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 78287 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Re: FLAC v1.4.x Performance Tests

Reply #150 –
Multi-thread tests only, with slower settings to stress test temperature and power limit.

-8p -A subdivide_tukey(9)

ktf-fast-math-noasm-manyflags-haswell
Total encoding time: 4:32.344, 87.91x realtime
2509390434 bytes

Case GCC 12.2.0
Total encoding time: 8:23.812, 47.52x realtime
2507621979 bytes

-8p -A subdivide_tukey(7)

Case GCC 12.2.0
Total encoding time: 5:12.453, 76.62x realtime
2507754046 bytes

Case GCC 7.3.0
Total encoding time: 5:22.157, 74.31x realtime
2507754044 bytes

So with the latest tweak, (9) is bigger than (7).

My previous tests showed that Case GCC 7.3.0 can make the CPU very hot, and exceed power limit with -8pe, but without -e the max temperature is similar to other builds in this test (83C max), also within power limit.

Still, these temperatures are far below TjMAX (100C) which will trigger thermal throttling.

Tests were done using Intel stock cooler at 27C ambient temperature.

Re: FLAC v1.4.x Performance Tests

Reply #151 –
Quote from: bennetng on
Another corpus with a more typical compression ratio. Faster overall speed than the previous corpus with an extreme ratio.

PCM (17 files)
4223331916 bytes

-8p

ktf-fast-math-noasm-manyflags-haswell
multi: 0:53.156, 450.40x realtime
single: 3:44.875, 106.46x realtime
2509518160 bytes
59.420%

Case GCC 12.2.0
multi: 1:01.547, 389.00x realtime
single: 4:17.797, 92.87x realtime
2508762363 bytes
59.402%
FLAC-git-90d7fdb3_20221012_Win_GCC122
multi: 1:00.453, 396.03x realtime
single: 4:19.515, 92.25x realtime
2509518286 bytes

Quote from: bennetng on
Multi-thread tests only, with slower settings to stress test temperature and power limit.

-8p -A subdivide_tukey(9)

ktf-fast-math-noasm-manyflags-haswell
Total encoding time: 4:32.344, 87.91x realtime
2509390434 bytes

Case GCC 12.2.0
Total encoding time: 8:23.812, 47.52x realtime
2507621979 bytes
FLAC-git-90d7fdb3_20221012_Win_GCC122
Total encoding time: 5:25.297, 73.59x realtime
2509390559 bytes

So, the older version at -8p produced smaller files than the newer version at -8p -A subdivide_tukey(9)

Re: FLAC v1.4.x Performance Tests

Reply #152 –
Quote from: bennetng on
-8p -A subdivide_tukey(9)

ktf-fast-math-noasm-manyflags-haswell
Total encoding time: 4:32.344, 87.91x realtime
2509390434 bytes

Case GCC 12.2.0
Total encoding time: 8:23.812, 47.52x realtime

Look at the times. Could it be that the guesstimations now fails to distinguish between functions to apply? Cf my uneducated outburst at 149 ...
@ktf : would this even be possible?

Re: FLAC v1.4.x Performance Tests

Reply #154 –
Quote from: Porcus on
Look at the times. Could it be that the guesstimations now fails to distinguish between functions to apply? Cf my uneducated outburst at 149 ...
@ktf : would this even be possible?
I'm not sure why, but the last commit is faulty. I've been doing some testing seeing the results here, and reverting that commit results in compression being back at 1.4.1 levels.
Music: sounds arranged such that they construct feelings.

