Fix suspected implicit casting problem in LinalgTileAndDistributePass.

* This was failing on gcc 9.3.1 and working on everything else (clang, msvc, gcc 9.3.0 (!)).
* As near as I can tell, it was an incomplete refactor that happens to be allowed via some implicit conversion path that I can't spot on other compilers.
* Fixes #4543.

iree/compiler/Conversion/LinalgToLLVM/LinalgTileAndDistributePass.cpp

diff --git a/iree/compiler/Conversion/LinalgToLLVM/LinalgTileAndDistributePass.cpp b/iree/compiler/Conversion/LinalgToLLVM/LinalgTileAndDistributePass.cpp
index 758569c..caab412 100644
--- a/iree/compiler/Conversion/LinalgToLLVM/LinalgTileAndDistributePass.cpp
+++ b/iree/compiler/Conversion/LinalgToLLVM/LinalgTileAndDistributePass.cpp
@@ -63,10 +63,10 @@
     // Find the parent FuncOp before tiling. If tiling succeeds, the op will be
     // erased.
     FuncOp funcOp = op->getParentOfType<FuncOp>();
-    SmallVector<Value, 4> tensorResults;
+    linalg::TiledLinalgOp tiledLinalgOp;
     if (!funcOp ||
-        failed(Base::matchAndRewriteBase(op, rewriter, tensorResults)) ||
-        !tensorResults.empty() ||
+        failed(Base::matchAndRewriteBase(op, rewriter, tiledLinalgOp)) ||
+        !tiledLinalgOp.tensorResults.empty() ||
         (funcOp->getAttr(getNumWorkgroupsFnAttrName()) &&
          failed(createNumWorkgroupsFromResultShape(
              rewriter, cast<linalg::LinalgOp>(op), funcOp,