Index: lib/Transforms/InstCombine/InstCombineCalls.cpp
===================================================================
--- lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -2083,6 +2083,20 @@
       if (match(Op0, m_ZeroInt()) || match(Op0, m_Undef()))
         return BinaryOperator::CreateLShr(Op1,
                                           ConstantExpr::getSub(WidthC, ShAmtC));
+
+      // try fshl((A & B), (C & D), E) -> lshr/shl
+      Value *A, *C;
+      const APInt *B, *D;
+      if (match(II->getArgOperand(0), m_c_And(m_Value(A), m_APInt(B))) &&
+          match(II->getArgOperand(1), m_c_And(m_Value(C), m_APInt(D))) &&
+          A == C) {
+        APInt BAndD = (*B & *D), E = dyn_cast<ConstantInt>(ShAmtC)->getValue();
+        if (BAndD.countLeadingZeros() >= E.getLimitedValue())
+          return BinaryOperator::CreateShl(Op0, ShAmtC);
+        if (BAndD.countTrailingZeros() >= BitWidth - E.getLimitedValue())
+          return BinaryOperator::CreateLShr(
+            Op1, ConstantInt::get(II->getContext(), APInt(BitWidth, BitWidth) - E));
+      }
     }
 
     // The shift amount (operand 2) of a funnel shift is modulo the bitwidth,
Index: test/Transforms/InstCombine/fsh.ll
===================================================================
--- test/Transforms/InstCombine/fsh.ll
+++ test/Transforms/InstCombine/fsh.ll
@@ -558,4 +558,41 @@
   %r = call i16 @llvm.fshr.i16(i16 %x, i16 %x, i16 8)
   ret i16 %r
 }
-
+define i32 @fshl_mask_args_same1(i32 %a) {
+; CHECK-LABEL: @fshl_mask_args_same1(
+; CHECK-NEXT:    [[TMP2:%.*]] = lshr i32 [[A:%.*]], 16
+; CHECK-NEXT:    ret i32 [[TMP2]]
+;
+  %tmp1 = and i32 %a, 4294901760 ; 0xffff0000
+  %tmp2 = call i32 @llvm.fshl.i32(i32 %tmp1, i32 %tmp1, i32 16)
+  ret i32 %tmp2
+}
+define i32 @fshl_mask_args_same2(i32 %a) {
+; CHECK-LABEL: @fshl_mask_args_same2(
+; CHECK-NEXT:    [[TMP1:%.*]] = shl i32 [[A:%.*]], 8
+; CHECK-NEXT:    [[TMP2:%.*]] = and i32 [[TMP1]], 65280
+; CHECK-NEXT:    ret i32 [[TMP2]]
+;
+  %tmp1 = and i32 %a, 255
+  %tmp2 = call i32 @llvm.fshl.i32(i32 %tmp1, i32 %tmp1, i32 8)
+  ret i32 %tmp2
+}
+define i32 @fshl_mask_args_same_3(i32 %a) {
+; CHECK-LABEL: @fshl_mask_args_same_3(
+; CHECK-NEXT:    [[TMP2:%.*]] = shl i32 [[A:%.*]], 24
+; CHECK-NEXT:    ret i32 [[TMP2]]
+;
+  %tmp1 = and i32 %a, 255
+  %tmp2 = call i32 @llvm.fshl.i32(i32 %tmp1, i32 %tmp1, i32 24)
+  ret i32 %tmp2
+}
+define i32 @fshl_mask_args_same4(i32 %a) {
+; CHECK-LABEL: @fshl_mask_args_same4(
+; CHECK-NEXT:    [[TMP1:%.*]] = lshr i32 [[A:%.*]], 8
+; CHECK-NEXT:    [[TMP2:%.*]] = and i32 [[TMP1]], 16711680
+; CHECK-NEXT:    ret i32 [[TMP2]]
+;
+  %tmp1 = and i32 %a, 4278190080 ; 0xff000000
+  %tmp2 = call i32 @llvm.fshl.i32(i32 %tmp1, i32 %tmp1, i32 24)
+  ret i32 %tmp2
+}