Hello everybody,

I have just implemented my first project in Julia on a 64 bit system.
I re-implemented a C program and even after optimizing my code using the 
profiler I can only reach roughly half of its speed.
Most of the CPU time is spend in methods from the Base.LinAlg.BLAS module.
Playing around with these functions, I noticed that there is a huge 
difference (roughly factor 2) depending on whether my input to
BLAS is Float32 or Float64 (which is obviously the slower one). 
Particularly I am referring to the axpy! function.
Since I am on a 64 bit system, every float that I directly specify (x = 
1.0) will be a Float64 by default,
and therefore, converting every float variable to Float32 will mess up my 
code considerably.
>From reading the mailing list and the github issues I have the impression 
that I can't simply set a Flag on my 64 bit system to compile
Julia in 32 bit mode.

Is there any way to make floats in Julia default to 32 bit on a 64 bit 
system?
At least in a lot of my work 32 bit precision is just fine and the speed I 
could potentially gain by simply switching is enormous.

Thank you in advance.

Best,

Hubert

Reply via email to