Today we continue reading the WRL research report on Swift Java Compiler. We were discussing alias analysis We will see a few more ways to relax memory dependencies. The basic Swift IR is retained. A process called store resolution is used to relax memory dependences by changing the global store inputs of some memory operations. In the case of the loop example mentioned earlier, rather than changing the stores of A and B to be the global store inputs, we keep them the original store inputs and move them out of the loop. Store resolution does a forward dataflow computation to build up information about memory operations, as follows.
We look at the sub-graph of the SSA graph where global stores are produced. We label such operations that produce them as StoreOps. We want to compute information at these nodes about sets of locations that are easy to classify. With the help of the language, we know particular fields are of certain types and array elements are of particular types. These different sets of location can be labeled LocTypes. At each node of the subgraph we want to find for each LocType, the most recent StoreOp that might have modified the locations of that type
From the dataflow, the state at each node maps each LocType to the most recent preceding StoreOp which might have modified locations of that LocType. The state entry has a default StoreOp that applies to most LocTypes. and miscellaneous ordered pairs of (LocType, StoreOp) for any locTypes that don't map to the default. Perhaps we can classify them to have more salient entries in the state.
#codingexercise

Double GetAlternateOddNumberRangeSumSquareRootTenthSquare (Double [] A)

{

if (A == null) return 0;

Return A.AlternateOddNumberRangeSumSquareRootTenthSquare();

}