[WebAssembly] Optimize away return instructions using fallthroughs.

This saves a small amount of code size, and is a first small step toward
passing values on the stack across block boundaries.

Differential Review: http://reviews.llvm.org/D20450

llvm-svn: 270294
diff --git a/llvm/lib/Target/WebAssembly/InstPrinter/WebAssemblyInstPrinter.cpp b/llvm/lib/Target/WebAssembly/InstPrinter/WebAssemblyInstPrinter.cpp
index a450a24b..267d716d 100644
--- a/llvm/lib/Target/WebAssembly/InstPrinter/WebAssemblyInstPrinter.cpp
+++ b/llvm/lib/Target/WebAssembly/InstPrinter/WebAssemblyInstPrinter.cpp
@@ -145,9 +145,9 @@
     if (int(WAReg) >= 0)
       printRegName(O, WAReg);
     else if (OpNo >= MII.get(MI->getOpcode()).getNumDefs())
-      O << "$pop" << (WAReg & INT32_MAX);
+      O << "$pop" << WebAssemblyFunctionInfo::getWARegStackId(WAReg);
     else if (WAReg != WebAssemblyFunctionInfo::UnusedReg)
-      O << "$push" << (WAReg & INT32_MAX);
+      O << "$push" << WebAssemblyFunctionInfo::getWARegStackId(WAReg);
     else
       O << "$drop";
     // Add a '=' suffix if this is a def.
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyAsmPrinter.cpp b/llvm/lib/Target/WebAssembly/WebAssemblyAsmPrinter.cpp
index d7a753d..64128bf 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyAsmPrinter.cpp
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyAsmPrinter.cpp
@@ -201,6 +201,30 @@
     // These represent values which are live into the function entry, so there's
     // no instruction to emit.
     break;
+  case WebAssembly::FALLTHROUGH_RETURN_I32:
+  case WebAssembly::FALLTHROUGH_RETURN_I64:
+  case WebAssembly::FALLTHROUGH_RETURN_F32:
+  case WebAssembly::FALLTHROUGH_RETURN_F64: {
+    // These instructions represent the implicit return at the end of a
+    // function body. The operand is always a pop.
+    assert(MFI->isVRegStackified(MI->getOperand(0).getReg()));
+
+    if (isVerbose()) {
+      OutStreamer->AddComment("fallthrough-return: $pop" +
+                              utostr(MFI->getWARegStackId(
+                                  MFI->getWAReg(MI->getOperand(0).getReg()))));
+      OutStreamer->AddBlankLine();
+    }
+    break;
+  }
+  case WebAssembly::FALLTHROUGH_RETURN_VOID:
+    // This instruction represents the implicit return at the end of a
+    // function body with no return value.
+    if (isVerbose()) {
+      OutStreamer->AddComment("fallthrough-return");
+      OutStreamer->AddBlankLine();
+    }
+    break;
   default: {
     WebAssemblyMCInstLower MCInstLowering(OutContext, *this);
     MCInst TmpInst;
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyInstrControl.td b/llvm/lib/Target/WebAssembly/WebAssemblyInstrControl.td
index ba4e1f9..444e275 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyInstrControl.td
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyInstrControl.td
@@ -71,6 +71,10 @@
 multiclass RETURN<WebAssemblyRegClass vt> {
   def RETURN_#vt : I<(outs), (ins vt:$val), [(WebAssemblyreturn vt:$val)],
                      "return  \t$val">;
+  // Equivalent to RETURN_#vt, for use at the end of a function when wasm
+  // semantics return by falling off the end of the block.
+  let isCodeGenOnly = 1 in
+  def FALLTHROUGH_RETURN_#vt : I<(outs), (ins vt:$val), []>;
 }
 
