encode Time: 2:00.89



flac git-92928f28 20221017
Copyright (C) 2000-2009  Josh Coalson, 2011-2022  Xiph.Org Foundation
flac comes with ABSOLUTELY NO WARRANTY.  This is free software, and you are
welcome to redistribute it under certain conditions.  Type `flac' for details.

the_fragile_album.wav: Verify OK, wrote 658115494 bytes, ratio=0.599

/usr/local/bin/flac -V -8 -e -p the_fragile_album.wav
Encode Time: 11:54.96



flac git-92928f28 20221017
Copyright (C) 2000-2009  Josh Coalson, 2011-2022  Xiph.Org Foundation
flac comes with ABSOLUTELY NO WARRANTY.  This is free software, and you are
welcome to redistribute it under certain conditions.  Type `flac' for details.

the_fragile_album.wav: Verify OK, wrote 659488936 bytes, ratio=0.600

/usr/local/bin/flac -V -8 -b 2304 the_fragile_album.wav
Encode Time: 0:58.16



flac git-92928f28 20221017
Copyright (C) 2000-2009  Josh Coalson, 2011-2022  Xiph.Org Foundation
flac comes with ABSOLUTELY NO WARRANTY.  This is free software, and you are
welcome to redistribute it under certain conditions.  Type `flac' for details.

the_fragile_album.wav: Verify OK, wrote 659082364 bytes, ratio=0.600

/usr/local/bin/flac -V -8 -A tukey(5e-1);partial_tukey(2);punchout_tukey(3) the_fragile_album.wav
Encode Time: 0:51.12

Re: FLAC v1.4.x Performance Tests

Reply #184 –
Compiling with --disable-asm-optimizations Clang slows the performance to half for 24-96 files against pure gcc.
Is --disable-asm-optimizations the right way atm.?
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: FLAC v1.4.x Performance Tests

Reply #185 –
Quote from: Wombat on
Ha! Didn't know that.
In frontah a 1.6GB flac decodes to wav in ~10 sec. with the Case and my manyflag fast-math version. I don't see a problem. You may have a better test.
flac 1.4.1 = Case GCC 12.2.0
flac git-0665053c 20221013 = fast-math-noasm-manyflags-haswell-git
8h46m48s single flac file (~3502MB) encoded with -l0 then decoded with -ts

flac 1.4.1
H:\>flac -ts H:\Image.flac
00:00:20,06

flac git-0665053c 20221013
H:\>flac -ts H:\Image.flac
00:00:23,49

Same file encoded with -l6 then decoded with -ts

flac 1.4.1
H:\>flac -ts H:\Image.flac
00:00:20,61

flac git-0665053c 20221013
H:\>flac -ts H:\Image.flac
00:00:21,39

Same file encoded with -l12 then decoded with -ts

flac 1.4.1
H:\>flac -ts H:\Image.flac
00:00:22,18

flac git-0665053c 20221013
H:\>flac -ts H:\Image.flac
00:00:22,95

Same file encoded with --lax -l32 then decoded with -ts

flac 1.4.1
H:\>flac -ts H:\Image.flac
00:00:25,00

flac git-0665053c 20221013
H:\>flac -ts H:\Image.flac
00:00:25,73

The timing was done using this method:
https://stackoverflow.com/a/9938411

I modified the script a bit, drag and drop a flac file into the cmd file to test in the same way I did. Even preset -5 uses -l8 so the differences should be small in most cases. Don't know it is CPU dependent or not.

Re: FLAC v1.4.x Performance Tests

Reply #188 –
Speed differences in decoding lower -l is real.

-l0

PS H:\> measure-command {h:\flac-case -ts image.flac | out-default}
TotalSeconds      : 20.2462909

PS H:\> measure-command {h:\flac-git -ts image.flac | out-default}
TotalSeconds      : 23.5196237

-l8

PS H:\> measure-command {h:\flac-case -ts image.flac | out-default}
TotalSeconds      : 21.362982

PS H:\> measure-command {h:\flac-git -ts image.flac | out-default}
TotalSeconds      : 22.1673663

Re: FLAC v1.4.x Performance Tests

Reply #189 –
So it seems in my test with Clang 16 vs GCC 12, Clang only seems to have the advantage with larger files being written to the storage device.  In my case, I'm using a single SSD in a ZFS pool.  I noticed encoding with the GCC compiled version, times were consistent within 1 second after multiple rounds of the same test, however, with the Clang compiled version, times seemed to vary more between rounds.  I wonder why Clang seems to do better when files are written to a disk...

