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

[llvmlistbot at llvm.org]

================
@@ -2928,18 +2928,39 @@ DiagnosedSilenceableFailure transform::tileToForallOpImpl(
     return diag;
   }
   rewriter.setInsertionPoint(tileableOp);
-  FailureOr<linalg::ForallTilingResult> maybeTilingResult = failure();
+  scf::SCFTilingOptions options;
+  options.setLoopType(scf::SCFTilingOptions::LoopType::ForallOp);
   if (!mixedNumThreads.empty()) {
-    maybeTilingResult =
-        linalg::tileToForallOp(rewriter, tileableOp, mixedNumThreads, mapping);
+    options.setNumThreads(mixedNumThreads);
   } else {
-    maybeTilingResult = linalg::tileToForallOpUsingTileSizes(
-        rewriter, tileableOp, mixedTileSizes, mapping);
+    SmallVector<Range> loopRanges = tileableOp.getIterationDomain(rewriter);
+    unsigned nLoops = loopRanges.size();
+    SmallVector<OpFoldResult> numThreads;
+    numThreads.reserve(nLoops);
+    AffineExpr s0, s1;
+    bindSymbols(rewriter.getContext(), s0, s1);
+    AffineExpr divExpr = s0.ceilDiv(s1);
+    for (int i = 0, e = std::min(mixedTileSizes.size(), loopRanges.size());
----------------
MaheshRavishankar wrote:

Yeah, thats a bit spooky if you dont know that it does that. Lets leave it more explicit.

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