Index: lib/Target/SystemZ/SystemZTargetTransformInfo.cpp
===================================================================
--- lib/Target/SystemZ/SystemZTargetTransformInfo.cpp
+++ lib/Target/SystemZ/SystemZTargetTransformInfo.cpp
@@ -899,6 +899,10 @@
        UserI->getOpcode() == Instruction::UDiv) &&
       UserI->getOperand(1) != FoldedValue)
     return false; // Not commutative, only RHS foldable.
+  // LoadOrTruncBits holds the number of effectively loaded bits, but 0 if an
+  // extension was made of the load.
+  unsigned LoadOrTruncBits =
+      ((SExtBits || ZExtBits) ? 0 : (TruncBits ? TruncBits : LoadedBits));
   switch (UserI->getOpcode()) {
   case Instruction::Add: // SE: 16->32, 16/32->64, z14:16->64. ZE: 32->64
   case Instruction::Sub:
@@ -910,6 +914,8 @@
         (SExtBits == 32 ||
          (SExtBits == 64 && ST->hasMiscellaneousExtensions2())))
       return true;
+    if (LoadOrTruncBits == 16)
+      return true;
     LLVM_FALLTHROUGH;
   case Instruction::SDiv:// SE: 32->64
     if (LoadedBits == 32 && SExtBits == 64)
@@ -929,10 +935,6 @@
     // case Instruction::FDiv:
 
     // All possible extensions of memory checked above.
-    if (SExtBits || ZExtBits)
-      return false;
-
-    unsigned LoadOrTruncBits = (TruncBits ? TruncBits : LoadedBits);
     return (LoadOrTruncBits == 32 || LoadOrTruncBits == 64);
     break;
   }
Index: test/Analysis/CostModel/SystemZ/memop-folding-int-arith.ll
===================================================================
--- test/Analysis/CostModel/SystemZ/memop-folding-int-arith.ll
+++ test/Analysis/CostModel/SystemZ/memop-folding-int-arith.ll
@@ -85,6 +85,37 @@
 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %10 = add i32 %sext_3, undef
 }
 
+define void @add_i16_mem16(i16 %Arg, i16* %Src1, i16* %Src2, i16* %Dst, i32* %Src32) {
+  %L1 = load i16, i16* %Src1
+  %S0 = add i16 %L1, %Arg
+  store volatile i16 %S0, i16* %Dst
+
+  %L2 = load i16, i16* %Src1
+  %L3 = load i16, i16* %Src2
+  %S1 = add i16 %L2, %L3
+  store volatile i16 %S1, i16* %Dst
+
+  ; Truncated load
+  %L32 = load i32, i32* %Src32
+  %tr = trunc i32 %L32 to i16
+  %S2 = add i16 %tr, %Arg
+  store volatile i16 %S2, i16* %Dst
+
+  ret void
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'add_i16_mem16':
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   %L1 = load i16, i16* %Src1
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %S0 = add i16 %L1, %Arg
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   store volatile i16 %S0, i16* %Dst
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   %L2 = load i16, i16* %Src1
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %L3 = load i16, i16* %Src2
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %S1 = add i16 %L2, %L3
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   store volatile i16 %S1, i16* %Dst
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   %L32 = load i32, i32* %Src32
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   %tr = trunc i32 %L32 to i16
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %S2 = add i16 %tr, %Arg
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   store volatile i16 %S2, i16* %Dst
+}
+
 define void @sub_lhs_mem() {
   %li32 = load i32, i32* undef
   sub i32 %li32, undef
@@ -228,6 +259,37 @@
 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %8 = sub i32 undef, %sext_3
 }
 
+define void @sub_i16_mem16(i16 %Arg, i16* %Src1, i16* %Src2, i16* %Dst, i32* %Src32) {
+  %L1 = load i16, i16* %Src1
+  %D0 = sub i16 %Arg, %L1
+  store volatile i16 %D0, i16* %Dst
+
+  %L2 = load i16, i16* %Src1
+  %L3 = load i16, i16* %Src2
+  %D1 = sub i16 %L2, %L3
+  store volatile i16 %D1, i16* %Dst
+
+  ; Truncated load
+  %L32 = load i32, i32* %Src32
+  %tr = trunc i32 %L32 to i16
+  %D2 = sub i16 %Arg, %tr
+  store volatile i16 %D2, i16* %Dst
+
+  ret void
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'sub_i16_mem16':
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   %L1 = load i16, i16* %Src1
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %D0 = sub i16 %Arg, %L1
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   store volatile i16 %D0, i16* %Dst
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %L2 = load i16, i16* %Src1
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   %L3 = load i16, i16* %Src2
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %D1 = sub i16 %L2, %L3
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   store volatile i16 %D1, i16* %Dst
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   %L32 = load i32, i32* %Src32
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   %tr = trunc i32 %L32 to i16
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %D2 = sub i16 %Arg, %tr
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   store volatile i16 %D2, i16* %Dst
+}
+
 define void @mul() {
   %li32 = load i32, i32* undef
   mul i32 %li32, undef
@@ -305,6 +367,37 @@
 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %10 = mul i32 %sext_3, undef
 }
 
+define void @mul_i16_mem16(i16 %Arg, i16* %Src1, i16* %Src2, i16* %Dst, i32* %Src32) {
+  %L1 = load i16, i16* %Src1
+  %P0 = mul i16 %Arg, %L1
+  store volatile i16 %P0, i16* %Dst
+
+  %L2 = load i16, i16* %Src1
+  %L3 = load i16, i16* %Src2
+  %P1 = mul i16 %L2, %L3
+  store volatile i16 %P1, i16* %Dst
+
+  ; Truncated load
+  %L32 = load i32, i32* %Src32
+  %tr = trunc i32 %L32 to i16
+  %P2 = mul i16 %Arg, %tr
+  store volatile i16 %P2, i16* %Dst
+
+  ret void
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'mul_i16_mem16':
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   %L1 = load i16, i16* %Src1
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %P0 = mul i16 %Arg, %L1
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   store volatile i16 %P0, i16* %Dst
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   %L2 = load i16, i16* %Src1
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %L3 = load i16, i16* %Src2
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %P1 = mul i16 %L2, %L3
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   store volatile i16 %P1, i16* %Dst
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   %L32 = load i32, i32* %Src32
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   %tr = trunc i32 %L32 to i16
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %P2 = mul i16 %Arg, %tr
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   store volatile i16 %P2, i16* %Dst
+}
+
 define void @sdiv_lhs(i32 %arg32, i64 %arg64) {
   %li32 = load i32, i32* undef
   sdiv i32 %li32, %arg32