Re: FLAC v1.4.x Performance Tests

Reply #155 –
Another corpus with completely different files.
Upper: ktf-fast-math-noasm-manyflags-haswell
Lower: Case GCC 12.2.0
Single thread, 31 files, 4278658628 bytes PCM size

-5
Total encoding time: 0:44.234, 548.34x realtime
2314954727 bytes
Total encoding time: 0:51.359, 472.27x realtime
2314955036 bytes

-6
Total encoding time: 0:54.000, 449.17x realtime
2310798337 bytes
Total encoding time: 1:01.671, 393.30x realtime
2310700059 bytes

-7
Total encoding time: 1:00.390, 401.64x realtime
2304079795 bytes
Total encoding time: 1:10.094, 346.04x realtime
2304028372 bytes

-8
Total encoding time: 1:19.453, 305.28x realtime
2303392314 bytes
Total encoding time: 1:37.187, 249.57x realtime
2302612016 bytes

-8 -A "tukey(7e-1);subdivide_tukey(3/2e-1)"
Total encoding time: 1:23.250, 291.35x realtime
2303079539 bytes
Total encoding time: 1:47.063, 226.55x realtime
2302352189 bytes

-8e
Total encoding time: 4:18.547, 93.81x realtime
2302133626 bytes
Total encoding time: 4:24.266, 91.78x realtime
2302133736 bytes

-8p
Total encoding time: 3:40.109, 110.19x realtime
2301560531 bytes
Total encoding time: 4:22.265, 92.48x realtime
2300997902 bytes

So -5 and -8e are the ones with marginally smaller sizes.

Not completely on topic but something interesting:
https://twitter.com/foone/status/1126996260026605568?lang=en
https://github.com/flyinghead/flycast/issues/644
Some 15 years ago I reported a similar issue on another Dreamcast emulator (Makaron) and the author talked about the same thing: precision differences between the Hitachi SH-4 CPU and generic x86 CPUs.

Re: FLAC v1.4.x Performance Tests

Reply #156 –
Quote from: Brazil2 on
This build is quite slow at decoding to WAV.
No difference here to older versions or even Case 12.2.0

Total:
  Decoded length: 21:29:32.933
  Opening time: 0:00.001
  Decoding time: 0:48.545
  Speed (x realtime): 1593.811
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: FLAC v1.4.x Performance Tests

Reply #157 –
Quote from: Wombat on
Total:
  Decoded length: 21:29:32.933
  Opening time: 0:00.001
  Decoding time: 0:48.545
  Speed (x realtime): 1593.811
Looks like a foo_benchmark report but foobar does not use flac.exe to decode files.

Re: FLAC v1.4.x Performance Tests

Reply #158 –
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.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: FLAC v1.4.x Performance Tests

Reply #159 –
Okay, I've done some testing with the last commit removed, which is now current flac git. See the image below

X

What you see here is three different compiles of FLAC 1.4.1 and one of current git, all with GCC 12.2. Each line has, from left to right, presets -8, -7, -6, -5 and -4.

The light blue one is FLAC 1.4.1 as is, quite similar to what can be downloaded from xiph.org. The dark blue line adds -march=x86-64-v3. That is a 'vendor-neutral' shorthand for including all SSE, AVX, AVX2 and FMA3 instruction set extensions. Most recent CPUs (less than about 6 years old) have those. You can see that is a bit faster, but not much. If you add --disable-asm-optimizations, you get the red line, which is much slower. This is because GCC isn't able to properly optimize.

Now, with the recent changes, you get the green line when using those last options: disabling of specially crafted SSE/AVX/FMA routines so the compiler can try to do better, combined with saying it can use all SSE, AVX, AVX2 and FMA.

For -8 the difference is rather small but for other presets it is quite interesting.
Music: sounds arranged such that they construct feelings.

Re: FLAC v1.4.x Performance Tests

Reply #160 –
The graph cannot show size differences, I mean even the dots obscure that; but: are the sizes so close to equal that we can be pretty sure the builds do, per file, (as good as) the same thing?
That none of the compiles would make round-offs in the model estimation that leads to them selecting a less elaborate compression?

Re: FLAC v1.4.x Performance Tests

Reply #161 –
There will always round-off differences, but these are no longer significant. With the builds posted here recently, a difference was clearly visible in my graphs. Don't have them around anymore though.
Music: sounds arranged such that they construct feelings.

