[wasm-simd][liftoff] Implement global get and set for S128

Most of the implementation work has been done as part of previous
patches, this finishes it by adding a new case for LoadType, and also
adding a test. The arm and arm64 implementation is new, and wasn't
required, since the Liftoff tests (in nooptimization variants) are
skipped on arm and arm64, and hence did not fail.

Bug: v8:9909
Change-Id: I01bd86d2e46de852bc067f44c802f66ac9e9b029
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2001561Reviewed-by: Clemens Backes <clemensb@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#65798}

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

@@ -372,20 +372,26 @@ void LiftoffAssembler::Load(LiftoffRegister dst, Register src_addr,
   }
   UseScratchRegisterScope temps(this);
   if (type.value() == LoadType::kF64Load ||
-      type.value() == LoadType::kF32Load) {
+      type.value() == LoadType::kF32Load ||
+      type.value() == LoadType::kS128Load) {
     Register actual_src_addr = liftoff::CalculateActualAddress(
         this, &temps, src_addr, offset_reg, offset_imm);
     if (type.value() == LoadType::kF64Load) {
       // Armv6 is not supported so Neon can be used to avoid alignment issues.
       CpuFeatureScope scope(this, NEON);
       vld1(Neon64, NeonListOperand(dst.fp()), NeonMemOperand(actual_src_addr));
-    } else {
+    } else if (type.value() == LoadType::kF32Load) {
       // TODO(arm): Use vld1 for f32 when implemented in simulator as used for
       // f64. It supports unaligned access.
       Register scratch =
           (actual_src_addr == src_addr) ? temps.Acquire() : actual_src_addr;
       ldr(scratch, MemOperand(actual_src_addr));
       vmov(liftoff::GetFloatRegister(dst.fp()), scratch);
+    } else {
+      // Armv6 is not supported so Neon can be used to avoid alignment issues.
+      CpuFeatureScope scope(this, NEON);
+      vld1(Neon8, NeonListOperand(dst.low_fp(), 2),
+           NeonMemOperand(actual_src_addr));
     }
   } else {
     MemOperand src_op =

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

@@ -296,6 +296,9 @@ void LiftoffAssembler::Load(LiftoffRegister dst, Register src_addr,
     case LoadType::kF64Load:
       Ldr(dst.fp().D(), src_op);
       break;
+    case LoadType::kS128Load:
+      Ldr(dst.fp().Q(), src_op);
+      break;
     default:
       UNREACHABLE();
   }

src/wasm/value-type.h

@@ -99,6 +99,8 @@ class LoadType {
         return kF32Load;
       case kWasmF64:
         return kF64Load;
+      case kWasmS128:
+        return kS128Load;
       default:
         UNREACHABLE();
     }

test/cctest/wasm/test-run-wasm-simd-liftoff.cc

@@ -34,6 +34,24 @@ WASM_SIMD_LIFTOFF_TEST(S128Local) {
   r.CheckUsedExecutionTier(ExecutionTier::kLiftoff);
 }
 
+WASM_SIMD_LIFTOFF_TEST(S128Global) {
+  WasmRunner<int32_t> r(ExecutionTier::kLiftoff, kNoLowerSimd);
+
+  int32_t* g0 = r.builder().AddGlobal<int32_t>(kWasmS128);
+  int32_t* g1 = r.builder().AddGlobal<int32_t>(kWasmS128);
+  BUILD(r, WASM_SET_GLOBAL(1, WASM_GET_GLOBAL(0)), WASM_ONE);
+
+  int32_t expected = 0x1234;
+  for (int i = 0; i < 4; i++) {
+    WriteLittleEndianValue<int32_t>(&g0[i], expected);
+  }
+  r.Call();
+  for (int i = 0; i < 4; i++) {
+    int32_t actual = ReadLittleEndianValue<int32_t>(&g1[i]);
+    CHECK_EQ(actual, expected);
+  }
+}
+
 #undef WASM_SIMD_LIFTOFF_TEST
 
 }  // namespace test_run_wasm_simd_liftoff