[AArch64] Fold adds with tprel_lo12_nc and secrel_lo12 into a following ldr/str

Differential Revision: https://reviews.llvm.org/D44355

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327316 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp

@@ -911,6 +911,16 @@ bool AArch64ExpandPseudo::expandMI(MachineBasicBlock &MBB,
     MI.eraseFromParent();
     return true;
   }
+  case AArch64::ADDlowTLS:
+    // Produce a plain ADD
+    BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(AArch64::ADDXri))
+        .add(MI.getOperand(0))
+        .add(MI.getOperand(1))
+        .add(MI.getOperand(2))
+        .addImm(0);
+    MI.eraseFromParent();
+    return true;
+
   case AArch64::MOVbaseTLS: {
     unsigned DstReg = MI.getOperand(0).getReg();
     auto SysReg = AArch64SysReg::TPIDR_EL0;

lib/Target/AArch64/AArch64ISelLowering.cpp

@@ -3918,10 +3918,7 @@ AArch64TargetLowering::LowerELFGlobalTLSAddress(SDValue Op,
                                    DAG.getTargetConstant(0, DL, MVT::i32)),
                 0);
     SDValue TPWithOff =
-        SDValue(DAG.getMachineNode(AArch64::ADDXri, DL, PtrVT, TPWithOff_lo,
-                                   LoVar,
-                                   DAG.getTargetConstant(0, DL, MVT::i32)),
-                0);
+        DAG.getNode(AArch64ISD::ADDlow, DL, PtrVT, TPWithOff_lo, LoVar);
     return TPWithOff;
   } else if (Model == TLSModel::InitialExec) {
     TPOff = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, AArch64II::MO_TLS);
@@ -3958,9 +3955,7 @@ AArch64TargetLowering::LowerELFGlobalTLSAddress(SDValue Op,
     TPOff = SDValue(DAG.getMachineNode(AArch64::ADDXri, DL, PtrVT, TPOff, HiVar,
                                        DAG.getTargetConstant(0, DL, MVT::i32)),
                     0);
-    TPOff = SDValue(DAG.getMachineNode(AArch64::ADDXri, DL, PtrVT, TPOff, LoVar,
-                                       DAG.getTargetConstant(0, DL, MVT::i32)),
-                    0);
+    TPOff = DAG.getNode(AArch64ISD::ADDlow, DL, PtrVT, TPOff, LoVar);
   } else if (Model == TLSModel::GeneralDynamic) {
     // The call needs a relocation too for linker relaxation. It doesn't make
     // sense to call it MO_PAGE or MO_PAGEOFF though so we need another copy of
@@ -4031,9 +4026,7 @@ AArch64TargetLowering::LowerWindowsGlobalTLSAddress(SDValue Op,
       SDValue(DAG.getMachineNode(AArch64::ADDXri, DL, PtrVT, TLS, TGAHi,
                                  DAG.getTargetConstant(0, DL, MVT::i32)),
               0);
-  Addr = SDValue(DAG.getMachineNode(AArch64::ADDXri, DL, PtrVT, Addr, TGALo,
-                                    DAG.getTargetConstant(0, DL, MVT::i32)),
-                 0);
+  Addr = DAG.getNode(AArch64ISD::ADDlow, DL, PtrVT, Addr, TGALo);
   return Addr;
 }
 

lib/Target/AArch64/AArch64InstrInfo.td

@@ -396,6 +396,15 @@ def MOVaddrEXT
              [(set GPR64:$dst, (AArch64addlow (AArch64adrp texternalsym:$hi),
                                             texternalsym:$low))]>,
       Sched<[WriteAdrAdr]>;
+// Normally AArch64addlow either gets folded into a following ldr/str,
+// or together with an adrp into MOVaddr above. For cases with TLS, it
+// might appear without either of them, so allow lowering it into a plain
+// add.
+def ADDlowTLS
+    : Pseudo<(outs GPR64:$dst), (ins GPR64:$src, i64imm:$low),
+             [(set GPR64:$dst, (AArch64addlow GPR64:$src,
+                                            tglobaltlsaddr:$low))]>,
+      Sched<[WriteAdr]>;
 
