Tweet
Paul, there is no need for 'glrunstomake'.
All that is needed is glloopsinasample and glsamples.
To get serious about code timing if glsamples is greater than 30 then we can use the normal distribution to determine confidence limits. For example we could have a 95% confidence that code executes within a lower and upper bound. This approach is much better when comparing two pieces of code. If the respective confidence limits overlap then the code with the lower average will, on occasion, be worse than the code with the higher average. In other words the case for the lower average code is weak. If there is no overlap and there is a good separation then we have a case for the lower average code.
When I have time I'll knock something out and drop it into the Source Code forum unless someone beats me to it.
Added: Having chosen a value for glsamples then glloopsinasample can be determined according to how long you are prepared to wait for some results.
All that is needed is glloopsinasample and glsamples.
To get serious about code timing if glsamples is greater than 30 then we can use the normal distribution to determine confidence limits. For example we could have a 95% confidence that code executes within a lower and upper bound. This approach is much better when comparing two pieces of code. If the respective confidence limits overlap then the code with the lower average will, on occasion, be worse than the code with the higher average. In other words the case for the lower average code is weak. If there is no overlap and there is a good separation then we have a case for the lower average code.
When I have time I'll knock something out and drop it into the Source Code forum unless someone beats me to it.
Added: Having chosen a value for glsamples then glloopsinasample can be determined according to how long you are prepared to wait for some results.
Comment