LLAMA's Heatmap mapping currently traces memory accesses for the entire blob regions. This may require an excessive amount of memory though. Since access patterns are often repretitive, we should add an option to allow the heatmap to only trace a subpart of the blobs and thus use less memory.
LLAMA's
Heatmap
mapping currently traces memory accesses for the entire blob regions. This may require an excessive amount of memory though. Since access patterns are often repretitive, we should add an option to allow the heatmap to only trace a subpart of the blobs and thus use less memory.