Merging r279933:

------------------------------------------------------------------------
r279933 | hfinkel | 2016-08-28 09:17:58 -0700 (Sun, 28 Aug 2016) | 4 lines

[PowerPC] Implement lowering for atomicrmw min/max/umin/umax

Implement lowering for atomicrmw min/max/umin/umax. Fixes PR28818.

------------------------------------------------------------------------

llvm-svn: 287810

5 files changed, 587 insertions, 5 deletions

diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index 34c9009..b33efb7 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -8385,7 +8385,9 @@ Instruction* PPCTargetLowering::emitTrailingFence(IRBuilder<> &Builder,
 MachineBasicBlock *
 PPCTargetLowering::EmitAtomicBinary(MachineInstr &MI, MachineBasicBlock *BB,
                                     unsigned AtomicSize,
-                                    unsigned BinOpcode) const {
+                                    unsigned BinOpcode,
+                                    unsigned CmpOpcode,
+                                    unsigned CmpPred) const {
   // This also handles ATOMIC_SWAP, indicated by BinOpcode==0.
   const TargetInstrInfo *TII = Subtarget.getInstrInfo();
 
@@ -8425,8 +8427,12 @@ PPCTargetLowering::EmitAtomicBinary(MachineInstr &MI, MachineBasicBlock *BB,
   DebugLoc dl = MI.getDebugLoc();
 
   MachineBasicBlock *loopMBB = F->CreateMachineBasicBlock(LLVM_BB);
+  MachineBasicBlock *loop2MBB =
+    CmpOpcode ? F->CreateMachineBasicBlock(LLVM_BB) : nullptr;
   MachineBasicBlock *exitMBB = F->CreateMachineBasicBlock(LLVM_BB);
   F->insert(It, loopMBB);
+  if (CmpOpcode)
+    F->insert(It, loop2MBB);
   F->insert(It, exitMBB);
   exitMBB->splice(exitMBB->begin(), BB,
                   std::next(MachineBasicBlock::iterator(MI)), BB->end());
@@ -8448,11 +8454,31 @@ PPCTargetLowering::EmitAtomicBinary(MachineInstr &MI, MachineBasicBlock *BB,
   //   st[wd]cx. r0, ptr
   //   bne- loopMBB
   //   fallthrough --> exitMBB
+
+  // For max/min...
+  //  loopMBB:
+  //   l[wd]arx dest, ptr
+  //   cmpl?[wd] incr, dest
+  //   bgt exitMBB
+  //  loop2MBB:
+  //   st[wd]cx. dest, ptr
+  //   bne- loopMBB
+  //   fallthrough --> exitMBB
+
   BB = loopMBB;
   BuildMI(BB, dl, TII->get(LoadMnemonic), dest)
     .addReg(ptrA).addReg(ptrB);
   if (BinOpcode)
     BuildMI(BB, dl, TII->get(BinOpcode), TmpReg).addReg(incr).addReg(dest);
+  if (CmpOpcode) {
+    BuildMI(BB, dl, TII->get(CmpOpcode), PPC::CR0)
+      .addReg(incr).addReg(dest);
+    BuildMI(BB, dl, TII->get(PPC::BCC))
+      .addImm(CmpPred).addReg(PPC::CR0).addMBB(exitMBB);
+    BB->addSuccessor(loop2MBB);
+    BB->addSuccessor(exitMBB);
+    BB = loop2MBB;
+  }
   BuildMI(BB, dl, TII->get(StoreMnemonic))
     .addReg(TmpReg).addReg(ptrA).addReg(ptrB);
   BuildMI(BB, dl, TII->get(PPC::BCC))
@@ -8470,10 +8496,13 @@ MachineBasicBlock *
 PPCTargetLowering::EmitPartwordAtomicBinary(MachineInstr &MI,
                                             MachineBasicBlock *BB,
                                             bool is8bit, // operation
-                                            unsigned BinOpcode) const {
+                                            unsigned BinOpcode,
+                                            unsigned CmpOpcode,
+                                            unsigned CmpPred) const {
   // If we support part-word atomic mnemonics, just use them
   if (Subtarget.hasPartwordAtomics())
-    return EmitAtomicBinary(MI, BB, is8bit ? 1 : 2, BinOpcode);
+    return EmitAtomicBinary(MI, BB, is8bit ? 1 : 2, BinOpcode,
+                            CmpOpcode, CmpPred);
 
