external_cl_half_max.py 370 B

12345678910111213
  1. from tinygrad.runtime.ops_gpu import CLDevice, CLProgram, compile_cl
  2. if __name__ == "__main__":
  3. dev = CLDevice()
  4. lib = compile_cl("""
  5. #pragma OPENCL EXTENSION cl_khr_fp16 : enable
  6. __kernel void test(__global half *out, __global half *a, __global half *b) {
  7. int gid = get_global_id(0);
  8. out[gid] = max(a[gid], b[gid]);
  9. }
  10. """)
  11. prg = CLProgram(dev, "test", lib)