Modular: TileTensor Part 1 - Compile-Time GPU Memory Layouts

AI Impact Summary

Modular's TileTensor introduces a new tensor type designed to simplify GPU kernel development by providing a compile-time abstraction for memory layouts. The core challenge addressed is the manual and error-prone process of calculating memory addresses for swizzled data access, particularly when targeting different GPU architectures with varying bank conflicts. This new abstraction allows for more robust and maintainable GPU code by ensuring layout correctness at compile time, eliminating runtime errors related to incorrect index computations.