Each round of test was done in order with these options:
Code: [Select]
flac -d *.flac
flac -V -8 *.wav
flac -V -8 -e *.wav
flac -V -8 -p *.wav
flac -V -8 -e -p *.wav
flac -V -8 -b 2304 *.wav
flac -V -8 -A "tukey(5e-1);partial_tukey(2);punchout_tukey(3)" *.wav


Nine Inch Nails - The Fragile (single file for whole album)
GCC 12.2.0 - R/W to disk
Code: [Select]
Decode Time: 0:37.55 
Encode Time: 1:34.83 
Encode Time: 1:56.33 
Encode Time: 1:53.33 
Encode Time: 7:09.77 
Encode Time: 1:30.12 
Encode Time: 1:24.01

Nine Inch Nails - The Fragile (single file for whole album)
Clang 16.0.0 - R/W to disk
Code: [Select]
Decode Time: 0:36.79 
Encode Time: 1:19.53 
Encode Time: 2:28.14 
Encode Time: 2:26.93 
Encode Time: 11:54.74 
Encode Time: 1:23.08 
Encode Time: 0:59.60

Nine Inch Nails - The Fragile (single file for whole album)
GCC 12.2.0 - R/W to ramdisk
Code: [Select]
Decode Time: 0:05.28 
Encode Time: 0:20.43 
Encode Time: 0:56.15 
Encode Time: 0:53.01 
Encode Time: 6:11.10 
Encode Time: 0:22.19 
Encode Time: 0:24.88

Nine Inch Nails - The Fragile (single file for whole album)
Clang 16.0.0 - R/W to ramdisk
Code: [Select]
Decode Time: 0:05.52 
Encode Time: 0:23.80 
Encode Time: 1:27.26 
Encode Time: 1:29.84 
Encode Time: 10:47.43 
Encode Time: 0:24.70 
Encode Time: 0:28.96

And because someone mentioned The Tea Party! 
This test mixed the album Transmission as individual tracks and Interzone Mantras as a single file.
GCC 12.2.0 R/W to disk
Code: [Select]
Decode Time: 0:42.11 
Encode Time: 1:18.81 
Encode Time: 1:58.11 
Encode Time: 1:52.34 
Encode Time: 7:26.50 
Encode Time: 1:25.81 
Encode Time: 1:27.76

Clang 16.0.0 R/W to disk
Code: [Select]
Decode Time: 0:43.53 
Encode Time: 1:27.22 
Encode Time: 2:30.46 
Encode Time: 2:35.19 
Encode Time: 12:20.65 
Encode Time: 1:15.82 
Encode Time: 1:00.24

GCC 12.2.0 - R/W to ramdisk
Code: [Select]
Decode Time: 0:05.37 
Encode Time: 0:20.91 
Encode Time: 0:57.80 
Encode Time: 0:53.64 
Encode Time: 6:24.09 
Encode Time: 0:22.74 
Encode Time: 0:25.49

Clang 16.0.0 - R/W ramdisk
Code: [Select]
Decode Time: 0:05.67 
Encode Time: 0:23.27 
Encode Time: 1:29.83 
Encode Time: 1:29.42 
Encode Time: 11:07.48 
Encode Time: 0:25.76 
Encode Time: 0:25.12

Re: FLAC v1.4.x Performance Tests

Reply #190 –
flac1013 = Wombat fast-math-noasm-manyflags-haswell-git
flac1021 = john33 flac-1.4.1-git-6abf272-20221021
flac141 = Case GCC 12.2.0

24-bit transcoding (96-352kHz)

PS H:\> measure-command{h:\flac1013 *.flac -fs8 -b16384 -A "subdivide_tukey(3/2e-1);welch;hann;flattop"}|select totalseconds

