[LoongArch] Add codegen support for atomicrmw min/max operation on LA64

[gonglingqin]

Summary

This patch is required by OpenMP. After applying this patch, OpenMP regression
test passed. To reduce review difficulty caused by too large patches,
atomicrmw min/max operations on LA32 will be added later.

[XiaodongLoong]

Please rebase code to latest commit. the failed test should be OK.

llvm/lib/Target/LoongArch/LoongArchExpandAtomicPseudoInsts.cpp

439–440

Align?

447–448

Align?

[gonglingqin]

llvm/lib/Target/LoongArch/LoongArchExpandAtomicPseudoInsts.cpp

439–440

Thanks, I'll normalize the code format.

447–448

Thanks, I'll normalize the code format.

[gonglingqin]

Rebase and normalize the code format.

[SixWeining]

LGTM and let's wait for others.

[xen0n]

llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp

2413–2414

This may have room for improvement:

Firstly, this value is statically known at compile-time, so it should probably be an immediate. I looked at the surrounding code and found out some similar expressions like this one, so you may choose to refactor later.

Secondly, we have ext.w.b and ext.w.h that is usable even in LA32 mode, so it may be very worthwhile to special-case these two sizes so that the sign extension can be done in one insn/cycle.

[gonglingqin]

llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp

2413–2414

Thanks for your advice! I will improve the relevant implementation in the later patch.

llvm/include/llvm/IR/IntrinsicsLoongArch.td

 }
 
 defm int_loongarch_masked_atomicrmw_xchg : MaskedAtomicRMWIntrinsics;
 defm int_loongarch_masked_atomicrmw_add : MaskedAtomicRMWIntrinsics;
 defm int_loongarch_masked_atomicrmw_sub : MaskedAtomicRMWIntrinsics;
 defm int_loongarch_masked_atomicrmw_nand : MaskedAtomicRMWIntrinsics;
 defm int_loongarch_masked_atomicrmw_umax : MaskedAtomicRMWIntrinsics;
 defm int_loongarch_masked_atomicrmw_umin : MaskedAtomicRMWIntrinsics;
+defm int_loongarch_masked_atomicrmw_max : MaskedAtomicRMWFiveOpIntrinsics;
+defm int_loongarch_masked_atomicrmw_min : MaskedAtomicRMWFiveOpIntrinsics;
 
 // @llvm.loongarch.masked.cmpxchg.i64.<p>(
 //   ptr addr, grlen cmpval, grlen newval, grlen mask, grlenimm ordering)
 defm int_loongarch_masked_cmpxchg : MaskedAtomicRMWFiveOpIntrinsics;
 
 //===----------------------------------------------------------------------===//
 // LoongArch BASE
 
 def int_loongarch_break : Intrinsic<[], [llvm_i32_ty], [ImmArg<ArgIndex<0>>]>;
 def int_loongarch_dbar : Intrinsic<[], [llvm_i32_ty], [ImmArg<ArgIndex<0>>]>;
 def int_loongarch_ibar : Intrinsic<[], [llvm_i32_ty], [ImmArg<ArgIndex<0>>]>;
 def int_loongarch_syscall : Intrinsic<[], [llvm_i32_ty], [ImmArg<ArgIndex<0>>]>;
 
 def int_loongarch_crc_w_d_w : Intrinsic<[llvm_i32_ty],
                                         [llvm_i64_ty, llvm_i32_ty]>;
 } // TargetPrefix = "loongarch"

llvm/lib/Target/LoongArch/LoongArchExpandAtomicPseudoInsts.cpp

case LoongArch::PseudoMaskedAtomicLoadUMin32:
     return expandAtomicMinMaxOp(MBB, MBBI, AtomicRMWInst::UMin, true, 32,
                                 NextMBBI);
   case LoongArch::PseudoCmpXchg32:
     return expandAtomicCmpXchg(MBB, MBBI, false, 32, NextMBBI);
   case LoongArch::PseudoCmpXchg64:
     return expandAtomicCmpXchg(MBB, MBBI, false, 64, NextMBBI);
   case LoongArch::PseudoMaskedCmpXchg32:
     return expandAtomicCmpXchg(MBB, MBBI, true, 32, NextMBBI);
+  case LoongArch::PseudoMaskedAtomicLoadMax32:
+    return expandAtomicMinMaxOp(MBB, MBBI, AtomicRMWInst::Max, true, 32,
+                                NextMBBI);
+  case LoongArch::PseudoMaskedAtomicLoadMin32:
+    return expandAtomicMinMaxOp(MBB, MBBI, AtomicRMWInst::Min, true, 32,
+                                NextMBBI);
   }
   return false;
 }
 