 } // isReMaterializable, isCodeGenOnly
 

test/CodeGen/AArch64/arm64-tls-execs.ll

@@ -40,11 +40,10 @@ define i32 @test_local_exec() {
 
 ; CHECK: mrs x[[R1:[0-9]+]], TPIDR_EL0
 ; CHECK: add x[[R2:[0-9]+]], x[[R1]], :tprel_hi12:local_exec_var
-; CHECK: add x[[R3:[0-9]+]], x[[R2]], :tprel_lo12_nc:local_exec_var
-; CHECK: ldr w0, [x[[R3]]]
+; CHECK: ldr w0, [x[[R2]], :tprel_lo12_nc:local_exec_var]
 
 ; CHECK-RELOC: R_AARCH64_TLSLE_ADD_TPREL_HI12
-; CHECK-RELOC: R_AARCH64_TLSLE_ADD_TPREL_LO12_NC
+; CHECK-RELOC: R_AARCH64_TLSLE_LDST32_TPREL_LO12_NC
   ret i32 %val
 }
 

test/CodeGen/AArch64/win-tls.ll

@@ -35,10 +35,7 @@ define i64 @getVar64() {
 
 ; CHECK: ldr [[TLS:x[0-9]+]], {{\[}}[[TLS_POINTER]], x[[TLS_INDEX]], lsl #3]
 ; CHECK: add [[TLS]], [[TLS]], :secrel_hi12:tlsVar
-; This add+ldr could also be folded into a single ldr with a :secrel_lo12:
-; offset.
-; CHECK: add [[TLS]], [[TLS]], :secrel_lo12:tlsVar
-; CHECK: ldr w0, {{\[}}[[TLS]]{{\]}}
+; CHECK: ldr w0, {{\[}}[[TLS]], :secrel_lo12:tlsVar{{\]}}
 
 ; CHECK-LABEL: getPtr
 ; CHECK: adrp [[TLS_INDEX_ADDR:x[0-9]+]], _tls_index
@@ -56,17 +53,12 @@ define i64 @getVar64() {
 
 ; CHECK: ldr [[TLS:x[0-9]+]], {{\[}}[[TLS_POINTER]], x[[TLS_INDEX]], lsl #3]
 ; CHECK: add [[TLS]], [[TLS]], :secrel_hi12:tlsVar
-; This add+ldr could also be folded into a single ldr with a :secrel_lo12:
-; offset.
-; CHECK: add [[TLS]], [[TLS]], :secrel_lo12:tlsVar
-; CHECK: str w0, {{\[}}[[TLS]]{{\]}}
+; CHECK: str w0, {{\[}}[[TLS]], :secrel_lo12:tlsVar{{\]}}
 
 ; CHECK-LABEL: getVar8
 ; CHECK: add [[TLS:x[0-9]+]], [[TLS]], :secrel_hi12:tlsVar8
-; CHECK: add [[TLS]], [[TLS]], :secrel_lo12:tlsVar8
-; CHECK: ldrb w0, {{\[}}[[TLS]]{{\]}}
+; CHECK: ldrb w0, {{\[}}[[TLS]], :secrel_lo12:tlsVar8{{\]}}
 
 ; CHECK-LABEL: getVar64
 ; CHECK: add [[TLS:x[0-9]+]], [[TLS]], :secrel_hi12:tlsVar64
-; CHECK: add [[TLS]], [[TLS]], :secrel_lo12:tlsVar64
-; CHECK: ldr x0, {{\[}}[[TLS]]{{\]}}
+; CHECK: ldr x0, {{\[}}[[TLS]], :secrel_lo12:tlsVar64{{\]}}