[X86] Extract HiPE prologue constants into metadata

[margnus1]

Summary

X86FrameLowering::adjustForHiPEPrologue() contains a hard-coded offset
into an Erlang Runtime System-internal data structure (the PCB). As the
layout of this data structure is prone to change, this poses problems
for maintaining compatibility.

To address this problem, we will have the compiler spit out this
information as module-level named metadata. For example (where
P_NSP_LIMIT is the offending offset):

!hipe.literals = !{ !2, !3, !4 }
!2 = !{ !"P_NSP_LIMIT", i32 152 }
!3 = !{ !"X86_LEAF_WORDS", i32 24 }
!4 = !{ !"AMD64_LEAF_WORDS", i32 24 }

[margnus1]

Ping. Oh, and the corresponding code for HiPE is currently being reviewed at Erlang/OTP's Github: https://github.com/erlang/otp/pull/1070

[margnus1]

Ping

[kostis]

We would appreciate some action on this one from the responsible reviewers...

[nadav]

Magnus, I don't have any problem with this change because it is specific to Erlang. I suggest CC-ing the original author of adjustForHiPEPrologue, and if that person has no objections commit the change. Relying on metadata is not idea, but I understand the constraint.

[yiannist]

It's totally fine; actually, it's even better than hard-coding those Erlang VM-specific values in the LLVM code. Thanks for working on this Magnus! :)

LGTM

[margnus1]

Excellent. I would appriciate if a reviewer could commit this for me, since I lack commit access.

llvm/trunk/lib/Target/X86/X86FrameLowering.cpp

void X86FrameLowering::adjustForSegmentedStacks(
   checkMBB->addSuccessor(allocMBB);
   checkMBB->addSuccessor(&PrologueMBB);
 
 #ifdef EXPENSIVE_CHECKS
   MF.verify();
 #endif
 }
 
+/// Lookup an ERTS parameter in the !hipe.literals named metadata node.
+/// HiPE provides Erlang Runtime System-internal parameters, such as PCB offsets
+/// to fields it needs, through a named metadata node "hipe.literals" containing
+/// name-value pairs.
+static unsigned getHiPELiteral(
+    NamedMDNode *HiPELiteralsMD, const StringRef LiteralName) {
+  for (int i = 0, e = HiPELiteralsMD->getNumOperands(); i != e; ++i) {
+    MDNode *Node = HiPELiteralsMD->getOperand(i);
+    if (Node->getNumOperands() != 2) continue;
+    MDString *NodeName = dyn_cast<MDString>(Node->getOperand(0));
+    ValueAsMetadata *NodeVal = dyn_cast<ValueAsMetadata>(Node->getOperand(1));
+    if (!NodeName || !NodeVal) continue;
+    ConstantInt *ValConst = dyn_cast_or_null<ConstantInt>(NodeVal->getValue());
+    if (ValConst && NodeName->getString() == LiteralName) {
+      return ValConst->getZExtValue();
+    }
+  }
+
+  report_fatal_error("HiPE literal " + LiteralName
+                     + " required but not provided");
+}
+
 /// Erlang programs may need a special prologue to handle the stack size they
 /// might need at runtime. That is because Erlang/OTP does not implement a C
 /// stack but uses a custom implementation of hybrid stack/heap architecture.
 /// (for more information see Eric Stenman's Ph.D. thesis:
 /// http://publications.uu.se/uu/fulltext/nbn_se_uu_diva-2688.pdf)
 ///
 /// CheckStack:
 ///       temp0 = sp - MaxStack
void X86FrameLowering::adjustForHiPEPrologue(
   MachineFrameInfo *MFI = MF.getFrameInfo();
   DebugLoc DL;
 
   // To support shrink-wrapping we would need to insert the new blocks
   // at the right place and update the branches to PrologueMBB.
   assert(&(*MF.begin()) == &PrologueMBB && "Shrink-wrapping not supported yet");
 
   // HiPE-specific values
-  const unsigned HipeLeafWords = 24;
+  NamedMDNode *HiPELiteralsMD = MF.getMMI().getModule()
+    ->getNamedMetadata("hipe.literals");
+  if (!HiPELiteralsMD)
+    report_fatal_error(
+        "Can't generate HiPE prologue without runtime parameters");
+  const unsigned HipeLeafWords
+    = getHiPELiteral(HiPELiteralsMD,
+                     Is64Bit ? "AMD64_LEAF_WORDS" : "X86_LEAF_WORDS");
   const unsigned CCRegisteredArgs = Is64Bit ? 6 : 5;
   const unsigned Guaranteed = HipeLeafWords * SlotSize;
   unsigned CallerStkArity = MF.getFunction()->arg_size() > CCRegisteredArgs ?
                             MF.getFunction()->arg_size() - CCRegisteredArgs : 0;
   unsigned MaxStack = MFI->getStackSize() + CallerStkArity*SlotSize + SlotSize;
 
   assert(STI.isTargetLinux() &&
          "HiPE prologue is only supported on Linux operating systems.");
for (const auto &LI : PrologueMBB.liveins()) {
       incStackMBB->addLiveIn(LI);
     }
 
