[mlir][sparse] Moved preparation of sparse pointer structures out of recursion

[wrengr]

Summary

Depends On D108670. Previously each recursive call of traverse checked whether the sparse pointer structures were empty (and so needed preparation). Now we prepare all the sparse pointer structures in one go, thereby simplifying the traverse function (and marginally improving performance).

[aartbik]

mlir/lib/ExecutionEngine/SparseUtils.cpp

188

please spell upperbound "r" out as rank = getRank() to be more consistent with surroundings
(also, I mostly use d++ and not ++d for ints, but I know where you are coming from with that ;-)

[wrengr]

[mlir][sparse] Moved preparation of sparse pointer structures out of recursion

[aartbik]

A rebase may actually give a conflict (or simply show nodiff after pulling in my recent change).
But approving the change formally here.

mlir/lib/ExecutionEngine/SparseUtils.cpp

for (uint64_t d = 0, s = 1, rank = getRank(); d < rank; d++) {
       if (sparsity[d] == kCompressed) {
         pointers[d].reserve(s + 1);
         indices[d].reserve(s);
         s = 1;
       } else {
         assert(sparsity[d] == kDense && "singleton not yet supported");
       }
     }
+    // Prepare sparse pointer structures for all dimensions.
+    for (uint64_t d = 0, rank = getRank(); d < rank; d++)
+      if (sparsity[d] == kCompressed)
+        pointers[d].push_back(0);
     // Then setup the tensor.
     traverse(tensor, sparsity, 0, nnz, 0);
   }
 
   virtual ~SparseTensorStorage() {}
 
   uint64_t getRank() const { return sizes.size(); }
 
private:
   void traverse(SparseTensor<V> *tensor, uint8_t *sparsity, uint64_t lo,
                 uint64_t hi, uint64_t d) {
     const std::vector<Element<V>> &elements = tensor->getElements();
     // Once dimensions are exhausted, insert the numerical values.
     if (d == getRank()) {
       values.push_back(lo < hi ? elements[lo].value : 0);
       return;
     }
-    // Prepare a sparse pointer structure at this dimension.
-    if (sparsity[d] == kCompressed && pointers[d].empty())
-      pointers[d].push_back(0);
     // Visit all elements in this interval.
     uint64_t full = 0;
     while (lo < hi) {
       // Find segment in interval with same index elements in this dimension.
       unsigned idx = elements[lo].indices[d];
       unsigned seg = lo + 1;
       while (seg < hi && elements[seg].indices[d] == idx)
         seg++;