 let isTerminator = 1, hasCtrlDep = 1, isBarrier = 1 in {
@@ -80,6 +84,10 @@
   defm : RETURN<F32>;
   defm : RETURN<F64>;
   def RETURN_VOID : I<(outs), (ins), [(WebAssemblyreturn)], "return">;
+
+  // This is to RETURN_VOID what FALLTHROUGH_RETURN_#vt is to RETURN_#vt.
+  let isCodeGenOnly = 1 in
+  def FALLTHROUGH_RETURN_VOID : I<(outs), (ins), []>;
 } // isReturn = 1
   def UNREACHABLE : I<(outs), (ins), [(trap)], "unreachable">;
 } // isTerminator = 1, hasCtrlDep = 1, isBarrier = 1
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyMachineFunctionInfo.h b/llvm/lib/Target/WebAssembly/WebAssemblyMachineFunctionInfo.h
index 79950a6..c2f695d 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyMachineFunctionInfo.h
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyMachineFunctionInfo.h
@@ -86,6 +86,12 @@
     assert(VReg = WARegs.size());
     WARegs.push_back(WAReg);
   }
+
+  // For a given stackified WAReg, return the id number to print with push/pop.
+  static unsigned getWARegStackId(unsigned Reg) {
+    assert(Reg & INT32_MIN);
+    return Reg & INT32_MAX;
+  }
 };
 
 } // end namespace llvm
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyPeephole.cpp b/llvm/lib/Target/WebAssembly/WebAssemblyPeephole.cpp
index 19c227d..56d44e6 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyPeephole.cpp
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyPeephole.cpp
@@ -12,17 +12,23 @@
 ///
 //===----------------------------------------------------------------------===//
 
-#include "WebAssembly.h"
 #include "MCTargetDesc/WebAssemblyMCTargetDesc.h"
+#include "WebAssembly.h"
 #include "WebAssemblyMachineFunctionInfo.h"
 #include "WebAssemblySubtarget.h"
 #include "llvm/Analysis/TargetLibraryInfo.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 using namespace llvm;
 
 #define DEBUG_TYPE "wasm-peephole"
 
+static cl::opt<bool> DisableWebAssemblyFallthroughReturnOpt(
+    "disable-wasm-fallthrough-return-opt", cl::Hidden,
+    cl::desc("WebAssembly: Disable fallthrough-return optimizations."),
+    cl::init(false));
+
 namespace {
 class WebAssemblyPeephole final : public MachineFunctionPass {
   const char *getPassName() const override {
@@ -50,8 +56,7 @@
 
 /// If desirable, rewrite NewReg to a drop register.
 static bool MaybeRewriteToDrop(unsigned OldReg, unsigned NewReg,
-                               MachineOperand &MO,
-                               WebAssemblyFunctionInfo &MFI,
+                               MachineOperand &MO, WebAssemblyFunctionInfo &MFI,
                                MachineRegisterInfo &MRI) {
   bool Changed = false;
   if (OldReg == NewReg) {
@@ -60,19 +65,50 @@
     MO.setReg(NewReg);
     MO.setIsDead();
     MFI.stackifyVReg(NewReg);
-    MFI.addWAReg(NewReg, WebAssemblyFunctionInfo::UnusedReg);
   }
   return Changed;
 }
 
+static bool MaybeRewriteToFallthrough(MachineInstr &MI, MachineBasicBlock &MBB,
+                                      const MachineFunction &MF,
+                                      WebAssemblyFunctionInfo &MFI,
+                                      MachineRegisterInfo &MRI,
+                                      const WebAssemblyInstrInfo &TII,
+                                      unsigned FallthroughOpc,
+                                      unsigned CopyLocalOpc) {
+  if (DisableWebAssemblyFallthroughReturnOpt)
+    return false;
+  if (&MBB != &MF.back())
+    return false;
+  if (&MI != &MBB.back())
+    return false;
+
+  // If the operand isn't stackified, insert a COPY_LOCAL to read the operand
+  // and stackify it.
+  MachineOperand &MO = MI.getOperand(0);
+  unsigned Reg = MO.getReg();
+  if (!MFI.isVRegStackified(Reg)) {
+    unsigned NewReg = MRI.createVirtualRegister(MRI.getRegClass(Reg));
+    BuildMI(MBB, MI, MI.getDebugLoc(), TII.get(CopyLocalOpc), NewReg)
+        .addReg(Reg);
+    MO.setReg(NewReg);
+    MFI.stackifyVReg(NewReg);
+  }
+
+  // Rewrite the return.
+  MI.setDesc(TII.get(FallthroughOpc));
+  return true;
+}
+
 bool WebAssemblyPeephole::runOnMachineFunction(MachineFunction &MF) {
   DEBUG({
-    dbgs() << "********** Store Results **********\n"
+    dbgs() << "********** Peephole **********\n"
            << "********** Function: " << MF.getName() << '\n';
   });
 
