Summary
Upgrade trueno dependency from 0.8.9 to 0.9.0 to get CUDA-tile GPU optimizations.
Current Version
Target Version
New Features in v0.9.0
- TensorView: Structured memory view with shape/stride metadata for GPU buffers
- PartitionView: Tiling strategy for 16×16 GPU workgroup distribution
- Tiled Reduction:
tiled_sum_2d, tiled_max_2d, tiled_min_2d algorithms
- ReduceOp trait: Custom reduction operations (SumOp, MaxOp, MinOp)
- Intel SDE Support: AVX-512 testing on non-AVX512 CPUs
- PTX Optimization Passes: FMA fusion (~33% instruction reduction)
Release Notes
https://github.com/paiml/trueno/releases/tag/v0.9.0
Migration
No breaking changes - drop-in upgrade.
Summary
Upgrade trueno dependency from 0.8.9 to 0.9.0 to get CUDA-tile GPU optimizations.
Current Version
Target Version
New Features in v0.9.0
tiled_sum_2d,tiled_max_2d,tiled_min_2dalgorithmsRelease Notes
https://github.com/paiml/trueno/releases/tag/v0.9.0
Migration
No breaking changes - drop-in upgrade.