double promotion warning in thrust

[correaa]

I changed some compilation parameters of my program and now I see this warning that might be easy to fix.

/usr/local/cuda/bin/../targets/x86_64-linux/include/thrust/detail/complex/catrigf.h: In function 'thrust::complex<float> thrust::detail::complex::casinhf(thrust::complex<float>)':
/usr/local/cuda/bin/../targets/x86_64-linux/include/thrust/detail/complex/catrigf.h:170:36: error: implicit conversion from 'float' to 'double' to match other operand of binary expression [-Werror=double-promotion]
  170 |   const float RECIP_EPSILON = 1.0 / FLT_EPSILON;
      |                              ~~~~~~^~~~~~~~~~~~~         

It happens with CUDA 11.3, 11.5 and 11.6 at least

[zishiwu123]

@correaa I think you might have used the Wdouble-promotion compilation flag: https://stackoverflow.com/questions/24348227/how-to-disable-double-precision-math-while-compiling-with-gcc-or-and-iar

Explicitly declaring the numerator to be a float should disable the warning.

Before:

thrust/thrust/detail/complex/catrigf.h

Line 170 in 39368d7

const float RECIP_EPSILON = 1.0 / FLT_EPSILON;

After

const float RECIP_EPSILON = 1.0f / FLT_EPSILON;

[zishiwu123]

This might also be related to NVIDIA/cccl#779