[LoongArch] Optimize codegen for ISD::{ROTL,ROTR} (#91174)

Author: heiher

llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp

@@ -1671,10 +1671,9 @@ static LoongArchISD::NodeType getLoongArchWOpcode(unsigned Opcode) {
     return LoongArchISD::SRA_W;
   case ISD::SRL:
     return LoongArchISD::SRL_W;
+  case ISD::ROTL:
   case ISD::ROTR:
     return LoongArchISD::ROTR_W;
-  case ISD::ROTL:
-    return LoongArchISD::ROTL_W;
   case ISD::CTTZ:
     return LoongArchISD::CTZ_W;
   case ISD::CTLZ:
@@ -1704,6 +1703,10 @@ static SDValue customLegalizeToWOp(SDNode *N, SelectionDAG &DAG, int NumOp,
   case 2: {
     NewOp0 = DAG.getNode(ExtOpc, DL, MVT::i64, N->getOperand(0));
     SDValue NewOp1 = DAG.getNode(ExtOpc, DL, MVT::i64, N->getOperand(1));
+    if (N->getOpcode() == ISD::ROTL) {
+      SDValue TmpOp = DAG.getConstant(32, DL, MVT::i64);
+      NewOp1 = DAG.getNode(ISD::SUB, DL, MVT::i64, TmpOp, NewOp1);
+    }
     NewRes = DAG.getNode(WOpcode, DL, MVT::i64, NewOp0, NewOp1);
     break;
   }
@@ -1841,7 +1844,6 @@ void LoongArchTargetLowering::ReplaceNodeResults(
   case ISD::SHL:
   case ISD::SRA:
   case ISD::SRL:
-  case ISD::ROTR:
     assert(VT == MVT::i32 && Subtarget.is64Bit() &&
            "Unexpected custom legalisation");
     if (N->getOperand(1).getOpcode() != ISD::Constant) {
@@ -1850,11 +1852,10 @@ void LoongArchTargetLowering::ReplaceNodeResults(
     }
     break;
   case ISD::ROTL:
-    ConstantSDNode *CN;
-    if ((CN = dyn_cast<ConstantSDNode>(N->getOperand(1)))) {
-      Results.push_back(customLegalizeToWOp(N, DAG, 2));
-      break;
-    }
+  case ISD::ROTR:
+    assert(VT == MVT::i32 && Subtarget.is64Bit() &&
+           "Unexpected custom legalisation");
+    Results.push_back(customLegalizeToWOp(N, DAG, 2));
     break;
   case ISD::FP_TO_SINT: {
     assert(VT == MVT::i32 && Subtarget.is64Bit() &&

llvm/lib/Target/LoongArch/LoongArchInstrInfo.td

@@ -85,7 +85,6 @@ def loongarch_sll_w : SDNode<"LoongArchISD::SLL_W", SDT_LoongArchIntBinOpW>;
 def loongarch_sra_w : SDNode<"LoongArchISD::SRA_W", SDT_LoongArchIntBinOpW>;
 def loongarch_srl_w : SDNode<"LoongArchISD::SRL_W", SDT_LoongArchIntBinOpW>;
 def loongarch_rotr_w : SDNode<"LoongArchISD::ROTR_W", SDT_LoongArchIntBinOpW>;
-def loongarch_rotl_w : SDNode<"LoongArchISD::ROTL_W", SDT_LoongArchIntBinOpW>;
 def loongarch_crc_w_b_w
     : SDNode<"LoongArchISD::CRC_W_B_W", SDT_LoongArchIntBinOpW, [SDNPHasChain]>;
 def loongarch_crc_w_h_w
@@ -1116,12 +1115,10 @@ def : PatGprGpr<srem, MOD_D>;
 def : PatGprGpr<urem, MOD_DU>;
 def : PatGprGpr<rotr, ROTR_D>;
 def : PatGprGpr<loongarch_rotr_w, ROTR_W>;
+def : PatGprGpr_32<rotr, ROTR_W>;
 def : PatGprImm<rotr, ROTRI_D, uimm6>;
 def : PatGprImm_32<rotr, ROTRI_W, uimm5>;
-def : Pat<(loongarch_rotl_w GPR:$rj, uimm5:$imm),
-          (ROTRI_W GPR:$rj, (ImmSubFrom32 uimm5:$imm))>;
-def : Pat<(sext_inreg (loongarch_rotl_w GPR:$rj, uimm5:$imm), i32),
-          (ROTRI_W GPR:$rj, (ImmSubFrom32 uimm5:$imm))>;
+def : PatGprImm<loongarch_rotr_w, ROTRI_W, uimm5>;
 // TODO: Select "_W[U]" instructions for i32xi32 if only lower 32 bits of the
 // product are used.
 def : PatGprGpr<mul, MUL_D>;

llvm/test/CodeGen/LoongArch/rotl-rotr.ll

@@ -2,8 +2,6 @@
 ; RUN: llc --mtriple=loongarch32 < %s | FileCheck %s --check-prefix=LA32
 ; RUN: llc --mtriple=loongarch64 < %s | FileCheck %s --check-prefix=LA64
 
