Check optimizations
Premature optimization is the root of all evil
- Is it faster / more accurate to calculate
e^{i M \phi}
by raisinge^{i \phi}
to the powerM
or from\text{sincos}(M \phi)
? - Is it faster to store results of trig functions and square roots in a
__local
lookup table or to calculate them on every iteration? - Does caching frequently-used read-only values in a
__local
array make any difference to speed versus usingprefetch()
? - Is
dot(float4, float4)
faster than a simple sum?
Edited by Chris Kerr