Add support for generating reg+reg preinc stores on PPC.

PPC will now generate STWUX and friends.

llvm-svn: 158698

llvm/lib/Target/PowerPC/PPCFrameLowering.cpp

@@ -368,9 +368,9 @@ void PPCFrameLowering::emitPrologue(MachineFunction &MF) const {
       BuildMI(MBB, MBBI, dl, TII.get(PPC::SUBFIC) ,PPC::R0)
         .addReg(PPC::R0, RegState::Kill)
         .addImm(NegFrameSize);
-      BuildMI(MBB, MBBI, dl, TII.get(PPC::STWUX))
+      BuildMI(MBB, MBBI, dl, TII.get(PPC::STWUX), PPC::R1)
         .addReg(PPC::R1, RegState::Kill)
-        .addReg(PPC::R1, RegState::Define)
+        .addReg(PPC::R1)
         .addReg(PPC::R0);
     } else if (isInt<16>(NegFrameSize)) {
       BuildMI(MBB, MBBI, dl, TII.get(PPC::STWU), PPC::R1)
@@ -383,9 +383,9 @@ void PPCFrameLowering::emitPrologue(MachineFunction &MF) const {
       BuildMI(MBB, MBBI, dl, TII.get(PPC::ORI), PPC::R0)
         .addReg(PPC::R0, RegState::Kill)
         .addImm(NegFrameSize & 0xFFFF);
-      BuildMI(MBB, MBBI, dl, TII.get(PPC::STWUX))
+      BuildMI(MBB, MBBI, dl, TII.get(PPC::STWUX), PPC::R1)
         .addReg(PPC::R1, RegState::Kill)
-        .addReg(PPC::R1, RegState::Define)
+        .addReg(PPC::R1)
         .addReg(PPC::R0);
     }
   } else {    // PPC64.
@@ -401,9 +401,9 @@ void PPCFrameLowering::emitPrologue(MachineFunction &MF) const {
       BuildMI(MBB, MBBI, dl, TII.get(PPC::SUBFIC8), PPC::X0)
         .addReg(PPC::X0)
         .addImm(NegFrameSize);
-      BuildMI(MBB, MBBI, dl, TII.get(PPC::STDUX))
+      BuildMI(MBB, MBBI, dl, TII.get(PPC::STDUX), PPC::X1)
         .addReg(PPC::X1, RegState::Kill)
-        .addReg(PPC::X1, RegState::Define)
+        .addReg(PPC::X1)
         .addReg(PPC::X0);
     } else if (isInt<16>(NegFrameSize)) {
       BuildMI(MBB, MBBI, dl, TII.get(PPC::STDU), PPC::X1)
@@ -416,9 +416,9 @@ void PPCFrameLowering::emitPrologue(MachineFunction &MF) const {
       BuildMI(MBB, MBBI, dl, TII.get(PPC::ORI8), PPC::X0)
         .addReg(PPC::X0, RegState::Kill)
         .addImm(NegFrameSize & 0xFFFF);
-      BuildMI(MBB, MBBI, dl, TII.get(PPC::STDUX))
+      BuildMI(MBB, MBBI, dl, TII.get(PPC::STDUX), PPC::X1)
         .addReg(PPC::X1, RegState::Kill)
-        .addReg(PPC::X1, RegState::Define)
+        .addReg(PPC::X1)
         .addReg(PPC::X0);
     }
   }

llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp

@@ -111,6 +111,18 @@ namespace {
     /// immediate field.  Because preinc imms have already been validated, just
     /// accept it.
     bool SelectAddrImmOffs(SDValue N, SDValue &Out) const {
+      if (isa<ConstantSDNode>(N)) {
+        Out = N;
+        return true;
+      }
+
+      return false;
+    }
+
+    /// SelectAddrIdxOffs - Return true if the operand is valid for a preinc
+    /// index field.  Because preinc imms have already been validated, just
+    /// accept it.
+    bool SelectAddrIdxOffs(SDValue N, SDValue &Out) const {
       Out = N;
       return true;
     }

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

@@ -1105,7 +1105,15 @@ bool PPCTargetLowering::getPreIndexedAddressParts(SDNode *N, SDValue &Base,
   if (VT.isVector())
     return false;
 
-  // TODO: Check reg+reg first.
+  if (SelectAddressRegReg(Ptr, Offset, Base, DAG)) {
+    if (isa<StoreSDNode>(N)) {
+      AM = ISD::PRE_INC;
+      return true;
+    }
+
+    // FIXME: reg+reg preinc loads
+    return false;
+  }
 