 static void doAtomicBinOpExpansion(const LoongArchInstrInfo *TII,
                                    MachineInstr &MI, DebugLoc DL,
                                    MachineBasicBlock *ThisMBB,
                                    MachineBasicBlock *LoopMBB,
bool LoongArchExpandAtomicPseudo::expandAtomicBinOp(
 
   LivePhysRegs LiveRegs;
   computeAndAddLiveIns(LiveRegs, *LoopMBB);
   computeAndAddLiveIns(LiveRegs, *DoneMBB);
 
   return true;
 }
 
+static void insertSext(const LoongArchInstrInfo *TII, DebugLoc DL,
+                       MachineBasicBlock *MBB, Register ValReg,
+                       Register ShamtReg) {
+  BuildMI(MBB, DL, TII->get(LoongArch::SLL_W), ValReg)
+      .addReg(ValReg)
+      .addReg(ShamtReg);
+  BuildMI(MBB, DL, TII->get(LoongArch::SRA_W), ValReg)
+      .addReg(ValReg)
+      .addReg(ShamtReg);
+}
+
 bool LoongArchExpandAtomicPseudo::expandAtomicMinMaxOp(
     MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI,
     AtomicRMWInst::BinOp BinOp, bool IsMasked, int Width,
     MachineBasicBlock::iterator &NextMBBI) {
   assert(IsMasked == true &&
          "Should only need to expand masked atomic max/min");
   assert(Width == 32 && "Should never need to expand masked 64-bit operations");
 
case AtomicRMWInst::UMax:
     break;
   // bgeu incr, scratch2, .looptail
   case AtomicRMWInst::UMin:
     BuildMI(LoopHeadMBB, DL, TII->get(LoongArch::BGEU))
         .addReg(IncrReg)
         .addReg(Scratch2Reg)
         .addMBB(LoopTailMBB);
     break;
+  case AtomicRMWInst::Max:
+    insertSext(TII, DL, LoopHeadMBB, Scratch2Reg, MI.getOperand(6).getReg());
+    // bge scratch2, incr, .looptail
+    BuildMI(LoopHeadMBB, DL, TII->get(LoongArch::BGE))
+        .addReg(Scratch2Reg)
+        .addReg(IncrReg)
+        .addMBB(LoopTailMBB);
+    break;
+  case AtomicRMWInst::Min:
+    insertSext(TII, DL, LoopHeadMBB, Scratch2Reg, MI.getOperand(6).getReg());
+    // bge incr, scratch2, .looptail
+    BuildMI(LoopHeadMBB, DL, TII->get(LoongArch::BGE))
+        .addReg(IncrReg)
+        .addReg(Scratch2Reg)
+        .addMBB(LoopTailMBB);
+    break;
     // TODO: support other AtomicRMWInst.
   }
 
   // .loopifbody:
   //   xor scratch1, destreg, incr
   //   and scratch1, scratch1, mask
   //   xor scratch1, destreg, scratch1
   insertMaskedMerge(TII, DL, LoopIfBodyMBB, Scratch1Reg, DestReg, IncrReg,

llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp

if (GRLen == 64) {
     case AtomicRMWInst::Sub:
       return Intrinsic::loongarch_masked_atomicrmw_sub_i64;
     case AtomicRMWInst::Nand:
       return Intrinsic::loongarch_masked_atomicrmw_nand_i64;
     case AtomicRMWInst::UMax:
       return Intrinsic::loongarch_masked_atomicrmw_umax_i64;
     case AtomicRMWInst::UMin:
       return Intrinsic::loongarch_masked_atomicrmw_umin_i64;
+    case AtomicRMWInst::Max:
+      return Intrinsic::loongarch_masked_atomicrmw_max_i64;
+    case AtomicRMWInst::Min:
+      return Intrinsic::loongarch_masked_atomicrmw_min_i64;
       // TODO: support other AtomicRMWInst.
     }
   }
 
