NEC's proprietary ncc, nc++, nfort compilers for the SX-Aurora Tsubasa support vector intrinsics. In Japanese.
[Read More]
Llama2 with bfloat16 on the SX-Aurora Vector Engine
llama2.c on the sx-aurora VE ran on bfloat16 data, performance measurements and comparison.
[Read More]
Llama2 on the SX-Aurora Vector Engine
Porting and tuning of llama2.c on the sx-aurora VE, performance measurements and comparison.
[Read More]
NEC compilers need no license any more
NEC's proprietary ncc, nc++, nfort compilers for the SX-Aurora Tsubasa can now be used without the need for a license.
[Read More]
Proprietary LLVM-VEC VE vectorizer for LLVM
NEC's proprietary ncc vectorizer is now usable from within a specially patched LLVM version.
[Read More]
Porting and Optimizing Molecular Docking Simulations on SX-Aurora Vector Engine (WSSP2021)
Presentation on porting of AutoDock to the SX-Aurora TSUBASA Vector Engine.
[Read More]
Debugging (A)VEO Programs
The post describes simple ways to debug AVEO program kernels that run on the vector engine.
[Read More]
veperf memory bandwidth, power and energy monitoring
The veperf monitoring tool displays live performance metrics of running VE programs. It has been extended to also show memory bandwidths as well as power and energy metrics.
[Read More]
Releasing LLVM-VE-RV 1.5.1
LLVM-VE-RV 1.5.1 is the new release of LLVM-VE for SX-Aurora featuring automatic loop vectorization and experimental OpenMP target offloading from VH to VE.
[Read More]
AVEO - Another/Alternative/Awesome VE Offloading
AVEO is a reimplementation of the VEO API with a completely different approach that solves several issues of VEO, especially reducing the call latency.
[Read More]