diff --git a/iree/compiler/Dialect/Flow/IR/FlowOps.cpp b/iree/compiler/Dialect/Flow/IR/FlowOps.cpp
index fa3384c57ea5..f61d44400823 100644
--- a/iree/compiler/Dialect/Flow/IR/FlowOps.cpp
+++ b/iree/compiler/Dialect/Flow/IR/FlowOps.cpp
@@ -827,12 +827,11 @@ static bool canDispatchRegionContainOp(Operation *op) {
                    constantValueAttr.dyn_cast<DenseElementsAttr>()) {
       // TODO(GH-4897): Non-splat constants seems to have an issue on the LLLVM
       // side. Uncomment after that is fixed.
-      // auto shapedType = constantOp.getType().cast<ShapedType>();
-      // uint64_t estimatedByteLength =
-      //     (shapedType.getNumElements() * shapedType.getElementTypeBitWidth())
-      //     / 8;
-      return denseAttr
-          .isSplat();  // || estimatedByteLength <= 256;  // or whatever
+      auto shapedType = constantOp.getType().cast<ShapedType>();
+      uint64_t estimatedByteLength =
+          (shapedType.getNumElements() * shapedType.getElementTypeBitWidth()) /
+          8;
+      return denseAttr.isSplat() || estimatedByteLength <= 256;  // or whatever
     } else if (constantType.isIntOrIndexOrFloat()) {
       return true;
     }
diff --git a/iree/compiler/Dialect/HAL/Target/LLVM/LLVMAOTTarget.cpp b/iree/compiler/Dialect/HAL/Target/LLVM/LLVMAOTTarget.cpp
index 483c0f3b002a..273d48abadd6 100644
--- a/iree/compiler/Dialect/HAL/Target/LLVM/LLVMAOTTarget.cpp
+++ b/iree/compiler/Dialect/HAL/Target/LLVM/LLVMAOTTarget.cpp
@@ -103,6 +103,35 @@ class LLVMAOTTargetBackend final : public TargetBackend {
         llvm::to_vector<8>(moduleOp.getOps<IREE::HAL::ExecutableOp>());
     if (sourceExecutableOps.size() <= 1) return success();
 
+    // Private symbols (i.e. llvm dialect private symbols) get deduped
+    // incorrectly by the link executables pass even though they should be
+    // treated as different symbols. For now just change the names of the
+    // private symbols to avoid conflicts.
+    unsigned moduleNumber = 0;
+    for (auto sourceExecutableOp : enumerate(sourceExecutableOps)) {
+      auto targetOps = llvm::to_vector<4>(
+          sourceExecutableOp.value().getOps<IREE::HAL::ExecutableTargetOp>());
+      for (auto targetOp : targetOps) {
+        if (!matchPattern(targetOp.target_backend_filter(), filter_pattern())) {
+          continue;
+        }
+
+        auto sourceModuleOp = targetOp.getInnerModule();
+        for (auto globalOp : sourceModuleOp.getOps<LLVM::GlobalOp>()) {
+          if (globalOp.linkage() != LLVM::Linkage::Private) {
+            continue;
+          }
+          auto disambiguateName =
+              llvm::formatv("{0}_{1}", globalOp.sym_name(), moduleNumber).str();
+          SymbolTableCollection symbolTable;
+          SymbolUserMap symbolUsers(symbolTable, sourceModuleOp);
+          symbolUsers.replaceAllUsesWith(globalOp, disambiguateName);
+          SymbolTable::setSymbolName(globalOp, disambiguateName);
+        }
+        moduleNumber++;
+      }
+    }
+
     // Guess a module name, if needed, to make the output files readable.
     auto moduleName = guessModuleName(moduleOp);
 
diff --git a/iree/test/e2e/linalg_tensor_ops/add.mlir b/iree/test/e2e/linalg_tensor_ops/add.mlir
index a9e31dd68ffe..6d5265676d8c 100644
--- a/iree/test/e2e/linalg_tensor_ops/add.mlir
+++ b/iree/test/e2e/linalg_tensor_ops/add.mlir
@@ -87,4 +87,4 @@ func @cst_plus_tensor() attributes { iree.module.export } {
    check.expect_eq_const(%1, dense<
      [[[2, 4, 6], [8, 10, 12]], [[14, 16, 18], [20, 22, 24]]]> : tensor<2x2x3xi32>) : tensor<2x2x3xi32>
    return
-}
\ No newline at end of file
+}