   if (GRLen == 32) {
     switch (BinOp) {
     default:
       llvm_unreachable("Unexpected AtomicRMW BinOp");
Value *LoongArchTargetLowering::emitMaskedAtomicRMWIntrinsic(
   if (GRLen == 64) {
     Incr = Builder.CreateSExt(Incr, Builder.getInt64Ty());
     Mask = Builder.CreateSExt(Mask, Builder.getInt64Ty());
     ShiftAmt = Builder.CreateSExt(ShiftAmt, Builder.getInt64Ty());
   }
 
   Value *Result;
 
+  // Must pass the shift amount needed to sign extend the loaded value prior
+  // to performing a signed comparison for min/max. ShiftAmt is the number of
+  // bits to shift the value into position. Pass GRLen-ShiftAmt-ValWidth, which
+  // is the number of bits to left+right shift the value in order to
+  // sign-extend.
+  if (AI->getOperation() == AtomicRMWInst::Min ||
+      AI->getOperation() == AtomicRMWInst::Max) {
+    const DataLayout &DL = AI->getModule()->getDataLayout();
+    unsigned ValWidth =
+        DL.getTypeStoreSizeInBits(AI->getValOperand()->getType());
+    Value *SextShamt =
+        Builder.CreateSub(Builder.getIntN(GRLen, GRLen - ValWidth), ShiftAmt);
+    Result = Builder.CreateCall(LlwOpScwLoop,
+                                {AlignedAddr, Incr, Mask, SextShamt, Ordering});
+  } else {
     Result =
         Builder.CreateCall(LlwOpScwLoop, {AlignedAddr, Incr, Mask, Ordering});
+  }
 
   if (GRLen == 64)
     Result = Builder.CreateTrunc(Result, Builder.getInt32Ty());
   return Result;
 }
 
 bool LoongArchTargetLowering::isFMAFasterThanFMulAndFAdd(
     const MachineFunction &MF, EVT VT) const {

llvm/lib/Target/LoongArch/LoongArchInstrInfo.td

class PseudoMaskedAMUMinUMax
   let mayLoad = 1;
   let mayStore = 1;
   let hasSideEffects = 0;
 }
 
 def PseudoMaskedAtomicLoadUMax32 : PseudoMaskedAMUMinUMax;
 def PseudoMaskedAtomicLoadUMin32 : PseudoMaskedAMUMinUMax;
 
+class PseudoMaskedAMMinMax
+    : Pseudo<(outs GPR:$res, GPR:$scratch1, GPR:$scratch2),
+             (ins GPR:$addr, GPR:$incr, GPR:$mask, grlenimm:$sextshamt,
+              grlenimm:$ordering)> {
+  let Constraints = "@earlyclobber $res,@earlyclobber $scratch1,"
+                    "@earlyclobber $scratch2";
+  let mayLoad = 1;
+  let mayStore = 1;
+  let hasSideEffects = 0;
+}
+
+def PseudoMaskedAtomicLoadMax32 : PseudoMaskedAMMinMax;
+def PseudoMaskedAtomicLoadMin32 : PseudoMaskedAMMinMax;
+
 /// Compare and exchange
 
 class PseudoCmpXchg
     : Pseudo<(outs GPR:$res, GPR:$scratch),
              (ins GPR:$addr, GPR:$cmpval, GPR:$newval)> {
   let Constraints = "@earlyclobber $res,@earlyclobber $scratch";
   let mayLoad = 1;
   let mayStore = 1;
   let hasSideEffects = 0;
 }
 
 def PseudoCmpXchg32 : PseudoCmpXchg;
 def PseudoCmpXchg64 : PseudoCmpXchg;
 
 def PseudoMaskedCmpXchg32
     : Pseudo<(outs GPR:$res, GPR:$scratch),
              (ins GPR:$addr, GPR:$cmpval, GPR:$newval, GPR:$mask,
               grlenimm:$ordering)> {
   let Constraints = "@earlyclobber $res,@earlyclobber $scratch";
   let mayLoad = 1;
   let mayStore = 1;
   let hasSideEffects = 0;
 }
 