     MF.push_front(incStackMBB);
     MF.push_front(stackCheckMBB);
 
     unsigned ScratchReg, SPReg, PReg, SPLimitOffset;
     unsigned LEAop, CMPop, CALLop;
+    SPLimitOffset = getHiPELiteral(HiPELiteralsMD, "P_NSP_LIMIT");
     if (Is64Bit) {
       SPReg = X86::RSP;
       PReg  = X86::RBP;
       LEAop = X86::LEA64r;
       CMPop = X86::CMP64rm;
       CALLop = X86::CALL64pcrel32;
-      SPLimitOffset = 0x90;
     } else {
       SPReg = X86::ESP;
       PReg  = X86::EBP;
       LEAop = X86::LEA32r;
       CMPop = X86::CMP32rm;
       CALLop = X86::CALLpcrel32;
-      SPLimitOffset = 0x4c;
     }
 
     ScratchReg = GetScratchRegister(Is64Bit, IsLP64, MF, true);
     assert(!MF.getRegInfo().isLiveIn(ScratchReg) &&
            "HiPE prologue scratch register is live-in");
 
     // Create new MBB for StackCheck:
     addRegOffset(BuildMI(stackCheckMBB, DL, TII.get(LEAop), ScratchReg),

llvm/trunk/test/CodeGen/X86/hipe-cc.ll

define cc 11 void @baz() nounwind {
   %tmp_clos2 = inttoptr i32 %tmp_clos to i32*
   %indirect_call = bitcast i32* %tmp_clos2 to void (i32, i32, i32)*
   ; CHECK:      movl $42, %eax
   ; CHECK-NEXT: jmpl *clos
   tail call cc 11 void %indirect_call(i32 undef, i32 undef, i32 42) nounwind
   ret void
 }
 
+!hipe.literals = !{ !0, !1, !2 }
+!0 = !{ !"P_NSP_LIMIT", i32 84 }
+!1 = !{ !"X86_LEAF_WORDS", i32 24 }
+!2 = !{ !"AMD64_LEAF_WORDS", i32 24 }
 @clos = external constant i32
 declare cc 11 void @bar(i32, i32, i32, i32, i32)

llvm/trunk/test/CodeGen/X86/hipe-cc64.ll

define cc 11 void @baz() nounwind {
   %tmp_clos2 = inttoptr i64 %tmp_clos to i64*
   %indirect_call = bitcast i64* %tmp_clos2 to void (i64, i64, i64)*
   ; CHECK:      movl $42, %esi
   ; CHECK-NEXT: jmpq *(%rax)
   tail call cc 11 void %indirect_call(i64 undef, i64 undef, i64 42) nounwind
   ret void
 }
 
+!hipe.literals = !{ !0, !1, !2 }
+!0 = !{ !"P_NSP_LIMIT", i32 160 }
+!1 = !{ !"X86_LEAF_WORDS", i32 24 }
+!2 = !{ !"AMD64_LEAF_WORDS", i32 24 }
 @clos = external constant i64
 declare cc 11 void @bar(i64, i64, i64, i64, i64, i64)

llvm/trunk/test/CodeGen/X86/hipe-prologue.ll

define {i32, i32} @test_basic(i32 %hp, i32 %p) {
   %1 = insertvalue {i32, i32} undef, i32 %hp, 0
   %2 = insertvalue {i32, i32} %1, i32 %p, 1
   ret {i32, i32} %1
 }
 
 define cc 11 {i32, i32} @test_basic_hipecc(i32 %hp, i32 %p) {
   ; X32-Linux-LABEL:       test_basic_hipecc:
   ; X32-Linux:       leal -140(%esp), %ebx
-  ; X32-Linux-NEXT:  cmpl 76(%ebp), %ebx
+  ; X32-Linux-NEXT:  cmpl 120(%ebp), %ebx
   ; X32-Linux-NEXT:  jb .LBB1_1
 
   ; X32-Linux:       ret
 
   ; X32-Linux:       .LBB1_1:
   ; X32-Linux-NEXT:  calll inc_stack_0
 
   ; X64-Linux-LABEL:       test_basic_hipecc:
-  ; X64-Linux:       leaq -232(%rsp), %r14
-  ; X64-Linux-NEXT:  cmpq 144(%rbp), %r14
+  ; X64-Linux:       leaq -184(%rsp), %r14
+  ; X64-Linux-NEXT:  cmpq 120(%rbp), %r14
   ; X64-Linux-NEXT:  jb .LBB1_1
 
   ; X64-Linux:       ret
 
   ; X64-Linux:       .LBB1_1:
   ; X64-Linux-NEXT:  callq inc_stack_0
 
   %mem = alloca i32, i32 10
define cc 11 {i32,i32,i32} @test_nocall_hipecc(i32 %hp,i32 %p,i32 %x,i32 %y) {
   %1 = add i32 %x, %y
   %2 = mul i32 42, %1
   %3 = sub i32 24, %2
   %4 = insertvalue {i32, i32, i32} undef, i32 %hp, 0
   %5 = insertvalue {i32, i32, i32} %4, i32 %p, 1
   %6 = insertvalue {i32, i32, i32} %5, i32 %p, 2
   ret {i32, i32, i32} %6
 }
+
+!hipe.literals = !{ !0, !1, !2 }
+!0 = !{ !"P_NSP_LIMIT", i32 120 }
+!1 = !{ !"X86_LEAF_WORDS", i32 24 }
+!2 = !{ !"AMD64_LEAF_WORDS", i32 18 }