   // This also handles ATOMIC_SWAP, indicated by BinOpcode==0.
   const TargetInstrInfo *TII = Subtarget.getInstrInfo();
@@ -8495,8 +8524,12 @@ PPCTargetLowering::EmitPartwordAtomicBinary(MachineInstr &MI,
   DebugLoc dl = MI.getDebugLoc();
 
   MachineBasicBlock *loopMBB = F->CreateMachineBasicBlock(LLVM_BB);
+  MachineBasicBlock *loop2MBB =
+    CmpOpcode ? F->CreateMachineBasicBlock(LLVM_BB) : nullptr;
   MachineBasicBlock *exitMBB = F->CreateMachineBasicBlock(LLVM_BB);
   F->insert(It, loopMBB);
+  if (CmpOpcode)
+    F->insert(It, loop2MBB);
   F->insert(It, exitMBB);
   exitMBB->splice(exitMBB->begin(), BB,
                   std::next(MachineBasicBlock::iterator(MI)), BB->end());
@@ -8581,6 +8614,32 @@ PPCTargetLowering::EmitPartwordAtomicBinary(MachineInstr &MI,
     .addReg(TmpDestReg).addReg(MaskReg);
   BuildMI(BB, dl, TII->get(is64bit ? PPC::AND8 : PPC::AND), Tmp3Reg)
     .addReg(TmpReg).addReg(MaskReg);
+  if (CmpOpcode) {
+    // For unsigned comparisons, we can directly compare the shifted values.
+    // For signed comparisons we shift and sign extend.
+    unsigned SReg = RegInfo.createVirtualRegister(RC);
+    BuildMI(BB, dl, TII->get(is64bit ? PPC::AND8 : PPC::AND), SReg)
+      .addReg(TmpDestReg).addReg(MaskReg);
+    unsigned ValueReg = SReg;
+    unsigned CmpReg = Incr2Reg;
+    if (CmpOpcode == PPC::CMPW) {
+      ValueReg = RegInfo.createVirtualRegister(RC);
+      BuildMI(BB, dl, TII->get(PPC::SRW), ValueReg)
+        .addReg(SReg).addReg(ShiftReg);
+      unsigned ValueSReg = RegInfo.createVirtualRegister(RC);
+      BuildMI(BB, dl, TII->get(is8bit ? PPC::EXTSB : PPC::EXTSH), ValueSReg)
+        .addReg(ValueReg);
+      ValueReg = ValueSReg;
+      CmpReg = incr;
+    }
+    BuildMI(BB, dl, TII->get(CmpOpcode), PPC::CR0)
+      .addReg(CmpReg).addReg(ValueReg);
+    BuildMI(BB, dl, TII->get(PPC::BCC))
+      .addImm(CmpPred).addReg(PPC::CR0).addMBB(exitMBB);
+    BB->addSuccessor(loop2MBB);
+    BB->addSuccessor(exitMBB);
+    BB = loop2MBB;
+  }
   BuildMI(BB, dl, TII->get(is64bit ? PPC::OR8 : PPC::OR), Tmp4Reg)
     .addReg(Tmp3Reg).addReg(Tmp2Reg);
   BuildMI(BB, dl, TII->get(PPC::STWCX))
@@ -9087,6 +9146,42 @@ PPCTargetLowering::EmitInstrWithCustomInserter(MachineInstr &MI,
   else if (MI.getOpcode() == PPC::ATOMIC_LOAD_SUB_I64)
     BB = EmitAtomicBinary(MI, BB, 8, PPC::SUBF8);
 