TotalSeconds
------------
  60.2480702


PS H:\> measure-command{h:\flac1021 *.flac -fs8 -b16384 -A "subdivide_tukey(3/2e-1);welch;hann;flattop"}|select totalseconds

TotalSeconds
------------
  56.8414738


PS H:\> measure-command{h:\flac141 *.flac -fs8 -b16384 -A "subdivide_tukey(3/2e-1);welch;hann;flattop"}|select totalseconds

TotalSeconds
------------
  55.1369771


16-bit transcoding (48kHz)

PS H:\> measure-command{h:\flac1013 *.flac -fs8}|select totalseconds

TotalSeconds
------------
   76.156825


PS H:\> measure-command{h:\flac1021 *.flac -fs8}|select totalseconds

TotalSeconds
------------
  86.8591274


PS H:\> measure-command{h:\flac141 *.flac -fs8}|select totalseconds

TotalSeconds
------------
  83.7913263


Decoding files encoded with -8 (16-bit 48kHz) For unknown reasons there is always a startup delay in the first, non-repeating decoding command despite using RAM disk, so both first and second runs are posted.

PS H:\> measure-command{h:\flac1013 -ts *.flac}|select totalseconds

TotalSeconds
------------
  19.7201548


PS H:\> measure-command{h:\flac1013 -ts *.flac}|select totalseconds

TotalSeconds
------------
  16.9045647


PS H:\> measure-command{h:\flac1021 -ts *.flac}|select totalseconds

TotalSeconds
------------
  20.4929391


PS H:\> measure-command{h:\flac1021 -ts *.flac}|select totalseconds

TotalSeconds
------------
  17.6664767


PS H:\> measure-command{h:\flac141 -ts *.flac}|select totalseconds

TotalSeconds
------------
  19.6514963


PS H:\> measure-command{h:\flac141 -ts *.flac}|select totalseconds

TotalSeconds
------------
  16.8442894


Decoding files encoded with -8 -b16384 (24-bit 96-352kHz)


PS H:\> measure-command{h:\flac1013 -ts *.flac}|select totalseconds

TotalSeconds
------------
   11.280577


PS H:\> measure-command{h:\flac1013 -ts *.flac}|select totalseconds

TotalSeconds
------------
   8.4720569


PS H:\> measure-command{h:\flac1021 -ts *.flac}|select totalseconds

TotalSeconds
------------
  11.5633101


PS H:\> measure-command{h:\flac1021 -ts *.flac}|select totalseconds

TotalSeconds
------------
   8.7760674


PS H:\> measure-command{h:\flac141 -ts *.flac}|select totalseconds

TotalSeconds
------------
  11.2297442


PS H:\> measure-command{h:\flac141 -ts *.flac}|select totalseconds

TotalSeconds
------------
   8.4221105
  
  
Decoding files encoded with -l0, mixed bit-depth and sample rate:

PS H:\> measure-command{h:\flac1013 -ts *.flac}|select totalseconds

TotalSeconds
------------
  14.9447091


PS H:\> measure-command{h:\flac1013 -ts *.flac}|select totalseconds

TotalSeconds
------------
  12.1270033


PS H:\> measure-command{h:\flac1021 -ts *.flac}|select totalseconds

TotalSeconds
------------
   13.809708


PS H:\> measure-command{h:\flac1021 -ts *.flac}|select totalseconds

TotalSeconds
------------
  10.9962077


PS H:\> measure-command{h:\flac141 -ts *.flac}|select totalseconds

TotalSeconds
------------
  13.2682311


PS H:\> measure-command{h:\flac141 -ts *.flac}|select totalseconds

TotalSeconds
------------
   10.469793

Re: FLAC v1.4.x Performance Tests

Reply #191 –
Seems that with the next release building without asm optimizations is good for 16bit only apps llike CUETools (besides HDCD).
Attached a current git versioin of both ways to compile.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: FLAC v1.4.x Performance Tests

Reply #192 –
Quote from: bennetng on
flac1013 = Wombat fast-math-noasm-manyflags-haswell-git
flac1021 = john33 flac-1.4.1-git-6abf272-20221021
flac141 = Case GCC 12.2.0

24-bit transcoding (96-352kHz)