   MachineRegisterInfo &MRI = MF.getRegInfo();
   WebAssemblyFunctionInfo &MFI = *MF.getInfo<WebAssemblyFunctionInfo>();
+  const auto &TII = *MF.getSubtarget<WebAssemblySubtarget>().getInstrInfo();
   const WebAssemblyTargetLowering &TLI =
       *MF.getSubtarget<WebAssemblySubtarget>().getTargetLowering();
   auto &LibInfo = getAnalysis<TargetLibraryInfoWrapperPass>().getTLI();
@@ -127,7 +163,34 @@
             }
           }
         }
+        break;
       }
+      // Optimize away an explicit void return at the end of the function.
+      case WebAssembly::RETURN_I32:
+        Changed |= MaybeRewriteToFallthrough(
+            MI, MBB, MF, MFI, MRI, TII, WebAssembly::FALLTHROUGH_RETURN_I32,
+            WebAssembly::COPY_LOCAL_I32);
+        break;
+      case WebAssembly::RETURN_I64:
+        Changed |= MaybeRewriteToFallthrough(
+            MI, MBB, MF, MFI, MRI, TII, WebAssembly::FALLTHROUGH_RETURN_I64,
+            WebAssembly::COPY_LOCAL_I64);
+        break;
+      case WebAssembly::RETURN_F32:
+        Changed |= MaybeRewriteToFallthrough(
+            MI, MBB, MF, MFI, MRI, TII, WebAssembly::FALLTHROUGH_RETURN_F32,
+            WebAssembly::COPY_LOCAL_F32);
+        break;
+      case WebAssembly::RETURN_F64:
+        Changed |= MaybeRewriteToFallthrough(
+            MI, MBB, MF, MFI, MRI, TII, WebAssembly::FALLTHROUGH_RETURN_F64,
+            WebAssembly::COPY_LOCAL_F64);
+        break;
+      case WebAssembly::RETURN_VOID:
+        if (!DisableWebAssemblyFallthroughReturnOpt &&
+            &MBB == &MF.back() && &MI == &MBB.back())
+          MI.setDesc(TII.get(WebAssembly::FALLTHROUGH_RETURN_VOID));
+        break;
       }
 
   return Changed;
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyTargetMachine.cpp b/llvm/lib/Target/WebAssembly/WebAssemblyTargetMachine.cpp
index e4b049d..32154af 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyTargetMachine.cpp
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyTargetMachine.cpp
@@ -225,10 +225,10 @@
   // Lower br_unless into br_if.
   addPass(createWebAssemblyLowerBrUnless());
 
-  // Create a mapping from LLVM CodeGen virtual registers to wasm registers.
-  addPass(createWebAssemblyRegNumbering());
-
   // Perform the very last peephole optimizations on the code.
   if (getOptLevel() != CodeGenOpt::None)
     addPass(createWebAssemblyPeephole());
+
+  // Create a mapping from LLVM CodeGen virtual registers to wasm registers.
+  addPass(createWebAssemblyRegNumbering());
 }
diff --git a/llvm/test/CodeGen/WebAssembly/address-offsets.ll b/llvm/test/CodeGen/WebAssembly/address-offsets.ll
index 875e229..6403b37 100644
--- a/llvm/test/CodeGen/WebAssembly/address-offsets.ll
+++ b/llvm/test/CodeGen/WebAssembly/address-offsets.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test folding constant offsets and symbols into load and store addresses under
 ; a variety of circumstances.
diff --git a/llvm/test/CodeGen/WebAssembly/byval.ll b/llvm/test/CodeGen/WebAssembly/byval.ll
index ffc3833..bc8f1eb8 100644
--- a/llvm/test/CodeGen/WebAssembly/byval.ll
+++ b/llvm/test/CodeGen/WebAssembly/byval.ll
@@ -1,5 +1,5 @@
-; RUN: llc < %s -asm-verbose=false -verify-machineinstrs | FileCheck %s
-; RUN: llc < %s -asm-verbose=false -verify-machineinstrs -fast-isel | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -verify-machineinstrs | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -verify-machineinstrs -fast-isel | FileCheck %s
 