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MIN_I8)
+    BB = EmitPartwordAtomicBinary(MI, BB, true, 0, PPC::CMPW, PPC::PRED_GE);
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MIN_I16)
+    BB = EmitPartwordAtomicBinary(MI, BB, false, 0, PPC::CMPW, PPC::PRED_GE);
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MIN_I32)
+    BB = EmitAtomicBinary(MI, BB, 4, 0, PPC::CMPW, PPC::PRED_GE);
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MIN_I64)
+    BB = EmitAtomicBinary(MI, BB, 8, 0, PPC::CMPD, PPC::PRED_GE);
+
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MAX_I8)
+    BB = EmitPartwordAtomicBinary(MI, BB, true, 0, PPC::CMPW, PPC::PRED_LE);
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MAX_I16)
+    BB = EmitPartwordAtomicBinary(MI, BB, false, 0, PPC::CMPW, PPC::PRED_LE);
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MAX_I32)
+    BB = EmitAtomicBinary(MI, BB, 4, 0, PPC::CMPW, PPC::PRED_LE);
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MAX_I64)
+    BB = EmitAtomicBinary(MI, BB, 8, 0, PPC::CMPD, PPC::PRED_LE);
+
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMIN_I8)
+    BB = EmitPartwordAtomicBinary(MI, BB, true, 0, PPC::CMPLW, PPC::PRED_GE);
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMIN_I16)
+    BB = EmitPartwordAtomicBinary(MI, BB, false, 0, PPC::CMPLW, PPC::PRED_GE);
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMIN_I32)
+    BB = EmitAtomicBinary(MI, BB, 4, 0, PPC::CMPLW, PPC::PRED_GE);
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMIN_I64)
+    BB = EmitAtomicBinary(MI, BB, 8, 0, PPC::CMPLD, PPC::PRED_GE);
+
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMAX_I8)
+    BB = EmitPartwordAtomicBinary(MI, BB, true, 0, PPC::CMPLW, PPC::PRED_LE);
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMAX_I16)
+    BB = EmitPartwordAtomicBinary(MI, BB, false, 0, PPC::CMPLW, PPC::PRED_LE);
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMAX_I32)
+    BB = EmitAtomicBinary(MI, BB, 4, 0, PPC::CMPLW, PPC::PRED_LE);
+  else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMAX_I64)
+    BB = EmitAtomicBinary(MI, BB, 8, 0, PPC::CMPLD, PPC::PRED_LE);
+
   else if (MI.getOpcode() == PPC::ATOMIC_SWAP_I8)
     BB = EmitPartwordAtomicBinary(MI, BB, true, 0);
   else if (MI.getOpcode() == PPC::ATOMIC_SWAP_I16)
diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.h b/llvm/lib/Target/PowerPC/PPCISelLowering.h
index 433aca1..e605045 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.h
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.h
@@ -585,11 +585,15 @@ namespace llvm {
     MachineBasicBlock *EmitAtomicBinary(MachineInstr &MI,
                                         MachineBasicBlock *MBB,
                                         unsigned AtomicSize,
-                                        unsigned BinOpcode) const;
+                                        unsigned BinOpcode,
+                                        unsigned CmpOpcode = 0,
+                                        unsigned CmpPred = 0) const;
     MachineBasicBlock *EmitPartwordAtomicBinary(MachineInstr &MI,
                                                 MachineBasicBlock *MBB,
                                                 bool is8bit,
-                                                unsigned Opcode) const;
+                                                unsigned Opcode,
+                                                unsigned CmpOpcode = 0,
+                                                unsigned CmpPred = 0) const;
 
     MachineBasicBlock *emitEHSjLjSetJmp(MachineInstr &MI,
                                         MachineBasicBlock *MBB) const;
diff --git a/llvm/lib/Target/PowerPC/PPCInstr64Bit.td b/llvm/lib/Target/PowerPC/PPCInstr64Bit.td
index e7eb8a1..5e514c8 100644
--- a/llvm/lib/Target/PowerPC/PPCInstr64Bit.td
+++ b/llvm/lib/Target/PowerPC/PPCInstr64Bit.td
@@ -224,6 +224,18 @@ let usesCustomInserter = 1 in {
     def ATOMIC_LOAD_NAND_I64 : Pseudo<
       (outs g8rc:$dst), (ins memrr:$ptr, g8rc:$incr), "#ATOMIC_LOAD_NAND_I64",
       [(set i64:$dst, (atomic_load_nand_64 xoaddr:$ptr, i64:$incr))]>;
+    def ATOMIC_LOAD_MIN_I64 : Pseudo<
+      (outs g8rc:$dst), (ins memrr:$ptr, g8rc:$incr), "#ATOMIC_LOAD_MIN_I64",
+      [(set i64:$dst, (atomic_load_min_64 xoaddr:$ptr, i64:$incr))]>;
+    def ATOMIC_LOAD_MAX_I64 : Pseudo<
+      (outs g8rc:$dst), (ins memrr:$ptr, g8rc:$incr), "#ATOMIC_LOAD_MAX_I64",
+      [(set i64:$dst, (atomic_load_max_64 xoaddr:$ptr, i64:$incr))]>;
+    def ATOMIC_LOAD_UMIN_I64 : Pseudo<
+      (outs g8rc:$dst), (ins memrr:$ptr, g8rc:$incr), "#ATOMIC_LOAD_UMIN_I64",
+      [(set i64:$dst, (atomic_load_umin_64 xoaddr:$ptr, i64:$incr))]>;
+    def ATOMIC_LOAD_UMAX_I64 : Pseudo<
+      (outs g8rc:$dst), (ins memrr:$ptr, g8rc:$incr), "#ATOMIC_LOAD_UMAX_I64",
+      [(set i64:$dst, (atomic_load_umax_64 xoaddr:$ptr, i64:$incr))]>;
 
