clang/test/CodeGenCUDA/Inputs/ocml-sample.cl

   1 #pragma OPENCL EXTENSION cl_khr_fp16 : enable
   2
   3 half do_f16_stuff(half a, half b, half c) {
   4   return __builtin_fmaf16(a, b, c) + 4.0h;
   5 }
   6
   7 float do_f32_stuff(float a, float b, float c) {
   8   return __builtin_fmaf(a, b, c) + 4.0f;
   9 }
  10
  11 double do_f64_stuff(double a, double b, double c) {
  12   return __builtin_fma(a, b, c) + 4.0;
  13 }
  14
  15 __attribute__((weak))
  16 float weak_do_f32_stuff(float a, float b, float c) {
  17   return c * (a / b);
  18 }