Slicing up the computation algorithm is often not necessary in the mainline code. Here is the reason....if the mainline code execution can be safely interrupted to so that a long interrupt can perform the calculation all in one go then that same calculation can be moved to the mainline code loop and executed as a single lump!!

I commented before that the calculation results rarely need to be on the spot time accurate relative to the time that the SMBus transactions arrive. Why....due to the long time the calcs will take and to a lesser degree to slow speed of SMBus itself.

This results in the fact that the calculations can be done in the the mainline code in a repeating manner and results stored into some global variables that the SMBus controller interrupt can access and process without hardly any clock stretching at all!! So what if the previously computed results are somewhat out of phase with the results request? And so what if the mainline code ends up doing some calcs that never have their results read!!

Please understand that I am a huge champion of not artificially slowing down the SMBus via clock stretching. Slowing down the bus has impacts on other bus devices and may come to hurt the activity within the device on the other end of the transaction.

Michael Karas