diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 2c87d120cb7a..c4cbeb007665 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -7426,6 +7426,14 @@ SDValue DAGCombiner::visitFADDForFMACombine(SDNode *N) {
   bool Aggressive = TLI.enableAggressiveFMAFusion(VT);
   bool LookThroughFPExt = TLI.isFPExtFree(VT);
 
+  // If we have two choices trying to fold (fadd (fmul u, v), (fmul x, y)),
+  // prefer to fold the multiply with fewer uses.
+  if (Aggressive && N0.getOpcode() == ISD::FMUL &&
+      N1.getOpcode() == ISD::FMUL) {
+    if (N0.getNode()->use_size() > N1.getNode()->use_size())
+      std::swap(N0, N1);
+  }
+
   // fold (fadd (fmul x, y), z) -> (fma x, y, z)
   if (N0.getOpcode() == ISD::FMUL &&
       (Aggressive || N0->hasOneUse())) {
diff --git a/llvm/test/CodeGen/NVPTX/fma-assoc.ll b/llvm/test/CodeGen/NVPTX/fma-assoc.ll
index fc04c61dd691..80a08a86316c 100644
--- a/llvm/test/CodeGen/NVPTX/fma-assoc.ll
+++ b/llvm/test/CodeGen/NVPTX/fma-assoc.ll
@@ -23,3 +23,16 @@ define ptx_device double @t1_f64(double %x, double %y, double %z,
   %d = fadd double %c, %z
   ret double %d
 }
+
+define double @two_choices(double %val1, double %val2) {
+; CHECK-LABEL: two_choices(
+; CHECK: mul.f64
+; CHECK-NOT: mul.f64
+; CHECK: fma.rn.f64
+  %1 = fmul double %val1, %val2
+  %2 = fmul double %1, %1
+  %3 = fadd double %1, %2
+
+  ret double %3
+}
+