gpu: a shared op layer over the backends, views as the currency, and a fourth backend to prove it - #2

Merged

yhirose merged 8 commits into

Sep 22, 2026

Commits on Sep 21, 2026

cuda_trace: what the CUDA backend asks of the driver, recorded with no GPU
yhirose
committed
65ca384
gpu: a shared layer over the backends, and the elementwise ops written once
yhirose
committed
eb55f46
gpu: eleven more single-kernel ops move to the shared layer
yhirose
committed
8855da6
gpu: every op takes views, and a backend declares only what it runs its own way
yhirose
committed
a610940
gpu: one null backend instead of a stub branch in every backend header
yhirose
committed
77737c6
docs: how the GPU layer is put together, and how to add an op or a backend
yhirose
committed
cadfbc3
gpu: residency decided once, and a conformance test over views
yhirose
committed
73f435b
gpu: a fourth backend, written the way the docs say one is
yhirose
committed
5911d39