[Mlir-commits] [mlir] [mlir][SCF] Deprecate `linalg::tileToForallOp` and `linalg::tileToForallOpUsingTileSizes` (PR #91878)

[Nicolas Vasilache]

================
@@ -582,21 +811,8 @@ mlir::scf::tileUsingSCF(RewriterBase &rewriter, TilingInterface op,
       -> LogicalResult {
     // 4a. Compute the `offsets` and `sizes` to use for tiling.
     SmallVector<OpFoldResult> offsets, sizes;
-    {
-      int materializedLoopNum = 0;
-      for (auto [tileSize, loopRange] :
-           llvm::zip_equal(tileSizes, iterationDomain)) {
-        if (isConstantIntValue(tileSize, 0)) {
-          offsets.push_back(loopRange.offset);
-          sizes.push_back(loopRange.size);
-          continue;
-        }
-        Value iv = ivs[materializedLoopNum++];
-        offsets.push_back(iv);
-        sizes.push_back(
-            getBoundedTileSize(rewriter, loc, loopRange, iv, tileSize));
-      }
-    }
+    std::tie(offsets, sizes) = getTileOffsetAndSizes(
----------------
nicolasvasilache wrote:

Indeed, generally good code-health wise to extract out this loop into its own named function.

https://github.com/llvm/llvm-project/pull/91878