anyone playing with AVX instructions yet?

[cbuchner1]

Hi,

has anyone tried doing some fractals with the new Intel AVX instruction set yet?

Scalar programming is so 20th century...

[cbuchner1]

oh, by the way what's really irking me about the SSE2 and AVX instruction sets is that transcendentals such as exp, log, sin, cos aren't available in hardware. You always have to implement them explicitly - much unlike CUDA or OpenCL where the GPU hardware will give you the result in a single instruction.

I remember seeing some Mandelbulb code that avoided transcendentals as much as possible (losing the feature of allowing variable exponents though), so this method is probably the way to go for a fast vectorized implementation.

[DarkBeam]

oh, by the way what's really irking me about the SSE2 and AVX instruction sets is that transcendentals such as exp, log, sin, cos aren't available in hardware. You always have to implement them explicitly - much unlike CUDA or OpenCL where the GPU hardware will give you the result in a single instruction.

I remember seeing some Mandelbulb code that avoided transcendentals as much as possible (losing the feature of allowing variable exponents though), so this method is probably the way to go for a fast vectorized implementation.

You can always use the good old floating point routines. I am very familiar about those