 target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
 target triple = "wasm32-unknown-unknown"
diff --git a/llvm/test/CodeGen/WebAssembly/call.ll b/llvm/test/CodeGen/WebAssembly/call.ll
index 9e05da0..bd5f7b8 100644
--- a/llvm/test/CodeGen/WebAssembly/call.ll
+++ b/llvm/test/CodeGen/WebAssembly/call.ll
@@ -1,5 +1,5 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
-; RUN: llc < %s -asm-verbose=false -fast-isel -fast-isel-abort=1 | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -fast-isel -fast-isel-abort=1 | FileCheck %s
 
 ; Test that basic call operations assemble as expected.
 
diff --git a/llvm/test/CodeGen/WebAssembly/cfg-stackify.ll b/llvm/test/CodeGen/WebAssembly/cfg-stackify.ll
index 8e3529a..d46898b 100644
--- a/llvm/test/CodeGen/WebAssembly/cfg-stackify.ll
+++ b/llvm/test/CodeGen/WebAssembly/cfg-stackify.ll
@@ -1,5 +1,5 @@
-; RUN: llc < %s -asm-verbose=false -disable-block-placement -verify-machineinstrs -fast-isel=false | FileCheck %s
-; RUN: llc < %s -asm-verbose=false -verify-machineinstrs -fast-isel=false | FileCheck -check-prefix=OPT %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -disable-block-placement -verify-machineinstrs -fast-isel=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -verify-machineinstrs -fast-isel=false | FileCheck -check-prefix=OPT %s
 
 ; Test the CFG stackifier pass.
 
diff --git a/llvm/test/CodeGen/WebAssembly/comparisons_f32.ll b/llvm/test/CodeGen/WebAssembly/comparisons_f32.ll
index 2d324f7..10e037d 100644
--- a/llvm/test/CodeGen/WebAssembly/comparisons_f32.ll
+++ b/llvm/test/CodeGen/WebAssembly/comparisons_f32.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that basic 32-bit floating-point comparison operations assemble as
 ; expected.
diff --git a/llvm/test/CodeGen/WebAssembly/comparisons_f64.ll b/llvm/test/CodeGen/WebAssembly/comparisons_f64.ll
index 22fbc1a..7d038a0 100644
--- a/llvm/test/CodeGen/WebAssembly/comparisons_f64.ll
+++ b/llvm/test/CodeGen/WebAssembly/comparisons_f64.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that basic 64-bit floating-point comparison operations assemble as
 ; expected.
diff --git a/llvm/test/CodeGen/WebAssembly/comparisons_i32.ll b/llvm/test/CodeGen/WebAssembly/comparisons_i32.ll
index 84b19bd..d2ba73f 100644
--- a/llvm/test/CodeGen/WebAssembly/comparisons_i32.ll
+++ b/llvm/test/CodeGen/WebAssembly/comparisons_i32.ll
@@ -1,5 +1,5 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
-; RUN: llc < %s -asm-verbose=false -fast-isel -fast-isel-abort=1 | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -fast-isel -fast-isel-abort=1 | FileCheck %s
 
 ; Test that basic 32-bit integer comparison operations assemble as expected.
 
diff --git a/llvm/test/CodeGen/WebAssembly/comparisons_i64.ll b/llvm/test/CodeGen/WebAssembly/comparisons_i64.ll
index 7d2b1d9..80950ae 100644
--- a/llvm/test/CodeGen/WebAssembly/comparisons_i64.ll
+++ b/llvm/test/CodeGen/WebAssembly/comparisons_i64.ll
@@ -1,5 +1,5 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
-; RUN: llc < %s -asm-verbose=false -fast-isel -fast-isel-abort=1 | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -fast-isel -fast-isel-abort=1 | FileCheck %s
 
