[wasm-simd] Fix max alignment for load extends

Load extends always load 64-bits. Previously, we were setting the max
alignment to be the size_log_2 of the load_type. For LoadExtends the
load_type indicates what the lane size to be extended is, *NOT* the size
to be loaded.

Bug: chromium:1082848
Change-Id: I0c4115ea6ec916211b03afdb83376ccc05c0c244
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2202721Reviewed-by: Bill Budge <bbudge@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#67815}

src/wasm/function-body-decoder-impl.h

@@ -2819,8 +2819,11 @@ class WasmFullDecoder : public WasmDecoder<validate> {
   int DecodeLoadTransformMem(LoadType type, LoadTransformationKind transform,
                              uint32_t opcode_length) {
     if (!CheckHasMemory()) return 0;
+    // Load extends always load 64-bits.
+    uint32_t max_alignment =
+        transform == LoadTransformationKind::kExtend ? 3 : type.size_log_2();
     MemoryAccessImmediate<validate> imm(this, this->pc_ + opcode_length,
-                                        type.size_log_2());
+                                        max_alignment);
     auto index = Pop(0, kWasmI32);
     auto* result = Push(kWasmS128);
     CALL_INTERFACE_IF_REACHABLE(LoadTransform, type, transform, imm, index,

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

@@ -3430,20 +3430,25 @@ void RunLoadExtendTest(ExecutionTier execution_tier, LowerSimd lower_simd,
   constexpr int lanes_s = 16 / sizeof(S);
   constexpr int lanes_t = 16 / sizeof(T);
   constexpr int mem_index = 16;  // Load from mem index 16 (bytes).
-  WasmRunner<int32_t> r(execution_tier, lower_simd);
-  S* memory = r.builder().AddMemoryElems<S>(kWasmPageSize / sizeof(S));
-  T* global = r.builder().AddGlobal<T>(kWasmS128);
-  BUILD(r, WASM_SET_GLOBAL(0, WASM_SIMD_LOAD_EXTEND(op, WASM_I32V(mem_index))),
-        WASM_ONE);
+  // Load extends always load 64 bits, so alignment values can be from 0 to 3.
+  for (byte alignment = 0; alignment <= 3; alignment++) {
+    WasmRunner<int32_t> r(execution_tier, lower_simd);
+    S* memory = r.builder().AddMemoryElems<S>(kWasmPageSize / sizeof(S));
+    T* global = r.builder().AddGlobal<T>(kWasmS128);
+    BUILD(r,
+          WASM_SET_GLOBAL(0, WASM_SIMD_LOAD_EXTEND_ALIGNMENT(
+                                 op, WASM_I32V(mem_index), alignment)),
+          WASM_ONE);
 
-  for (S x : compiler::ValueHelper::GetVector<S>()) {
-    for (int i = 0; i < lanes_s; i++) {
-      // 16-th byte in memory is lanes-th element (size T) of memory.
-      r.builder().WriteMemory(&memory[lanes_s + i], x);
-    }
-    r.Call();
-    for (int i = 0; i < lanes_t; i++) {
-      CHECK_EQ(static_cast<T>(x), ReadLittleEndianValue<T>(&global[i]));
+    for (S x : compiler::ValueHelper::GetVector<S>()) {
+      for (int i = 0; i < lanes_s; i++) {
+        // 16-th byte in memory is lanes-th element (size T) of memory.
+        r.builder().WriteMemory(&memory[lanes_s + i], x);
+      }
+      r.Call();
+      for (int i = 0; i < lanes_t; i++) {
+        CHECK_EQ(static_cast<T>(x), ReadLittleEndianValue<T>(&global[i]));
+      }
     }
   }
 }

test/common/wasm/wasm-macro-gen.h

@@ -815,6 +815,8 @@ inline WasmOpcode LoadStoreOpcodeOf(MachineType type, bool store) {
   index, WASM_SIMD_OP(opcode), ZERO_ALIGNMENT, offset
 #define WASM_SIMD_LOAD_EXTEND(opcode, index) \
   index, WASM_SIMD_OP(opcode), ZERO_ALIGNMENT, ZERO_OFFSET
+#define WASM_SIMD_LOAD_EXTEND_ALIGNMENT(opcode, index, alignment) \
+  index, WASM_SIMD_OP(opcode), alignment, ZERO_OFFSET
 
 //------------------------------------------------------------------------------
 // Compilation Hints.