+class PseudoMaskedAMMinMaxPat<Intrinsic intrin, Pseudo AMInst>
+    : Pat<(intrin GPR:$addr, GPR:$incr, GPR:$mask, GPR:$shiftamt,
+           timm:$ordering),
+          (AMInst GPR:$addr, GPR:$incr, GPR:$mask, GPR:$shiftamt,
+           timm:$ordering)>;
+
 class AtomicPat<Intrinsic intrin, Pseudo AMInst>
     : Pat<(intrin GPR:$addr, GPR:$incr, GPR:$mask, timm:$ordering),
           (AMInst GPR:$addr, GPR:$incr, GPR:$mask, timm:$ordering)>;
 
 let Predicates = [IsLA64] in {
 def : AtomicPat<int_loongarch_masked_atomicrmw_xchg_i64,
                 PseudoMaskedAtomicSwap32>;
 def : Pat<(atomic_swap_32 GPR:$addr, GPR:$incr),
def : Pat<(atomic_load_umin_32 GPR:$rj, GPR:$rk),
           (AMMIN_DB_WU GPR:$rk, GPR:$rj)>;
 def : Pat<(atomic_load_umin_64 GPR:$rj, GPR:$rk),
           (AMMIN_DB_DU GPR:$rk, GPR:$rj)>;
 def : Pat<(atomic_load_umax_32 GPR:$rj, GPR:$rk),
           (AMMAX_DB_WU GPR:$rk, GPR:$rj)>;
 def : Pat<(atomic_load_umax_64 GPR:$rj, GPR:$rk),
           (AMMAX_DB_DU GPR:$rk, GPR:$rj)>;
 
+def : Pat<(atomic_load_min_32 GPR:$rj, GPR:$rk),
+          (AMMIN_DB_W GPR:$rk, GPR:$rj)>;
+def : Pat<(atomic_load_min_64 GPR:$rj, GPR:$rk),
+          (AMMIN_DB_D GPR:$rk, GPR:$rj)>;
+def : Pat<(atomic_load_max_32 GPR:$rj, GPR:$rk),
+          (AMMAX_DB_W GPR:$rk, GPR:$rj)>;
+def : Pat<(atomic_load_max_64 GPR:$rj, GPR:$rk),
+          (AMMAX_DB_D GPR:$rk, GPR:$rj)>;
+
 def : AtomicPat<int_loongarch_masked_atomicrmw_umax_i64,
                 PseudoMaskedAtomicLoadUMax32>;
 def : AtomicPat<int_loongarch_masked_atomicrmw_umin_i64,
                 PseudoMaskedAtomicLoadUMin32>;
 
 def : Pat<(atomic_cmp_swap_64 GPR:$addr, GPR:$cmp, GPR:$new),
           (PseudoCmpXchg64 GPR:$addr, GPR:$cmp, GPR:$new)>;
 def : Pat<(int_loongarch_masked_cmpxchg_i64
             GPR:$addr, GPR:$cmpval, GPR:$newval, GPR:$mask, timm:$ordering),
           (PseudoMaskedCmpXchg32
             GPR:$addr, GPR:$cmpval, GPR:$newval, GPR:$mask, timm:$ordering)>;
 def : Pat<(atomic_cmp_swap_32 GPR:$addr, GPR:$cmp, GPR:$new),
           (PseudoCmpXchg32 GPR:$addr, GPR:$cmp, GPR:$new)>;
+
+def : PseudoMaskedAMMinMaxPat<int_loongarch_masked_atomicrmw_max_i64,
+                              PseudoMaskedAtomicLoadMax32>;
+def : PseudoMaskedAMMinMaxPat<int_loongarch_masked_atomicrmw_min_i64,
+                              PseudoMaskedAtomicLoadMin32>;
 } // Predicates = [IsLA64]
 
 defm : PseudoBinPat<"atomic_load_nand_32", PseudoAtomicLoadNand32>;
 