 ; Test that basic 64-bit integer comparison operations assemble as expected.
 
diff --git a/llvm/test/CodeGen/WebAssembly/conv.ll b/llvm/test/CodeGen/WebAssembly/conv.ll
index 1a4bd72..27cebb1 100644
--- a/llvm/test/CodeGen/WebAssembly/conv.ll
+++ b/llvm/test/CodeGen/WebAssembly/conv.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that basic conversion operations assemble as expected.
 
diff --git a/llvm/test/CodeGen/WebAssembly/f32.ll b/llvm/test/CodeGen/WebAssembly/f32.ll
index c32a7c3..1c1d819 100644
--- a/llvm/test/CodeGen/WebAssembly/f32.ll
+++ b/llvm/test/CodeGen/WebAssembly/f32.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that basic 32-bit floating-point operations assemble as expected.
 
diff --git a/llvm/test/CodeGen/WebAssembly/f64.ll b/llvm/test/CodeGen/WebAssembly/f64.ll
index 9228499..670f3f0 100644
--- a/llvm/test/CodeGen/WebAssembly/f64.ll
+++ b/llvm/test/CodeGen/WebAssembly/f64.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that basic 64-bit floating-point operations assemble as expected.
 
diff --git a/llvm/test/CodeGen/WebAssembly/frem.ll b/llvm/test/CodeGen/WebAssembly/frem.ll
index b8c80fbe..b874522 100644
--- a/llvm/test/CodeGen/WebAssembly/frem.ll
+++ b/llvm/test/CodeGen/WebAssembly/frem.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that the frem instruction works.
 
diff --git a/llvm/test/CodeGen/WebAssembly/func.ll b/llvm/test/CodeGen/WebAssembly/func.ll
index b7122a3..71c00a4 100644
--- a/llvm/test/CodeGen/WebAssembly/func.ll
+++ b/llvm/test/CodeGen/WebAssembly/func.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that basic functions assemble as expected.
 
diff --git a/llvm/test/CodeGen/WebAssembly/global.ll b/llvm/test/CodeGen/WebAssembly/global.ll
index 704fb3f..1d24035 100644
--- a/llvm/test/CodeGen/WebAssembly/global.ll
+++ b/llvm/test/CodeGen/WebAssembly/global.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that globals assemble as expected.
 
diff --git a/llvm/test/CodeGen/WebAssembly/i128.ll b/llvm/test/CodeGen/WebAssembly/i128.ll
index 5c82752..29bf787 100644
--- a/llvm/test/CodeGen/WebAssembly/i128.ll
+++ b/llvm/test/CodeGen/WebAssembly/i128.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that basic 128-bit integer operations assemble as expected.
 
diff --git a/llvm/test/CodeGen/WebAssembly/i32-load-store-alignment.ll b/llvm/test/CodeGen/WebAssembly/i32-load-store-alignment.ll
index e7732e7..b254413 100644
--- a/llvm/test/CodeGen/WebAssembly/i32-load-store-alignment.ll
+++ b/llvm/test/CodeGen/WebAssembly/i32-load-store-alignment.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test loads and stores with custom alignment values.
 
diff --git a/llvm/test/CodeGen/WebAssembly/i32.ll b/llvm/test/CodeGen/WebAssembly/i32.ll
index 945a447..a07dd02 100644
--- a/llvm/test/CodeGen/WebAssembly/i32.ll
+++ b/llvm/test/CodeGen/WebAssembly/i32.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that basic 32-bit integer operations assemble as expected.
 
diff --git a/llvm/test/CodeGen/WebAssembly/i64-load-store-alignment.ll b/llvm/test/CodeGen/WebAssembly/i64-load-store-alignment.ll
index fb38209..b2fb962 100644
--- a/llvm/test/CodeGen/WebAssembly/i64-load-store-alignment.ll
+++ b/llvm/test/CodeGen/WebAssembly/i64-load-store-alignment.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test loads and stores with custom alignment values.
 
