rdtsc test case accuracy on x86_64.

[sowani]

rdtsc test case accuracy on x86_64.

[ongardie]

Hey @sowani, sorry for the long delay. Why'd you close this? It seems like a valid complaint to me.

How about changing:

EXPECT_LT(a + 1000 * 1000, b);

to:

EXPECT_LT(a + 1000 * 900, b);

Would that be good enough for most reasonable CPUs?

[sowani]

@ongardie Thanks for responding and re-opening this issue. I closed it as I didn't get any response and did not hope to get any, so your response was indeed welcome!

I tried the changes you suggested and obtained a few results. The test fails even with "900". Typical values are as follows:
Run 1:
build/Core/TimeTest.cc:263: Failure
Expected: (a * 1000 * 900) < (b), actual: 876232819846779456 vs 7871584397350686
Run 2:
build/Core/TimeTest.cc:263: Failure
Expected: (a * 1000 * 900) < (b), actual: 896691502889979456 vs 7871607129220134
Run 3:
build/Core/TimeTest.cc:263: Failure
Expected: (a * 1000 * 900) < (b), actual: 977104264638579456 vs 7871696476734194

Thanks,
Atul.

[sowani]

With some trial-and-error, I found that the multiplication factor of 500 works okay on ppc64le (a + 1000 * 500).

[ongardie]

I'm confused. What CPU produced the runs you showed above? I don't even know how to read that either, is it a vs b or a * 100 * 900 vs b? I still don't get why you'd multiply a by anything. In any case, those look off by much more than 1000 * 500.

[sowani]

It is "a *+*" - there was a typo in my original posting. It's not "a * 1000 * 500" but "a + 1000 * 500". Sorry for the typo.

…

On Fri, Feb 10, 2017 at 12:26 AM, Diego Ongaro ***@***.***> wrote: I'm confused. What CPU produced the runs you showed above <#225 (comment)>? I don't even know how to read that either, is it a vs b or a * 100 * 900 vs b? I still don't get why you'd multiply a by anything. In any case, those look off by much more than 1000 * 500. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#225 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AK9wb6YvRbUjIYhASpu5fkCxXWF1VCiSks5ra2F1gaJpZM4LkOof> .

[ongardie]

Closed by #229.

[sowani]

Thank you very much @ongardie !