   // LDU/STU use reg+imm*4, others use reg+imm.
   if (VT != MVT::i64) {

llvm/lib/Target/PowerPC/PPCInstr64Bit.td

@@ -680,10 +680,41 @@ def STDU : DSForm_1a<62, 1, (outs ptr_rc:$ea_res), (ins G8RC:$rS,
                     RegConstraint<"$ptrreg = $ea_res">, NoEncode<"$ea_res">,
                     isPPC64;
 
-let mayStore = 1 in
+
-def STDUX : XForm_8<31, 181, (outs), (ins G8RC:$rS, memrr:$dst),
+def STBUX8 : XForm_8<31, 247, (outs ptr_rc:$ea_res),
-                   "stdux $rS, $dst", LdStSTD,
+                              (ins G8RC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
-                   []>, isPPC64;
+                    "stbux $rS, $ptroff, $ptrreg", LdStStore,
+                    [(set ptr_rc:$ea_res,
+                       (pre_truncsti8 G8RC:$rS,
+                                      ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
+                    RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
+                    PPC970_DGroup_Cracked;
+
+def STHUX8 : XForm_8<31, 439, (outs ptr_rc:$ea_res),
+                              (ins G8RC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
+                    "sthux $rS, $ptroff, $ptrreg", LdStStore,
+                    [(set ptr_rc:$ea_res,
+                       (pre_truncsti16 G8RC:$rS,
+                                       ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
+                    RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
+                    PPC970_DGroup_Cracked;
+
+def STWUX8 : XForm_8<31, 183, (outs ptr_rc:$ea_res),
+                              (ins G8RC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
+                    "stwux $rS, $ptroff, $ptrreg", LdStStore,
+                    [(set ptr_rc:$ea_res,
+                       (pre_truncsti32 G8RC:$rS,
+                                       ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
+                    RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
+                    PPC970_DGroup_Cracked;
+
+def STDUX : XForm_8<31, 181, (outs ptr_rc:$ea_res),
+                              (ins G8RC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
+                    "stdux $rS, $ptroff, $ptrreg", LdStStore,
+                    [(set ptr_rc:$ea_res,
+                       (pre_store G8RC:$rS, ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
+                    RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
+                    PPC970_DGroup_Cracked, isPPC64;
 
 // STD_32/STDX_32 - Just like STD/STDX, but uses a '32-bit' input register.
 def STD_32  : DSForm_1<62, 0, (outs), (ins GPRC:$rT, memrix:$dst),

llvm/lib/Target/PowerPC/PPCInstrInfo.td

@@ -349,6 +349,7 @@ def ixaddr : ComplexPattern<iPTR, 2, "SelectAddrImmShift", [], []>; // "std"
 
 /// This is just the offset part of iaddr, used for preinc.
 def iaddroff : ComplexPattern<iPTR, 1, "SelectAddrImmOffs", [], []>;
+def xaddroff : ComplexPattern<iPTR, 1, "SelectAddrIdxOffs", [], []>;
 
 //===----------------------------------------------------------------------===//
 // PowerPC Instruction Predicate Definitions.
@@ -822,12 +823,49 @@ def STWX  : XForm_8<31, 151, (outs), (ins GPRC:$rS, memrr:$dst),
                    "stwx $rS, $dst", LdStStore,
                    [(store GPRC:$rS, xaddr:$dst)]>,
                    PPC970_DGroup_Cracked;
-                   
+ 
-let mayStore = 1 in {
+def STBUX : XForm_8<31, 247, (outs ptr_rc:$ea_res),
-def STWUX : XForm_8<31, 183, (outs), (ins GPRC:$rS, GPRC:$rA, GPRC:$rB),
+                             (ins GPRC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
-                   "stwux $rS, $rA, $rB", LdStStore,
+                   "stbux $rS, $ptroff, $ptrreg", LdStStore,
-                   []>;
+                   [(set ptr_rc:$ea_res,
-}
+                      (pre_truncsti8 GPRC:$rS,
+                                     ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
+                   RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
+                   PPC970_DGroup_Cracked;
+ 
+def STHUX : XForm_8<31, 439, (outs ptr_rc:$ea_res),
+                             (ins GPRC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
+                   "sthux $rS, $ptroff, $ptrreg", LdStStore,
+                   [(set ptr_rc:$ea_res,
+                      (pre_truncsti16 GPRC:$rS,
+                                      ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
+                   RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
+                   PPC970_DGroup_Cracked;
+                 
+def STWUX : XForm_8<31, 183, (outs ptr_rc:$ea_res),
+                             (ins GPRC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
+                   "stwux $rS, $ptroff, $ptrreg", LdStStore,
+                   [(set ptr_rc:$ea_res,
+                      (pre_store GPRC:$rS, ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
+                   RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
+                   PPC970_DGroup_Cracked;
+
+def STFSUX : XForm_8<31, 695, (outs ptr_rc:$ea_res),
+                              (ins F4RC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
+                    "stfsux $rS, $ptroff, $ptrreg", LdStStore,
+                    [(set ptr_rc:$ea_res,
+                       (pre_store F4RC:$rS, ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
+                    RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
+                    PPC970_DGroup_Cracked;
+
+def STFDUX : XForm_8<31, 759, (outs ptr_rc:$ea_res),
+                              (ins F8RC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
+                    "stfdux $rS, $ptroff, $ptrreg", LdStStore,
+                    [(set ptr_rc:$ea_res,
+                       (pre_store F8RC:$rS, ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
+                    RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
+                    PPC970_DGroup_Cracked;
+
 def STHBRX: XForm_8<31, 918, (outs), (ins GPRC:$rS, memrr:$dst),
                    "sthbrx $rS, $dst", LdStStore,
                    [(PPCstbrx GPRC:$rS, xoaddr:$dst, i16)]>, 