diff --git a/llvm/test/CodeGen/WebAssembly/i64.ll b/llvm/test/CodeGen/WebAssembly/i64.ll
index 85a8ae3..93e32bfc 100644
--- a/llvm/test/CodeGen/WebAssembly/i64.ll
+++ b/llvm/test/CodeGen/WebAssembly/i64.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that basic 64-bit integer operations assemble as expected.
 
diff --git a/llvm/test/CodeGen/WebAssembly/immediates.ll b/llvm/test/CodeGen/WebAssembly/immediates.ll
index 0ccfe09..3d11f94 100644
--- a/llvm/test/CodeGen/WebAssembly/immediates.ll
+++ b/llvm/test/CodeGen/WebAssembly/immediates.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that basic immediates assemble as expected.
 
diff --git a/llvm/test/CodeGen/WebAssembly/inline-asm.ll b/llvm/test/CodeGen/WebAssembly/inline-asm.ll
index 0d12e03..d36c32b 100644
--- a/llvm/test/CodeGen/WebAssembly/inline-asm.ll
+++ b/llvm/test/CodeGen/WebAssembly/inline-asm.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false -no-integrated-as | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -no-integrated-as | FileCheck %s
 
 ; Test basic inline assembly. Pass -no-integrated-as since these aren't
 ; actually valid assembly syntax.
diff --git a/llvm/test/CodeGen/WebAssembly/legalize.ll b/llvm/test/CodeGen/WebAssembly/legalize.ll
index 5feb2e8..5cbfb8a 100644
--- a/llvm/test/CodeGen/WebAssembly/legalize.ll
+++ b/llvm/test/CodeGen/WebAssembly/legalize.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test various types and operators that need to be legalized.
 
diff --git a/llvm/test/CodeGen/WebAssembly/load-ext.ll b/llvm/test/CodeGen/WebAssembly/load-ext.ll
index d52df33..48a7ce7 100644
--- a/llvm/test/CodeGen/WebAssembly/load-ext.ll
+++ b/llvm/test/CodeGen/WebAssembly/load-ext.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that extending loads are assembled properly.
 
diff --git a/llvm/test/CodeGen/WebAssembly/load-store-i1.ll b/llvm/test/CodeGen/WebAssembly/load-store-i1.ll
index 2b8ab4a..2a2318f 100644
--- a/llvm/test/CodeGen/WebAssembly/load-store-i1.ll
+++ b/llvm/test/CodeGen/WebAssembly/load-store-i1.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that i1 extending loads and truncating stores are assembled properly.
 
diff --git a/llvm/test/CodeGen/WebAssembly/load.ll b/llvm/test/CodeGen/WebAssembly/load.ll
index 776dba3..a8e174e 100644
--- a/llvm/test/CodeGen/WebAssembly/load.ll
+++ b/llvm/test/CodeGen/WebAssembly/load.ll
@@ -1,5 +1,5 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
-; RUN: llc < %s -asm-verbose=false -fast-isel -fast-isel-abort=1 | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -fast-isel -fast-isel-abort=1 | FileCheck %s
 
 ; Test that basic loads are assembled properly.
 
diff --git a/llvm/test/CodeGen/WebAssembly/mem-intrinsics.ll b/llvm/test/CodeGen/WebAssembly/mem-intrinsics.ll
index a49ff57..71787fe 100644
--- a/llvm/test/CodeGen/WebAssembly/mem-intrinsics.ll
+++ b/llvm/test/CodeGen/WebAssembly/mem-intrinsics.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test memcpy, memmove, and memset intrinsics.
 
diff --git a/llvm/test/CodeGen/WebAssembly/memory-addr32.ll b/llvm/test/CodeGen/WebAssembly/memory-addr32.ll
index 2c2d075..583201b 100644
--- a/llvm/test/CodeGen/WebAssembly/memory-addr32.ll
+++ b/llvm/test/CodeGen/WebAssembly/memory-addr32.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that basic memory operations assemble as expected with 32-bit addresses.
 
