The local register allocator has to spill dirty callee saved registers before a

call that might throw. The landing pad assumes that all registers are in stack
slots.

We used to spill those dirty CSRs after the call, and the stack slots would be
wrong when arriving at the landing pad.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@102770 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/RegAllocLocal.cpp b/lib/CodeGen/RegAllocLocal.cpp
index 1885fabab857b483daa4a6c75c6835b1b0949fd6..920ba5732adce6d32f515c3ac7825a5c66e0c8a1 100644 (file)
--- a/lib/CodeGen/RegAllocLocal.cpp
+++ b/lib/CodeGen/RegAllocLocal.cpp
@@ -189,6 +189,9 @@ namespace {
     ///
     void removePhysReg(unsigned PhysReg);
 
+    void storeVirtReg(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
+                      unsigned VirtReg, unsigned PhysReg, bool isKill);
+
     /// spillVirtReg - This method spills the value specified by PhysReg into
     /// the virtual register slot specified by VirtReg.  It then updates the RA
     /// data structures to indicate the fact that PhysReg is now available.
@@ -286,6 +289,17 @@ void RALocal::removePhysReg(unsigned PhysReg) {
     PhysRegsUseOrder.erase(It);
 }
 
+/// storeVirtReg - Store a virtual register to its assigned stack slot.
+void RALocal::storeVirtReg(MachineBasicBlock &MBB,
+                           MachineBasicBlock::iterator I,
+                           unsigned VirtReg, unsigned PhysReg,
+                           bool isKill) {
+  const TargetRegisterClass *RC = MF->getRegInfo().getRegClass(VirtReg);
+  int FrameIndex = getStackSpaceFor(VirtReg, RC);
+  DEBUG(dbgs() << " to stack slot #" << FrameIndex);
+  TII->storeRegToStackSlot(MBB, I, PhysReg, isKill, FrameIndex, RC);
+  ++NumStores;   // Update statistics
+}
 
 /// spillVirtReg - This method spills the value specified by PhysReg into the
 /// virtual register slot specified by VirtReg.  It then updates the RA data
@@ -309,15 +323,11 @@ void RALocal::spillVirtReg(MachineBasicBlock &MBB,
     // Otherwise, there is a virtual register corresponding to this physical
     // register.  We only need to spill it into its stack slot if it has been
     // modified.
-    const TargetRegisterClass *RC = MF->getRegInfo().getRegClass(VirtReg);
-    int FrameIndex = getStackSpaceFor(VirtReg, RC);
-    DEBUG(dbgs() << " to stack slot #" << FrameIndex);
     // If the instruction reads the register that's spilled, (e.g. this can
     // happen if it is a move to a physical register), then the spill
     // instruction is not a kill.
     bool isKill = !(I != MBB.end() && I->readsRegister(PhysReg));
-    TII->storeRegToStackSlot(MBB, I, PhysReg, isKill, FrameIndex, RC);
-    ++NumStores;   // Update statistics
+    storeVirtReg(MBB, I, VirtReg, PhysReg, isKill);
   }
 
   getVirt2PhysRegMapSlot(VirtReg) = 0;   // VirtReg no longer available
@@ -801,9 +811,12 @@ void RALocal::AllocateBasicBlock(MachineBasicBlock &MBB) {
         dbgs() << "\nStarting RegAlloc of: " << *MI;
         dbgs() << "  Regs have values: ";
         for (unsigned i = 0; i != TRI->getNumRegs(); ++i)
-          if (PhysRegsUsed[i] != -1 && PhysRegsUsed[i] != -2)
+          if (PhysRegsUsed[i] != -1 && PhysRegsUsed[i] != -2) {
+            if (PhysRegsUsed[i] && isVirtRegModified(PhysRegsUsed[i]))
+              dbgs() << "*";
             dbgs() << "[" << TRI->getName(i)
                    << ",%reg" << PhysRegsUsed[i] << "] ";
+          }
         dbgs() << '\n';
       });
 
@@ -1092,6 +1105,20 @@ void RALocal::AllocateBasicBlock(MachineBasicBlock &MBB) {
       }
     }
     
