[WIP] Demo a functional problem from D81647 with a fix and test case

[reames]

Summary

When spilling gc operands, we have to mark them as load and store. They were originally defs and we need to preserve that.

This is semantically important as a relocating collector *will* update them sometimes.

[dantrushin]

I will incorporate this into D81647, since fix will be even simpler there.

llvm/lib/CodeGen/FixupStatepointCallerSaved.cpp

238

Defs are never seen here by construction of OpsToSpill

248

Can iterate over MI.defs() (not operands()) and delete these checks

llvm/lib/CodeGen/FixupStatepointCallerSaved.cpp

     OpsToSpill.push_back(MI.getNumOperands());
     unsigned CurOpIdx = 0;
 
+    SmallSet<int, 8> WrittenSlotIndices;
     for (unsigned I = 0; I < MI.getNumOperands(); ++I) {
       MachineOperand &MO = MI.getOperand(I);
       if (I == OpsToSpill[CurOpIdx]) {

         MIB.addFrameIndex(FI);
         MIB.addImm(0);
         ++CurOpIdx;
+        if (MO.isDef())
+          WrittenSlotIndices.insert(FI);
       } else
         MIB.add(MO);
     }
     assert(CurOpIdx == (OpsToSpill.size() - 1) && "Not all operands processed");
+
+    // Note: This loop exists only to make a patch separably from D81647, once
+    // that lands, remove this loop.
+    for (MachineOperand &MO : MI.operands())
+      if (MO.isReg() && !MO.isImplicit() && MO.isDef() &&
+          RegToSlotIdx.count(MO.getReg()))
+        WrittenSlotIndices.insert(RegToSlotIdx[MO.getReg()]);
+
     // Add mem operands.
     NewMI->setMemRefs(MF, MI.memoperands());
     for (auto It : RegToSlotIdx) {
       int FrameIndex = It.second;
       auto PtrInfo = MachinePointerInfo::getFixedStack(MF, FrameIndex);
-      auto *MMO = MF.getMachineMemOperand(PtrInfo, MachineMemOperand::MOLoad,
+      MachineMemOperand::Flags Flags = MachineMemOperand::MOLoad;
+      if (WrittenSlotIndices.count(FrameIndex))
+        Flags |= MachineMemOperand::MOStore;
+      auto *MMO = MF.getMachineMemOperand(PtrInfo, Flags,
                                           getRegisterSize(TRI, It.first),
                                           MFI.getObjectAlign(FrameIndex));
       NewMI->addMemOperand(MF, MMO);

