Fix thumbv4t indirect calls

So there are a couple of issues with indirect calls on thumbv4t. First, the most
'obvious' instruction, 'blx' isn't available until v5t. And secondly, the
next-most-obvious sequence: 'mov lr, pc; bx rN' doesn't DTRT in thumb code
because the saved off pc has its thumb bit cleared, so when the callee returns
we end up in ARM mode.... yuck.

The solution is to 'bl' to a nearby landing pad with a 'bx rN' in it.

We could cut down on code size by sharing the landing pads between call sites
that are close enough, but for the moment let's do correctness first and look at
performance later.


Patch by: Iain Sandoe

http://reviews.llvm.org/D6519


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223380 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/ARM/ARMAsmPrinter.cpp

@@ -120,6 +120,23 @@ bool ARMAsmPrinter::runOnMachineFunction(MachineFunction &MF) {
   // Emit the rest of the function body.
   EmitFunctionBody();
 
+  // If we need V4T thumb mode Register Indirect Jump pads, emit them.
+  // These are created per function, rather than per TU, since it's
+  // relatively easy to exceed the thumb branch range within a TU.
+  if (! ThumbIndirectPads.empty()) {
+    OutStreamer.EmitAssemblerFlag(MCAF_Code16);
+    EmitAlignment(1);
+    for (unsigned i = 0, e = ThumbIndirectPads.size(); i < e; i++) {
+      OutStreamer.EmitLabel(ThumbIndirectPads[i].second);
+      EmitToStreamer(OutStreamer, MCInstBuilder(ARM::tBX)
+        .addReg(ThumbIndirectPads[i].first)
+        // Add predicate operands.
+        .addImm(ARMCC::AL)
+        .addReg(0));
+    }
+    ThumbIndirectPads.clear();
+  }
+
   // We didn't modify anything.
   return false;
 }
@@ -1282,18 +1299,34 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
     return;
   }
   case ARM::tBX_CALL: {
-    EmitToStreamer(OutStreamer, MCInstBuilder(ARM::tMOVr)
-      .addReg(ARM::LR)
-      .addReg(ARM::PC)
-      // Add predicate operands.
-      .addImm(ARMCC::AL)
-      .addReg(0));
+    if (Subtarget->hasV5TOps())
+      llvm_unreachable("Expected BLX to be selected for v5t+");
 
-    EmitToStreamer(OutStreamer, MCInstBuilder(ARM::tBX)
-      .addReg(MI->getOperand(0).getReg())
-      // Add predicate operands.
-      .addImm(ARMCC::AL)
-      .addReg(0));
+    // On ARM v4t, when doing a call from thumb mode, we need to ensure
+    // that the saved lr has its LSB set correctly (the arch doesn't
+    // have blx).
+    // So here we generate a bl to a small jump pad that does bx rN.
+    // The jump pads are emitted after the function body.
+
+    unsigned TReg = MI->getOperand(0).getReg();
+    MCSymbol *TRegSym = nullptr;
+    for (unsigned i = 0, e = ThumbIndirectPads.size(); i < e; i++) {
+      if (ThumbIndirectPads[i].first == TReg) {
+        TRegSym = ThumbIndirectPads[i].second;
+        break;
+      }
+    }
+
+    if (!TRegSym) {
+      TRegSym = OutContext.CreateTempSymbol();
+      ThumbIndirectPads.push_back(std::make_pair(TReg, TRegSym));
+    }
+
+    // Create a link-saving branch to the Reg Indirect Jump Pad.
+    EmitToStreamer(OutStreamer, MCInstBuilder(ARM::tBL)
+        // Predicate comes first here.
+        .addImm(ARMCC::AL).addReg(0)
+        .addExpr(MCSymbolRefExpr::Create(TRegSym, OutContext)));
     return;
   }
   case ARM::BMOVPCRX_CALL: {

lib/Target/ARM/ARMAsmPrinter.h

@@ -20,6 +20,7 @@ class ARMFunctionInfo;
 class MCOperand;
 class MachineConstantPool;
 class MachineOperand;
+class MCSymbol;
 