     def ATOMIC_CMP_SWAP_I64 : Pseudo<
       (outs g8rc:$dst), (ins memrr:$ptr, g8rc:$old, g8rc:$new), "#ATOMIC_CMP_SWAP_I64",
diff --git a/llvm/lib/Target/PowerPC/PPCInstrInfo.td b/llvm/lib/Target/PowerPC/PPCInstrInfo.td
index 4a42a94..a40d4e1 100644
--- a/llvm/lib/Target/PowerPC/PPCInstrInfo.td
+++ b/llvm/lib/Target/PowerPC/PPCInstrInfo.td
@@ -1509,6 +1509,18 @@ let usesCustomInserter = 1 in {
     def ATOMIC_LOAD_NAND_I8 : Pseudo<
       (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_NAND_I8",
       [(set i32:$dst, (atomic_load_nand_8 xoaddr:$ptr, i32:$incr))]>;
+    def ATOMIC_LOAD_MIN_I8 : Pseudo<
+      (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_MIN_I8",
+      [(set i32:$dst, (atomic_load_min_8 xoaddr:$ptr, i32:$incr))]>;
+    def ATOMIC_LOAD_MAX_I8 : Pseudo<
+      (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_MAX_I8",
+      [(set i32:$dst, (atomic_load_max_8 xoaddr:$ptr, i32:$incr))]>;
+    def ATOMIC_LOAD_UMIN_I8 : Pseudo<
+      (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_UMIN_I8",
+      [(set i32:$dst, (atomic_load_umin_8 xoaddr:$ptr, i32:$incr))]>;
+    def ATOMIC_LOAD_UMAX_I8 : Pseudo<
+      (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_UMAX_I8",
+      [(set i32:$dst, (atomic_load_umax_8 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_ADD_I16 : Pseudo<
       (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_ADD_I16",
       [(set i32:$dst, (atomic_load_add_16 xoaddr:$ptr, i32:$incr))]>;
@@ -1527,6 +1539,18 @@ let usesCustomInserter = 1 in {
     def ATOMIC_LOAD_NAND_I16 : Pseudo<
       (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_NAND_I16",
       [(set i32:$dst, (atomic_load_nand_16 xoaddr:$ptr, i32:$incr))]>;
+    def ATOMIC_LOAD_MIN_I16 : Pseudo<
+      (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_MIN_I16",
+      [(set i32:$dst, (atomic_load_min_16 xoaddr:$ptr, i32:$incr))]>;
+    def ATOMIC_LOAD_MAX_I16 : Pseudo<
+      (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_MAX_I16",
+      [(set i32:$dst, (atomic_load_max_16 xoaddr:$ptr, i32:$incr))]>;
+    def ATOMIC_LOAD_UMIN_I16 : Pseudo<
+      (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_UMIN_I16",
+      [(set i32:$dst, (atomic_load_umin_16 xoaddr:$ptr, i32:$incr))]>;
+    def ATOMIC_LOAD_UMAX_I16 : Pseudo<
+      (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_UMAX_I16",
+      [(set i32:$dst, (atomic_load_umax_16 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_ADD_I32 : Pseudo<
       (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_ADD_I32",
       [(set i32:$dst, (atomic_load_add_32 xoaddr:$ptr, i32:$incr))]>;
@@ -1545,6 +1569,18 @@ let usesCustomInserter = 1 in {
     def ATOMIC_LOAD_NAND_I32 : Pseudo<
       (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_NAND_I32",
       [(set i32:$dst, (atomic_load_nand_32 xoaddr:$ptr, i32:$incr))]>;
+    def ATOMIC_LOAD_MIN_I32 : Pseudo<
+      (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_MIN_I32",
+      [(set i32:$dst, (atomic_load_min_32 xoaddr:$ptr, i32:$incr))]>;
+    def ATOMIC_LOAD_MAX_I32 : Pseudo<
+      (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_MAX_I32",
+      [(set i32:$dst, (atomic_load_max_32 xoaddr:$ptr, i32:$incr))]>;
+    def ATOMIC_LOAD_UMIN_I32 : Pseudo<
+      (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_UMIN_I32",
+      [(set i32:$dst, (atomic_load_umin_32 xoaddr:$ptr, i32:$incr))]>;
+    def ATOMIC_LOAD_UMAX_I32 : Pseudo<
+      (outs gprc:$dst), (ins memrr:$ptr, gprc:$incr), "#ATOMIC_LOAD_UMAX_I32",
+      [(set i32:$dst, (atomic_load_umax_32 xoaddr:$ptr, i32:$incr))]>;
 
