"""
Example 6: GPU Profiling (Optional - requires GPU hardware)
Real-world scenario: A machine learning engineer uses GPU libraries like PyTorch
or TensorFlow and wants to understand how much time is spent on GPU vs CPU.
Use case: Identify GPU bottlenecks and optimization opportunities.
Note: This example requires a GPU. It's safe to run on CPU - it will just show
minimal GPU time. The example is included for teams with GPU hardware.
"""
import numpy as np
def cpu_matrix_multiply(size: int = 100) -> np.ndarray:
"""Pure CPU matrix multiplication using NumPy."""
a = np.random.randn(size, size)
b = np.random.randn(size, size)
# np.dot uses BLAS (optimized C code) but runs on CPU
result = np.dot(a, b)
return result
def demonstrate_gpu_potential():
"""
Show where GPU could help.
Note: This doesn't actually use GPU without PyTorch/TensorFlow installed.
It's here to show the pattern.
"""
print("="*60)
print("GPU Profiling Example")
print("="*60)
print("\n1. CPU Matrix Multiplication")
result = cpu_matrix_multiply(size=100)
print(f" Shape: {result.shape}")
# Demonstrate the pattern (without actual GPU code)
print("\n2. Pattern for GPU profiling with PyTorch:")
print("""
# If you had PyTorch:
import torch
# CPU computation
cpu_tensor = torch.randn(100, 100)
cpu_result = torch.mm(cpu_tensor, cpu_tensor)
# GPU computation (if available)
if torch.cuda.is_available():
gpu_tensor = torch.randn(100, 100, device='cuda')
gpu_result = torch.mm(gpu_tensor, gpu_tensor)
# Scalene would show:
# - Python time: Control flow, setup
# - C time: NumPy/torch CPU operations
# - GPU time: Computation on GPU (if you have GPU monitoring)
""")
print("\n3. Pattern for GPU profiling with TensorFlow:")
print("""
# If you had TensorFlow:
import tensorflow as tf
@tf.function # Compile to GPU
def gpu_computation():
a = tf.random.normal((100, 100))
b = tf.random.normal((100, 100))
return tf.matmul(a, b)
result = gpu_computation()
# Scalene shows time breakdown between Python, C, and GPU
""")
print("\n" + "="*60)
print("GPU Profiling Benefits:")
print(" - See CPU vs GPU time split")
print(" - Identify if GPU is utilized")
print(" - Find bottlenecks (CPU waiting for GPU, etc.)")
print(" - Optimize data transfer between CPU/GPU")
print("="*60)
print("\nTo enable GPU profiling:")
print(" - Install NVIDIA drivers (for NVIDIA GPU)")
print(" - Profile with: python -m scalene --gpu examples/6_gpu_profiling.py")
print("\nNote: Requires NVIDIA GPU support. CPU fallback is safe.")
if __name__ == "__main__":
demonstrate_gpu_potential()