llvm/lib/Target/PowerPC/PPCRegisterInfo.cpp

@@ -328,14 +328,14 @@ void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II,
   // address of new allocated space.
   if (LP64) {
     if (requiresRegisterScavenging(MF)) // FIXME (64-bit): Use "true" part.
-      BuildMI(MBB, II, dl, TII.get(PPC::STDUX))
+      BuildMI(MBB, II, dl, TII.get(PPC::STDUX), PPC::X1)
         .addReg(Reg, RegState::Kill)
-        .addReg(PPC::X1, RegState::Define)
+        .addReg(PPC::X1)
         .addReg(MI.getOperand(1).getReg());
     else
-      BuildMI(MBB, II, dl, TII.get(PPC::STDUX))
+      BuildMI(MBB, II, dl, TII.get(PPC::STDUX), PPC::X1)
         .addReg(PPC::X0, RegState::Kill)
-        .addReg(PPC::X1, RegState::Define)
+        .addReg(PPC::X1)
         .addReg(MI.getOperand(1).getReg());
 
     if (!MI.getOperand(1).isKill())
@@ -349,9 +349,9 @@ void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II,
         .addImm(maxCallFrameSize)
         .addReg(MI.getOperand(1).getReg(), RegState::ImplicitKill);
   } else {
-    BuildMI(MBB, II, dl, TII.get(PPC::STWUX))
+    BuildMI(MBB, II, dl, TII.get(PPC::STWUX), PPC::R1)
       .addReg(Reg, RegState::Kill)
-      .addReg(PPC::R1, RegState::Define)
+      .addReg(PPC::R1)
       .addReg(MI.getOperand(1).getReg());
 
     if (!MI.getOperand(1).isKill())

llvm/test/CodeGen/PowerPC/stwux.ll

@@ -0,0 +1,47 @@
+target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v128:128:128-n32:64"
+target triple = "powerpc64-unknown-linux-gnu"
+; RUN: llc < %s | FileCheck %s
+
+@multvec_i = external unnamed_addr global [100 x i32], align 4
+
+define fastcc void @subs_STMultiExceptIntern() nounwind {
+entry:
+  br i1 undef, label %while.body.lr.ph, label %return
+
+while.body.lr.ph:                                 ; preds = %entry
+  br label %while.body
+
+while.body:                                       ; preds = %if.end12, %while.body.lr.ph
+  %i.0240 = phi i32 [ -1, %while.body.lr.ph ], [ %i.1, %if.end12 ]
+  br i1 undef, label %if.end12, label %if.then
+
+if.then:                                          ; preds = %while.body
+  br label %if.end12
+
+if.end12:                                         ; preds = %if.then, %while.body
+  %i.1 = phi i32 [ %i.0240, %while.body ], [ undef, %if.then ]
+  br i1 undef, label %while.body, label %while.end
+
+while.end:                                        ; preds = %if.end12
+  br i1 undef, label %return, label %if.end15
+
+if.end15:                                         ; preds = %while.end
+  %idxprom.i.i230 = sext i32 %i.1 to i64
+  %arrayidx18 = getelementptr inbounds [100 x i32]* @multvec_i, i64 0, i64 %idxprom.i.i230
+  store i32 0, i32* %arrayidx18, align 4
+  br i1 undef, label %while.body21, label %while.end90
+
+while.body21:                                     ; preds = %if.end15
+  unreachable
+
+while.end90:                                      ; preds = %if.end15
+  store i32 0, i32* %arrayidx18, align 4
+  br label %return
+
+return:                                           ; preds = %while.end90, %while.end, %entry
+  ret void
+
+; CHECK: @subs_STMultiExceptIntern
+; CHECK: stwux
+}
+