     def ATOMIC_CMP_SWAP_I8 : Pseudo<
       (outs gprc:$dst), (ins memrr:$ptr, gprc:$old, gprc:$new), "#ATOMIC_CMP_SWAP_I8",
diff --git a/llvm/test/CodeGen/PowerPC/atomic-minmax.ll b/llvm/test/CodeGen/PowerPC/atomic-minmax.ll
new file mode 100644
index 0000000..5b9a153
--- /dev/null
+++ b/llvm/test/CodeGen/PowerPC/atomic-minmax.ll
@@ -0,0 +1,435 @@
+; RUN: llc < %s | FileCheck %s
+target datalayout = "E-m:e-i64:64-n32:64"
+target triple = "powerpc64-unknown-linux-gnu"
+
+define void @a32min(i32* nocapture dereferenceable(4) %minimum, i32 %val) #0 {
+entry:
+  %0 = atomicrmw min i32* %minimum, i32 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a32min
+; CHECK: lwarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmpw 4, [[OLDV]]
+; CHECK: bgelr 0
+; CHECK: stwcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @a32max(i32* nocapture dereferenceable(4) %minimum, i32 %val) #0 {
+entry:
+  %0 = atomicrmw max i32* %minimum, i32 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a32max
+; CHECK: lwarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmpw 4, [[OLDV]]
+; CHECK: blelr 0
+; CHECK: stwcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @a32umin(i32* nocapture dereferenceable(4) %minimum, i32 %val) #0 {
+entry:
+  %0 = atomicrmw umin i32* %minimum, i32 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a32umin
+; CHECK: lwarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmplw 4, [[OLDV]]
+; CHECK: bgelr 0
+; CHECK: stwcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @a32umax(i32* nocapture dereferenceable(4) %minimum, i32 %val) #0 {
+entry:
+  %0 = atomicrmw umax i32* %minimum, i32 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a32umax
+; CHECK: lwarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmplw 4, [[OLDV]]
+; CHECK: blelr 0
+; CHECK: stwcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @a16min(i16* nocapture dereferenceable(4) %minimum, i16 %val) #1 {
+entry:
+  %0 = atomicrmw min i16* %minimum, i16 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a16min
+; CHECK: lharx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmpw 4, [[OLDV]]
+; CHECK: bgelr 0
+; CHECK: sthcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @a16max(i16* nocapture dereferenceable(4) %minimum, i16 %val) #1 {
+entry:
+  %0 = atomicrmw max i16* %minimum, i16 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a16max
+; CHECK: lharx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmpw 4, [[OLDV]]
+; CHECK: blelr 0
+; CHECK: sthcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @a16umin(i16* nocapture dereferenceable(4) %minimum, i16 %val) #1 {
+entry:
+  %0 = atomicrmw umin i16* %minimum, i16 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a16umin
+; CHECK: lharx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmplw 4, [[OLDV]]
+; CHECK: bgelr 0
+; CHECK: sthcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @a16umax(i16* nocapture dereferenceable(4) %minimum, i16 %val) #1 {
+entry:
+  %0 = atomicrmw umax i16* %minimum, i16 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a16umax
+; CHECK: lharx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmplw 4, [[OLDV]]
+; CHECK: blelr 0
+; CHECK: sthcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @a8min(i8* nocapture dereferenceable(4) %minimum, i8 %val) #1 {
+entry:
