Tensor lifting of OpenMP loops enables mapping scientific kernels to AI Engines, matching multicore CPU performance at lower energy and delivering up to 40% speedup with 15% energy reduction when hybridizing CPU and NPU.
Programmer productivity and performance on amd’s ai engines: Offloading fortran intrinsics via mlir a case-study
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.DC 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Lifting to tensors when compiling scientific computing workloads for AI Engines
Tensor lifting of OpenMP loops enables mapping scientific kernels to AI Engines, matching multicore CPU performance at lower energy and delivering up to 40% speedup with 15% energy reduction when hybridizing CPU and NPU.