[GlobalISel] Improve stack slot tracking in dbg.values

[fdeazeve]

Summary

For IR like:

%alloca = alloca ...
dbg.value(%alloca, !myvar, OP_deref(<other_ops>))

GlobalISel lowers it to MIR:

%some_reg = G_FRAME_INDEX <stack_slot>
DBG_VALUE %some_reg, !myvar, OP_deref(<other_ops>)

In other words, if the value of !myvar can be obtained by
dereferencing an alloca, in MIR we say that the _location_ of a variable
is obtained by dereferencing register %some_reg (plus some
<other_ops>).

We can instead remove the use of %some_reg: the location of !myvar
_is_ <stack_slot> (plus some <other_ops>). This patch implements
this transformation, which improves debug information handling in O0, as
these registers hardly ever survive register allocation.

A note about testing: similar to what was done in D76934
(f24e2e9eebde4b7a1d), this patch exposed a bug in the Builder class when
using -debug, where we tried to print an incomplete instruction. The
changes in MachineIRBuilder.cpp address that.

[fdeazeve]

llvm/test/CodeGen/AArch64/GlobalISel/debug-insts.ll

41

what is this meaning of this dbg.value with 4 arguments? I can't find it anywhere in the docs

[aprantl]

llvm/test/CodeGen/AArch64/GlobalISel/debug-insts.ll

40

Can you update the test to not hardcode !17, !18?

41

This is an old form of dbg.value with an offset field. Let's update this to the 3 argument form by dropping it.

[fdeazeve]

I'll submit a separate patch for the test cleanup

llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp

assert(DI.getVariable()->isValidLocationForIntrinsic(
                MIRBuilder.getDebugLoc()) &&
            "Expected inlined-at fields to agree");
     if (!V || DI.hasArgList()) {
       // DI cannot produce a valid DBG_VALUE, so produce an undef DBG_VALUE to
       // terminate any prior location.
       MIRBuilder.buildIndirectDbgValue(0, DI.getVariable(), DI.getExpression());
     } else if (const auto *CI = dyn_cast<Constant>(V)) {
       MIRBuilder.buildConstDbgValue(*CI, DI.getVariable(), DI.getExpression());
+    } else if (auto *AI = dyn_cast<AllocaInst>(V);
+               AI && AI->isStaticAlloca() &&
+               DI.getExpression()->startsWithDeref()) {
+      // If the value is an alloca and the expression starts with a
+      // dereference, track a stack slot instead of a register, as registers
+      // may be clobbered.
+      auto ExprOperands = DI.getExpression()->getElements();
+      auto *ExprDerefRemoved =
+          DIExpression::get(AI->getContext(), ExprOperands.drop_front());
+      MIRBuilder.buildFIDbgValue(getOrCreateFrameIndex(*AI), DI.getVariable(),
+                                 ExprDerefRemoved);
     } else {
       for (Register Reg : getOrCreateVRegs(*V)) {
         // FIXME: This does not handle register-indirect values at offset 0. The
         // direct/indirect thing shouldn't really be handled by something as
         // implicit as reg+noreg vs reg+imm in the first place, but it seems
         // pretty baked in right now.
         MIRBuilder.buildDirectDbgValue(Reg, DI.getVariable(), DI.getExpression());
       }

llvm/lib/CodeGen/GlobalISel/MachineIRBuilder.cpp

 MachineInstrBuilder MachineIRBuilder::buildFIDbgValue(int FI,
                                                       const MDNode *Variable,
                                                       const MDNode *Expr) {
   assert(isa<DILocalVariable>(Variable) && "not a variable");
   assert(cast<DIExpression>(Expr)->isValid() && "not an expression");
   assert(
       cast<DILocalVariable>(Variable)->isValidLocationForIntrinsic(getDL()) &&
       "Expected inlined-at fields to agree");
-  return buildInstr(TargetOpcode::DBG_VALUE)
+  return insertInstr(buildInstrNoInsert(TargetOpcode::DBG_VALUE)
                          .addFrameIndex(FI)
                          .addImm(0)
                          .addMetadata(Variable)
-      .addMetadata(Expr);
+                         .addMetadata(Expr));
 }
 
 MachineInstrBuilder MachineIRBuilder::buildConstDbgValue(const Constant &C,
                                                          const MDNode *Variable,
                                                          const MDNode *Expr) {
   assert(isa<DILocalVariable>(Variable) && "not a variable");
   assert(cast<DIExpression>(Expr)->isValid() && "not an expression");
   assert(

llvm/test/CodeGen/AArch64/GlobalISel/debug-insts.ll

entry:
   %vla.addr = alloca i32, i32 %in
   call void @llvm.dbg.declare(metadata ptr %vla.addr, metadata !14, metadata !DIExpression()), !dbg !15
   ret void, !dbg !15
 }
 
 @gv = global i32 zeroinitializer
 
 ; CHECK-LABEL: name: debug_value
+; CHECK: stack:
+; CHECK:    - { id: {{.*}}, name: addr
 ; CHECK: [[IN:%[0-9]+]]:_(s32) = COPY $w0
 define void @debug_value(i32 %in) #0 !dbg !16 {
+; CHECK: G_FRAME_INDEX %[[stack_slot:.*]]
   %addr = alloca i32
 ; CHECK: DBG_VALUE [[IN]](s32), $noreg, !17, !DIExpression(), debug-location !18
   call void @llvm.dbg.value(metadata i32 %in, i64 0, metadata !17, metadata !DIExpression()), !dbg !18
   store i32 %in, ptr %addr
-; CHECK: DBG_VALUE %1(p0), $noreg, !17, !DIExpression(DW_OP_deref), debug-location !18
+; CHECK: DBG_VALUE %[[stack_slot]], 0, !17, !DIExpression(), debug-location !18
   call void @llvm.dbg.value(metadata ptr %addr, i64 0, metadata !17, metadata !DIExpression(DW_OP_deref)), !dbg !18
 ; CHECK: DBG_VALUE 123, 0, !17, !DIExpression(), debug-location !18
   call void @llvm.dbg.value(metadata i32 123, i64 0, metadata !17, metadata !DIExpression()), !dbg !18
 ; CHECK: DBG_VALUE float 1.000000e+00, 0, !17, !DIExpression(), debug-location !18
   call void @llvm.dbg.value(metadata float 1.000000e+00, i64 0, metadata !17, metadata !DIExpression()), !dbg !18
 ; CHECK: DBG_VALUE 0, 0, !17, !DIExpression(), debug-location !18
   call void @llvm.dbg.value(metadata ptr null, i64 0, metadata !17, metadata !DIExpression()), !dbg !18
 ; CHECK: DBG_VALUE $noreg, 0, !17, !DIExpression(), debug-location !18
   call void @llvm.dbg.value(metadata ptr @gv, i64 0, metadata !17, metadata !DIExpression()), !dbg !18