[CSKY] Optimize 'llvm.cttz.i32' and 'llvm.ctlz.i32'

[zixuan-wu]

llvm/lib/Target/CSKY/CSKYISelLowering.cpp

1406

I think no more need of Ty->isIntegerTy(Subtarget.XLen)

llvm/lib/Target/CSKY/CSKYISelLowering.h

SDValue getStaticTLSAddr(GlobalAddressSDNode *N, SelectionDAG &DAG,
                            bool UseGOT) const;
   SDValue getDynamicTLSAddr(GlobalAddressSDNode *N, SelectionDAG &DAG) const;
 
   CCAssignFn *CCAssignFnForCall(CallingConv::ID CC, bool IsVarArg) const;
   CCAssignFn *CCAssignFnForReturn(CallingConv::ID CC, bool IsVarArg) const;
 
   bool decomposeMulByConstant(LLVMContext &Context, EVT VT,
                               SDValue C) const override;
+  bool isCheapToSpeculateCttz(Type *Ty) const override;
+  bool isCheapToSpeculateCtlz(Type *Ty) const override;
 };
 
 } // namespace llvm
 
 #endif // LLVM_LIB_TARGET_CSKY_CSKYISELLOWERING_H

llvm/lib/Target/CSKY/CSKYISelLowering.cpp

if (auto *ConstNode = dyn_cast<ConstantSDNode>(C.getNode())) {
     // instruction will be generated against the above 3 cases. We leave it
     // unchanged on sub targets with MULT32, since not sure it is better.
     if (!Subtarget.hasE2() && (-1 - Imm).isPowerOf2())
       return true;
   }
 
   return false;
 }
+
+bool CSKYTargetLowering::isCheapToSpeculateCttz(Type *Ty) const {
+  return Ty->isIntegerTy(Subtarget.XLen) && Subtarget.has2E3();
+}
+
+bool CSKYTargetLowering::isCheapToSpeculateCtlz(Type *Ty) const {
+  return Ty->isIntegerTy(Subtarget.XLen) && Subtarget.hasE2();
+}

llvm/test/CodeGen/CSKY/intrinsic.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -verify-machineinstrs -csky-no-aliases -mattr=+e2 -mattr=+2e3 < %s -mtriple=csky | FileCheck %s
 
 define i32 @ctlz_0(i32 %x) {
 ; CHECK-LABEL: ctlz_0:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    bez32 a0, .LBB0_2
-; CHECK-NEXT:  # %bb.1: # %cond.false
 ; CHECK-NEXT:    ff1.32 a0, a0
 ; CHECK-NEXT:    rts16
-; CHECK-NEXT:  .LBB0_2:
-; CHECK-NEXT:    movi16 a0, 32
-; CHECK-NEXT:    rts16
 entry:
   %nlz = call i32 @llvm.ctlz.i32(i32 %x, i1 0)
   ret i32 %nlz
 }
 
 define i32 @ctlz_1(i32 %x) {
 ; CHECK-LABEL: ctlz_1:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    ff1.32 a0, a0
 ; CHECK-NEXT:    rts16
 entry:
   %nlz = call i32 @llvm.ctlz.i32(i32 %x, i1 1)
   ret i32 %nlz
 }
 
 define i32 @cttz_0(i32 %x) {
 ; CHECK-LABEL: cttz_0:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    bez32 a0, .LBB2_2
-; CHECK-NEXT:  # %bb.1: # %cond.false
 ; CHECK-NEXT:    brev32 a0, a0
 ; CHECK-NEXT:    ff1.32 a0, a0
 ; CHECK-NEXT:    rts16
-; CHECK-NEXT:  .LBB2_2:
-; CHECK-NEXT:    movi16 a0, 32
-; CHECK-NEXT:    rts16
 entry:
   %ntz = call i32 @llvm.cttz.i32(i32 %x, i1 0)
   ret i32 %ntz
 }
 
 define i32 @cttz_1(i32 %x) {
 ; CHECK-LABEL: cttz_1:
 ; CHECK:       # %bb.0: # %entry