PR for llvm/llvm-project#66984

llvm/lib/CodeGen/StackColoring.cpp

@@ -370,37 +370,6 @@ STATISTIC(EscapedAllocas, "Number of allocas that escaped the lifetime region");
 // If in RPO ordering chosen to walk the CFG  we happen to visit the b[k]
 // before visiting the memcpy block (which will contain the lifetime start
 // for "b" then it will appear that 'b' has a degenerate lifetime.
-//
-// Handle Windows Exception with LifetimeStartOnFirstUse:
-// -----------------
-//
-// There was a bug for using LifetimeStartOnFirstUse in win32.
-// class Type1 {
-// ...
-// ~Type1(){ write memory;}
-// }
-// ...
-// try{
-// Type1 V
-// ...
-// } catch (Type2 X){
-// ...
-// }
-// For variable X in catch(X), we put point pX=&(&X) into ConservativeSlots
-// to prevent using LifetimeStartOnFirstUse. Because pX may merged with
-// object V which may call destructor after implicitly writing pX. All these
-// are done in C++ EH runtime libs (through CxxThrowException), and can't
-// obviously check it in IR level.
-//
-// The loader of pX, without obvious writing IR, is usually the first LOAD MI
-// in EHPad, Some like:
-// bb.x.catch.i (landing-pad, ehfunclet-entry):
-// ; predecessors: %bb...
-//   successors: %bb...
-//  %n:gr32 = MOV32rm %stack.pX ...
-//  ...
-// The Type2** %stack.pX will only be written in EH runtime libs, so we
-// check the StoreSlots to screen it out.
 
 namespace {
 
@@ -462,9 +431,6 @@ class StackColoring : public MachineFunctionPass {
   /// slots lifetime-start-on-first-use is disabled).
   BitVector ConservativeSlots;
 
-  /// Record the FI slots referenced by a 'may write to memory'.
-  BitVector StoreSlots;
-
   /// Number of iterations taken during data flow analysis.
   unsigned NumIterations;
 
@@ -660,13 +626,10 @@ unsigned StackColoring::collectMarkers(unsigned NumSlot) {
   InterestingSlots.resize(NumSlot);
   ConservativeSlots.clear();
   ConservativeSlots.resize(NumSlot);
-  StoreSlots.clear();
-  StoreSlots.resize(NumSlot);
 
   // number of start and end lifetime ops for each slot
   SmallVector<int, 8> NumStartLifetimes(NumSlot, 0);
   SmallVector<int, 8> NumEndLifetimes(NumSlot, 0);
-  SmallVector<int, 8> NumLoadInCatchPad(NumSlot, 0);
 
   // Step 1: collect markers and populate the "InterestingSlots"
   // and "ConservativeSlots" sets.
@@ -722,13 +685,6 @@ unsigned StackColoring::collectMarkers(unsigned NumSlot) {
           if (! BetweenStartEnd.test(Slot)) {
             ConservativeSlots.set(Slot);
           }
-          // Here we check the StoreSlots to screen catch point out. For more
-          // information, please refer "Handle Windows Exception with
-          // LifetimeStartOnFirstUse" at the head of this file.
-          if (MI.mayStore())
-            StoreSlots.set(Slot);
-          if (MF->getWinEHFuncInfo() && MBB->isEHPad() && MI.mayLoad())
-            NumLoadInCatchPad[Slot] += 1;
         }
       }
     }
@@ -739,14 +695,24 @@ unsigned StackColoring::collectMarkers(unsigned NumSlot) {
     return 0;
   }
 
-  // 1) PR27903: slots with multiple start or end lifetime ops are not
+  // PR27903: slots with multiple start or end lifetime ops are not
   // safe to enable for "lifetime-start-on-first-use".
-  // 2) And also not safe for variable X in catch(X) in windows.
   for (unsigned slot = 0; slot < NumSlot; ++slot) {
-    if (NumStartLifetimes[slot] > 1 || NumEndLifetimes[slot] > 1 ||
-        (NumLoadInCatchPad[slot] > 1 && !StoreSlots.test(slot)))
+    if (NumStartLifetimes[slot] > 1 || NumEndLifetimes[slot] > 1)
       ConservativeSlots.set(slot);
   }
+
+  // The write to the catch object by the personality function is not propely
+  // modeled in IR: It happens before any cleanuppads are executed, even if the
+  // first mention of the catch object is in a catchpad. As such, mark catch
+  // object slots as conservative, so they are excluded from first-use analysis.
+  if (WinEHFuncInfo *EHInfo = MF->getWinEHFuncInfo())
+    for (WinEHTryBlockMapEntry &TBME : EHInfo->TryBlockMap)
+      for (WinEHHandlerType &H : TBME.HandlerArray)
+        if (H.CatchObj.FrameIndex != std::numeric_limits<int>::max() &&
+            H.CatchObj.FrameIndex >= 0)
+          ConservativeSlots.set(H.CatchObj.FrameIndex);
+
   LLVM_DEBUG(dumpBV("Conservative slots", ConservativeSlots));
 
