[Mlir-commits] [mlir] [MLIR][mesh] Mesh fixes (PR #124724)

[Renato Golin]

================
@@ -796,8 +790,12 @@ spmdizeFuncOp(FunctionOpInterface op, IRMapping &spmdizationMap,
   // Snapshot the original blocks to not mess up the iteration when adding new
   // blocks.
   SmallVector<Block *> originalBlocks;
-  llvm::transform(op.getBlocks(), std::back_inserter(originalBlocks),
-                  [](Block &b) { return &b; });
+  for (Block &b : op.getBlocks()) {
----------------
rengolin wrote:

So, if you have two blocks with shard ops and one in the middle without (but still dominating the last one), you don't SPMDize the middle block?

https://github.com/llvm/llvm-project/pull/124724