# Generated from the displayed GPU Lab example; no GPU timings are measured.
# Run: python example.py --device cpu  (default; no CPU/GPU transfers)
# Or:  python example.py --device cuda (compatible NVIDIA GPU and PyTorch required)
# Explicit CPU-only calculations remain on CPU in either mode.
import argparse
import torch

_lab_parser = argparse.ArgumentParser(description="Run this GPU Lab PyTorch example.")
_lab_parser.add_argument("--device", choices=["cpu", "cuda"], default="cpu")
_lab_device = _lab_parser.parse_args().device
if _lab_device == "cuda" and not torch.cuda.is_available():
    _lab_parser.error("CUDA mode needs an available NVIDIA GPU and compatible PyTorch installation.")
if _lab_device == "cpu":
    print("CPU mode checks values and layouts; it does not perform CPU-GPU transfers.")

# The displayed example, with its CUDA device requests selected by --device.
import torch

row = torch.tensor([1.0003 if k < 8 else -1.0 for k in range(16)],
                   dtype=torch.float32)
converted = row.to(torch.float16).float()
total = torch.zeros((), dtype=torch.float32)
reference = torch.zeros((), dtype=torch.float32)
for k in range(16):
    total = total + converted[k] * 1.0
    reference = reference + row[k] * 1.0
print("Converted-input sum:", total.item())
print("Original-input sum:", reference.item())