[DebugInfo] Prevent salvaged debug vars from being marked artificial

We were using compiler-generated source location (i.e. line number
0) on `debug_value` instructions were emitted by salvage debug info.
But it turned out that IRGen will attach DW_AT_artificial on the
associated debug variables, which are hidden in debugger by default.
This patch prevent this issue by using the source location from the
just-deleted instruction for these `debug_value` instructions.

This indirectly triggered another bug where some of the LLVM
!DIExpression-s we generated are actually invalid -- more specifically,
the fragment inside is covering the whole debug variable. The reason it was
not previously caught is because LLVM verifier skips any debug variable
that is marked artificial, and all debug variables that have illegal fragment
are falling under this category. Thus, this patch also fixes this issue
by not generating the DW_OP_LLVM_fragment part if it's illegal.

Author: mshockwave

lib/IRGen/IRGenDebugInfo.cpp

@@ -2542,6 +2542,23 @@ void IRGenDebugInfoImpl::emitDbgIntrinsic(
       !llvm::FindDbgDeclareUses(Storage).empty())
     return;
 
+  // Fragment DIExpression cannot cover the whole variable
+  // or going out-of-bound.
+  if (auto Fragment = Expr->getFragmentInfo())
+    if (auto VarSize = Var->getSizeInBits()) {
+      unsigned FragSize = Fragment->SizeInBits;
+      unsigned FragOffset = Fragment->OffsetInBits;
+      if (FragOffset + FragSize > *VarSize ||
+          FragSize == *VarSize) {
+        // Drop the fragment part
+        assert(Expr->isValid());
+        // Since this expression is valid, DW_OP_LLVM_fragment
+        // and its arguments must be the last 3 elements.
+        auto OrigElements = Expr->getElements();
+        Expr = DBuilder.createExpression(OrigElements.drop_back(3));
+      }
+    }
+
   // A dbg.declare is only meaningful if there is a single alloca for
   // the variable that is live throughout the function.
   if (auto *Alloca = dyn_cast<llvm::AllocaInst>(Storage)) {

lib/SILOptimizer/Utils/InstOptUtils.cpp

@@ -2119,8 +2119,7 @@ void swift::salvageDebugInfo(SILInstruction *I) {
               DestVal.getDefiningInstruction())) {
         if (auto VarInfo = ASI->getVarInfo())
           SILBuilder(SI, ASI->getDebugScope())
-              .createDebugValue(RegularLocation::getAutoGeneratedLocation(),
-                                SI->getSrc(), *VarInfo);
+              .createDebugValue(SI->getLoc(), SI->getSrc(), *VarInfo);
       }
   }
 }

test/DebugInfo/debug_info_expression.sil

@@ -8,6 +8,10 @@ struct MyStruct {
   var y: Builtin.Int64
 }
 
+struct SmallStruct {
+  var z : Builtin.Int64
+}
+
 sil_scope 1 { loc "file.swift":7:6 parent @test_fragment : $@convention(thin) () -> () }
 
 // Testing op_fragment w/ debug_value_addr
@@ -16,6 +20,8 @@ bb0:
   %2 = alloc_stack $MyStruct, var, name "my_struct", loc "file.swift":8:9, scope 1
   // CHECK: %[[MY_STRUCT:.+]] = alloca %{{.*}}MyStruct
   // CHECK: llvm.dbg.declare(metadata {{.*}}* %[[MY_STRUCT]], metadata ![[VAR_DECL_MD:[0-9]+]]
+  // CHECK: %[[SMALL_STRUCT:.+]] = alloca %{{.*}}SmallStruct
+  // CHECK: llvm.dbg.declare(metadata {{.*}}* %[[SMALL_STRUCT]], metadata ![[SMALL_VAR_DECL_MD:[0-9]+]]
   %3 = struct_element_addr %2 : $*MyStruct, #MyStruct.x, loc "file.swift":9:17, scope 1
   // CHECK: %[[FIELD_X:.*]] = getelementptr {{.*}} %[[MY_STRUCT]]
   // CHECK-SIL: debug_value_addr %{{[0-9]+}} : $*Builtin.Int64
@@ -25,6 +31,17 @@ bb0:
   // CHECK: llvm.dbg.value(metadata {{.*}}* %[[FIELD_X]], metadata ![[VAR_DECL_MD]]
   // CHECK-SAME:           !DIExpression(DW_OP_deref, DW_OP_LLVM_fragment, 0, 64)
   // CHECK-NOT:           ), !dbg ![[VAR_DECL_MD]]
+
+  %4 = alloc_stack $SmallStruct, var, name "small_struct", loc "file.swift":10:11, scope 1
+  %5 = struct_element_addr %4 : $*SmallStruct, #SmallStruct.z, loc "file.swift":11:13, scope 1
+  // CHECK: %[[FIELD_Z:.*]] = getelementptr {{.*}} %[[SMALL_STRUCT]]
+  // If the fragment covers the whole struct, we're not generating the
+  // DW_OP_LLVM_fragment part.
+  // CHECK: llvm.dbg.value(metadata {{.*}}* %[[FIELD_Z]], metadata ![[SMALL_VAR_DECL_MD]]
+  // CHECK-SAME:           !DIExpression(DW_OP_deref)
+  debug_value_addr %5 : $*Builtin.Int64, var, (name "small_struct", loc "file.swift":10:11, scope 1), type $SmallStruct, expr op_fragment:#SmallStruct.z, loc "file.swift":11:13, scope 1
+  dealloc_stack %4 : $*SmallStruct
+
   dealloc_stack %2 : $*MyStruct
   %r = tuple()
   return %r : $()

test/DebugInfo/sroa_mem2reg.sil

@@ -2,6 +2,8 @@
 // RUN: %target-sil-opt -enable-sil-verify-all -sil-print-debuginfo -sroa -mem2reg %s -o %t.sil
 // RUN: %FileCheck --check-prefix=CHECK-MEM2REG %s --input-file=%t.sil
 // RUN: %target-swiftc_driver -Xfrontend -disable-debugger-shadow-copies -g -emit-ir %t.sil | %FileCheck --check-prefix=CHECK-IR %s
+// RUN: %target-swiftc_driver -Xfrontend -disable-debugger-shadow-copies -g -c %t.sil -o %t.o
+// RUN: %llvm-dwarfdump --debug-info %t.o | %FileCheck --check-prefix=CHECK-DWARF %s
 sil_stage canonical
 
 import Builtin
@@ -78,3 +80,10 @@ bb0(%0 : $Int64, %1 : $Int64):
 // CHECK-IR-SAME:                            line: 7
 // CHECK-IR: ![[ARG2_MD]] = !DILocalVariable(name: "in_y", arg: 2
 // CHECK-IR-SAME:                            line: 7
+
+// CHECK-DWARF-LABEL:   DW_AT_name    ("foo")
+// CHECK-DWARF:       DW_TAG_variable
+// CHECK-DWARF:         DW_AT_name  ("my_struct")
+// Shouldn't be marked artificial
+// CHECK-DWARF-NOT:     DW_AT_artificial      (true)
+// CHECK-DWARF:       DW_TAG_{{.*}}