+  %0 = atomicrmw min i8* %minimum, i8 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a8min
+; CHECK: lbarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmpw 4, [[OLDV]]
+; CHECK: bgelr 0
+; CHECK: stbcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @a8max(i8* nocapture dereferenceable(4) %minimum, i8 %val) #1 {
+entry:
+  %0 = atomicrmw max i8* %minimum, i8 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a8max
+; CHECK: lbarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmpw 4, [[OLDV]]
+; CHECK: blelr 0
+; CHECK: stbcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @a8umin(i8* nocapture dereferenceable(4) %minimum, i8 %val) #1 {
+entry:
+  %0 = atomicrmw umin i8* %minimum, i8 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a8umin
+; CHECK: lbarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmplw 4, [[OLDV]]
+; CHECK: bgelr 0
+; CHECK: stbcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @a8umax(i8* nocapture dereferenceable(4) %minimum, i8 %val) #1 {
+entry:
+  %0 = atomicrmw umax i8* %minimum, i8 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a8umax
+; CHECK: lbarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmplw 4, [[OLDV]]
+; CHECK: blelr 0
+; CHECK: stbcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @a64min(i64* nocapture dereferenceable(4) %minimum, i64 %val) #0 {
+entry:
+  %0 = atomicrmw min i64* %minimum, i64 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a64min
+; CHECK: ldarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmpd 4, [[OLDV]]
+; CHECK: bgelr 0
+; CHECK: stdcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @a64max(i64* nocapture dereferenceable(4) %minimum, i64 %val) #0 {
+entry:
+  %0 = atomicrmw max i64* %minimum, i64 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a64max
+; CHECK: ldarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmpd 4, [[OLDV]]
+; CHECK: blelr 0
+; CHECK: stdcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @a64umin(i64* nocapture dereferenceable(4) %minimum, i64 %val) #0 {
+entry:
+  %0 = atomicrmw umin i64* %minimum, i64 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a64umin
+; CHECK: ldarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmpld 4, [[OLDV]]
+; CHECK: bgelr 0
+; CHECK: stdcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @a64umax(i64* nocapture dereferenceable(4) %minimum, i64 %val) #0 {
+entry:
+  %0 = atomicrmw umax i64* %minimum, i64 %val monotonic
+  ret void
+
+; CHECK-LABEL: @a64umax
+; CHECK: ldarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: cmpld 4, [[OLDV]]
+; CHECK: blelr 0
+; CHECK: stdcx. 4, 0, 3
+; CHECK: bne 0,
+; CHECK: blr
+}
+
+define void @ae16min(i16* nocapture dereferenceable(4) %minimum, i16 %val) #0 {
+entry:
+  %0 = atomicrmw min i16* %minimum, i16 %val monotonic
+  ret void
+
+; CHECK-LABEL: @ae16min
+; CHECK-DAG: rlwinm [[SA1:[0-9]+]], 3, 3, 27, 27
+; CHECK-DAG: li [[M1:[0-9]+]], 0
+; CHECK-DAG: rldicr 3, 3, 0, 61
+; CHECK-DAG: xori [[SA:[0-9]+]], [[SA1]], 16
+; CHECK-DAG: ori [[M2:[0-9]+]], [[M1]], 65535
+; CHECK-DAG: slw [[SV:[0-9]+]], 4, [[SA]]
+; CHECK-DAG: slw [[M:[0-9]+]], [[M2]], [[SA]]
+; CHECK-DAG: and [[SMV:[0-9]+]], [[SV]], [[M]]
+; CHECK: lwarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: and [[MOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: srw [[SMOLDV:[0-9]+]], [[MOLDV]], [[SA]]