 let Predicates = [IsLA32] in {
 def : AtomicPat<int_loongarch_masked_atomicrmw_xchg_i32,
                 PseudoMaskedAtomicSwap32>;
 defm : PseudoBinPat<"atomic_swap_32", PseudoAtomicSwap32>;

llvm/test/CodeGen/LoongArch/ir-instruction/atomicrmw-minmax.ll

 ; LA64-LABEL: atomicrmw_umin_i64_acquire:
 ; LA64:       # %bb.0:
 ; LA64-NEXT:    ammin_db.du $a2, $a1, $a0
 ; LA64-NEXT:    move $a0, $a2
 ; LA64-NEXT:    ret
   %1 = atomicrmw umin ptr %a, i64 %b acquire
   ret i64 %1
 }
+
+define i8 @atomicrmw_max_i8_acquire(ptr %a, i8 %b) nounwind {
+; LA64-LABEL: atomicrmw_max_i8_acquire:
+; LA64:       # %bb.0:
+; LA64-NEXT:    addi.w $a2, $zero, -4
+; LA64-NEXT:    and $a2, $a0, $a2
+; LA64-NEXT:    slli.d $a0, $a0, 3
+; LA64-NEXT:    ori $a3, $zero, 255
+; LA64-NEXT:    sll.w $a3, $a3, $a0
+; LA64-NEXT:    addi.w $a3, $a3, 0
+; LA64-NEXT:    ext.w.b $a1, $a1
+; LA64-NEXT:    sll.w $a1, $a1, $a0
+; LA64-NEXT:    addi.w $a1, $a1, 0
+; LA64-NEXT:    andi $a4, $a0, 24
+; LA64-NEXT:    xori $a4, $a4, 56
+; LA64-NEXT:  .LBB8_1: # =>This Inner Loop Header: Depth=1
+; LA64-NEXT:    dbar 0
+; LA64-NEXT:    ll.w $a5, $a2, 0
+; LA64-NEXT:    and $a7, $a5, $a3
+; LA64-NEXT:    move $a6, $a5
+; LA64-NEXT:    sll.w $a7, $a7, $a4
+; LA64-NEXT:    sra.w $a7, $a7, $a4
+; LA64-NEXT:    bge $a7, $a1, .LBB8_3
+; LA64-NEXT:  # %bb.2: # in Loop: Header=BB8_1 Depth=1
+; LA64-NEXT:    xor $a6, $a5, $a1
+; LA64-NEXT:    and $a6, $a6, $a3
+; LA64-NEXT:    xor $a6, $a5, $a6
+; LA64-NEXT:  .LBB8_3: # in Loop: Header=BB8_1 Depth=1
+; LA64-NEXT:    sc.w $a6, $a2, 0
+; LA64-NEXT:    beqz $a6, .LBB8_1
+; LA64-NEXT:  # %bb.4:
+; LA64-NEXT:    dbar 1792
+; LA64-NEXT:  # %bb.5:
+; LA64-NEXT:    srl.w $a0, $a5, $a0
+; LA64-NEXT:    ret
+  %1 = atomicrmw max ptr %a, i8 %b acquire
+  ret i8 %1
+}
+
+define i16 @atomicrmw_max_i16_acquire(ptr %a, i16 %b) nounwind {
+; LA64-LABEL: atomicrmw_max_i16_acquire:
+; LA64:       # %bb.0:
+; LA64-NEXT:    addi.w $a2, $zero, -4
+; LA64-NEXT:    and $a2, $a0, $a2
+; LA64-NEXT:    slli.d $a0, $a0, 3
+; LA64-NEXT:    andi $a3, $a0, 24
+; LA64-NEXT:    ori $a4, $zero, 48
+; LA64-NEXT:    sub.d $a3, $a4, $a3
+; LA64-NEXT:    lu12i.w $a4, 15
+; LA64-NEXT:    ori $a4, $a4, 4095
+; LA64-NEXT:    sll.w $a4, $a4, $a0
+; LA64-NEXT:    addi.w $a4, $a4, 0
+; LA64-NEXT:    ext.w.h $a1, $a1
+; LA64-NEXT:    sll.w $a1, $a1, $a0
