PPC: Support dynamic allocas with large alignment

Support for dynamic stack alignments in the PPC backend has been unfinished, in
part because it depends on dynamic stack realignment (which I only just
recently implemented fully). Now we can also support dynamic allocas with
higher than the default target stack alignment (16 bytes).

In order to round-up the requested size to the maximum requested alignment, we
need an additional register to hold the rounded-up size. We're already using one
scavenged register to hold the previous stack-pointer value (which needs to be
stored with the signal-safe stdux update), and so when we have dynamic allocas
and a large alignment, we allocate two emergency spill slots for the scavenger.

llvm-svn: 186562

lib/Target/PowerPC/PPCFrameLowering.cpp

@@ -1237,8 +1237,12 @@ PPCFrameLowering::addScavengingSpillSlot(MachineFunction &MF,
                                                        RC->getAlignment(),
                                                        false));
 
+    // Might we have over-aligned allocas?
+    bool HasAlVars = MFI->hasVarSizedObjects() &&
+                     MFI->getMaxAlignment() > getStackAlignment();
+
     // These kinds of spills might need two registers.
-    if (spillsCR(MF) || spillsVRSAVE(MF))
+    if (spillsCR(MF) || spillsVRSAVE(MF) || HasAlVars)
       RS->addScavengingFrameIndex(MFI->CreateStackObject(RC->getSize(),
                                                          RC->getAlignment(),
                                                          false));

lib/Target/PowerPC/PPCRegisterInfo.cpp

@@ -269,8 +269,8 @@ void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II) const {
   // Get stack alignments.
   unsigned TargetAlign = MF.getTarget().getFrameLowering()->getStackAlignment();
   unsigned MaxAlign = MFI->getMaxAlignment();
-  if (MaxAlign > TargetAlign)
+  assert((maxCallFrameSize & (MaxAlign-1)) == 0 &&
-    report_fatal_error("Dynamic alloca with large aligns not supported");
+         "Maximum call-frame size not sufficiently aligned");
 