Re: FLAC v1.4.x Performance Tests

Reply #162 –
Quote from: ktf on
For -8 the difference is rather small

In part because the first axis is speed (not time). It seems that if green is X seconds faster than blue at -7, then it would be about X seconds faster at -8 as well. And not far from X for -5 either. (Quick and dirty "calculations" from quick and dirty graph reading - you got the actual times?)

Anyway: does this give any information on what part of the job the fast compile actually does fast?
(Does it matter? Maybe not, but one is curious eh?)

Quote from: ktf on
There will always round-off differences, but these are no longer significant.
Looks that way! But visuals sometimes make for optical illusions.

Re: FLAC v1.4.x Performance Tests

Reply #163 –
Quote from: Porcus on
In part because the first axis is speed (not time). It seems that if green is X seconds faster than blue at -7, then it would be about X seconds faster at -8 as well. And not far from X for -5 either. (Quick and dirty "calculations" from quick and dirty graph reading - you got the actual times?)

Anyway: does this give any information on what part of the job the fast compile actually does fast?
(Does it matter? Maybe not, but one is curious eh?)
The pixel resolution is good up to .002% size difference and I am seeing something like 190/195x vs 210x in -8. Relative speed differences may vary in different CPUs so I would rather do my own tests later on.

In general, if the compiler is allowed to change anything which is logically correct with unlimited precision, then additional steps should be made to limit the intermediate values within a smaller range to avoid significant loss of accuracy.

Re: FLAC v1.4.x Performance Tests

Reply #164 –
Size is back to 1.4.1 with current git and speed is inbetween.
-8 -p is now at ~123x and ~1199x realtime.
reference libFLAC git-0665053c 20221013 attached.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: FLAC v1.4.x Performance Tests

Reply #165 –
fast-math-noasm-manyflags-haswell-git
Same test condition as Reply #155

-5
Total encoding time: 0:41.406, 585.79x realtime
2314954944 bytes

-6
Total encoding time: 0:51.437, 471.55x realtime
2310700076 bytes

-7
Total encoding time: 1:00.016, 404.14x realtime
2304028346 bytes

-8
Total encoding time: 1:23.453, 290.64x realtime
2302612081 bytes

-8 -A "tukey(7e-1);subdivide_tukey(3/2e-1)"
Total encoding time: 1:33.125, 260.46x realtime
2302352273 bytes

-8e
Total encoding time: 4:06.922, 98.23x realtime
2302133843 bytes

-8p
Total encoding time: 3:58.109, 101.86x realtime
2300997970 bytes

So around 10-20% faster than Case GCC 12.2.0 with almost identical file sizes.

Re: FLAC v1.4.x Performance Tests

Reply #166 –
Time for 1.4.2?  ;)
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: FLAC v1.4.x Performance Tests

Reply #167 –
@ktf: Pardon my ignorance, just to make sure I understood what's going on right now:
There was a coding error that crept in which you found and corrected as reported in #154.
This glitch resulted in higher encoding speed @ a little worse compression. After correction, the encoded file size is back to what is to be expected and we lost some speed.
-> But the FLACs produced by the faulty binary are still fine? (at least they have the same MD5 hash as the ones from the reference encoder)

Re: FLAC v1.4.x Performance Tests

Reply #169 –
@ktf  - whenever I specify CFLAGS or CXXFLAGS environmental variables, I see some of the default optimizations disappear in the makefile such as -O3 and -funroll-loops.  I also see they are prepended, so any conflicting flags will be overridden by the defaults


So after messing with compile flags, these seem to get me the fastest encode times.  There may be others I haven't tried that might improve things further.  Using gcc-11 seemed to have a slight edge over gcc-12.  I also have to manually edit the Makefile to remove the -fstack-protector-strong flags.
Code: [Select]
export CC="/usr/bin/gcc-11"  
export CXX="/usr/bin/g++-11"
export CFLAGS="-march=native -O3 -funroll-loops -pipe -flto -fomit-frame-pointer -fno-stack-protector"
export CXXFLAGS="-march=native -O3 -funroll-loops -pipe -flto -fomit-frame-pointer -fno-stack-protector"
export LDFLAGS="-Wl,-s"
./configure --disable-asm-optimizations --disable-altivec

