[wasm-simd][liftoff][arm64] Implement i64x2.bitmask

Move the current code sequence in TurboFan to a macro-assembler helper
function to allow Liftoff to reuse it.

Bug: v8:10997
Change-Id: I6205350897a4afc7ca9d0f84fd514be24508aef0
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2620905Reviewed-by: Clemens Backes <clemensb@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#72054}

src/codegen/arm64/macro-assembler-arm64.cc

@@ -3408,6 +3408,16 @@ void TurboAssembler::StoreReturnAddressInWasmExitFrame(Label* return_location) {
   Str(x17, MemOperand(fp, WasmExitFrameConstants::kCallingPCOffset));
 }
 
+void TurboAssembler::I64x2BitMask(Register dst, VRegister src) {
+  UseScratchRegisterScope scope(this);
+  VRegister tmp1 = scope.AcquireV(kFormat2D);
+  Register tmp2 = scope.AcquireX();
+  Ushr(tmp1.V2D(), src.V2D(), 63);
+  Mov(dst.X(), tmp1.D(), 0);
+  Mov(tmp2.X(), tmp1.D(), 1);
+  Add(dst.W(), dst.W(), Operand(tmp2.W(), LSL, 1));
+}
+
 }  // namespace internal
 }  // namespace v8
 

src/codegen/arm64/macro-assembler-arm64.h

@@ -1371,6 +1371,11 @@ class V8_EXPORT_PRIVATE TurboAssembler : public TurboAssemblerBase {
 
   void StoreReturnAddressInWasmExitFrame(Label* return_location);
 
+  // Wasm SIMD helpers. These instructions don't have direct lowering to native
+  // instructions. These helpers allow us to define the optimal code sequence,
+  // and be used in both TurboFan and Liftoff.
+  void I64x2BitMask(Register dst, VRegister src);
+
  protected:
   // The actual Push and Pop implementations. These don't generate any code
   // other than that required for the push or pop. This allows

src/compiler/backend/arm64/code-generator-arm64.cc

@@ -2222,15 +2222,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       break;
     }
     case kArm64I64x2BitMask: {
-      UseScratchRegisterScope scope(tasm());
-      Register dst = i.OutputRegister32();
-      VRegister src = i.InputSimd128Register(0);
-      VRegister tmp1 = scope.AcquireV(kFormat2D);
-      Register tmp2 = scope.AcquireX();
-      __ Ushr(tmp1.V2D(), src.V2D(), 63);
-      __ Mov(dst.X(), tmp1.D(), 0);
-      __ Mov(tmp2.X(), tmp1.D(), 1);
-      __ Add(dst.W(), dst.W(), Operand(tmp2.W(), LSL, 1));
+      __ I64x2BitMask(i.OutputRegister32(), i.InputSimd128Register(0));
       break;
     }
     case kArm64I32x4Splat: {

src/wasm/baseline/arm64/liftoff-assembler-arm64.h

@@ -2014,7 +2014,7 @@ void LiftoffAssembler::emit_i64x2_extmul_high_i32x4_u(LiftoffRegister dst,
 
 void LiftoffAssembler::emit_i64x2_bitmask(LiftoffRegister dst,
                                           LiftoffRegister src) {
-  bailout(kSimd, "i64x2_bitmask");
+  I64x2BitMask(dst.gp(), src.fp());
 }
 
 void LiftoffAssembler::emit_i32x4_splat(LiftoffRegister dst,