[Mlir-commits] [mlir] [mlir][memref] `memref.subview`: Verify result strides with rank reductions (PR #80158)

[Jacques Pienaar]

================
@@ -2756,17 +2756,26 @@ static bool haveCompatibleOffsets(MemRefType t1, MemRefType t2) {
 }
 
 /// Return true if `t1` and `t2` have equal strides (both dynamic or of same
-/// static value).
-static bool haveCompatibleStrides(MemRefType t1, MemRefType t2) {
+/// static value). Dimensions of `t1` may be dropped in `t2`; these must be
+/// marked as dropped in `droppedDims`.
+static bool haveCompatibleStrides(MemRefType t1, MemRefType t2,
+                                  const llvm::SmallBitVector &droppedDims) {
+  assert(t1.getRank() == droppedDims.size() && "incorrect number of bits");
+  assert(t1.getRank() - t2.getRank() == droppedDims.count() &&
+         "incorrect number of dropped dims");
   int64_t t1Offset, t2Offset;
   SmallVector<int64_t> t1Strides, t2Strides;
   auto res1 = getStridesAndOffset(t1, t1Strides, t1Offset);
   auto res2 = getStridesAndOffset(t2, t2Strides, t2Offset);
   if (failed(res1) || failed(res2))
     return false;
-  for (auto [s1, s2] : llvm::zip_equal(t1Strides, t2Strides))
-    if (s1 != s2)
+  for (int64_t i = 0, j = 0; i < t1.getRank(); ++i) {
----------------
jpienaar wrote:

Nit: if rank doesn't change during iteration, hoist out (https://llvm.org/docs/CodingStandards.html#don-t-evaluate-end-every-time-through-a-loop)

https://github.com/llvm/llvm-project/pull/80158