flac v1.3.4 - default Debian build.
Code: [Select]
time flac -V -8
File-01.wav: Verify OK, wrote 24439771 bytes, ratio=0.727
File-02.wav: Verify OK, wrote 24699891 bytes, ratio=0.562
File-03.wav: Verify OK, wrote 39523493 bytes, ratio=0.701
File-04.wav: Verify OK, wrote 40045704 bytes, ratio=0.818
File-05.wav: Verify OK, wrote 15922434 bytes, ratio=0.366

real	0m4.398s
user	0m4.144s
sys	0m0.244s

 
time flac -V -8 -e *.wav
File-01.wav: Verify OK, wrote 24435335 bytes, ratio=0.727
File-02.wav: Verify OK, wrote 24662789 bytes, ratio=0.562
File-03.wav: Verify OK, wrote 39491754 bytes, ratio=0.700
File-04.wav: Verify OK, wrote 40039188 bytes, ratio=0.818
File-05.wav: Verify OK, wrote 15782279 bytes, ratio=0.363

real	0m10.772s
user	0m10.523s
sys	0m0.240s


time flac -V -8 -p *.wav
File-01.wav: Verify OK, wrote 24419397 bytes, ratio=0.727
File-02.wav: Verify OK, wrote 24686505 bytes, ratio=0.562
File-03.wav: Verify OK, wrote 39490413 bytes, ratio=0.700
File-04.wav: Verify OK, wrote 40014362 bytes, ratio=0.817
File-05.wav: Verify OK, wrote 15900707 bytes, ratio=0.366

real	0m9.328s
user	0m9.006s
sys	0m0.316s


time flac -V -8 -e -p *.wav
File-01.wav: Verify OK, wrote 24415349 bytes, ratio=0.727
File-02.wav: Verify OK, wrote 24649517 bytes, ratio=0.561
File-03.wav: Verify OK, wrote 39456681 bytes, ratio=0.699
File-04.wav: Verify OK, wrote 40008204 bytes, ratio=0.817
File-05.wav: Verify OK, wrote 15754777 bytes, ratio=0.362

real	1m0.028s
user	0m59.751s
sys	0m0.276s


time flac -V -8 -e -p -b 2304 *.wav
File-01.wav: Verify OK, wrote 24435705 bytes, ratio=0.727
File-02.wav: Verify OK, wrote 24730440 bytes, ratio=0.563
File-03.wav: Verify OK, wrote 39359937 bytes, ratio=0.698
File-04.wav: Verify OK, wrote 40034300 bytes, ratio=0.818
File-05.wav: Verify OK, wrote 15796351 bytes, ratio=0.363

real	1m15.970s
user	1m15.594s
sys	0m0.376s

 
time -V -8 -b 2304 *.wav
File-01.wav: Verify OK, wrote 24465811 bytes, ratio=0.728
File-02.wav: Verify OK, wrote 24780187 bytes, ratio=0.564
File-03.wav: Verify OK, wrote 39445204 bytes, ratio=0.699
File-04.wav: Verify OK, wrote 40077690 bytes, ratio=0.819
File-05.wav: Verify OK, wrote 15970568 bytes, ratio=0.367

real	0m4.687s
user	0m4.353s
sys	0m0.320s

 
time flac -V -8 -A "tukey(5e-1);partial_tukey(2);punchout_tukey(3)" *.wav
File-01.wav: Verify OK, wrote 24439771 bytes, ratio=0.727
File-02.wav: Verify OK, wrote 24699891 bytes, ratio=0.562
File-03.wav: Verify OK, wrote 39523493 bytes, ratio=0.701
File-04.wav: Verify OK, wrote 40045704 bytes, ratio=0.818
File-05.wav: Verify OK, wrote 15922434 bytes, ratio=0.366