+; LA64-NEXT:    addi.w $a1, $a1, 0
+; LA64-NEXT:  .LBB9_1: # =>This Inner Loop Header: Depth=1
+; LA64-NEXT:    dbar 0
+; LA64-NEXT:    ll.w $a5, $a2, 0
+; LA64-NEXT:    and $a7, $a5, $a4
+; LA64-NEXT:    move $a6, $a5
+; LA64-NEXT:    sll.w $a7, $a7, $a3
+; LA64-NEXT:    sra.w $a7, $a7, $a3
+; LA64-NEXT:    bge $a7, $a1, .LBB9_3
+; LA64-NEXT:  # %bb.2: # in Loop: Header=BB9_1 Depth=1
+; LA64-NEXT:    xor $a6, $a5, $a1
+; LA64-NEXT:    and $a6, $a6, $a4
+; LA64-NEXT:    xor $a6, $a5, $a6
+; LA64-NEXT:  .LBB9_3: # in Loop: Header=BB9_1 Depth=1
+; LA64-NEXT:    sc.w $a6, $a2, 0
+; LA64-NEXT:    beqz $a6, .LBB9_1
+; LA64-NEXT:  # %bb.4:
+; LA64-NEXT:    dbar 1792
+; LA64-NEXT:  # %bb.5:
+; LA64-NEXT:    srl.w $a0, $a5, $a0
+; LA64-NEXT:    ret
+  %1 = atomicrmw max ptr %a, i16 %b acquire
+  ret i16 %1
+}
+
+define i32 @atomicrmw_max_i32_acquire(ptr %a, i32 %b) nounwind {
+; LA64-LABEL: atomicrmw_max_i32_acquire:
+; LA64:       # %bb.0:
+; LA64-NEXT:    ammax_db.w $a2, $a1, $a0
+; LA64-NEXT:    move $a0, $a2
+; LA64-NEXT:    ret
+  %1 = atomicrmw max ptr %a, i32 %b acquire
+  ret i32 %1
+}
+
+define i64 @atomicrmw_max_i64_acquire(ptr %a, i64 %b) nounwind {
+; LA64-LABEL: atomicrmw_max_i64_acquire:
+; LA64:       # %bb.0:
+; LA64-NEXT:    ammax_db.d $a2, $a1, $a0
+; LA64-NEXT:    move $a0, $a2
+; LA64-NEXT:    ret
+  %1 = atomicrmw max ptr %a, i64 %b acquire
+  ret i64 %1
+}
+
+define i8 @atomicrmw_min_i8_acquire(ptr %a, i8 %b) nounwind {
+; LA64-LABEL: atomicrmw_min_i8_acquire:
+; LA64:       # %bb.0:
+; LA64-NEXT:    addi.w $a2, $zero, -4
+; LA64-NEXT:    and $a2, $a0, $a2
+; LA64-NEXT:    slli.d $a0, $a0, 3
+; LA64-NEXT:    ori $a3, $zero, 255
+; LA64-NEXT:    sll.w $a3, $a3, $a0
+; LA64-NEXT:    addi.w $a3, $a3, 0
+; LA64-NEXT:    ext.w.b $a1, $a1
+; LA64-NEXT:    sll.w $a1, $a1, $a0
+; LA64-NEXT:    addi.w $a1, $a1, 0
+; LA64-NEXT:    andi $a4, $a0, 24
+; LA64-NEXT:    xori $a4, $a4, 56
+; LA64-NEXT:  .LBB12_1: # =>This Inner Loop Header: Depth=1
+; LA64-NEXT:    dbar 0
+; LA64-NEXT:    ll.w $a5, $a2, 0
+; LA64-NEXT:    and $a7, $a5, $a3
+; LA64-NEXT:    move $a6, $a5
+; LA64-NEXT:    sll.w $a7, $a7, $a4
+; LA64-NEXT:    sra.w $a7, $a7, $a4
+; LA64-NEXT:    bge $a1, $a7, .LBB12_3
+; LA64-NEXT:  # %bb.2: # in Loop: Header=BB12_1 Depth=1
+; LA64-NEXT:    xor $a6, $a5, $a1
