A plan to model CUDA in Bend and pursue a certified compiler
The poster says they'll set an AutoResearch goal to minimize the number of parallel steps a compiled program takes.
TLDR
The poster says they plan to formally model CUDA in Bend, then set a goal to use AutoResearch to pursue a certified compiler that minimizes the parallel steps a compiled program takes. They acknowledge the effort could fail for reasons they don't understand.
Combined views
—
1 Source, first seen ago
