[Mlir-commits] [mlir] [mlir][linalg] Add support for scalable vectorization of linalg.mmt4d (PR #146531)

[Diego Caballero]

================
@@ -470,6 +481,22 @@ Value VectorizationState::getOrCreateMaskFor(
     return Value();
   }
 
+  if (assumeDynamicDimsMatchVecSizes) {
+    // While for _dynamic_ dim sizes we can _assume_ that the corresponding
+    // vector sizes match, we still need to check the _static_ dim sizes. Only
+    // then we can be 100% sure that masking is not required.
+    if (llvm::all_of(llvm::zip(permutedStaticSizes, maskType.getShape()),
+                     [](auto it) {
+                       return std::get<0>(it) == ShapedType::kDynamic
+                                  ? true
+                                  : std::get<0>(it) == std::get<1>(it);
+                     })) {
+      LDBG("Masking is not needed for masking map: " << maskingMap << "\n");
----------------
dcaballe wrote:

Probably good to make this debug message more specific about the assumption made

https://github.com/llvm/llvm-project/pull/146531