llvm/test/CodeGen/X86/statepoint-vreg-fixup-csr.ll

+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+; RUN: llc -max-registers-for-gc-values=5 -stop-after=fixup-statepoint-caller-saved -disable-peephole < %s | FileCheck %s
+; -diable-peephole is a workaround for llvm bug 46916
+
+target datalayout = "e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-pc-linux-gnu"
+
+declare void @foo()
+declare void @consume(i8 addrspace(1)*)
+
+; Make sure that spilled gc operands get marked w/MOStore
+define void @test(i8 addrspace(1)* %a, i8 addrspace(1)* %b, i8 addrspace(1)* %c, i8 addrspace(1)* %d, i8 addrspace(1)* %e, i8 addrspace(1)* %f, i8 addrspace(1)* %g, i8 addrspace(1)* %h) gc "statepoint-example" {
+  ; CHECK-LABEL: name: test
+  ; CHECK: bb.0.entry:
+  ; CHECK:   liveins: $rcx, $rdi, $rdx, $rsi, $r8, $r9
+  ; CHECK:   renamable $r15 = COPY $r9
+  ; CHECK:   renamable $r13 = COPY $r8
+  ; CHECK:   renamable $rbx = COPY $rcx
+  ; CHECK:   renamable $r12 = MOV64rm %fixed-stack.1, 1, $noreg, 0, $noreg :: (load 8 from %fixed-stack.1, align 16)
+  ; CHECK:   renamable $r14 = MOV64rm %fixed-stack.0, 1, $noreg, 0, $noreg :: (load 8 from %fixed-stack.0)
+  ; CHECK:   MOV64mr %stack.1, 1, $noreg, 0, $noreg, killed renamable $rsi :: (store 8 into %stack.1)
+  ; CHECK:   MOV64mr %stack.0, 1, $noreg, 0, $noreg, killed renamable $rdx :: (store 8 into %stack.0)
+  ; CHECK:   MOV64mr %stack.2, 1, $noreg, 0, $noreg, killed renamable $rdi :: (store 8 into %stack.2)
+  ; CHECK:   ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   renamable $r14, renamable $r12, renamable $r15, renamable $r13, dead renamable $rbx = STATEPOINT 0, 0, 0, @foo, 2, 0, 2, 0, 2, 0, killed renamable $r14, renamable $r14(tied-def 0), killed renamable $r12, renamable $r12(tied-def 1), killed renamable $r15, renamable $r15(tied-def 2), killed renamable $r13, renamable $r13(tied-def 3), killed renamable $rbx, renamable $rbx(tied-def 4), 1, 8, %stack.0, 0, 1, 8, %stack.0, 0, 1, 8, %stack.1, 0, 1, 8, %stack.1, 0, 1, 8, %stack.2, 0, 1, 8, %stack.2, 0, csr_64, implicit-def $rsp, implicit-def $ssp :: (volatile load store 8 on %stack.0), (volatile load store 8 on %stack.1), (volatile load store 8 on %stack.2)
+  ; CHECK:   ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   renamable $rdi = MOV64rm %stack.2, 1, $noreg, 0, $noreg :: (load 8 from %stack.2)
+  ; CHECK:   ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+  ; CHECK:   ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   renamable $rdi = MOV64rm %stack.1, 1, $noreg, 0, $noreg :: (load 8 from %stack.1)
+  ; CHECK:   ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+  ; CHECK:   ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   renamable $rbx = MOV64rm %stack.0, 1, $noreg, 0, $noreg :: (load 8 from %stack.0)
+  ; CHECK:   ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   $rdi = COPY renamable $rbx
+  ; CHECK:   CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+  ; CHECK:   ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   $rdi = COPY killed renamable $rbx
+  ; CHECK:   CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+  ; CHECK:   ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   $rdi = COPY killed renamable $r13
+  ; CHECK:   CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+  ; CHECK:   ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   $rdi = COPY killed renamable $r15
+  ; CHECK:   CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+  ; CHECK:   ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   $rdi = COPY killed renamable $r12
+  ; CHECK:   CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+  ; CHECK:   ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   $rdi = COPY killed renamable $r14
+  ; CHECK:   CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+  ; CHECK:   ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   RET 0
+entry:
+  %safepoint_token = tail call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @foo, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i8 addrspace(1)* %a, i8 addrspace(1)* %b, i8 addrspace(1)* %c, i8 addrspace(1)* %d, i8 addrspace(1)* %e, i8 addrspace(1)* %f, i8 addrspace(1)* %g, i8 addrspace(1)* %h)]
+  %a1 = call i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %safepoint_token,  i32 0, i32 0)
+  call void @consume(i8 addrspace(1)* %a1)
+  %b1 = call i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %safepoint_token,  i32 1, i32 1)
+  call void @consume(i8 addrspace(1)* %b1)
+  %c1 = call i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %safepoint_token,  i32 2, i32 2)
+  call void @consume(i8 addrspace(1)* %c1)
+  %d1 = call i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %safepoint_token,  i32 3, i32 3)
+  call void @consume(i8 addrspace(1)* %c1)
+  %e1 = call i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %safepoint_token,  i32 4, i32 4)
+  call void @consume(i8 addrspace(1)* %e1)
+  %f1 = call i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %safepoint_token,  i32 5, i32 5)
+  call void @consume(i8 addrspace(1)* %f1)
+  %g1 = call i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %safepoint_token,  i32 6, i32 6)
+  call void @consume(i8 addrspace(1)* %g1)
+  %h1 = call i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %safepoint_token,  i32 7, i32 7)
+  call void @consume(i8 addrspace(1)* %h1)
+  ret void
+}
+
+declare token @llvm.experimental.gc.statepoint.p0f_i1f(i64, i32, i1 ()*, i32, i32, ...)
+declare token @llvm.experimental.gc.statepoint.p0f_isVoidf(i64, i32, void ()*, i32, i32, ...)
+declare i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token, i32, i32)