On 03/26/2015 04:25 PM, Peter Zijlstra wrote:
On Fri, Feb 27, 2015 at 04:54:13PM +0100, Vincent Guittot wrote:
Add the SD_PREFER_SIBLING flag for SMT level in order to ensure that the scheduler will put at least 1 task per core.
Signed-off-by: Vincent Guittot vincent.guittot@linaro.org Reviewed-by: Preeti U. Murthy preeti@linux.vnet.ibm.com
Preeti, what benchmarks did you use on power8 smt to verify performance?
I'm seeing a slight but statistically significant regression on my ivb-ep kernel build when I match the build concurrency to my core count.
/me goes run (and install, its a fairly new box) more benches.
I use ebizzy benchmark to test performance regressions. But for this particular patch, I recollect that I did a code walk through to verify the correctness of the patch and not a performance test. Let me run ebizzy against this patch and verify.
Regards Preeti U Murthy