Please see the Phoronix Test Suite results HERE for a guide showing how I found this bug.
I am in the process of running updated tests that enable both cores on my Core 2 Duo but keep the 800 Mhz speed limit for use in making benchmarks related to the recent Phoronix article on Cortex A9s vs. a single-core Atom at 1Ghz (http://www.phoronix.com/scan.php?pag..._arm_x32&num=1). My first run was using my 4-year old notebook with a single core disabled on the CPU and the other core limited to 800 MHz. Next, I re-ran the benchmarks with both cores activated, but both cores also being limited to 800 Mhz. Of course, I have had some big speedups in multi-threaded applications... with the notable exception of n-queens!
I have made multiple runs of the n-queens benchmarks, and as you can see from the results page, there is basically zero speedup between the "Core_2_Crippled" system with only one core turned on vs. the dual-core setup. Look at the other benchmarks that do use multi-threading and you'll definitely see that multi-threading is working on this machine, just not on the N-queens benchmark.
Please note: I have confirmed that the binary being executed is the 'qn24b_openmp' binary that is supposed to be multi-threaded through OpenMP. I have run the binary manually and through the standard PTS runs, and it does not use my second CPU core whatsoever. I think there is a bug in PTS that should get fixed. I'm running version 4.0.1 BTW.
I am in the process of running updated tests that enable both cores on my Core 2 Duo but keep the 800 Mhz speed limit for use in making benchmarks related to the recent Phoronix article on Cortex A9s vs. a single-core Atom at 1Ghz (http://www.phoronix.com/scan.php?pag..._arm_x32&num=1). My first run was using my 4-year old notebook with a single core disabled on the CPU and the other core limited to 800 MHz. Next, I re-ran the benchmarks with both cores activated, but both cores also being limited to 800 Mhz. Of course, I have had some big speedups in multi-threaded applications... with the notable exception of n-queens!
I have made multiple runs of the n-queens benchmarks, and as you can see from the results page, there is basically zero speedup between the "Core_2_Crippled" system with only one core turned on vs. the dual-core setup. Look at the other benchmarks that do use multi-threading and you'll definitely see that multi-threading is working on this machine, just not on the N-queens benchmark.
Please note: I have confirmed that the binary being executed is the 'qn24b_openmp' binary that is supposed to be multi-threaded through OpenMP. I have run the binary manually and through the standard PTS runs, and it does not use my second CPU core whatsoever. I think there is a bug in PTS that should get fixed. I'm running version 4.0.1 BTW.
Comment