[Mlir-commits] [mlir] [mlir][vector] Add verification for incorrect vector.extract (PR #115824)

[Kunwar Grover]

================
@@ -1339,6 +1339,83 @@ bool ExtractOp::isCompatibleReturnTypes(TypeRange l, TypeRange r) {
   return l == r;
 }
 
+// Common verification rules for `InsertOp` and `ExtractOp` involving indices
+// and shapes. `indexedType` is the vector type being indexed by the operation,
+// i.e., the destination type in `InsertOp` and the source type in `ExtractOp`.
+// `nonIndexedType` is the inserted or extracted type by an `InsertOp` or and
+// `ExtractOp`, respectively.
+static LogicalResult verifyInsertExtractIndicesAndShapes(Operation *op,
+                                                         VectorType indexedType,
+                                                         int64_t numIndices,
+                                                         Type nonIndexedType) {
+  assert((isa<InsertOp>(op) || isa<ExtractOp>(op)) &&
+         "Expected InsertOp or ExtractOp");
+
+  std::string nonIndexedStr = isa<InsertOp>(op) ? "inserted" : "extracted";
+  std::string indexedStr = isa<InsertOp>(op) ? "destination" : "source";
+  int64_t indexedRank = indexedType.getRank();
+  if (numIndices > indexedRank) {
+    return op->emitOpError()
+           << "expected a number of indices no greater than the " << indexedStr
+           << " vector rank";
+  }
+
+  if (auto nonIndexedVecType = dyn_cast<VectorType>(nonIndexedType)) {
+    // Vector case, including meaningful cases such as:
+    //  * 0-D vector:
+    //    * vector.extract %src[2]: vector<f32> from vector<8xf32)
+    //    * vector.insert %src, %dst[3]: vector<f32> into vector<8xf32>
+    //  * One-element vector:
+    //    * vector.extract %src[4]: vector<1xf32> from vector<8xf32>
+    //    * vector.insert %src, %dst[1]: vector<1xf32> into vector<8xf32>
+    //    * vector.extract %src[7]: vector<1xf32> from vector<8x1xf32>
+    //    * vector.insert %src, %dst[5]: vector<1xf32> into vector<8x1xf32>
----------------
Groverkss wrote:

Note that Possibility 2 is what the documentation documents:

https://mlir.llvm.org/docs/Dialects/Vector/#vectorextract-vectorextractop

```
Takes an n-D vector and a k-D position and extracts the (n-k)-D vector at the proper position. Degenerates to an element type if n-k is zero.
``` 

Suprisingly, vector.insert documentation diverges from this:

https://mlir.llvm.org/docs/Dialects/Vector/#vectorinsert-vectorinsertop

```
Takes an n-D source vector, an (n+k)-D destination vector and a k-D position and inserts the n-D source into the (n+k)-D destination at the proper position. Degenerates to a scalar or a 0-d vector source type when n = 0.
```

I'm not sure what degenerates to "0-d source type" means, but we should make it same as vector.extract and it would be easier.


https://github.com/llvm/llvm-project/pull/115824