From: Peter R. <p.r...@sh...> - 2011-05-17 13:05:31
|
As an obvious follow-up, I have repeated my previously posted tests on Linux (gcc 4.5.2). I get: Code version with double argument: 1) No optimisation = 0.27665 2) -O1 = 0.132152 3) -O2 = 0.270331 4) -O1 & -O2 = 0.132152 Code version with 1D vector argument (but semantically identical): 1) No optimisation = 0.27665 2) -O1 = 0.132152 3) -O2 = 3.16427 4) -O1 & -O2 = 0.132152 So the same inconsistencies appear with Linux although the exact numbers are different. So this suggests an up-stream issue/feature? As an aside, I am curious why two Intel boxes give different results depending on the OS. In what way are the different OSs using nominally the same hardware to produce different results? P. |