   // Step 2: compute begin/end sets for each block

llvm/test/CodeGen/X86/stack-coloring-wineh.ll

@@ -0,0 +1,190 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3
+; RUN: llc -mtriple=i686-windows-msvc < %s | FileCheck %s --check-prefix=I686
+; RUN: llc -mtriple=x86_64-windows-msvc < %s | FileCheck %s --check-prefix=X86_64
+
+@type_info = external global ptr
+
+; Make sure %a1 and %a2 don't share the same stack offset.
+define void @pr66984(ptr %arg) personality ptr @__CxxFrameHandler3 {
+; I686-LABEL: pr66984:
+; I686:       # %bb.0: # %bb
+; I686-NEXT:    pushl %ebp
+; I686-NEXT:    movl %esp, %ebp
+; I686-NEXT:    pushl %ebx
+; I686-NEXT:    pushl %edi
+; I686-NEXT:    pushl %esi
+; I686-NEXT:    subl $24, %esp
+; I686-NEXT:    movl %esp, -28(%ebp)
+; I686-NEXT:    movl $-1, -16(%ebp)
+; I686-NEXT:    leal -24(%ebp), %eax
+; I686-NEXT:    movl $___ehhandler$pr66984, -20(%ebp)
+; I686-NEXT:    movl %fs:0, %ecx
+; I686-NEXT:    movl %ecx, -24(%ebp)
+; I686-NEXT:    movl %eax, %fs:0
+; I686-NEXT:    movl $1, -16(%ebp)
+; I686-NEXT:    calll _throw
+; I686-NEXT:  # %bb.1: # %bb14
+; I686-NEXT:  LBB0_3: # Block address taken
+; I686-NEXT:    # %bb17
+; I686-NEXT:    addl $12, %ebp
+; I686-NEXT:    jmp LBB0_4
+; I686-NEXT:  LBB0_4: # %exit
+; I686-NEXT:  $ehgcr_0_4:
+; I686-NEXT:    movl -24(%ebp), %eax
+; I686-NEXT:    movl %eax, %fs:0
+; I686-NEXT:    addl $24, %esp
+; I686-NEXT:    popl %esi
+; I686-NEXT:    popl %edi
+; I686-NEXT:    popl %ebx
+; I686-NEXT:    popl %ebp
+; I686-NEXT:    retl
+; I686-NEXT:    .def "?catch$2@?0?pr66984@4HA";
+; I686-NEXT:    .scl 3;
+; I686-NEXT:    .type 32;
+; I686-NEXT:    .endef
+; I686-NEXT:    .p2align 4, 0x90
+; I686-NEXT:  "?catch$2@?0?pr66984@4HA":
+; I686-NEXT:  LBB0_2: # %bb17
+; I686-NEXT:    pushl %ebp
+; I686-NEXT:    addl $12, %ebp
+; I686-NEXT:    movl %esp, -28(%ebp)
+; I686-NEXT:    movl -36(%ebp), %ecx
+; I686-NEXT:    movl $2, -16(%ebp)
+; I686-NEXT:    calll _cleanup
+; I686-NEXT:    movl $LBB0_3, %eax
+; I686-NEXT:    popl %ebp
+; I686-NEXT:    retl # CATCHRET
+; I686-NEXT:    .def "?dtor$5@?0?pr66984@4HA";
+; I686-NEXT:    .scl 3;
+; I686-NEXT:    .type 32;
+; I686-NEXT:    .endef
+; I686-NEXT:    .p2align 4, 0x90
+; I686-NEXT:  "?dtor$5@?0?pr66984@4HA":
+; I686-NEXT:  LBB0_5: # %bb8
+; I686-NEXT:    pushl %ebp
+; I686-NEXT:    addl $12, %ebp
+; I686-NEXT:    movl 8(%ebp), %eax
+; I686-NEXT:    movl %eax, -32(%ebp)
+; I686-NEXT:    leal -32(%ebp), %ecx
+; I686-NEXT:    calll _foo
+; I686-NEXT:    popl %ebp
+; I686-NEXT:    retl # CLEANUPRET
+; I686-NEXT:  Lfunc_end0:
+;
+; X86_64-LABEL: pr66984:
+; X86_64:       # %bb.0: # %bb
+; X86_64-NEXT:    pushq %rbp
+; X86_64-NEXT:    .seh_pushreg %rbp
+; X86_64-NEXT:    subq $64, %rsp
+; X86_64-NEXT:    .seh_stackalloc 64
+; X86_64-NEXT:    leaq {{[0-9]+}}(%rsp), %rbp
+; X86_64-NEXT:    .seh_setframe %rbp, 64
+; X86_64-NEXT:    .seh_endprologue
+; X86_64-NEXT:    movq $-2, -16(%rbp)
+; X86_64-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X86_64-NEXT:  .Ltmp0:
+; X86_64-NEXT:    callq throw
+; X86_64-NEXT:  .Ltmp1:
+; X86_64-NEXT:  # %bb.1: # %bb14
+; X86_64-NEXT:  .LBB0_3: # Block address taken
+; X86_64-NEXT:    # %exit
+; X86_64-NEXT:  $ehgcr_0_3:
+; X86_64-NEXT:    nop
+; X86_64-NEXT:    addq $64, %rsp
+; X86_64-NEXT:    popq %rbp
+; X86_64-NEXT:    retq
+; X86_64-NEXT:    .seh_handlerdata
+; X86_64-NEXT:    .long ($cppxdata$pr66984)@IMGREL