diff --git a/llvm/test/CodeGen/WebAssembly/memory-addr64.ll b/llvm/test/CodeGen/WebAssembly/memory-addr64.ll
index 804274d..dc6da61 100644
--- a/llvm/test/CodeGen/WebAssembly/memory-addr64.ll
+++ b/llvm/test/CodeGen/WebAssembly/memory-addr64.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that basic memory operations assemble as expected with 64-bit addresses.
 
diff --git a/llvm/test/CodeGen/WebAssembly/phi.ll b/llvm/test/CodeGen/WebAssembly/phi.ll
index 00e5859..747ae5c 100644
--- a/llvm/test/CodeGen/WebAssembly/phi.ll
+++ b/llvm/test/CodeGen/WebAssembly/phi.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false -verify-machineinstrs | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -verify-machineinstrs | FileCheck %s
 
 ; Test that phis are lowered.
 
diff --git a/llvm/test/CodeGen/WebAssembly/reg-stackify.ll b/llvm/test/CodeGen/WebAssembly/reg-stackify.ll
index f112534..2f70a56 100644
--- a/llvm/test/CodeGen/WebAssembly/reg-stackify.ll
+++ b/llvm/test/CodeGen/WebAssembly/reg-stackify.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false -verify-machineinstrs | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -verify-machineinstrs | FileCheck %s
 
 ; Test the register stackifier pass.
 
diff --git a/llvm/test/CodeGen/WebAssembly/return-int32.ll b/llvm/test/CodeGen/WebAssembly/return-int32.ll
index cbc9ec9..9e663b9 100644
--- a/llvm/test/CodeGen/WebAssembly/return-int32.ll
+++ b/llvm/test/CodeGen/WebAssembly/return-int32.ll
@@ -5,7 +5,30 @@
 target triple = "wasm32-unknown-unknown"
 
 ; CHECK-LABEL: return_i32:
-; CHECK: return $0{{$}}
+; CHECK-NEXT:  .param i32{{$}}
+; CHECK-NEXT:  .result i32{{$}}
+; CHECK-NEXT:  copy_local  $push0=, $0
+; CHECK-NEXT:  .endfunc{{$}}
 define i32 @return_i32(i32 %p) {
   ret i32 %p
 }
+
+; CHECK-LABEL: return_i32_twice:
+; CHECK:      store
+; CHECK-NEXT: i32.const $push[[L0:[^,]+]]=, 1{{$}}
+; CHECK-NEXT: return $pop[[L0]]{{$}}
+; CHECK:      store
+; CHECK-NEXT: i32.const $push{{[^,]+}}=, 3{{$}}
+; CHECK-NEXT: .endfunc{{$}}
+define i32 @return_i32_twice(i32 %a) {
+  %b = icmp ne i32 %a, 0
+  br i1 %b, label %true, label %false
+
+true:
+  store i32 0, i32* null
+  ret i32 1 
+
+false:
+  store i32 2, i32* null
+  ret i32 3
+}
diff --git a/llvm/test/CodeGen/WebAssembly/return-void.ll b/llvm/test/CodeGen/WebAssembly/return-void.ll
index cf4d168..c3a600f 100644
--- a/llvm/test/CodeGen/WebAssembly/return-void.ll
+++ b/llvm/test/CodeGen/WebAssembly/return-void.ll
@@ -5,7 +5,25 @@
 target triple = "wasm32-unknown-unknown"
 
 ; CHECK-LABEL: return_void:
-; CHECK: return{{$}}
+; CHECK-NEXT: .endfunc{{$}}
 define void @return_void() {
   ret void
 }
+
+; CHECK-LABEL: return_void_twice:
+; CHECK:      store
+; CHECK-NEXT: return{{$}}
+; CHECK:      store
+; CHECK-NEXT: .endfunc{{$}}
+define void @return_void_twice(i32 %a) {
+  %b = icmp ne i32 %a, 0
+  br i1 %b, label %true, label %false
+
+true:
+  store i32 0, i32* null
+  ret void
+
+false:
+  store i32 1, i32* null
+  ret void
+}
diff --git a/llvm/test/CodeGen/WebAssembly/returned.ll b/llvm/test/CodeGen/WebAssembly/returned.ll
index 8f590c0..a277928 100644
--- a/llvm/test/CodeGen/WebAssembly/returned.ll
+++ b/llvm/test/CodeGen/WebAssembly/returned.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that the "returned" attribute is optimized effectively.
 
