Index: clang/include/clang/Basic/LangOptions.def
===================================================================
--- clang/include/clang/Basic/LangOptions.def
+++ clang/include/clang/Basic/LangOptions.def
@@ -145,6 +145,7 @@
 LANGOPT(NoMathBuiltin     , 1, 0, "disable math builtin functions")
 LANGOPT(GNUAsm            , 1, 1, "GNU-style inline assembly")
 LANGOPT(Coroutines        , 1, 0, "C++20 coroutines")
+LANGOPT(SplitCoroRamp     , 1, 0, "Split Ramp function for C++20 coroutines")
 LANGOPT(DllExportInlines  , 1, 1, "dllexported classes dllexport inline methods")
 LANGOPT(RelaxedTemplateTemplateArgs, 1, 0, "C++17 relaxed matching of template template arguments")
 
Index: clang/include/clang/Driver/Options.td
===================================================================
--- clang/include/clang/Driver/Options.td
+++ clang/include/clang/Driver/Options.td
@@ -1075,6 +1075,8 @@
   PosFlag<SetTrue, [CC1Option], "Enable support for the C++ Coroutines TS">,
   NegFlag<SetFalse>>;
 
+defm split_coroutine_ramp_function : OptInFFlag<"split-coroutine-ramp", "Enable split ramp function for C++ Coroutine function.">;
+
 def fembed_bitcode_EQ : Joined<["-"], "fembed-bitcode=">,
     Group<f_Group>, Flags<[NoXarchOption, CC1Option, CC1AsOption]>, MetaVarName<"<option>">,
     HelpText<"Embed LLVM bitcode (option: off, all, bitcode, marker)">,
Index: clang/lib/CodeGen/CGCoroutine.cpp
===================================================================
--- clang/lib/CodeGen/CGCoroutine.cpp
+++ clang/lib/CodeGen/CGCoroutine.cpp
@@ -172,6 +172,29 @@
     RValue RV;
   };
 }
+static void emitFakeSuspendExpression(CodeGenFunction &CGF, CGCoroData &Coro) {
+  SmallString<32> Prefix{"coro.fake.suspend"};
+  BasicBlock *ReadyBlock = CGF.createBasicBlock(Prefix + Twine(".ready"));
+  BasicBlock *SuspendBlock = CGF.createBasicBlock(Prefix + Twine(".suspend"));
+  BasicBlock *CleanupBlock = CGF.createBasicBlock(Prefix + Twine(".cleanup"));
+  CGF.EmitBlock(SuspendBlock);
+
+  auto &Builder = CGF.Builder;
+
+  llvm::Function *CoroSuspend =
+      CGF.CGM.getIntrinsic(llvm::Intrinsic::coro_suspend);
+  auto *SuspendResult = Builder.CreateCall(
+      CoroSuspend, {llvm::ConstantTokenNone::get(CGF.getLLVMContext()),
+                    Builder.getInt1(false)});
+
+  // Create a switch capturing three possible continuations.
+  auto *Switch = Builder.CreateSwitch(SuspendResult, Coro.SuspendBB, 2);
+  Switch->addCase(Builder.getInt8(0), ReadyBlock);
+  Switch->addCase(Builder.getInt8(1), CleanupBlock);
+
+  CGF.EmitBlock(CleanupBlock);
+  CGF.EmitBlock(ReadyBlock);
+}
 static LValueOrRValue emitSuspendExpression(CodeGenFunction &CGF, CGCoroData &Coro,
                                     CoroutineSuspendExpr const &S,
                                     AwaitKind Kind, AggValueSlot aggSlot,
@@ -631,6 +654,12 @@
 
     CurCoro.Data->CurrentAwaitKind = AwaitKind::Init;
     CurCoro.Data->ExceptionHandler = S.getExceptionHandler();
+
+    // Emit fake suspend in the front of initial suspend to reduce the size for
+    // ramp function in which case the initial suspend isn't always suspend.
+    if (getLangOpts().SplitCoroRamp)
+      emitFakeSuspendExpression(*this, *CurCoro.Data);
+
     EmitStmt(S.getInitSuspendStmt());
     CurCoro.Data->FinalJD = getJumpDestInCurrentScope(FinalBB);
 
