[llvm] r309343 - [X86] Fix latent bug in sibcall eligibility logic

[Reid Kleckner via llvm-commits]

Author: rnk
Date: Thu Jul 27 17:58:35 2017
New Revision: 309343

URL: http://llvm.org/viewvc/llvm-project?rev=309343&view=rev
Log:
[X86] Fix latent bug in sibcall eligibility logic

The X86 tail call eligibility logic was correct when it was written, but
the addition of inalloca and argument copy elision broke its
assumptions. It was assuming that fixed stack objects were immutable.

Currently, we aim to emit a tail call if no arguments have to be
re-arranged in memory. This code would trace the outgoing argument
values back to check if they are loads from an incoming stack object.
If the stack argument is immutable, then we won't need to store it back
to the stack when we tail call.

Fortunately, stack objects track their mutability, so we can just make
the obvious check to fix the bug.

This was http://crbug.com/749826

Added:
    llvm/trunk/test/CodeGen/X86/tail-call-mutable-memarg.ll
Modified:
    llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=309343&r1=309342&r2=309343&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Thu Jul 27 17:58:35 2017
@@ -3983,6 +3983,13 @@ bool MatchingStackOffset(SDValue Arg, un
   if (Offset != MFI.getObjectOffset(FI))
     return false;
 
+  // If this is not byval, check that the argument stack object is immutable.
+  // inalloca and argument copy elision can create mutable argument stack
+  // objects. Byval objects can be mutated, but a byval call intends to pass the
+  // mutated memory.
+  if (!Flags.isByVal() && !MFI.isImmutableObjectIndex(FI))
+    return false;
+
   if (VA.getLocVT().getSizeInBits() > Arg.getValueSizeInBits()) {
     // If the argument location is wider than the argument type, check that any
     // extension flags match.

Added: llvm/trunk/test/CodeGen/X86/tail-call-mutable-memarg.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/tail-call-mutable-memarg.ll?rev=309343&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/X86/tail-call-mutable-memarg.ll (added)
+++ llvm/trunk/test/CodeGen/X86/tail-call-mutable-memarg.ll Thu Jul 27 17:58:35 2017
@@ -0,0 +1,42 @@
+; RUN: llc < %s | FileCheck %s
+
+; Make sure we check that forwarded memory arguments are not modified when tail
+; calling. inalloca and copy arg elimination make argument slots mutable.
+
+target datalayout = "e-m:x-p:32:32-i64:64-f80:32-n8:16:32-a:0:32-S32"
+target triple = "i386-pc-windows-msvc19.0.24215"
+
+declare x86_stdcallcc void @tail_std(i32)
+declare void @capture(i32*)
+
+define x86_thiscallcc void @inalloca(i32* %this, i32* inalloca %args) {
+entry:
+  %val = load i32, i32* %args
+  store i32 0, i32* %args
+  tail call x86_stdcallcc void @tail_std(i32 %val)
+  ret void
+}
+
+; CHECK-LABEL: _inalloca:                              # @inalloca
+; CHECK:         movl    4(%esp), %[[reg:[^ ]*]]
+; CHECK:         movl    $0, 4(%esp)
+; CHECK:         pushl   %[[reg]]
+; CHECK:         calll   _tail_std at 4
+; CHECK:         retl    $4
+
+define x86_stdcallcc void @copy_elide(i32 %arg) {
+entry:
+  %arg.ptr = alloca i32
+  store i32 %arg, i32* %arg.ptr
+  call void @capture(i32* %arg.ptr)
+  tail call x86_stdcallcc void @tail_std(i32 %arg)
+  ret void
+}
+
+; CHECK-LABEL: _copy_elide at 4:                          # @copy_elide
+; CHECK:         leal    {{[0-9]+}}(%esp), %[[reg:[^ ]*]]
+; CHECK:         pushl   %[[reg]]
+; CHECK:         calll   _capture
+; ...
+; CHECK:         calll   _tail_std at 4
+; CHECK:         retl    $4