diff --git a/mlir/lib/Conversion/TosaToLinalg/TosaToLinalgNamed.cpp b/mlir/lib/Conversion/TosaToLinalg/TosaToLinalgNamed.cpp
--- a/mlir/lib/Conversion/TosaToLinalg/TosaToLinalgNamed.cpp
+++ b/mlir/lib/Conversion/TosaToLinalg/TosaToLinalgNamed.cpp
@@ -63,27 +63,25 @@
       highIndices, padValue);
 }
 
-static mlir::Value makeIntBiasAdd(PatternRewriter &rewriter, Location loc,
-                                  ShapedType resultTy, Value bias, Value conv,
-                                  Value result,
-                                  ArrayRef<AffineMap> indexingMaps) {
-  result = rewriter
-               .create<linalg::GenericOp>(
-                   loc, resultTy, ValueRange({bias, conv}), result,
-                   indexingMaps, getNParallelLoopsAttrs(resultTy.getRank()),
-                   [](OpBuilder &builder, Location loc, ValueRange args) {
-                     Value biasVal = args[0];
-                     Type resType = args[1].getType();
-                     if (resType != biasVal.getType()) {
-                       biasVal = builder.create<arith::ExtSIOp>(loc, resType,
-                                                                biasVal);
-                     }
-                     Value added =
-                         builder.create<arith::AddIOp>(loc, biasVal, args[1]);
-                     builder.create<linalg::YieldOp>(loc, added);
-                   })
-               .getResult(0);
-  return result;
+static mlir::Value
+linalgIntBroadcastExtSIAdd(PatternRewriter &rewriter, Location loc, Value bias,
+                           Value conv, Value result,
+                           ArrayRef<AffineMap> indexingMaps) {
+  ShapedType resultTy = conv.getType().cast<ShapedType>();
+  return rewriter
+      .create<linalg::GenericOp>(
+          loc, resultTy, ValueRange({bias, conv}), result, indexingMaps,
+          getNParallelLoopsAttrs(resultTy.getRank()),
+          [](OpBuilder &builder, Location loc, ValueRange args) {
+            Value biasVal = args[0];
+            Type resType = args[1].getType();
+            if (resType != biasVal.getType()) {
+              biasVal = builder.create<arith::ExtSIOp>(loc, resType, biasVal);
+            }
+            Value added = builder.create<arith::AddIOp>(loc, biasVal, args[1]);
+            builder.create<linalg::YieldOp>(loc, added);
+          })
+      .getResult(0);
 }
 
 static mlir::Value reifyConstantDim(int64_t attr,
@@ -313,8 +311,8 @@
                   loc, resultTy, ValueRange{input, weight, iZpVal, kZpVal},
                   ValueRange{zeroTensor}, strideAttr, dilationAttr)
               ->getResult(0);
-      Value result = makeIntBiasAdd(rewriter, loc, resultTy, bias, conv,
-                                    biasEmptyTensor, indexingMaps);
+      Value result = linalgIntBroadcastExtSIAdd(rewriter, loc, bias, conv,
+                                                biasEmptyTensor, indexingMaps);
       rewriter.replaceOp(op, result);
       return success();
     }
@@ -491,8 +489,8 @@
       createDepthwiseConvCollapseMap(resultRank, reassociationMap, rewriter);
       Value convReshape = rewriter.create<tensor::CollapseShapeOp>(
           loc, resultTy, conv, reassociationMap);
-      Value result = makeIntBiasAdd(rewriter, loc, resultTy, bias, convReshape,
-                                    biasEmptyTensor, indexingMaps);
+      Value result = linalgIntBroadcastExtSIAdd(
+          rewriter, loc, bias, convReshape, biasEmptyTensor, indexingMaps);
       rewriter.replaceOp(op, result);
     }
     return success();
@@ -663,8 +661,8 @@
                 ValueRange{input, transposedWeight, inputZp, outputZp},
                 zeroTensor)
             ->getResult(0);
-    Value result = makeIntBiasAdd(rewriter, loc, outputTy, bias, matmul,
-                                  biasEmptyTensor, indexingMaps);
+    Value result = linalgIntBroadcastExtSIAdd(rewriter, loc, bias, matmul,
+                                              biasEmptyTensor, indexingMaps);
     rewriter.replaceOp(op, result);
     return success();
   }