diff --git a/clang/lib/CodeGen/CGStmt.cpp b/clang/lib/CodeGen/CGStmt.cpp
index e46cfb433ab8..d9fd406ad64e 100644
--- a/clang/lib/CodeGen/CGStmt.cpp
+++ b/clang/lib/CodeGen/CGStmt.cpp
@@ -1130,7 +1130,14 @@ void CodeGenFunction::EmitWhileStmt(const WhileStmt &S,
     if (!Weights && CGM.getCodeGenOpts().OptimizationLevel)
       BoolCondVal = emitCondLikelihoodViaExpectIntrinsic(
           BoolCondVal, Stmt::getLikelihood(S.getBody()));
-    Builder.CreateCondBr(BoolCondVal, LoopBody, ExitBlock, Weights);
+    auto *I = Builder.CreateCondBr(BoolCondVal, LoopBody, ExitBlock, Weights);
+    // Key Instructions: Emit the condition and branch as separate atoms to
+    // match existing loop stepping behaviour. FIXME: We could have the branch
+    // as the backup location for the condition, which would probably be a
+    // better experience. Explore this later.
+    if (auto *I = dyn_cast<llvm::Instruction>(BoolCondVal))
+      addInstToNewSourceAtom(I, nullptr);
+    addInstToNewSourceAtom(I, nullptr);
 
     if (ExitBlock != LoopExit.getBlock()) {
       EmitBlock(ExitBlock);
diff --git a/clang/test/KeyInstructions/while.c b/clang/test/KeyInstructions/while.c
new file mode 100644
index 000000000000..f1bb91520315
--- /dev/null
+++ b/clang/test/KeyInstructions/while.c
@@ -0,0 +1,34 @@
+// RUN: %clang -gkey-instructions -x c++ -std=c++17 %s -gmlt -S -emit-llvm -o - \
+// RUN: | FileCheck %s --implicit-check-not atomGroup --implicit-check-not atomRank
+
+// RUN: %clang -gkey-instructions -x c %s -gmlt -S -emit-llvm -o -  \
+// RUN: | FileCheck %s --implicit-check-not atomGroup --implicit-check-not atomRank
+
+// Perennial quesiton: should the `dec` be in its own source atom or not
+// (currently it is).
+
+// We've made the cmp and br separate source atoms for now, to match existing
+// behaviour in this case:
+// 1. while (
+// 2.   int i = --End
+// 3.   ) {
+// 4.   useValue(i);
+// 5. }
+// Without Key Instructions we go: 2, 1[, 4, 2, 1]+
+// Without separating cmp and br with Key Instructions we'd get:
+// 1[, 4, 1]+. If we made the cmp higher precedence than the
+// br and had them in the same group, we could get:
+// 2, [4, 2]+ which might be nicer. FIXME: do that later.
+
+void a(int A) {
+// CHECK: %dec = add nsw i32 %0, -1, !dbg [[G1R2:!.*]]
+// CHECK: store i32 %dec, ptr %A.addr{{.*}}, !dbg [[G1R1:!.*]]
+// CHECK: %tobool = icmp ne i32 %dec, 0, !dbg [[G2R1:!.*]]
+// CHECK: br i1 %tobool, label %while.body, label %while.end, !dbg [[G3R1:!.*]]
+    while (--A) { };
+}
+
+// CHECK: [[G1R2]] = !DILocation({{.*}}, atomGroup: 1, atomRank: 2)
+// CHECK: [[G1R1]] = !DILocation({{.*}}, atomGroup: 1, atomRank: 1)
+// CHECK: [[G2R1]] = !DILocation({{.*}}, atomGroup: 2, atomRank: 1)
+// CHECK: [[G3R1]] = !DILocation({{.*}}, atomGroup: 3, atomRank: 1)