Using Intrinsec functions in double precision

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi.

I am trying to implement a simple calculation algorithm on a GPU (GTX 295).

Because I need very good accuracy, I am using double precision.

I wanted to know if nvcc is using GPU native/intrinsec exp and log functions (in double precision) if I compile a code like that (in the kernel):

R[index] = exp(Y[index])+log(U[index]);

Of course, I am using -arch sm_13 and -O4 when compiling, and R,Y,U are double type arrays.

Thanks in advance :)

Lags

2

(1) Error bounds for all the math functions are documented in the appendices of the Programming Guide.
(2) There is no direct hardware support for double-precision math functions, double-precision exp() and log() are pure software implementations.
(3) The source code for most math functions is user visible and can be found in the math_function*.h header files.

3

Thank you for this reply.

This respond to my interrogations. I just hope further instructions will be added in double precision in the near future…

Lags