+    // If this instruction is a call, make sure there are no dirty registers. The
+    // call might throw an exception, and the landing pad expects to find all
+    // registers in stack slots.
+    if (TID.isCall())
+      for (unsigned i = 0, e = TRI->getNumRegs(); i != e; ++i) {
+        if (PhysRegsUsed[i] <= 0) continue;
+        unsigned VirtReg = PhysRegsUsed[i];
+        if (!isVirtRegModified(VirtReg)) continue;
+        DEBUG(dbgs() << "  Storing dirty %reg" << VirtReg);
+        storeVirtReg(MBB, MI, VirtReg, i, false);
+        markVirtRegModified(VirtReg, false);
+        DEBUG(dbgs() << " because the call might throw\n");
+      }
+
     // Finally, if this is a noop copy instruction, zap it.  (Except that if
     // the copy is dead, it must be kept to avoid messing up liveness info for
     // the register scavenger.  See pr4100.)

diff --git a/test/CodeGen/X86/2010-04-30-LocalAlloc-LandingPad.ll b/test/CodeGen/X86/2010-04-30-LocalAlloc-LandingPad.ll
new file mode 100644 (file)
index 0000000..f5048af
--- /dev/null
+++ b/test/CodeGen/X86/2010-04-30-LocalAlloc-LandingPad.ll
@@ -0,0 +1,143 @@
+; RUN: llc < %s -O0 -regalloc=local -relocation-model=pic -disable-fp-elim | FileCheck %s
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32"
+target triple = "i386-apple-darwin10.0.0"
+
+%struct.S = type { [2 x i8*] }
+
+@_ZTIi = external constant i8*                    ; <i8**> [#uses=1]
+@.str = internal constant [4 x i8] c"%p\0A\00"    ; <[4 x i8]*> [#uses=1]
+@llvm.used = appending global [1 x i8*] [i8* bitcast (i8* (%struct.S*, i32, %struct.S*)* @_Z4test1SiS_ to i8*)], section "llvm.metadata" ; <[1 x i8*]*> [#uses=0]
+
+; Verify that %esi gets spilled before the call.
+; CHECK: Z4test1SiS
+; CHECK: movl %esi,{{.*}}(%ebp) 
+; CHECK: call __Z6throwsv
+
+define i8* @_Z4test1SiS_(%struct.S* byval %s1, i32 %n, %struct.S* byval %s2) ssp {
+entry:
+  %retval = alloca i8*, align 4                   ; <i8**> [#uses=2]
+  %n.addr = alloca i32, align 4                   ; <i32*> [#uses=1]
+  %_rethrow = alloca i8*                          ; <i8**> [#uses=4]
+  %0 = alloca i32, align 4                        ; <i32*> [#uses=1]
+  %cleanup.dst = alloca i32                       ; <i32*> [#uses=3]
+  %cleanup.dst7 = alloca i32                      ; <i32*> [#uses=6]
+  store i32 %n, i32* %n.addr
+  invoke void @_Z6throwsv()
+          to label %invoke.cont unwind label %try.handler
+
+invoke.cont:                                      ; preds = %entry
+  store i32 1, i32* %cleanup.dst7
+  br label %finally
+
+terminate.handler:                                ; preds = %match.end
+  %exc = call i8* @llvm.eh.exception()            ; <i8*> [#uses=1]
+  %1 = call i32 (i8*, i8*, ...)* @llvm.eh.selector(i8* %exc, i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*), i32 1) ; <i32> [#uses=0]
+  call void @_ZSt9terminatev() noreturn nounwind
+  unreachable
+
+try.handler:                                      ; preds = %entry
+  %exc1 = call i8* @llvm.eh.exception()           ; <i8*> [#uses=3]
+  %selector = call i32 (i8*, i8*, ...)* @llvm.eh.selector(i8* %exc1, i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*), i8* bitcast (i8** @_ZTIi to i8*), i8* null) ; <i32> [#uses=1]
+  %2 = call i32 @llvm.eh.typeid.for(i8* bitcast (i8** @_ZTIi to i8*)) ; <i32> [#uses=1]
+  %3 = icmp eq i32 %selector, %2                  ; <i1> [#uses=1]
+  br i1 %3, label %match, label %catch.next
+
+match:                                            ; preds = %try.handler
+  %4 = call i8* @__cxa_begin_catch(i8* %exc1)     ; <i8*> [#uses=1]
+  %5 = bitcast i8* %4 to i32*                     ; <i32*> [#uses=1]