diff --git a/llvm/test/CodeGen/WebAssembly/select.ll b/llvm/test/CodeGen/WebAssembly/select.ll
index 6e26da5..06837e4 100644
--- a/llvm/test/CodeGen/WebAssembly/select.ll
+++ b/llvm/test/CodeGen/WebAssembly/select.ll
@@ -1,5 +1,5 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
-; RUN: llc < %s -asm-verbose=false -fast-isel -fast-isel-abort=1 | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -fast-isel -fast-isel-abort=1 | FileCheck %s
 
 ; Test that wasm select instruction is selected from LLVM select instruction.
 
diff --git a/llvm/test/CodeGen/WebAssembly/signext-zeroext.ll b/llvm/test/CodeGen/WebAssembly/signext-zeroext.ll
index 27e524b..f9561da 100644
--- a/llvm/test/CodeGen/WebAssembly/signext-zeroext.ll
+++ b/llvm/test/CodeGen/WebAssembly/signext-zeroext.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test zeroext and signext ABI keywords
 
diff --git a/llvm/test/CodeGen/WebAssembly/store-results.ll b/llvm/test/CodeGen/WebAssembly/store-results.ll
index b2dfc70..121ee91 100644
--- a/llvm/test/CodeGen/WebAssembly/store-results.ll
+++ b/llvm/test/CodeGen/WebAssembly/store-results.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Test that the wasm-store-results pass makes users of stored values use the
 ; result of store expressions to reduce get_local/set_local traffic.
diff --git a/llvm/test/CodeGen/WebAssembly/store.ll b/llvm/test/CodeGen/WebAssembly/store.ll
index 03bd28f..3ff8488 100644
--- a/llvm/test/CodeGen/WebAssembly/store.ll
+++ b/llvm/test/CodeGen/WebAssembly/store.ll
@@ -1,5 +1,5 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
-; RUN: llc < %s -asm-verbose=false -fast-isel -fast-isel-abort=1 | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -fast-isel -fast-isel-abort=1 | FileCheck %s
 
 ; Test that basic stores are assembled properly.
 
diff --git a/llvm/test/CodeGen/WebAssembly/switch.ll b/llvm/test/CodeGen/WebAssembly/switch.ll
index 20af644..8355bc8 100644
--- a/llvm/test/CodeGen/WebAssembly/switch.ll
+++ b/llvm/test/CodeGen/WebAssembly/switch.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false -disable-block-placement -verify-machineinstrs | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -disable-block-placement -verify-machineinstrs | FileCheck %s
 
 ; Test switch instructions. Block placement is disabled because it reorders
 ; the blocks in a way that isn't interesting here.
diff --git a/llvm/test/CodeGen/WebAssembly/unused-argument.ll b/llvm/test/CodeGen/WebAssembly/unused-argument.ll
index bfbb7cb..ff943b2 100644
--- a/llvm/test/CodeGen/WebAssembly/unused-argument.ll
+++ b/llvm/test/CodeGen/WebAssembly/unused-argument.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 ; Make sure that argument offsets are correct even if some arguments are unused.
 
diff --git a/llvm/test/CodeGen/WebAssembly/userstack.ll b/llvm/test/CodeGen/WebAssembly/userstack.ll
index 90b3eef..ee2bbfb 100644
--- a/llvm/test/CodeGen/WebAssembly/userstack.ll
+++ b/llvm/test/CodeGen/WebAssembly/userstack.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s
 
 target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
 target triple = "wasm32-unknown-unknown"
diff --git a/llvm/test/CodeGen/WebAssembly/varargs.ll b/llvm/test/CodeGen/WebAssembly/varargs.ll
index 990db2d..483d452 100644
--- a/llvm/test/CodeGen/WebAssembly/varargs.ll
+++ b/llvm/test/CodeGen/WebAssembly/varargs.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=false -verify-machineinstrs | FileCheck %s
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -verify-machineinstrs | FileCheck %s
 
 ; Test varargs constructs.