PS H:\> measure-command{h:\flac1013 *.flac -fs8 -b16384 -A "subdivide_tukey(3/2e-1);welch;hann;flattop"}|select totalseconds

TotalSeconds
------------
  60.2480702


PS H:\> measure-command{h:\flac1021 *.flac -fs8 -b16384 -A "subdivide_tukey(3/2e-1);welch;hann;flattop"}|select totalseconds

TotalSeconds
------------
  56.8414738


PS H:\> measure-command{h:\flac141 *.flac -fs8 -b16384 -A "subdivide_tukey(3/2e-1);welch;hann;flattop"}|select totalseconds

TotalSeconds
------------
  55.1369771
PS H:\> measure-command{h:\flac1021wombat *.flac -fs8 -b16384 -A "subdivide_tukey(3/2e-1);welch;hann;flattop"}|select totalseconds

TotalSeconds
------------
  55.8933077


PS H:\> measure-command{h:\flac1021wombat-noasm *.flac -fs8 -b16384 -A "subdivide_tukey(3/2e-1);welch;hann;flattop"}|select totalseconds

TotalSeconds
------------
  60.1707838

Same files in same sample rates, but 16-bit

PS H:\> measure-command{h:\flac1021wombat *.flac -fs8 -b16384 -A "subdivide_tukey(3/2e-1);welch;hann;flattop"}|select totalseconds

TotalSeconds
------------
  37.7704419


PS H:\> measure-command{h:\flac1021wombat-noasm *.flac -fs8 -b16384 -A "subdivide_tukey(3/2e-1);welch;hann;flattop"}|select totalseconds

TotalSeconds
------------
  35.4706544

Re: FLAC v1.4.x Performance Tests

Reply #193 –
Deathblow with -p

24/48, 8 wav files, multi-thread, -8p

flac1021wombat-noasm
Total encoding time: 1:12.704, 176.12x realtime

flac1021wombat
Total encoding time: 0:59.812, 214.08x realtime

Same files but 16/48

flac1021wombat-noasm
Total encoding time: 0:29.735, 430.62x realtime

flac1021wombat
Total encoding time: 0:29.656, 431.77x realtime

Re: FLAC v1.4.x Performance Tests

Reply #194 –
I tested again my AMD Ryzen 5 3600X with different builds to see if there is some encoding speed up.  8)

Code: [Select]
flac -8p

Source
Code: [Select]
Codec      :     PCM (WAV)
Duration   :     57:21:749
Sample rate:     48000 Hz
Channels   :     2
Bits per sample: 16

flac 1.4.1-win64 Xiph
Code: [Select]
wrote 425812106 bytes, ratio=0,644
Global  Time =    56.570

flac-1.4.1-win64-znver3 (Case)
Code: [Select]
wrote 425812103 bytes, ratio=0,644
Global  Time =    54.792

FLAC-1.4.1-git-6abf272-20221021 (john33)
Built on October 21, 2022, GCC 12.2.0
(Code Base : 1.4.1) (-Ofast -m64 -march=haswell)
Code: [Select]
wrote 425812106 bytes, ratio=0,644
Global  Time =    57.293

FLAC-1.4.1-git-6abf272-20221021 (john33)
Built on October 21, 2022, GCC 12.2.0
(Code Base : 1.4.1) (-Ofast -m64 -march=znver2)
Code: [Select]
wrote 425812106 bytes, ratio=0,644
Global  Time =    52.017

Re: FLAC v1.4.x Performance Tests

Reply #195 –
@music_1 : If you test -8, -8r8 and -8e rather than -8p, what happens?
Asking because -p brute-forces part of the process, -e a different one. Of course -8 is faster than either (and -8e is not much useful anymore!), but it is interesting to see whether the order of compiles stays the same. If not, then one makes part of the job more efficient and another a different part of the job.
 

