Index: lib/Target/NDS32/NDS32InstrFormats.td
===================================================================
--- lib/Target/NDS32/NDS32InstrFormats.td
+++ lib/Target/NDS32/NDS32InstrFormats.td
@@ -66,6 +66,17 @@
   let Inst{5-0} = sub_opcode;
 }
 
+class ALU_2Form_r<dag outs, dag ins, string asmstr, list<dag> pattern,
+                  bits<6> sub_opcode>
+  : 32BitInst<0b100001, outs, ins, asmstr, pattern> {
+  bits<5> Rt;
+  bits<5> Ra;
+  let Inst{24-20} = Rt;
+  let Inst{19-15} = Ra;
+  let Inst{14-6} = 0;
+  let Inst{5-0} = sub_opcode;
+}
+
 class 32Bit_5R5R15Imm<bits<6> opcode, dag outs, dag ins,
                       string asmstr, list<dag> pattern>
   : 32BitInst<opcode, outs, ins, asmstr, pattern> {
Index: lib/Target/NDS32/NDS32InstrInfo.td
===================================================================
--- lib/Target/NDS32/NDS32InstrInfo.td
+++ lib/Target/NDS32/NDS32InstrInfo.td
@@ -303,6 +303,16 @@
 
 
 //===----------------------------------------------------------------------===//
+// Performance Extension Instructions
+//
+
+def CLZ : ALU_2Form_r<(outs GPR:$Rt), (ins GPR:$Ra),
+                      "clz\t$Rt, $Ra",
+                      [(set GPR:$Rt, (ctlz GPR:$Ra))],
+                      0b000111>;
+
+
+//===----------------------------------------------------------------------===//
 // Load/Store Instructions
 //
 