+; LA64-NEXT:    and $a6, $a6, $a3
+; LA64-NEXT:    xor $a6, $a5, $a6
+; LA64-NEXT:  .LBB12_3: # in Loop: Header=BB12_1 Depth=1
+; LA64-NEXT:    sc.w $a6, $a2, 0
+; LA64-NEXT:    beqz $a6, .LBB12_1
+; LA64-NEXT:  # %bb.4:
+; LA64-NEXT:    dbar 1792
+; LA64-NEXT:  # %bb.5:
+; LA64-NEXT:    srl.w $a0, $a5, $a0
+; LA64-NEXT:    ret
+  %1 = atomicrmw min ptr %a, i8 %b acquire
+  ret i8 %1
+}
+
+define i16 @atomicrmw_min_i16_acquire(ptr %a, i16 %b) nounwind {
+; LA64-LABEL: atomicrmw_min_i16_acquire:
+; LA64:       # %bb.0:
+; LA64-NEXT:    addi.w $a2, $zero, -4
+; LA64-NEXT:    and $a2, $a0, $a2
+; LA64-NEXT:    slli.d $a0, $a0, 3
+; LA64-NEXT:    andi $a3, $a0, 24
+; LA64-NEXT:    ori $a4, $zero, 48
+; LA64-NEXT:    sub.d $a3, $a4, $a3
+; LA64-NEXT:    lu12i.w $a4, 15
+; LA64-NEXT:    ori $a4, $a4, 4095
+; LA64-NEXT:    sll.w $a4, $a4, $a0
+; LA64-NEXT:    addi.w $a4, $a4, 0
+; LA64-NEXT:    ext.w.h $a1, $a1
+; LA64-NEXT:    sll.w $a1, $a1, $a0
+; LA64-NEXT:    addi.w $a1, $a1, 0
+; LA64-NEXT:  .LBB13_1: # =>This Inner Loop Header: Depth=1
+; LA64-NEXT:    dbar 0
+; LA64-NEXT:    ll.w $a5, $a2, 0
+; LA64-NEXT:    and $a7, $a5, $a4
+; LA64-NEXT:    move $a6, $a5
+; LA64-NEXT:    sll.w $a7, $a7, $a3
+; LA64-NEXT:    sra.w $a7, $a7, $a3
+; LA64-NEXT:    bge $a1, $a7, .LBB13_3
+; LA64-NEXT:  # %bb.2: # in Loop: Header=BB13_1 Depth=1
+; LA64-NEXT:    xor $a6, $a5, $a1
+; LA64-NEXT:    and $a6, $a6, $a4
+; LA64-NEXT:    xor $a6, $a5, $a6
+; LA64-NEXT:  .LBB13_3: # in Loop: Header=BB13_1 Depth=1
+; LA64-NEXT:    sc.w $a6, $a2, 0
+; LA64-NEXT:    beqz $a6, .LBB13_1
+; LA64-NEXT:  # %bb.4:
+; LA64-NEXT:    dbar 1792
+; LA64-NEXT:  # %bb.5:
+; LA64-NEXT:    srl.w $a0, $a5, $a0
+; LA64-NEXT:    ret
+  %1 = atomicrmw min ptr %a, i16 %b acquire
+  ret i16 %1
+}
+
+define i32 @atomicrmw_min_i32_acquire(ptr %a, i32 %b) nounwind {
+; LA64-LABEL: atomicrmw_min_i32_acquire:
+; LA64:       # %bb.0:
+; LA64-NEXT:    ammin_db.w $a2, $a1, $a0
+; LA64-NEXT:    move $a0, $a2
+; LA64-NEXT:    ret
+  %1 = atomicrmw min ptr %a, i32 %b acquire
+  ret i32 %1
+}
+
+define i64 @atomicrmw_min_i64_acquire(ptr %a, i64 %b) nounwind {
+; LA64-LABEL: atomicrmw_min_i64_acquire:
+; LA64:       # %bb.0:
+; LA64-NEXT:    ammin_db.d $a2, $a1, $a0
+; LA64-NEXT:    move $a0, $a2
+; LA64-NEXT:    ret
+  %1 = atomicrmw min ptr %a, i64 %b acquire
+  ret i64 %1
+}