real	0m4.421s
user	0m4.119s
sys	0m0.292s


flac git-0665053c 20221013
Code: [Select]
time flac -V -8 *.wav
File-01.wav: Verify OK, wrote 24434850 bytes, ratio=0.727
File-02.wav: Verify OK, wrote 24616849 bytes, ratio=0.561
File-03.wav: Verify OK, wrote 39438963 bytes, ratio=0.699
File-04.wav: Verify OK, wrote 40041267 bytes, ratio=0.818
File-05.wav: Verify OK, wrote 15382115 bytes, ratio=0.354

real	0m4.620s
user	0m4.294s
sys	0m0.292s
 

time flac -V -8 -e *.wav
File-01.wav: Verify OK, wrote 24431769 bytes, ratio=0.727
File-02.wav: Verify OK, wrote 24615103 bytes, ratio=0.561
File-03.wav: Verify OK, wrote 39419532 bytes, ratio=0.699
File-04.wav: Verify OK, wrote 40036367 bytes, ratio=0.818
File-05.wav: Verify OK, wrote 15367594 bytes, ratio=0.354

real	0m12.232s
user	0m11.943s
sys	0m0.276s
 

time flac -V -8 -p *.wav
File-01.wav: Verify OK, wrote 24416121 bytes, ratio=0.727
File-02.wav: Verify OK, wrote 24604533 bytes, ratio=0.560
File-03.wav: Verify OK, wrote 39416318 bytes, ratio=0.699
File-04.wav: Verify OK, wrote 40010908 bytes, ratio=0.817
File-05.wav: Verify OK, wrote 15358013 bytes, ratio=0.353

real	0m11.245s
user	0m10.935s
sys	0m0.293s
 

time flac -V -8 -e -p *.wav
File-01.wav: Verify OK, wrote 24412782 bytes, ratio=0.726
File-02.wav: Verify OK, wrote 24602298 bytes, ratio=0.560
File-03.wav: Verify OK, wrote 39387742 bytes, ratio=0.698
File-04.wav: Verify OK, wrote 40005678 bytes, ratio=0.817
File-05.wav: Verify OK, wrote 15343571 bytes, ratio=0.353

real	1m19.520s
user	1m18.936s
sys	0m0.408s
 

time flac -V -8 -e -p -b 2304 *.wav
File-01.wav: Verify OK, wrote 24433519 bytes, ratio=0.727
File-02.wav: Verify OK, wrote 24697968 bytes, ratio=0.562
File-03.wav: Verify OK, wrote 39305681 bytes, ratio=0.697
File-04.wav: Verify OK, wrote 40031779 bytes, ratio=0.818
File-05.wav: Verify OK, wrote 15459132 bytes, ratio=0.356

real	1m45.081s
user	1m44.448s
sys	0m0.416s
 

time flac -V -8 -b 2304 *.wav
File-01.wav: Verify OK, wrote 24460714 bytes, ratio=0.728
File-02.wav: Verify OK, wrote 24720984 bytes, ratio=0.563
File-03.wav: Verify OK, wrote 39357368 bytes, ratio=0.698
File-04.wav: Verify OK, wrote 40072725 bytes, ratio=0.818
File-05.wav: Verify OK, wrote 15503783 bytes, ratio=0.357

real	0m4.857s
user	0m4.491s
sys	0m0.356s
 

time flac -V -8 -A "tukey(5e-1);partial_tukey(2);punchout_tukey(3)" *.wav
File-01.wav: Verify OK, wrote 24438161 bytes, ratio=0.727
File-02.wav: Verify OK, wrote 24619892 bytes, ratio=0.561
File-03.wav: Verify OK, wrote 39500724 bytes, ratio=0.700
File-04.wav: Verify OK, wrote 40043969 bytes, ratio=0.818
File-05.wav: Verify OK, wrote 15389983 bytes, ratio=0.354

