I agree! Paradoxically, my BMI2 compile of Cfish_2017-03-25 is running just fine against Cfish_2017-01-04_BMI2, I will report the result later. But the popcnt compiles are trouble, for whatever reason.
pohl4711 wrote:
Gusev wrote:Bad news. I stopped my 20"+0.5" test of the Cfish_x64_240317_modern_620gcc compile at the losing score of +0-41=26 against AsmFishW_2017-03-15_popcnt. I observed that Cfish spent its time much faster early on, then played on the increment and got in trouble. I will compare behaviors of my BMI2 compiles of Ronald's Cfish_2017-01-04 and the new Cfish_240317.
Oops. So, a testrun of Cfish makes no sense at the moment.
I will give Andscacs 0.90 a try, when the actual Stockfish-Testrun is finished. Perhaps it can take the place of Fire 4 as an opponent for my Stockfish-Testruns.
pohl4711 wrote:
I will give Andscacs 0.90 a try, when the actual Stockfish-Testrun is finished. Perhaps it can take the place of Fire 4 as an opponent for my Stockfish-Testruns.
Only the master branch is identical (to Cfish 2017-01-04). And then there are other branches. I am testing the compile built from the branch updated most recently as of 2017-03-26, 12 hours ago at that time, i.e. on 2017-03-25.
MikeB wrote:Correct, right now it is identical to cfish by Ronald.
Gusev wrote:Only the master branch is identical (to Cfish 2017-01-04). And then there are other branches. I am testing the compile built from the branch updated most recently as of 2017-03-26, 12 hours ago at that time, i.e. on 2017-03-25.
MikeB wrote:Correct, right now it is identical to cfish by Ronald.
This new Cfish BMI2 compile is still running ~7 Elo points ahead of the previously tested Cfish 2017-01-04 after 1000+ games on two of my computers (4-core and 8-core), so I made it available at http://www.firenzina.org/cfish_2017-03-25_bmi2.zip. Enjoy!
MikeB wrote:
Gusev wrote:Only the master branch is identical (to Cfish 2017-01-04). And then there are other branches. I am testing the compile built from the branch updated most recently as of 2017-03-26, 12 hours ago at that time, i.e. on 2017-03-25.
MikeB wrote:Correct, right now it is identical to cfish by Ronald.
BTW, would you like to port your latest McBrain novelties to C and into Cfish? The resulting fish would likely pull noticeably away from its lively pack (school).
MikeB wrote:
Gusev wrote:Only the master branch is identical (to Cfish 2017-01-04). And then there are other branches. I am testing the compile built from the branch updated most recently as of 2017-03-26, 12 hours ago at that time, i.e. on 2017-03-25.
MikeB wrote:Correct, right now it is identical to cfish by Ronald.
Update on my two 2046-game tests of cfish_2017-03-25_BMI2 v. cfish_2017-01-04_BMI2:
4CPU: After 872 games, cfish_2017-03-25 has the lead, +91-77=704, +6 Elo
8CPU: After 534 games, cfish_2017-03-25 is ahead, +59-37=438, +14 Elo
Gusev wrote:This new Cfish BMI2 compile is still running ~7 Elo points ahead of the previously tested Cfish 2017-01-04 after 1000+ games on two of my computers (4-core and 8-core), so I made it available at http://www.firenzina.org/cfish_2017-03-25_bmi2.zip. Enjoy!
MikeB wrote:
Gusev wrote:Only the master branch is identical (to Cfish 2017-01-04). And then there are other branches. I am testing the compile built from the branch updated most recently as of 2017-03-26, 12 hours ago at that time, i.e. on 2017-03-25.
MikeB wrote:Correct, right now it is identical to cfish by Ronald.
The 8CPU test of Cfish_2017-03-25_bmi2 is complete, the files are available at http://www.firenzina.org/FOSCEC2017Test05.zip. Cfish_2017-03-25_bmi2 narrowly defeated Cfish_2017-01-04, +249-236=1561, 50.32%, +2 Elo. The 4CPU test continues. After 1416 games out of 2046, Cfish_2017-03-25_bmi2 is slightly ahead, +160-143=1113, +4 Elo.
