[Mlir-commits] [mlir] [mlir][transform] Fix handling of transitive include in interpreter. (PR #67560)

[Oleksandr Alex Zinenko]

Ingo =?utf-8?q?Müller?= <ingomueller at google.com>,
Ingo =?utf-8?q?Müller?= <ingomueller at google.com>,
Ingo =?utf-8?q?Müller?= <ingomueller at google.com>,
Ingo =?utf-8?q?Müller?= <ingomueller at google.com>,
Ingo =?utf-8?q?Müller?= <ingomueller at google.com>,
Ingo =?utf-8?q?Müller?= <ingomueller at google.com>,
Ingo =?utf-8?q?Müller?= <ingomueller at google.com>,
Ingo =?utf-8?q?Müller?= <ingomueller at google.com>,
Ingo =?utf-8?q?Müller?= <ingomueller at google.com>,
Ingo =?utf-8?q?Müller?= <ingomueller at google.com>
Message-ID:
In-Reply-To: <llvm/llvm-project/pull/67560/mlir at github.com>


================
@@ -302,80 +304,268 @@ static void performOptionalDebugActions(
     transform->removeAttr(kTransformDialectTagAttrName);
 }
 
-/// Replaces external symbols in `block` with their (non-external) definitions
-/// from the given module.
-static LogicalResult defineDeclaredSymbols(Block &block, ModuleOp definitions) {
-  MLIRContext &ctx = *definitions->getContext();
-  auto consumedName =
-      StringAttr::get(&ctx, transform::TransformDialect::kArgConsumedAttrName);
-  auto readOnlyName =
-      StringAttr::get(&ctx, transform::TransformDialect::kArgReadOnlyAttrName);
-
-  for (Operation &op : llvm::make_early_inc_range(block)) {
-    LLVM_DEBUG(DBGS() << op << "\n");
-    auto symbol = dyn_cast<SymbolOpInterface>(op);
-    if (!symbol)
-      continue;
-    if (symbol->getNumRegions() == 1 && !symbol->getRegion(0).empty())
-      continue;
-
-    LLVM_DEBUG(DBGS() << "looking for definition of symbol "
-                      << symbol.getNameAttr() << ":");
-    SymbolTable symbolTable(definitions);
-    Operation *externalSymbol = symbolTable.lookup(symbol.getNameAttr());
-    if (!externalSymbol || externalSymbol->getNumRegions() != 1 ||
-        externalSymbol->getRegion(0).empty()) {
-      LLVM_DEBUG(llvm::dbgs() << "not found\n");
-      continue;
+/// Rename `op` to avoid a collision with `otherOp`. `symbolTable` and
+/// `otherSymbolTable` are the symbol tables of the two ops, respectively.
+/// `uniqueId` is used to generate a unique name in the context of the caller.
+LogicalResult renameToUnique(SymbolOpInterface op, SymbolOpInterface otherOp,
+                             SymbolTable &symbolTable,
+                             SymbolTable &otherSymbolTable, int &uniqueId) {
+  assert(symbolTable.lookup(op.getNameAttr()) == op &&
+         "symbol table does not contain op");
+  assert(otherSymbolTable.lookup(otherOp.getNameAttr()) == otherOp &&
+         "other symbol table does not contain other op");
+
+  // Determine new name that is unique in both symbol tables.
+  StringAttr oldName = op.getNameAttr();
+  StringAttr newName;
+  {
+    MLIRContext *context = op->getContext();
+    SmallString<64> prefix = oldName.getValue();
+    prefix.push_back('_');
+    while (true) {
+      newName = StringAttr::get(context, prefix + Twine(uniqueId++));
+      if (!symbolTable.lookup(newName) && !otherSymbolTable.lookup(newName)) {
+        break;
+      }
     }
----------------
ftynse wrote:

Having a chain of commits, each with a small amount of code to avoid touching multiple subsystems. Here, it would be changing the SymbolTable in a separate commit. Some project follow this style, e.g., to have different code owners review them separately. LLVM/MLIR prefer having larger cross-cutting commits to see the overall motivation for changes.

No action is necessary (pull requests will be squashed into one commit anyway) so I resolved the discussion.

https://github.com/llvm/llvm-project/pull/67560