Application/algorithm efficiency - MIPS versus DMIPS

[PGoll]

Dear all,

We are trying to have a rough comparison measure of efficiency of a set of algorithms running across different CPU architectures and trying to understand how we can express it in an architecture independent ways. We could measure the MIPS from Intel's vTune or similar tool for the algorithms running on Intel's CPU.  However, we can not use the MIPS to compare the same figure for the same algorithms to that of  ARM core. I am thinking of figuring out a rough way to convert MIPS to DMIPS and use DMIPS as a way to compare the efficiency of the algorithm across different CPU cores. I do understand the host of limitations of such a measure (wait states, Cache etc.).   Now, (1) do you think it is a reasonable method to do so or is there any better way to do this ? (2) how to convert MIPS to approximate DMIPS for different CPU cores ?  Many thanks for your support !

[Vitaly_S_Intel]

Well, in my understanding DMIPS is a result score of Dhrystone benchmark which is supposed to characterize the system (not a particular workload). So, you can run this benchmark on Intel Architecture and on ARM and compare the resulting score. Now, if you try to measure Dhrystone benchmark under VTune or similar tool, you should be able to get MIPS number for this benchmark running on a particular system. So, you can now get a ratio between MIPS and DMIPS for the particular system which then can be used for some kind of DMIPS representation for particular workload converting MIPS to DMIPS.

[PGoll]

Thank you !