OK, I figured it out. It seems that either (a) standard algebraic
computations such as 2*a will be cast to double in some cases or (b)
this happens in cases like powf(a, 2). I replaced all numbers with a
float number (e.g. 2.0f) and that error is not raised anymore.
On Wed, Jun 29, 2011 at 11:22 AM, Thomas Wiecki <Thomas_Wiecki(a)brown.edu> wrote:
I always got these warnings while compiling, but now it seems they
lead to a compile error:
pycuda.driver.CompileError: nvcc said it demoted types in source code
it compiled--this is likely not what you want.
My function is an elementwise kernel (so c code), all input arguments
and variables are floats. In the function I use logf, expf, sinf,
floorf, fmaxf etc. Would any of these return a double? How can I find