Re: FLAC v1.4.x Performance Tests

Reply #196 –
Quote from: bennetng on
Deathblow with -p

24/48, 8 wav files, multi-thread, -8p

flac1021wombat-noasm
Total encoding time: 1:12.704, 176.12x realtime

flac1021wombat
Total encoding time: 0:59.812, 214.08x realtime

Same files but 16/48

flac1021wombat-noasm
Total encoding time: 0:29.735, 430.62x realtime

flac1021wombat
Total encoding time: 0:29.656, 431.77x realtime
flac1021znver2john33
24/48
Total encoding time: 1:03.859, 200.51x realtime
16/48
Total encoding time: 0:29.985, 427.03x realtime

My i3-12100 must be a remarked Ryzen  :))

Re: FLAC v1.4.x Performance Tests

Reply #197 –
flac -8

flac 1.4.1-win64 Xiph
Code: [Select]
wrote 426124832 bytes, ratio=0,645
Global  Time =    18.246

flac-1.4.1-win64-znver3 (Case)
Code: [Select]
wrote 426124828 bytes, ratio=0,645
Global  Time =    17.863

FLAC-1.4.1-git-6abf272-20221021 (john33)
Built on October 21, 2022, GCC 12.2.0
(Code Base : 1.4.1) (-Ofast -m64 -march=haswell)
Code: [Select]
wrote 425812106 bytes, ratio=0,644
Global  Time =    17.792

FLAC-1.4.1-git-6abf272-20221021 (john33)
Built on October 21, 2022, GCC 12.2.0
(Code Base : 1.4.1) (-Ofast -m64 -march=znver2)
Code: [Select]
wrote 426124836 bytes, ratio=0,645
Global  Time =    17.647

flac -8r8

flac 1.4.1-win64 Xiph
Code: [Select]
wrote 426124602 bytes, ratio=0,645
Global  Time =    20.921

flac-1.4.1-win64-znver3 (Case)
Code: [Select]
wrote 426124598 bytes, ratio=0,645
Global  Time =    20.196

FLAC-1.4.1-git-6abf272-20221021 (john33)
Built on October 21, 2022, GCC 12.2.0
(Code Base : 1.4.1) (-Ofast -m64 -march=haswell)
Code: [Select]
wrote 426124606 bytes, ratio=0,645
Global  Time =    20.341

FLAC-1.4.1-git-6abf272-20221021 (john33)
Built on October 21, 2022, GCC 12.2.0
(Code Base : 1.4.1) (-Ofast -m64 -march=znver2)
Code: [Select]
wrote 426124606 bytes, ratio=0,645
Global  Time =    20.960

flac -8e

flac 1.4.1-win64 Xiph
Code: [Select]
wrote 426050030 bytes, ratio=0,645
Global  Time =    51.351

flac-1.4.1-win64-znver3 (Case)
Code: [Select]
wrote 426050026 bytes, ratio=0,645
Global  Time =    52.222

FLAC-1.4.1-git-6abf272-20221021 (john33)
Built on October 21, 2022, GCC 12.2.0
(Code Base : 1.4.1) (-Ofast -m64 -march=haswell)
Code: [Select]
wrote 426050035 bytes, ratio=0,645
Global  Time =    50.218

FLAC-1.4.1-git-6abf272-20221021 (john33)
Built on October 21, 2022, GCC 12.2.0
(Code Base : 1.4.1) (-Ofast -m64 -march=znver2)
Code: [Select]
wrote 426050035 bytes, ratio=0,645
Global  Time =    51.782

Re: FLAC v1.4.x Performance Tests

Reply #199 –
Quote from: bennetng on
Thanks. Here are results with some AVX-only builds.

Case GCC 12.2.0
Total encoding time: 1:11.218, 30.19x realtime
425513472 bytes

http://www.rarewares.org/files/lossless/flac-1.4.1-x64-znver2-GCC1220.zip
Total encoding time: 1:13.328, 29.32x realtime
425513429 bytes

znver3
Total encoding time: 1:11.891, 29.91x realtime
425513429 bytes

http://www.rarewares.org/files/lossless/flac-1.4.1-x64-AVX2%20-GCC1220.zip
Total encoding time: 1:12.250, 29.76x realtime
425513472 bytes

Case Haswell
Total encoding time: 1:16.328, 28.17x realtime
425513511 bytes

It seems that the Ryzen builds have no compatibility issue with my Intel CPU.
No joke. Looks like a znver3 build would ever be better.