"Boosting Python Numeric Computations with Dynamic C++ Integration"
summary
Builds a C++ extension with pybind11 and a simple setup to compare identical for‑loop matrix multiplies called from Python, reporting a 79x improvement at n=100 and ~57x at n=1000 without algorithmic changes. Shows zerocopy via zero-copy data exchange and smart‑pointer references so tensors on GPU need not be copied, notes support for clang/GCC/LLVM/NVIDIA compilers, and emphasizes profiling to extract hot kernels before porting to C++.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:14Performance
- 1:58Realtime Processing
- 2:56Why is Python Slow
- 4:26Numeric Computation Libraries
- 6:15Simple Applications
- 8:40Deep Learning
- 15:35Future of Programming
- 18:57Concerns
related talks
01h 21mAccelerating Python With C++
01h 02mWriting Python Bindings for C++ Libraries: Easy-to-use Performance
29mC++ ♥ Python: from modules to hybrid applications
34mAccelerating GPU Algorithms in pure Python
31m"Optimizing Scientific Python with C++, CUDA, and Serverless Compute"
55mAccelerated Python: The Community and Ecosystem
30mCUDA in Your Python: Effective Parallel Programming on the GPU
25mHow to bring your C++ project to Python land