+; CHECK: extsh [[SESMOLDV:[0-9]+]], [[SMOLDV]]
+; CHECK: cmpw 0, 4, [[SESMOLDV]]
+; CHECK: bgelr 0
+; CHECK: andc [[NOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: or [[NEWV:[0-9]+]], [[SMV]], [[NOLDV]]
+; CHECK: stwcx. [[NEWV]], 0, 3
+; CHECK: bne  0,
+; CHECK: blr
+}
+
+define void @ae16max(i16* nocapture dereferenceable(4) %minimum, i16 %val) #0 {
+entry:
+  %0 = atomicrmw max i16* %minimum, i16 %val monotonic
+  ret void
+
+; CHECK-LABEL: @ae16max
+; CHECK-DAG: rlwinm [[SA1:[0-9]+]], 3, 3, 27, 27
+; CHECK-DAG: li [[M1:[0-9]+]], 0
+; CHECK-DAG: rldicr 3, 3, 0, 61
+; CHECK-DAG: xori [[SA:[0-9]+]], [[SA1]], 16
+; CHECK-DAG: ori [[M2:[0-9]+]], [[M1]], 65535
+; CHECK-DAG: slw [[SV:[0-9]+]], 4, [[SA]]
+; CHECK-DAG: slw [[M:[0-9]+]], [[M2]], [[SA]]
+; CHECK-DAG: and [[SMV:[0-9]+]], [[SV]], [[M]]
+; CHECK: lwarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: and [[MOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: srw [[SMOLDV:[0-9]+]], [[MOLDV]], [[SA]]
+; CHECK: extsh [[SESMOLDV:[0-9]+]], [[SMOLDV]]
+; CHECK: cmpw 0, 4, [[SESMOLDV]]
+; CHECK: blelr 0
+; CHECK: andc [[NOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: or [[NEWV:[0-9]+]], [[SMV]], [[NOLDV]]
+; CHECK: stwcx. [[NEWV]], 0, 3
+; CHECK: bne  0,
+; CHECK: blr
+}
+
+define void @ae16umin(i16* nocapture dereferenceable(4) %minimum, i16 %val) #0 {
+entry:
+  %0 = atomicrmw umin i16* %minimum, i16 %val monotonic
+  ret void
+
+; CHECK-LABEL: @ae16umin
+; CHECK-DAG: rlwinm [[SA1:[0-9]+]], 3, 3, 27, 27
+; CHECK-DAG: li [[M1:[0-9]+]], 0
+; CHECK-DAG: rldicr 3, 3, 0, 61
+; CHECK-DAG: xori [[SA:[0-9]+]], [[SA1]], 16
+; CHECK-DAG: ori [[M2:[0-9]+]], [[M1]], 65535
+; CHECK-DAG: slw [[SV:[0-9]+]], 4, [[SA]]
+; CHECK-DAG: slw [[M:[0-9]+]], [[M2]], [[SA]]
+; CHECK-DAG: and [[SMV:[0-9]+]], [[SV]], [[M]]
+; CHECK: lwarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: and [[MOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: cmplw 0, 4, [[MOLDV]]
+; CHECK: bgelr 0
+; CHECK: andc [[NOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: or [[NEWV:[0-9]+]], [[SMV]], [[NOLDV]]
+; CHECK: stwcx. [[NEWV]], 0, 3
+; CHECK: bne  0,
+; CHECK: blr
+}
+
+define void @ae16umax(i16* nocapture dereferenceable(4) %minimum, i16 %val) #0 {
+entry:
+  %0 = atomicrmw umax i16* %minimum, i16 %val monotonic
+  ret void
+
+; CHECK-LABEL: @ae16umax
+; CHECK-DAG: rlwinm [[SA1:[0-9]+]], 3, 3, 27, 27
+; CHECK-DAG: li [[M1:[0-9]+]], 0
+; CHECK-DAG: rldicr 3, 3, 0, 61
+; CHECK-DAG: xori [[SA:[0-9]+]], [[SA1]], 16
+; CHECK-DAG: ori [[M2:[0-9]+]], [[M1]], 65535
+; CHECK-DAG: slw [[SV:[0-9]+]], 4, [[SA]]
+; CHECK-DAG: slw [[M:[0-9]+]], [[M2]], [[SA]]
+; CHECK-DAG: and [[SMV:[0-9]+]], [[SV]], [[M]]
+; CHECK: lwarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: and [[MOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: cmplw 0, 4, [[MOLDV]]
+; CHECK: blelr 0
+; CHECK: andc [[NOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: or [[NEWV:[0-9]+]], [[SMV]], [[NOLDV]]
+; CHECK: stwcx. [[NEWV]], 0, 3
+; CHECK: bne  0,
+; CHECK: blr
+}
+
+define void @ae8min(i8* nocapture dereferenceable(4) %minimum, i8 %val) #0 {
+entry:
+  %0 = atomicrmw min i8* %minimum, i8 %val monotonic
+  ret void
+
+; CHECK-LABEL: @ae8min
+; CHECK-DAG: rlwinm [[SA1:[0-9]+]], 3, 3, 27, 28
+; CHECK-DAG: li [[M1:[0-9]+]], 255