 namespace ARM {
   enum DW_ISA {
@@ -45,6 +46,11 @@ class LLVM_LIBRARY_VISIBILITY ARMAsmPrinter : public AsmPrinter {
   /// InConstantPool - Maintain state when emitting a sequence of constant
   /// pool entries so we can properly mark them as data regions.
   bool InConstantPool;
+
+  /// ThumbIndirectPads - These maintain a per-function list of jump pad
+  /// labels used for ARMv4t thumb code to make register indirect calls.
+  SmallVector<std::pair<unsigned, MCSymbol*>, 4> ThumbIndirectPads;
+
 public:
   explicit ARMAsmPrinter(TargetMachine &TM, MCStreamer &Streamer)
     : AsmPrinter(TM, Streamer), AFI(nullptr), MCP(nullptr),

test/CodeGen/ARM/none-macho-v4t.ll

@@ -11,11 +11,15 @@ define void @test_call() {
 ; CHECK: [[PC_LABEL:LPC[0-9]+_[0-9]+]]:
 ; CHECK-NEXT: add r[[CALLEE_STUB]], pc
 ; CHECK: ldr [[CALLEE:r[0-9]+]], [r[[CALLEE_STUB]]]
-; CHECK: mov lr, pc
-; CHECK: bx [[CALLEE]]
+; CHECK-NOT: mov lr, pc
+; CHECK: bl [[INDIRECT_PAD:Ltmp[0-9]+]]
 
 ; CHECK: [[LITPOOL]]:
 ; CHECK-NEXT: .long L_callee$non_lazy_ptr-([[PC_LABEL]]+4)
+
+; CHECK: [[INDIRECT_PAD]]:
+; CHECK: bx [[CALLEE]]
+
   call void @callee()
   ret void
 }

test/CodeGen/ARM/thumb_indirect_calls.ll

@@ -0,0 +1,40 @@
+; RUN: llc -mtriple=thumbv4t-eabi %s -o - | FileCheck ---check-prefix=CHECK -check-prefix=CHECK-V4T %s
+; RUN: llc -mtriple=thumbv5t-eabi %s -o - | FileCheck ---check-prefix=CHECK -check-prefix=CHECK-V5T %s
+
+@f = common global void (i32)* null, align 4
+
+; CHECK-LABEL foo:
+define void @foo(i32 %x) {
+entry:
+  %0 = load void (i32)** @f, align 4
+  tail call void %0(i32 %x)
+  ret void
+
+; CHECK: ldr [[TMP:r[0-3]]], [[F:\.[A-Z0-9_]+]]
+; CHECK: ldr [[CALLEE:r[0-3]]], {{\[}}[[TMP]]{{\]}}
+
+; CHECK-V4T-NOT: blx
+; CHECK-V4T: bl [[INDIRECT_PAD:\.Ltmp[0-9]+]]
+; CHECK-V4T: [[F]]:
+; CHECK-V4T: [[INDIRECT_PAD]]:
+; CHECK-V4T-NEXT: bx [[CALLEE]]
+; CHECK-V5T: blx [[CALLEE]]
+}
+
+; CHECK-LABEL bar:
+define void @bar(void (i32)* nocapture %g, i32 %x, void (i32)* nocapture %h) {
+entry:
+  tail call void %g(i32 %x)
+  tail call void %h(i32 %x)
+  ret void
+
+; CHECK-V4T: bl [[INDIRECT_PAD1:\.Ltmp[0-9]+]]
+; CHECK-V4T: bl [[INDIRECT_PAD2:\.Ltmp[0-9]+]]
+; CHECK-V4T: [[INDIRECT_PAD1]]:
+; CHECK-V4T-NEXT: bx
+; CHECK-V4T: [[INDIRECT_PAD2]]:
+; CHECK-V4T-NEXT: bx
+; CHECK-V5T: blx
+; CHECK-V5T: blx
+}
+