NVCC 12.0 lacking support for floating point template parameters with C++20

[barnack]

I tried reporting through the bug reporting page, but I get an error whenever trying to submit the form. Hopefully someone from NVIDIA will notice here.

Defining a template with a float or double non-type parameter leads to the following error: floating-point template parameter is nonstandard. Using NVCC 12.0 even with the “-std=c++20” flag enabled. Floating-point template parameters are allowed as of C++20

Steps:
Compile the following file with “nvcc.exe -std=c++20”:
template <float v> struct s {};

[Yuki_Ni]

It was be supported soon in next CUDA 12.x minor version soon . Thanks for reporting this.

Edited , this will be resolved since CUDA version 12.1 and so on .

yni@node2:~/yni/tmp$ /usr/local/cuda-12.0/bin/nvcc -std=c++20 -c -x cu test.c 
test.c(1): error: floating-point template parameter is nonstandard

1 error detected in the compilation of "test.c".
yni@node2:~/yni/tmp$ 
yni@node2:~/yni/tmp$ /usr/local/cuda-12.1/bin/nvcc -std=c++20 -c -x cu test.c 
yni@node2:~/yni/tmp$ 
yni@node2:~/yni/tmp$ /usr/local/cuda-12.2/bin/nvcc -std=c++20 -c -x cu test.c 
yni@node2:~/yni/tmp$ 
yni@node2:~/yni/tmp$ cat test.c 
template <float v> struct s {};
yni@node2:~/yni/tmp$ 
yni@node2:~/yni/tmp$