Improve the effectiveness of BDCE's debug info salvaging

This patch improves the effectiveness of BDCE's debug info salvaging
by processing the instructions in reverse order and delaying
dropAllReferences until after debug info salvaging. This allows
salvaging of entire chains of deleted instructions!

Previously we would remove all references from an instruction, which
would make it impossible to use that instruction to salvage a later
instruction in the instruction stream, because its operands were
already removed.

This reapplies the previous patch with a fix for a use-after-free.

Differential Revision: https://reviews.llvm.org/D110568

llvm/lib/Transforms/Scalar/BDCE.cpp

@@ -106,9 +106,7 @@ static bool bitTrackingDCE(Function &F, DemandedBits &DB) {
         (I.getType()->isIntOrIntVectorTy() &&
          DB.getDemandedBits(&I).isNullValue() &&
          wouldInstructionBeTriviallyDead(&I))) {
-      salvageDebugInfo(I);
       Worklist.push_back(&I);
-      I.dropAllReferences();
       Changed = true;
       continue;
     }
@@ -155,6 +153,11 @@ static bool bitTrackingDCE(Function &F, DemandedBits &DB) {
     }
   }
 
+  for (Instruction *&I : llvm::reverse(Worklist)) {
+    salvageDebugInfo(*I);
+    I->dropAllReferences();
+  }
+
   for (Instruction *&I : Worklist) {
     ++NumRemoved;
     I->eraseFromParent();

llvm/test/Transforms/Util/salvage-debuginfo.ll

@@ -1,4 +1,5 @@
 ; RUN: opt -adce %s -S -o - | FileCheck %s
+; RUN: opt -bdce %s -S -o - | FileCheck %s
 target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-apple-macosx"
 define void @f(i32) !dbg !8 {