real	0m5.573s
user	0m5.337s
sys	0m0.225s

Compression results.  File names appended with version and options used.
ex..  B = -b 2304, A = -A "tukey(5e-1);partial_tukey(2);punchout_tukey(3)"
Code: [Select]
24412782 File-01.flac-1.4.1-EP
24415349 File-01.flac-1.3.4-EP
24416121 File-01.flac-1.4.1-P
24419397 File-01.flac-1.3.4-P
24431769 File-01.flac-1.4.1-E
24433519 File-01.flac-1.4.1-EPB
24434850 File-01.flac-1.4.1
24435335 File-01.flac-1.3.4-E
24435705 File-01.flac-1.3.4-EPB
24438161 File-01.flac-1.4.1-A
24439771 File-01.flac-1.3.4
24439771 File-01.flac-1.3.4-A
24460714 File-01.flac-1.4.1-B
24465811 File-01.flac-1.3.4-B
33605420 File-01.wav
 
24602298 File-02.flac-1.4.1-EP
24604533 File-02.flac-1.4.1-P
24615103 File-02.flac-1.4.1-E
24616849 File-02.flac-1.4.1
24619892 File-02.flac-1.4.1-A
24649517 File-02.flac-1.3.4-EP
24662789 File-02.flac-1.3.4-E
24686505 File-02.flac-1.3.4-P
24697968 File-02.flac-1.4.1-EPB
24699891 File-02.flac-1.3.4
24699891 File-02.flac-1.3.4-A
24720984 File-02.flac-1.4.1-B
24730440 File-02.flac-1.3.4-EPB
24780187 File-02.flac-1.3.4-B
43911884 File-02.wav
 
39305681 File-03.flac-1.4.1-EPB
39357368 File-03.flac-1.4.1-B
39359937 File-03.flac-1.3.4-EPB
39387742 File-03.flac-1.4.1-EP
39416318 File-03.flac-1.4.1-P
39419532 File-03.flac-1.4.1-E
39438963 File-03.flac-1.4.1
39445204 File-03.flac-1.3.4-B
39456681 File-03.flac-1.3.4-EP
39490413 File-03.flac-1.3.4-P
39491754 File-03.flac-1.3.4-E
39500724 File-03.flac-1.4.1-A
39523493 File-03.flac-1.3.4
39523493 File-03.flac-1.3.4-A
56417468 File-03.wav
 
40005678 File-04.flac-1.4.1-EP
40008204 File-04.flac-1.3.4-EP
40010908 File-04.flac-1.4.1-P
40014362 File-04.flac-1.3.4-P
40031779 File-04.flac-1.4.1-EPB
40034300 File-04.flac-1.3.4-EPB
40036367 File-04.flac-1.4.1-E
40039188 File-04.flac-1.3.4-E
40041267 File-04.flac-1.4.1
40043969 File-04.flac-1.4.1-A
40045704 File-04.flac-1.3.4
40045704 File-04.flac-1.3.4-A
40072725 File-04.flac-1.4.1-B
40077690 File-04.flac-1.3.4-B
48963980 File-04.wav
 
15343571 File-05.flac-1.4.1-EP
15358013 File-05.flac-1.4.1-P
15367594 File-05.flac-1.4.1-E
15382115 File-05.flac-1.4.1
15389983 File-05.flac-1.4.1-A
15459132 File-05.flac-1.4.1-EPB
15503783 File-05.flac-1.4.1-B
15754777 File-05.flac-1.3.4-EP
15782279 File-05.flac-1.3.4-E
15796351 File-05.flac-1.3.4-EPB
15900707 File-05.flac-1.3.4-P
15922434 File-05.flac-1.3.4
15922434 File-05.flac-1.3.4-A
15970568 File-05.flac-1.3.4-B
43467356 File-05.wav

Re: FLAC v1.4.x Performance Tests

Reply #170 –
It would be interesting to check the "fake hi-res improvement" as well. Same corpus as Reply #155. Speed not shown as I don't have enough RAM drive space.

