This isn’t a related metric.
As a result of graphics pipelines rework objects via matrix multiplication, it all the time takes the similar quantity of labor to remodel an object / its vertices from any place / orientation / Cartesian coordinate system to some other: one matrix multiply. This doesn’t get costlier if the matrix has to invert or trade some axes; it is simply the identical directions working on totally different numbers.
Meaning we do not add or save operations by utilizing any specific coordinate conference.
Equally, when changing a mouse delta to a rotation or different gameplay impact, we all the time multiply by some scaling issue: a “mouse sensitivity”. Multiplying by a damaging issue prices the identical as multiplying by a optimistic one, by way of the CPU directions emitted from the compiler: one float multiply or fused-multiply-add the place relevant. So even the instances the place mouse enter must be negated don’t essentially add an additional operation. (You might write code that forces the compiler to do that inversion as a second step, however you would additionally simply write code that does it in a single). We’ll additionally often add an non-obligatory negation within the settings menu for folk who like utilizing inverted controls anyway, in any coordinate scheme.
You could get micro-optimizations out of your knowledge structure — arranging e.g. rows or columns of your matrix in order that they’re simple to make use of SIMD directions on. However that may be achieved in any coordinate conference.
So: no mixture of conventions like “Y up / down” or “Z up / down” or “Digital camera faces z+ / z-” will demand essentially extra operations than some other. Select your conventions based mostly on what makes probably the most intuitive sense in your asset and modifying workflows, to not attempt to squeeze a micro-optimization out of it.
