Stockfish 2.2.2 JA update available

[Jim Ablett]

For all the windows users having problems with the Win 64 Gcc compiler builds,
I have made some Intel versions for you to try >

http://dl.dropbox.com/u/5047625/stockfi ... l64-ja.zip

Jim.

[Graham Banks]

For testing purposes, can 2.2.1 and 2.2.2 be combined?
Are the 32-bit compiles changed as well?

For testing purposes the expected improvement of 2.2.2 is far below the error bar of most of the testers, so I'd recommend to do not bother retesting....at least if you are not really getting bored today

Would you be okay with combining 2.2.1 results with 2.2.2 though? No problem if the answer is no, but would like to know.

[mcostalba]

Would you be okay with combining 2.2.1 results with 2.2.2 though? No problem if the answer is no, but would like to know.

Yes it is, you can safely migrate all the results of 2.2.1 to 2.2.2, in your case (32 bits) I am even not sure that the 2.2.2 is faster.

[Graham Banks]

Would you be okay with combining 2.2.1 results with 2.2.2 though? No problem if the answer is no, but would like to know.

Yes it is, you can safely migrate all the results of 2.2.1 to 2.2.2, in your case (32 bits) I am even not sure that the 2.2.2 is faster.

Thanks Marco. Yes - I'll be the only CCRL tester testing the 32-bit version. The others all do 64-bit.

[Jouni]

I got 8-9% speed-up with bench command. But after 2000 testposition 2.2.2 seems to be equal or slower than 2.2.1! So I run my own benchmark: only 1,5-1,8 % real speedup As I earlier noticed, something is really wrong in Stockfish benchmark, but what?

Jouni

[Vinvin]

Sad ... 8-9% speedup would mean about +9 Elo pts.

I got 8-9% speed-up with bench command. But after 2000 testposition 2.2.2 seems to be equal or slower than 2.2.1! So I run my own benchmark: only 1,5-1,8 % real speedup As I earlier noticed, something is really wrong in Stockfish benchmark, but what?

Jouni

[viriathus]

Hi, this is my stockish bench results on Linux:

Code: Select all

Linux 3.1.9-2-ARCH #1 SMP PREEMPT
x86_64 AMD Phenom(tm) II X4 940 Processor AuthenticAMD GNU/Linux

gcc --version
gcc (GCC) 4.6.2 20111223 (prerelease)

Code: Select all

git pull
make profile-build ARCH=x86-64-modern

Code: Select all

Stockfish  120117 64bit SSE4.2
Total time (ms) : 3779
Nodes searched  : 5216701
Nodes/second    : 1380444

Code: Select all

./stockfish-222-64-ja-sse42
Illegal instruction

Stockfish 2.2.2 JA (64bits)
Total time (ms) : 4294
Nodes searched  : 5447426
Nodes/second    : 1268613

Stockfish 2.2.2 JA (32bits)
Total time (ms) : 5700
Nodes searched  : 5447426
Nodes/second    : 955688

[zullil]

Hi, this is my stockish bench results on Linux:

Code: Select all

Linux 3.1.9-2-ARCH #1 SMP PREEMPT
x86_64 AMD Phenom(tm) II X4 940 Processor AuthenticAMD GNU/Linux

gcc --version
gcc (GCC) 4.6.2 20111223 (prerelease)

Code: Select all

git pull
make profile-build ARCH=x86-64-modern

Code: Select all

Stockfish  120117 64bit SSE4.2
Total time (ms) : 3779
Nodes searched  : 5216701
Nodes/second    : 1380444

Several questions:

0) Your "Nodes searched" number seems strange. Has Marco made changes since the 2.2.2 release?

1) What is your CPU's speed in GHz?

2) Do you get a faster executable if you simply do

Code: Select all

make build ARCH=x86-64-modern

(I do. The PGO binary is actually a bit slower than the non-PGO).

3) Do you get a faster executable if you add -fomit-frame-pointer to the optimization flags in the Makefile?

Here's what I see:

Code: Select all

gcc version 4.6.2 (GCC) 

Code: Select all

./stockfish-2.2.2 bench

===========================
Total time (ms) : 4383
Nodes searched  : 5447426
Nodes/second    : 1242853

I'm using a Mac Pro with two 2.26 GHz (E5520) Quad-core Intel Xeon "Gainstown" cpus.

[viriathus]

Several questions:

0) Your "Nodes searched" number seems strange. Has Marco made changes since the 2.2.2 release?

I don't know. Yes, why my compilation dont have 5447426 nodes?. I only do "git pull" from a directory with stockfish git. After this, I exec "make clean" and "make ..."

1) What is your CPU's speed in GHz?

Code: Select all

processor	: 3
vendor_id	: AuthenticAMD
cpu family	: 16
model		: 4
model name	: AMD Phenom(tm) II X4 940 Processor
stepping	: 2
cpu MHz		: 3005.759
cache size	: 512 KB
physical id	: 0
siblings	: 4
core id		: 3
cpu cores	: 4
apicid		: 3
initial apicid	: 3
fpu		: yes
fpu_exception	: yes
cpuid level	: 5
wp		: yes
flags		: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm 3dnowext 3dnow constant_tsc rep_good nopl nonstop_tsc extd_apicid pni monitor cx16 popcnt lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt npt lbrv svm_lock nrip_save
bogomips	: 6015.37
TLB size	: 1024 4K pages
clflush size	: 64
cache_alignment	: 64
address sizes	: 48 bits physical, 48 bits virtual
power management: ts ttp tm stc 100mhzsteps hwpstate

2) Do you get a faster executable if you simply do

Code: Select all

make build ARCH=x86-64-modern

(I do. The PGO binary is actually a bit slower than the non-PGO).

After clean and build with your params:
Total time (ms) : 4046
Nodes searched : 5447426
Nodes/second : 1346373

Hey! nodes searched look fine!

3) Do you get a faster executable if you add -fomit-frame-pointer to the optimization flags in the Makefile?

Here's what I see:

Code: Select all

gcc version 4.6.2 (GCC) 

Code: Select all

./stockfish-2.2.2 bench

===========================
Total time (ms) : 4383
Nodes searched  : 5447426
Nodes/second    : 1242853

I'm using a Mac Pro with two 2.26 GHz (E5520) Quad-core Intel Xeon "Gainstown" cpus.

-fomit-frame-pointer added into makefile and build with:

Code: Select all

make profile-build ARCH=x86-64-modern

Total time (ms) : 4285
Nodes searched : 5447426
Nodes/second : 1271277

After:

Code: Select all

git reset --hard HEAD
git pull
make profile-build ARCH=x86-64-modern
./stockfish bench

Ops!, now look correct:

Total time (ms) : 4272
Nodes searched : 5447426
Nodes/second : 1275146

Maybe I must run always git reset and pull ...

[viriathus]

Ok, after:

Code: Select all

rm -rf Stockfish/
git clone https://github.com/mcostalba/Stockfish.git
cd Stockfish/
vim Makefile (added -fomit-frame-pointer on CXXFLAGS)
make build ARCH=x86-64-modern
./stockfish bench

I don't understand nothing:

into gnome-terminal:

Code: Select all

Total time (ms) : 4052
Nodes searched  : 5447426
Nodes/second    : 1344379

the same binary, from TTY2:

Code: Select all

Total time (ms) : 4220
Nodes searched  : 5447426
Nodes/second    : 1290859

I go to bed