I built a GPU compute library that's actually simple. No CUDA, no complex setup.
Python:
import adamah
adamah.init()
adamah.put("x", [1, 2, 3, 4])
adamah.sin("y", "x", 4)
print(adamah.get("y"))
c:
inject("x", data, n);
vop1(VOP_SIN, "y", "x", n);
extract("y", result, n);
Features:
Named buffers (auto-managed, no alloc/free)
Full math: trig, calculus, reduce, linear algebra
Sparse memory maps (scatter/gather)
FFI-ready for any language
Single header + one .c file
Built on Vulkan so it runs everywhere (not just NVIDIA).
pip install adamah
I built a GPU compute library that's actually simple. No CUDA, no complex setup. Python: import adamah adamah.init() adamah.put("x", [1, 2, 3, 4]) adamah.sin("y", "x", 4) print(adamah.get("y")) c: inject("x", data, n); vop1(VOP_SIN, "y", "x", n); extract("y", result, n); Features:
Named buffers (auto-managed, no alloc/free) Full math: trig, calculus, reduce, linear algebra Sparse memory maps (scatter/gather) FFI-ready for any language Single header + one .c file
Built on Vulkan so it runs everywhere (not just NVIDIA). pip install adamah