RetroArch 88200Hz, highest quality, 24-bit, no dither, no RG (i.e. with intersample over induced clipping)

Case GCC 12.2.0 (-8)
5606032877 bytes

fast-math-noasm-manyflags-haswell-git (-8)
5606121407 bytes

v1.3.2 Xiph x64 (-8)
7179114110 bytes

v1.3.2 (-8p)
7174033037 bytes

v1.3.2 (-8e)
7026991603 bytes

RetroArch 88200Hz, lower quality, 24-bit, no dither, with RG (i.e. no clipping but with some ultrasonic leakage)

Case GCC 12.2.0 (-8)
6946871439 bytes

fast-math-noasm-manyflags-haswell-git (-8)
6946870597 bytes

v1.3.2 (-8)
7200099834 bytes

v1.3.2 (-8p)
7196121884 bytes

v1.3.2 (-8e)
7138462138 bytes

RetroArch 88200Hz, normal quality, 24-bit, no dither, with RG (i.e. no clipping but with minor ultrasonic leakage)

Case GCC 12.2.0 (-8)
6231623126 bytes

fast-math-noasm-manyflags-haswell-git (-8)
6231620978 bytes

I don't have a lot of "real" hi-res files to test apart from the free ones. Files above 48kHz may use -b beyond 4608 so others can test it too.

Re: FLAC v1.4.x Performance Tests

Reply #171 –
One tiny thing is that git version generated files are some bits larger only because of the longer version number string.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: FLAC v1.4.x Performance Tests

Reply #172 –
In this case the file size difference should be consistant among different settings, but for example the post right above it:

RetroArch 88200Hz, normal quality, 24-bit, no dither, with RG (i.e. no clipping but with minor ultrasonic leakage)

Case GCC 12.2.0 (-8)
6231623126 bytes

fast-math-noasm-manyflags-haswell-git (-8)
6231620978 bytes

The git version is 2148 bytes smaller, while in another test:

RetroArch 88200Hz, highest quality, 24-bit, no dither, no RG (i.e. with intersample over induced clipping)

Case GCC 12.2.0 (-8)
5606032877 bytes

fast-math-noasm-manyflags-haswell-git (-8)
5606121407 bytes

The git version is 88530 bytes bigger. Negligible differences, but not consistent.

Re: FLAC v1.4.x Performance Tests

Reply #173 –
Sorry, was not related to these files especialy. The longer git string is 6 or 7 bits. Didn't check it further.

This is for 18.6GB -8 -p HiBitrate files i used for RG testing:

19.713.863.889 Bytes flac 1.4.1
19.713.858.079 Bytes fast-math-noasm-manyflags-haswell-git
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!
 

Re: FLAC v1.4.x Performance Tests

Reply #174 –
http://www.2l.no/hires/
Code: [Select]
http://www.lindberg.no/hires/test/2L-145/2L-45_stereo_01_FLAC_352k_24b.flac
http://www.lindberg.no/hires/test/2L-139/2L-139_stereo_FLAC_176k_24b_01.flac
http://www.lindberg.no/hires/test/2L-106/2L-106_stereo_PCM-96k_MAGNIFICAT_04.flac
http://www.lindberg.no/hires/test/2L38_01_96kHz.flac



Upper: fast-math-noasm-manyflags-haswell-git
Lower: Case GCC 12.2.0
1579150352 bytes PCM size

-8
Total encoding time: 0:31.921, 49.25x realtime
772176807 bytes
Total encoding time: 0:30.422, 51.67x realtime
772176835 bytes

-8e
Total encoding time: 2:08.391, 12.24x realtime
772158544 bytes
Total encoding time: 1:47.328, 14.64x realtime
772158545 bytes

-8p
Total encoding time: 2:39.969, 9.82x realtime
771495942 bytes
Total encoding time: 2:10.266, 12.06x realtime
771495961 bytes

-8 -b16384
Total encoding time: 0:31.125, 50.51x realtime
769125599 bytes
Total encoding time: 0:29.109, 54.01x realtime
769125592 bytes

-8 -b16384 -A subdivide_tukey(5)
Total encoding time: 1:01.484, 25.57x realtime
769070756 bytes
Total encoding time: 0:52.516, 29.93x realtime
769070764 bytes

Case GCC 12.2.0 is consistently faster with these hi-res files on my i3-12100.