[Mlir-commits] [mlir] [mlir][EmitC] Expand the MemRefToEmitC pass - Lowering `CopyOp` (PR #151206)

[Gil Rapaport]

================
@@ -55,35 +65,33 @@ struct ConvertMemRefToEmitCPass
       return signalPassFailure();
 
     mlir::ModuleOp module = getOperation();
+    llvm::SmallVector<StringRef> requiredHeaders;
     module.walk([&](mlir::emitc::CallOpaqueOp callOp) {
-      if (callOp.getCallee() != alignedAllocFunctionName &&
-          callOp.getCallee() != mallocFunctionName) {
+      StringRef expectedHeader;
+      if (callOp.getCallee() == alignedAllocFunctionName ||
+          callOp.getCallee() == mallocFunctionName)
+        expectedHeader = options.lowerToCpp ? cppStandardLibraryHeader
+                                            : cStandardLibraryHeader;
+      else if (callOp.getCallee() == memcpyFunctionName)
+        expectedHeader =
+            options.lowerToCpp ? cppStringLibraryHeader : cStringLibraryHeader;
+      else
         return mlir::WalkResult::advance();
-      }
+      requiredHeaders.push_back(expectedHeader);
+      return mlir::WalkResult::advance();
+    });
+    for (StringRef expectedHeader : requiredHeaders) {
+      bool headerFound = llvm::any_of(*module.getBody(), [&](Operation &op) {
----------------
aniragil wrote:

This scans all module-level ops for each expected header. Can instead scan once in advance to find all available headers, then when scanning `call_opaque`s can skip those whose requested header is already included.

https://github.com/llvm/llvm-project/pull/151206