MeasuringStallsWithSEP < Main

MispredictedBranchesTest was used for tests

(random case)

Counter	random case	random %	predictable case	predictable %
RESOURCE_STALLS.ANY	822738037	100%	1049479394	100%
RESOURCE_STALLS.BR_MISS_CLEAR	200541074	24%	178300913	17%
RESOURCE_STALLS.ROB_FULL	362559	0%	358405	0%
RESOURCE_STALLS.LD_ST	1196932	0%	182299050	17%
RESOURCE_STALLS.RS_FULL	620809986	75%	678978611	64%

(totals compare to 'ANY' within 1%)

And yet, there are MORE stalls for the predictable case, that runs faster...

#!/bin/bash
APP="./br r"

sep -start -c -ec "RESOURCE_STALLS.ANY" -out 1_ANY -app taskset -args "-c 1 $APP"
sep -start -c -ec "RESOURCE_STALLS.BR_MISS_CLEAR" -out 2_BR_MISS -app taskset -args "-c 1 $APP"
sep -start -c -ec "RESOURCE_STALLS.ROB_FULL" -out 3_ROB_FULL -app taskset -args "-c 1 $APP"
sep -start -c -ec "RESOURCE_STALLS.LD_ST" -out 4_LD_ST -app taskset -args "-c 1 $APP"
sep -start -c -ec "RESOURCE_STALLS.RS_FULL" -out 5_RS_FULL -app taskset -args "-c 1 $APP"

-- MattWalsh - 24 Aug 2006

Topic revision: r2 - 28 Aug 2006 - MattWalsh