Index: test/CodeGen/NDS32/count-zero.ll
===================================================================
--- /dev/null
+++ test/CodeGen/NDS32/count-zero.ll
@@ -0,0 +1,115 @@
+; RUN: llc < %s | FileCheck %s
+target datalayout = "e-m:e-p:32:32-i64:64-a:0:32-n32-S64"
+target triple = "nds32le---elf"
+
+%struct.anon = type { i32, i32 }
+
+@tab = local_unnamed_addr global [8 x %struct.anon] [%struct.anon { i32 -2147483648, i32 32 }, %struct.anon { i32 -1515870811, i32 1 }, %struct.anon { i32 1515870810, i32 2 }, %struct.anon { i32 -889323520, i32 18 }, %struct.anon { i32 32768, i32 16 }, %struct.anon { i32 42405, i32 1 }, %struct.anon { i32 23130, i32 2 }, %struct.anon { i32 3232, i32 6 }], align 4
+
+; Function Attrs: noreturn nounwind
+define i32 @count_zero() local_unnamed_addr #0 {
+entry:
+  %0 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 0, i32 0), align 4
+  %1 = tail call i32 @llvm.cttz.i32(i32 %0, i1 true)
+; CHECK: clz     $r1, $r1
+  %2 = add nuw nsw i32 %1, 1
+  %iszero = icmp eq i32 %0, 0
+  %ffs = select i1 %iszero, i32 0, i32 %2
+  %3 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 0, i32 1), align 4
+  %cmp2 = icmp eq i32 %ffs, %3
+  br i1 %cmp2, label %for.cond, label %if.then
+
+for.cond:                                         ; preds = %entry
+  %4 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 1, i32 0), align 4
+  %5 = tail call i32 @llvm.cttz.i32(i32 %4, i1 true)
+; CHECK: clz     $r3, $r3
+  %6 = add nuw nsw i32 %5, 1
+  %iszero.1 = icmp eq i32 %4, 0
+  %ffs.1 = select i1 %iszero.1, i32 0, i32 %6
+  %7 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 1, i32 1), align 4
+  %cmp2.1 = icmp eq i32 %ffs.1, %7
+  br i1 %cmp2.1, label %for.cond.1, label %if.then
+
+if.then:                                          ; preds = %for.cond.6, %for.cond.5, %for.cond.4, %for.cond.3, %for.cond.2, %for.cond.1, %for.cond, %entry
+  tail call void @abort() #3
+  unreachable
+
+for.cond.1:                                       ; preds = %for.cond
+  %8 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 2, i32 0), align 4
+  %9 = tail call i32 @llvm.cttz.i32(i32 %8, i1 true)
+; CHECK: clz     $r1, $r1
+  %10 = add nuw nsw i32 %9, 1
+  %iszero.2 = icmp eq i32 %8, 0
+  %ffs.2 = select i1 %iszero.2, i32 0, i32 %10
+  %11 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 2, i32 1), align 4
+  %cmp2.2 = icmp eq i32 %ffs.2, %11
+  br i1 %cmp2.2, label %for.cond.2, label %if.then
+
+for.cond.2:                                       ; preds = %for.cond.1
+  %12 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 3, i32 0), align 4
+  %13 = tail call i32 @llvm.cttz.i32(i32 %12, i1 true)
+; CHECK: clz     $r3, $r3
+  %14 = add nuw nsw i32 %13, 1
+  %iszero.3 = icmp eq i32 %12, 0
+  %ffs.3 = select i1 %iszero.3, i32 0, i32 %14
+  %15 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 3, i32 1), align 4
+  %cmp2.3 = icmp eq i32 %ffs.3, %15
+  br i1 %cmp2.3, label %for.cond.3, label %if.then
+
+for.cond.3:                                       ; preds = %for.cond.2
+  %16 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 4, i32 0), align 4
+  %17 = tail call i32 @llvm.cttz.i32(i32 %16, i1 true)
+; CHECK: clz     $r1, $r1
+  %18 = add nuw nsw i32 %17, 1
+  %iszero.4 = icmp eq i32 %16, 0
+  %ffs.4 = select i1 %iszero.4, i32 0, i32 %18
+  %19 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 4, i32 1), align 4
+  %cmp2.4 = icmp eq i32 %ffs.4, %19
+  br i1 %cmp2.4, label %for.cond.4, label %if.then
+
+for.cond.4:                                       ; preds = %for.cond.3
+  %20 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 5, i32 0), align 4
+  %21 = tail call i32 @llvm.cttz.i32(i32 %20, i1 true)
+; CHECK: clz     $r3, $r3
+  %22 = add nuw nsw i32 %21, 1
+  %iszero.5 = icmp eq i32 %20, 0
+  %ffs.5 = select i1 %iszero.5, i32 0, i32 %22
+  %23 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 5, i32 1), align 4
+  %cmp2.5 = icmp eq i32 %ffs.5, %23
+  br i1 %cmp2.5, label %for.cond.5, label %if.then
+
+for.cond.5:                                       ; preds = %for.cond.4
+  %24 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 6, i32 0), align 4
+  %25 = tail call i32 @llvm.cttz.i32(i32 %24, i1 true)
+; CHECK: clz     $r1, $r1
+  %26 = add nuw nsw i32 %25, 1
+  %iszero.6 = icmp eq i32 %24, 0
+  %ffs.6 = select i1 %iszero.6, i32 0, i32 %26
+  %27 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 6, i32 1), align 4
+  %cmp2.6 = icmp eq i32 %ffs.6, %27
+  br i1 %cmp2.6, label %for.cond.6, label %if.then
+
+for.cond.6:                                       ; preds = %for.cond.5
+  %28 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 7, i32 0), align 4
+  %29 = tail call i32 @llvm.cttz.i32(i32 %28, i1 true)
+; CHECK: clz     $r3, $r3
+  %30 = add nuw nsw i32 %29, 1
+  %iszero.7 = icmp eq i32 %28, 0
+  %ffs.7 = select i1 %iszero.7, i32 0, i32 %30
+  %31 = load i32, i32* getelementptr inbounds ([8 x %struct.anon], [8 x %struct.anon]* @tab, i32 0, i32 7, i32 1), align 4
+  %cmp2.7 = icmp eq i32 %ffs.7, %31
+  br i1 %cmp2.7, label %for.cond.7, label %if.then
+
+for.cond.7:                                       ; preds = %for.cond.6
+  tail call void @exit(i32 0) #3
+  unreachable
+}
+
+; Function Attrs: nounwind readnone
+declare i32 @llvm.cttz.i32(i32, i1) #1
+
+; Function Attrs: noreturn
+declare void @abort() local_unnamed_addr #2
+
+; Function Attrs: noreturn
+declare void @exit(i32) local_unnamed_addr #2