+; CHECK-DAG: rldicr 3, 3, 0, 61
+; CHECK-DAG: xori [[SA:[0-9]+]], [[SA1]], 24
+; CHECK-DAG: slw [[SV:[0-9]+]], 4, [[SA]]
+; CHECK-DAG: slw [[M:[0-9]+]], [[M1]], [[SA]]
+; CHECK-DAG: and [[SMV:[0-9]+]], [[SV]], [[M]]
+; CHECK: lwarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: and [[MOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: srw [[SMOLDV:[0-9]+]], [[MOLDV]], [[SA]]
+; CHECK: extsb [[SESMOLDV:[0-9]+]], [[SMOLDV]]
+; CHECK: cmpw 0, 4, [[SESMOLDV]]
+; CHECK: bgelr 0
+; CHECK: andc [[NOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: or [[NEWV:[0-9]+]], [[SMV]], [[NOLDV]]
+; CHECK: stwcx. [[NEWV]], 0, 3
+; CHECK: bne  0,
+; CHECK: blr
+}
+
+define void @ae8max(i8* nocapture dereferenceable(4) %minimum, i8 %val) #0 {
+entry:
+  %0 = atomicrmw max i8* %minimum, i8 %val monotonic
+  ret void
+
+; CHECK-LABEL: @ae8max
+; CHECK-DAG: rlwinm [[SA1:[0-9]+]], 3, 3, 27, 28
+; CHECK-DAG: li [[M1:[0-9]+]], 255
+; CHECK-DAG: rldicr 3, 3, 0, 61
+; CHECK-DAG: xori [[SA:[0-9]+]], [[SA1]], 24
+; CHECK-DAG: slw [[SV:[0-9]+]], 4, [[SA]]
+; CHECK-DAG: slw [[M:[0-9]+]], [[M1]], [[SA]]
+; CHECK-DAG: and [[SMV:[0-9]+]], [[SV]], [[M]]
+; CHECK: lwarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: and [[MOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: srw [[SMOLDV:[0-9]+]], [[MOLDV]], [[SA]]
+; CHECK: extsb [[SESMOLDV:[0-9]+]], [[SMOLDV]]
+; CHECK: cmpw 0, 4, [[SESMOLDV]]
+; CHECK: blelr 0
+; CHECK: andc [[NOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: or [[NEWV:[0-9]+]], [[SMV]], [[NOLDV]]
+; CHECK: stwcx. [[NEWV]], 0, 3
+; CHECK: bne  0,
+; CHECK: blr
+}
+
+define void @ae8umin(i8* nocapture dereferenceable(4) %minimum, i8 %val) #0 {
+entry:
+  %0 = atomicrmw umin i8* %minimum, i8 %val monotonic
+  ret void
+
+; CHECK-LABEL: @ae8umin
+; CHECK-DAG: rlwinm [[SA1:[0-9]+]], 3, 3, 27, 28
+; CHECK-DAG: li [[M1:[0-9]+]], 255
+; CHECK-DAG: rldicr 3, 3, 0, 61
+; CHECK-DAG: xori [[SA:[0-9]+]], [[SA1]], 24
+; CHECK-DAG: slw [[SV:[0-9]+]], 4, [[SA]]
+; CHECK-DAG: slw [[M:[0-9]+]], [[M1]], [[SA]]
+; CHECK-DAG: and [[SMV:[0-9]+]], [[SV]], [[M]]
+; CHECK: lwarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: and [[MOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: cmplw 0, 4, [[MOLDV]]
+; CHECK: bgelr 0
+; CHECK: andc [[NOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: or [[NEWV:[0-9]+]], [[SMV]], [[NOLDV]]
+; CHECK: stwcx. [[NEWV]], 0, 3
+; CHECK: bne  0,
+; CHECK: blr
+}
+
+define void @ae8umax(i8* nocapture dereferenceable(4) %minimum, i8 %val) #0 {
+entry:
+  %0 = atomicrmw umax i8* %minimum, i8 %val monotonic
+  ret void
+
+; CHECK-LABEL: @ae8umax
+; CHECK-DAG: rlwinm [[SA1:[0-9]+]], 3, 3, 27, 28
+; CHECK-DAG: li [[M1:[0-9]+]], 255
+; CHECK-DAG: rldicr 3, 3, 0, 61
+; CHECK-DAG: xori [[SA:[0-9]+]], [[SA1]], 24
+; CHECK-DAG: slw [[SV:[0-9]+]], 4, [[SA]]
+; CHECK-DAG: slw [[M:[0-9]+]], [[M1]], [[SA]]
+; CHECK-DAG: and [[SMV:[0-9]+]], [[SV]], [[M]]
+; CHECK: lwarx [[OLDV:[0-9]+]], 0, 3
+; CHECK: and [[MOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: cmplw 0, 4, [[MOLDV]]
+; CHECK: blelr 0
+; CHECK: andc [[NOLDV:[0-9]+]], [[OLDV]], [[M]]
+; CHECK: or [[NEWV:[0-9]+]], [[SMV]], [[NOLDV]]
+; CHECK: stwcx. [[NEWV]], 0, 3
+; CHECK: bne  0,
+; CHECK: blr
+}
+
+attributes #0 = { nounwind "target-cpu"="ppc64" }
+attributes #1 = { nounwind "target-cpu"="pwr8" }
+