+  %6 = load i32* %5                               ; <i32> [#uses=1]
+  store i32 %6, i32* %0
+  %call = invoke i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i32 0, i32 0), %struct.S* %s2)
+          to label %invoke.cont2 unwind label %match.handler ; <i32> [#uses=0]
+
+invoke.cont2:                                     ; preds = %match
+  store i32 1, i32* %cleanup.dst
+  br label %match.end
+
+match.handler:                                    ; preds = %match
+  %exc3 = call i8* @llvm.eh.exception()           ; <i8*> [#uses=2]
+  %7 = call i32 (i8*, i8*, ...)* @llvm.eh.selector(i8* %exc3, i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*), i32 0) ; <i32> [#uses=0]
+  store i8* %exc3, i8** %_rethrow
+  store i32 2, i32* %cleanup.dst
+  br label %match.end
+
+cleanup.pad:                                      ; preds = %cleanup.switch
+  store i32 1, i32* %cleanup.dst7
+  br label %finally
+
+cleanup.pad4:                                     ; preds = %cleanup.switch
+  store i32 2, i32* %cleanup.dst7
+  br label %finally
+
+match.end:                                        ; preds = %match.handler, %invoke.cont2
+  invoke void @__cxa_end_catch()
+          to label %invoke.cont5 unwind label %terminate.handler
+
+invoke.cont5:                                     ; preds = %match.end
+  br label %cleanup.switch
+
+cleanup.switch:                                   ; preds = %invoke.cont5
+  %tmp = load i32* %cleanup.dst                   ; <i32> [#uses=1]
+  switch i32 %tmp, label %cleanup.end [
+    i32 1, label %cleanup.pad
+    i32 2, label %cleanup.pad4
+  ]
+
+cleanup.end:                                      ; preds = %cleanup.switch
+  %exc6 = call i8* @llvm.eh.exception()           ; <i8*> [#uses=1]
+  store i8* %exc6, i8** %_rethrow
+  store i32 2, i32* %cleanup.dst7
+  br label %finally
+
+catch.next:                                       ; preds = %try.handler
+  store i8* %exc1, i8** %_rethrow
+  store i32 2, i32* %cleanup.dst7
+  br label %finally
+
+finally:                                          ; preds = %catch.next, %cleanup.end, %cleanup.pad4, %cleanup.pad, %invoke.cont
+  br label %cleanup.switch9
+
+cleanup.switch9:                                  ; preds = %finally
+  %tmp8 = load i32* %cleanup.dst7                 ; <i32> [#uses=1]
+  switch i32 %tmp8, label %cleanup.end10 [
+    i32 1, label %finally.end
+    i32 2, label %finally.throw
+  ]
+
+cleanup.end10:                                    ; preds = %cleanup.switch9
+  br label %finally.end
+
+finally.throw:                                    ; preds = %cleanup.switch9
+  %8 = load i8** %_rethrow                        ; <i8*> [#uses=1]
+  call void @_Unwind_Resume_or_Rethrow(i8* %8)
+  unreachable
+
+finally.end:                                      ; preds = %cleanup.end10, %cleanup.switch9
+  %tmp11 = getelementptr inbounds %struct.S* %s1, i32 0, i32 0 ; <[2 x i8*]*> [#uses=1]
+  %arraydecay = getelementptr inbounds [2 x i8*]* %tmp11, i32 0, i32 0 ; <i8**> [#uses=1]
+  %arrayidx = getelementptr inbounds i8** %arraydecay, i32 1 ; <i8**> [#uses=1]
+  %tmp12 = load i8** %arrayidx                    ; <i8*> [#uses=1]
+  store i8* %tmp12, i8** %retval
+  %9 = load i8** %retval                          ; <i8*> [#uses=1]
+  ret i8* %9
+}
+
+declare void @_Z6throwsv() ssp
+
+declare i32 @__gxx_personality_v0(...)
+
+declare i8* @llvm.eh.exception() nounwind readonly
+
+declare i32 @llvm.eh.selector(i8*, i8*, ...) nounwind
+
+declare void @_ZSt9terminatev()
+
+declare void @_Unwind_Resume_or_Rethrow(i8*)
+
+declare i32 @llvm.eh.typeid.for(i8*) nounwind
+
+declare i8* @__cxa_begin_catch(i8*)
+
+declare i32 @printf(i8*, ...)
+
+declare void @__cxa_end_catch()