   // Determine the previous frame's address.  If FrameSize can't be
   // represented as 16 bits or we need special alignment, then we load the
@@ -295,40 +295,62 @@ void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II) const {
       .addImm(0)
       .addReg(PPC::R1);
   }
-  
+
+  bool KillNegSizeReg = MI.getOperand(1).isKill();
+  unsigned NegSizeReg = MI.getOperand(1).getReg();
+
   // Grow the stack and update the stack pointer link, then determine the
   // address of new allocated space.
   if (LP64) {
+    if (MaxAlign > TargetAlign) {
+      unsigned UnalNegSizeReg = NegSizeReg;
+      NegSizeReg = MF.getRegInfo().createVirtualRegister(G8RC);
+
+      // Unfortunately, there is no andi, only andi., and we can't insert that
+      // here because we might clobber cr0 while it is live.
+      BuildMI(MBB, II, dl, TII.get(PPC::LI8), NegSizeReg)
+        .addImm(~(MaxAlign-1));
+
+      unsigned NegSizeReg1 = NegSizeReg;
+      NegSizeReg = MF.getRegInfo().createVirtualRegister(G8RC);
+      BuildMI(MBB, II, dl, TII.get(PPC::AND8), NegSizeReg)
+        .addReg(UnalNegSizeReg, getKillRegState(KillNegSizeReg))
+        .addReg(NegSizeReg1, RegState::Kill);
+      KillNegSizeReg = true;
+    }
+
     BuildMI(MBB, II, dl, TII.get(PPC::STDUX), PPC::X1)
       .addReg(Reg, RegState::Kill)
       .addReg(PPC::X1)
-      .addReg(MI.getOperand(1).getReg());
+      .addReg(NegSizeReg, getKillRegState(KillNegSizeReg));
-    if (!MI.getOperand(1).isKill())
+    BuildMI(MBB, II, dl, TII.get(PPC::ADDI8), MI.getOperand(0).getReg())
-      BuildMI(MBB, II, dl, TII.get(PPC::ADDI8), MI.getOperand(0).getReg())
+      .addReg(PPC::X1)
-        .addReg(PPC::X1)
+      .addImm(maxCallFrameSize);
-        .addImm(maxCallFrameSize);
-    else
-      // Implicitly kill the register.
-      BuildMI(MBB, II, dl, TII.get(PPC::ADDI8), MI.getOperand(0).getReg())
-        .addReg(PPC::X1)
-        .addImm(maxCallFrameSize)
-        .addReg(MI.getOperand(1).getReg(), RegState::ImplicitKill);
   } else {
+    if (MaxAlign > TargetAlign) {
+      unsigned UnalNegSizeReg = NegSizeReg;
+      NegSizeReg = MF.getRegInfo().createVirtualRegister(GPRC);
+
+      // Unfortunately, there is no andi, only andi., and we can't insert that
+      // here because we might clobber cr0 while it is live.
+      BuildMI(MBB, II, dl, TII.get(PPC::LI), NegSizeReg)
+        .addImm(~(MaxAlign-1));
+
+      unsigned NegSizeReg1 = NegSizeReg;
+      NegSizeReg = MF.getRegInfo().createVirtualRegister(GPRC);
+      BuildMI(MBB, II, dl, TII.get(PPC::AND), NegSizeReg)
+        .addReg(UnalNegSizeReg, getKillRegState(KillNegSizeReg))
+        .addReg(NegSizeReg1, RegState::Kill);
+      KillNegSizeReg = true;
+    }
+
     BuildMI(MBB, II, dl, TII.get(PPC::STWUX), PPC::R1)
       .addReg(Reg, RegState::Kill)
       .addReg(PPC::R1)
-      .addReg(MI.getOperand(1).getReg());
+      .addReg(NegSizeReg, getKillRegState(KillNegSizeReg));
-
+    BuildMI(MBB, II, dl, TII.get(PPC::ADDI), MI.getOperand(0).getReg())
-    if (!MI.getOperand(1).isKill())
+      .addReg(PPC::R1)
-      BuildMI(MBB, II, dl, TII.get(PPC::ADDI), MI.getOperand(0).getReg())
+      .addImm(maxCallFrameSize);
-        .addReg(PPC::R1)
-        .addImm(maxCallFrameSize);
-    else
-      // Implicitly kill the register.
-      BuildMI(MBB, II, dl, TII.get(PPC::ADDI), MI.getOperand(0).getReg())
-        .addReg(PPC::R1)
-        .addImm(maxCallFrameSize)
-        .addReg(MI.getOperand(1).getReg(), RegState::ImplicitKill);
   }
   
   // Discard the DYNALLOC instruction.

test/CodeGen/PowerPC/dyn-alloca-aligned.ll

@@ -0,0 +1,39 @@
+; RUN: llc -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr7 < %s | FileCheck %s
+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-f128:128:128-v128:128:128-n32:64"
+target triple = "powerpc64-unknown-linux-gnu"
+
+%struct.s = type { i32, i32 }
+
+declare void @bar(i32*, i32*) #0
+
+define void @goo(%struct.s* byval nocapture readonly %a, i32 signext %n) #0 {
+entry:
+  %0 = zext i32 %n to i64
+  %vla = alloca i32, i64 %0, align 128
+  %vla1 = alloca i32, i64 %0, align 128
+  %a2 = getelementptr inbounds %struct.s* %a, i64 0, i32 0
+  %1 = load i32* %a2, align 4, !tbaa !0
+  store i32 %1, i32* %vla1, align 128, !tbaa !0
+  %b = getelementptr inbounds %struct.s* %a, i64 0, i32 1
+  %2 = load i32* %b, align 4, !tbaa !0
+  %arrayidx3 = getelementptr inbounds i32* %vla1, i64 1
+  store i32 %2, i32* %arrayidx3, align 4, !tbaa !0
+  call void @bar(i32* %vla1, i32* %vla) #0
+  ret void
+
+; CHECK-LABEL: @goo
+
+; CHECK-DAG: li [[REG1:[0-9]+]], -128
+; CHECK-DAG: neg [[REG2:[0-9]+]],
+; CHECK: and [[REG1]], [[REG2]], [[REG1]]
+; CHECK: stdux {{[0-9]+}}, 1, [[REG1]]
+
+; CHECK: blr
+
+}
+
+attributes #0 = { nounwind }
+
+!0 = metadata !{metadata !"int", metadata !1}
+!1 = metadata !{metadata !"omnipotent char", metadata !2}
+!2 = metadata !{metadata !"Simple C/C++ TBAA"}