Automating High Level Control F'low Transformations For Dsp Memory Management

M.F.X.B. van Swaaij, F. Franssen, Francky Catthoor, H.J. De Man · 2005

Silage code corresponding to original procedural code written by system designer [4]. (right) Procural code as generated by the control flow optimization tool. tangent diff mean grid_diff path 64 --- 64 8.404.992 ------------ 64 131.328 ---------- 1 131.328 ---------- 1 511 ---- 1 i x 64 511 511 diff mean grid_diff path Figure 7. (left) PDG of the kernel of the CRD algorithm. (right) Optimal polytope placement. dencies according to the presented model, leads to the PDG of figure 7. The result of the placement optimization can be found in figure 7. In this figure, the polytopes, but not their placement, correspond to those extracted from the initial applicative code. Their dependencies are indicated for a single i; j-pair by arrows. The bold curved arrows indicate the dependencies on the 64 signal instances diff [i][j][]. Note that the placed polytopes together form again a polytope in which each integer point contains an operation (maximum density). Furthermore, dependency-variancy a...

Read the paper · More papers on PaperTik