[wasm-simd][arm64] Skip top 64-bit write in MOVI if identical.

The MOVI instruction writes an immediate into every lane of a vector. In
order to support all immediates, we split it into two 64-bit
writes. However if the top and bottom 64 bits are identical, we can skip
the second write.

Bug: v8:11033
Change-Id: Idfb74471de815d278a84c78d204d534d19214b2c
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2507691Reviewed-by: Bill Budge <bbudge@chromium.org>
Reviewed-by: Zhi An Ng <zhin@chromium.org>
Commit-Queue: Pierre Langlois <pierre.langlois@arm.com>
Cr-Commit-Position: refs/heads/master@{#70987}

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

@@ -503,13 +503,15 @@ void TurboAssembler::Movi(const VRegister& vd, uint64_t imm, Shift shift,
 }
 
 void TurboAssembler::Movi(const VRegister& vd, uint64_t hi, uint64_t lo) {
-  // TODO(all): Move 128-bit values in a more efficient way.
+  // TODO(v8:11033): Move 128-bit values in a more efficient way.
   DCHECK(vd.Is128Bits());
-  UseScratchRegisterScope temps(this);
   Movi(vd.V2D(), lo);
-  Register temp = temps.AcquireX();
-  Mov(temp, hi);
-  Ins(vd.V2D(), 1, temp);
+  if (lo != hi) {
+    UseScratchRegisterScope temps(this);
+    Register temp = temps.AcquireX();
+    Mov(temp, hi);
+    Ins(vd.V2D(), 1, temp);
+  }
 }
 
 void TurboAssembler::Mvn(const Register& rd, const Operand& operand) {