How is instructions per cycles calculated

[Shintaro]

Hi,
I was wondering if anyone knew how IPC (Instructions per cycle) was calculated for CPU's. Or is it CPI (Cycles Per Instruction)?
Just looking for a simple example. I can dig in to the "Why" later.

It seems to be a bit of a rabbit warren.

The reason I ask is that on the 86Box GitHub page it say's the following:
"Most emulation logic is executed in a single thread, therefore generally systems with better IPC (instructions per clock) should be able to emulate higher clock speeds."

Cheers,
Craig

 

[x BlueRobot]

From what I remember, usually you say instructions per a cycle, and that is largely dependent on the CPU architecture itself.

Some instructions will take more CPU cycles than others, hence why when optimising a program as much as possible, you'll take into take account how that program is compiled for that target architecture.

The compiler itself will play a large role into the efficiency of a program as well.

 

[Digerati]

I think it really depends on the specific instruction. I note some instructions can take place in a 1/2 cycle too. One instruction on the "flip" (on), another on the "flop" (off).

I am not sure it is even "calculated" but rather "counted". Calculated, in this scenario (to me anyway) leaves room for different variables to be introduced. And that leaves room for marketing weenies to make all sorts of accurate, or totally hyped-up claims.