@@ -687,6 +716,13 @@
   // resume and destroy parts of the coroutine should not include them.
   llvm::Function *CoroEnd = CGM.getIntrinsic(llvm::Intrinsic::coro_end);
   Builder.CreateCall(CoroEnd, {NullPtr, Builder.getFalse()});
+  if (getLangOpts().SplitCoroRamp) {
+    // Emit Call to resume function to keep program behavior
+    // if we try to split ramp function t
+    llvm::Function *CallToResume =
+        CGM.getIntrinsic(llvm::Intrinsic::coro_resume);
+    Builder.CreateCall(CallToResume, {CurCoro.Data->CoroBegin});
+  }
 
   if (Stmt *Ret = S.getReturnStmt())
     EmitStmt(Ret);
Index: clang/lib/Driver/ToolChains/Clang.cpp
===================================================================
--- clang/lib/Driver/ToolChains/Clang.cpp
+++ clang/lib/Driver/ToolChains/Clang.cpp
@@ -5835,6 +5835,12 @@
     CmdArgs.push_back("-fcoroutines-ts");
   }
 
+  if (Args.hasFlag(options::OPT_fsplit_coroutine_ramp_function,
+                   options::OPT_fno_split_coroutine_ramp_function, false) &&
+      types::isCXX(InputType)) {
+    CmdArgs.push_back("-fsplit-coroutine-ramp");
+  }
+
   Args.AddLastArg(CmdArgs, options::OPT_fdouble_square_bracket_attributes,
                   options::OPT_fno_double_square_bracket_attributes);
 
Index: clang/lib/Frontend/CompilerInvocation.cpp
===================================================================
--- clang/lib/Frontend/CompilerInvocation.cpp
+++ clang/lib/Frontend/CompilerInvocation.cpp
@@ -3688,6 +3688,7 @@
 
   Opts.Blocks = Args.hasArg(OPT_fblocks) || (Opts.OpenCL
     && Opts.OpenCLVersion == 200);
+  Opts.SplitCoroRamp = Args.hasArg(OPT_fsplit_coroutine_ramp_function);
 
   Opts.ConvergentFunctions = Opts.OpenCL || (Opts.CUDA && Opts.CUDAIsDevice) ||
                              Opts.SYCLIsDevice ||
Index: clang/test/CodeGenCoroutines/coro-split-ramp.cpp
===================================================================
--- /dev/null
+++ clang/test/CodeGenCoroutines/coro-split-ramp.cpp
@@ -0,0 +1,56 @@
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -emit-llvm -fcoroutines-ts \
+// RUN:   -fexperimental-new-pass-manager -O0 %s -fsplit-coroutine-ramp -o - | FileCheck %s
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -emit-llvm -fcoroutines-ts \
+// RUN:   -O0 %s -fsplit-coroutine-ramp -o - | FileCheck %s
+
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -emit-llvm -fcoroutines-ts \
+// RUN:   -fexperimental-new-pass-manager -O0 %s -o - | FileCheck %s --check-prefix=CHECK-NOSPLIT
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -emit-llvm -fcoroutines-ts \
+// RUN:   -O0 %s -o - | FileCheck %s --check-prefix=CHECK-NOSPLIT
+namespace std {
+namespace experimental {
+
+struct handle {};
+
+struct awaitable {
+  bool await_ready() noexcept { return true; }
+  inline void await_suspend(handle) noexcept {}
+  bool await_resume() noexcept { return true; }
+};
+
+template <typename T>
+struct coroutine_handle {
+  static handle from_address(void *address) noexcept { return {}; }
+};
+
+template <typename T = void>
+struct coroutine_traits {
+  struct promise_type {
+    awaitable initial_suspend() { return {}; }
+    awaitable final_suspend() noexcept { return {}; }
+    void return_void() {}
+    T get_return_object() { return T(); }
+    void unhandled_exception() {}
+  };
+};
+} // namespace experimental
+} // namespace std
+
+// The number of indexes would add one if we try to split ramp function
+// CHECK: %_Z3foov.Frame = {{.*}}i2
+// CHECK-LABEL: @_Z3foov
+// CHECK-LABEL: AfterCoroEnd:
+// CHECK:          %[[CASTING:.*]] = bitcast
+// CHECK:          %[[GEP:.*]] = getelementptr inbounds { i8*, i8* }, { i8*, i8* }* %[[CASTING]], i32 0, i32 0
+// CHECK:          %[[LOAD:.*]] = load i8*, i8** %[[GEP]], align
+// CHECK:          %[[FUNC_POINTER:.*]] = bitcast i8* %[[LOAD]] to void (i8*)*
+// CHECK:          call fastcc void %[[FUNC_POINTER]]
+// CHECK-NEXT:          ret void
+
+// CHECK-NOSPLIT: %_Z3foov.Frame = {{.*}}i1
+// CHECK-NOSPLIT-LABEL: @_Z3foov
+// CHECK-NOSPLIT-LABEL: AfterCoroEnd:
+// CHECK-NOSPLIT-NOT:         call fastcc void %
+// CHECK-NOSPLIT-NEXT:      ret void
+
+void foo() { co_return; }