How acos/asin/atan in pytorch handle half input on gpu?

I see that acos for gpu has implementation here.

But when I tried to compile the following code

// Type your code here, or load an example.
#include <cuda.h>
#include <cuda_fp16.h>
#include <cuda_runtime.h>

__global__ void asin_kernel(half *input, half *output, int size) {
    int tid = blockIdx.x * blockDim.x + threadIdx.x;
    if (tid < size)  {
        output[tid] = ::acos(input[tid]);

I got error

<source>(9): error: more than one instance of overloaded function "acos" matches the argument list:
            function "std::acos(long double)" (declared at line 91 of /opt/compiler-explorer/gcc-10.2.0/include/c++/10.2.0/cmath)
            function "std::acos(float)" (declared at line 87 of /opt/compiler-explorer/gcc-10.2.0/include/c++/10.2.0/cmath)
            argument types are: (half)
          output[tid] = ::acos(input[tid]);

1 error detected in the compilation of "<source>".
Compiler returned: 2

how does pytorch get half input with acos work?