[wasm] Implement 128-bit endian swap for simd type

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

 // Copyright 2016 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "src/assembler-inl.h"
 #include "test/cctest/cctest.h"
 #include "test/cctest/compiler/value-helper.h"
 #include "test/cctest/wasm/wasm-run-utils.h"
 #include "test/common/wasm/wasm-macro-gen.h"
 
@@ skipping 2191 matching lines @@
   int32_t* memory = r.module().AddMemoryElems<int32_t>(4);
 
   BUILD(r,
         WASM_STORE_MEM(MachineType::Simd128(), WASM_ZERO,
                        WASM_LOAD_MEM(MachineType::Simd128(), WASM_ZERO)),
         WASM_SIMD_I32x4_EXTRACT_LANE(
             0, WASM_LOAD_MEM(MachineType::Simd128(), WASM_ZERO)));
 
   FOR_INT32_INPUTS(i) {
     int32_t expected = *i;
+#if defined(V8_TARGET_LITTLE_ENDIAN)
     r.module().WriteMemory(&memory[0], expected);
+#else
+    r.module().WriteMemory(&memory[3], expected);

[Clemens Hammacher, 2017/04/26 18:40:05]

This does not look like the right fix. Storing the lanes in different order
would make endianess observable, breaking platform independence of wasm code.
I am no expert in the SIMD proposal, though.

Bill, can you comment on that?

[bbudge, 2017/04/26 20:14:12]

You're right Clemens. SIMD shouldn't be any different than the rest of
WebAssembly. Loads and stores are always in little-endian order, so this test
shouldn't change.

+#endif
     CHECK_EQ(expected, r.Call());
   }
 }
 #endif  // V8_TARGET_ARCH_ARM || SIMD_LOWERING_TARGET