+; X86_64-NEXT:    .text
+; X86_64-NEXT:    .seh_endproc
+; X86_64-NEXT:    .def "?catch$2@?0?pr66984@4HA";
+; X86_64-NEXT:    .scl 3;
+; X86_64-NEXT:    .type 32;
+; X86_64-NEXT:    .endef
+; X86_64-NEXT:    .p2align 4, 0x90
+; X86_64-NEXT:  "?catch$2@?0?pr66984@4HA":
+; X86_64-NEXT:  .seh_proc "?catch$2@?0?pr66984@4HA"
+; X86_64-NEXT:    .seh_handler __CxxFrameHandler3, @unwind, @except
+; X86_64-NEXT:  .LBB0_2: # %bb17
+; X86_64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp)
+; X86_64-NEXT:    pushq %rbp
+; X86_64-NEXT:    .seh_pushreg %rbp
+; X86_64-NEXT:    subq $32, %rsp
+; X86_64-NEXT:    .seh_stackalloc 32
+; X86_64-NEXT:    leaq 64(%rdx), %rbp
+; X86_64-NEXT:    .seh_endprologue
+; X86_64-NEXT:    movq -8(%rbp), %rcx
+; X86_64-NEXT:    callq cleanup
+; X86_64-NEXT:    leaq .LBB0_3(%rip), %rax
+; X86_64-NEXT:    addq $32, %rsp
+; X86_64-NEXT:    popq %rbp
+; X86_64-NEXT:    retq # CATCHRET
+; X86_64-NEXT:    .seh_handlerdata
+; X86_64-NEXT:    .long ($cppxdata$pr66984)@IMGREL
+; X86_64-NEXT:    .text
+; X86_64-NEXT:    .seh_endproc
+; X86_64-NEXT:    .def "?dtor$4@?0?pr66984@4HA";
+; X86_64-NEXT:    .scl 3;
+; X86_64-NEXT:    .type 32;
+; X86_64-NEXT:    .endef
+; X86_64-NEXT:    .p2align 4, 0x90
+; X86_64-NEXT:  "?dtor$4@?0?pr66984@4HA":
+; X86_64-NEXT:  .seh_proc "?dtor$4@?0?pr66984@4HA"
+; X86_64-NEXT:  .LBB0_4: # %bb8
+; X86_64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp)
+; X86_64-NEXT:    pushq %rbp
+; X86_64-NEXT:    .seh_pushreg %rbp
+; X86_64-NEXT:    subq $32, %rsp
+; X86_64-NEXT:    .seh_stackalloc 32
+; X86_64-NEXT:    leaq 64(%rdx), %rbp
+; X86_64-NEXT:    .seh_endprologue
+; X86_64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
+; X86_64-NEXT:    movq %rax, -32(%rbp)
+; X86_64-NEXT:    leaq -32(%rbp), %rcx
+; X86_64-NEXT:    callq foo
+; X86_64-NEXT:    nop
+; X86_64-NEXT:    addq $32, %rsp
+; X86_64-NEXT:    popq %rbp
+; X86_64-NEXT:    retq # CLEANUPRET
+; X86_64-NEXT:  .Lfunc_end0:
+; X86_64-NEXT:    .seh_handlerdata
+; X86_64-NEXT:    .text
+; X86_64-NEXT:    .seh_endproc
+bb:
+  %a1 = alloca ptr, align 4
+  %a2 = alloca ptr, align 4
+  call void @llvm.lifetime.start.p0(i64 4, ptr nonnull %a2)
+  invoke void @throw()
+          to label %bb14 unwind label %bb8
+
+bb8:                                              ; preds = %bb7
+  %i9 = cleanuppad within none []
+  call void @llvm.lifetime.start.p0(i64 4, ptr nonnull %a1)
+  store ptr %arg, ptr %a1, align 4
+  call fastcc void @foo(ptr %a1) [ "funclet"(token %i9) ]
+  call void @llvm.lifetime.end.p0(i64 4, ptr nonnull %a1)
+  cleanupret from %i9 unwind label %bb15
+
+bb14:                                             ; preds = %bb7
+  unreachable
+
+bb15:                                             ; preds = %bb13, %bb5
+  %cs = catchswitch within none [label %bb17] unwind to caller
+
+bb17:                                             ; preds = %bb15
+  %cp = catchpad within %cs [ptr @type_info, i32 8, ptr %a2]
+  %p = load ptr, ptr %a2, align 4
+  call fastcc void @cleanup(ptr %p) [ "funclet"(token %cp) ]
+  catchret from %cp to label %exit
+
+exit:
+  call void @llvm.lifetime.end.p0(i64 4, ptr nonnull %a2)
+  ret void
+}
+
+declare i32 @__CxxFrameHandler3(...)
+declare void @throw()
+declare void @cleanup(ptr)
+declare void @foo(ptr)
+declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture)
+declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture)