Uppsala Architecture Research Team

Software Optimization

Using our measurement and modeling technologies we can quickly provide insight into where and how to optimize software for better power and efficiency.

Memory System Optimization

Eliminating Cache Pollution

Resource Efficient Prefetching

Our memory modeling tools allow us to identify the key memory instructions in an application, and use our understanding of how they interact with the memory hierarchy to optimize their performance.

Reductions in cache miss ratio from changing cache polluting instructions to non-temporal accesses.

DVFS Optimization

Green Governors: A Framework for Continuously Adaptive DVFS
Towards more efficient execution: A decoupled access-execute approach
A compiler automated decoupled access-execute approach

Our power models allow us to determine the optimal DVFS settings for applications based on low-cost performance counter data. We can have applied this knowledge to the Linux DVFS regulator and task-based runtime scheduling to optimize application power efficiency.

The impact of decoupling access and execute: blue regions (access) can be run at lowest frequency without hurting performance.