-;; TODO: Add optimization to ISD::ROTL
-
 define signext i32 @rotl_32(i32 signext %x, i32 signext %y) nounwind {
 ; LA32-LABEL: rotl_32:
 ; LA32:       # %bb.0:
@@ -14,10 +12,9 @@ define signext i32 @rotl_32(i32 signext %x, i32 signext %y) nounwind {
 ;
 ; LA64-LABEL: rotl_32:
 ; LA64:       # %bb.0:
-; LA64-NEXT:    sll.w $a2, $a0, $a1
-; LA64-NEXT:    sub.d $a1, $zero, $a1
-; LA64-NEXT:    srl.w $a0, $a0, $a1
-; LA64-NEXT:    or $a0, $a2, $a0
+; LA64-NEXT:    ori $a2, $zero, 32
+; LA64-NEXT:    sub.d $a1, $a2, $a1
+; LA64-NEXT:    rotr.w $a0, $a0, $a1
 ; LA64-NEXT:    ret
   %z = sub i32 32, %y
   %b = shl i32 %x, %y
@@ -152,10 +149,9 @@ define signext i32 @rotl_32_mask(i32 signext %x, i32 signext %y) nounwind {
 ;
 ; LA64-LABEL: rotl_32_mask:
 ; LA64:       # %bb.0:
-; LA64-NEXT:    sll.w $a2, $a0, $a1
-; LA64-NEXT:    sub.d $a1, $zero, $a1
-; LA64-NEXT:    srl.w $a0, $a0, $a1
-; LA64-NEXT:    or $a0, $a2, $a0
+; LA64-NEXT:    ori $a2, $zero, 32
+; LA64-NEXT:    sub.d $a1, $a2, $a1
+; LA64-NEXT:    rotr.w $a0, $a0, $a1
 ; LA64-NEXT:    ret
   %z = sub i32 0, %y
   %and = and i32 %z, 31
@@ -174,10 +170,9 @@ define signext i32 @rotl_32_mask_and_63_and_31(i32 signext %x, i32 signext %y) n
 ;
 ; LA64-LABEL: rotl_32_mask_and_63_and_31:
 ; LA64:       # %bb.0:
-; LA64-NEXT:    sll.w $a2, $a0, $a1
-; LA64-NEXT:    sub.d $a1, $zero, $a1
-; LA64-NEXT:    srl.w $a0, $a0, $a1
-; LA64-NEXT:    or $a0, $a2, $a0
+; LA64-NEXT:    ori $a2, $zero, 32
+; LA64-NEXT:    sub.d $a1, $a2, $a1
+; LA64-NEXT:    rotr.w $a0, $a0, $a1
 ; LA64-NEXT:    ret
   %a = and i32 %y, 63
   %b = shl i32 %x, %a
@@ -197,10 +192,9 @@ define signext i32 @rotl_32_mask_or_64_or_32(i32 signext %x, i32 signext %y) nou
 ;
 ; LA64-LABEL: rotl_32_mask_or_64_or_32:
 ; LA64:       # %bb.0:
-; LA64-NEXT:    sll.w $a2, $a0, $a1
-; LA64-NEXT:    sub.d $a1, $zero, $a1
-; LA64-NEXT:    srl.w $a0, $a0, $a1
-; LA64-NEXT:    or $a0, $a2, $a0
+; LA64-NEXT:    ori $a2, $zero, 32
+; LA64-NEXT:    sub.d $a1, $a2, $a1
+; LA64-NEXT:    rotr.w $a0, $a0, $a1
 ; LA64-NEXT:    ret
   %a = or i32 %y, 64
   %b = shl i32 %x, %a
@@ -591,10 +585,7 @@ define signext i32 @rotr_i32_fshr(i32 signext %a) nounwind {
 ;
 ; LA64-LABEL: rotr_i32_fshr:
 ; LA64:       # %bb.0:
-; LA64-NEXT:    slli.d $a1, $a0, 20
-; LA64-NEXT:    bstrpick.d $a0, $a0, 31, 12
-; LA64-NEXT:    or $a0, $a0, $a1
-; LA64-NEXT:    addi.w $a0, $a0, 0
+; LA64-NEXT:    rotri.w $a0, $a0, 12
 ; LA64-NEXT:    ret
   %or = tail call i32 @llvm.fshr.i32(i32 %a, i32 %a, i32 12)
   ret i32 %or