[wasm-simd][ia32][x64] Fix swizzle with constant masks

We optimized swizzle with constant mask, but failed to actually swizzle
using the masks...

Bug: v8:10992
Change-Id: If655fdad1e17e92b62e8a2eaabbf1f8d82e4d5e4
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2822951Reviewed-by: Deepti Gandluri <gdeepti@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#73943}

src/codegen/ia32/macro-assembler-ia32.cc

@@ -913,7 +913,7 @@ void TurboAssembler::I8x16Swizzle(XMMRegister dst, XMMRegister src,
                                   XMMRegister mask, XMMRegister scratch,
                                   Register tmp, bool omit_add) {
   if (omit_add) {
-    Pshufb(dst, src, scratch);
+    Pshufb(dst, src, mask);
     return;
   }
 

src/codegen/x64/macro-assembler-x64.cc

@@ -2355,7 +2355,7 @@ void TurboAssembler::I8x16Swizzle(XMMRegister dst, XMMRegister src,
   if (omit_add) {
     // We have determined that the indices are immediates, and they are either
     // within bounds, or the top bit is set, so we can omit the add.
-    Pshufb(dst, src, kScratchDoubleReg);
+    Pshufb(dst, src, mask);
     return;
   }
 

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

@@ -2226,6 +2226,30 @@ WASM_SIMD_TEST(I8x16Swizzle) {
       CHECK_EQ(ReadLittleEndianValue<uint8_t>(&dst[i]), si.expected[i]);
     }
   }
+
+  {
+    // We have an optimization for constant indices, test this case.
+    for (SwizzleTestArgs si : swizzle_test_vector) {
+      WasmRunner<int32_t> r(execution_tier, lower_simd);
+      uint8_t* dst = r.builder().AddGlobal<uint8_t>(kWasmS128);
+      uint8_t* src0 = r.builder().AddGlobal<uint8_t>(kWasmS128);
+      BUILD(r,
+            WASM_GLOBAL_SET(
+                0, WASM_SIMD_BINOP(kExprI8x16Swizzle, WASM_GLOBAL_GET(1),
+                                   WASM_SIMD_CONSTANT(si.indices))),
+            WASM_ONE);
+
+      for (int i = 0; i < kSimd128Size; i++) {
+        WriteLittleEndianValue<uint8_t>(&src0[i], si.input[i]);
+      }
+
+      CHECK_EQ(1, r.Call());
+
+      for (int i = 0; i < kSimd128Size; i++) {
+        CHECK_EQ(ReadLittleEndianValue<uint8_t>(&dst[i]), si.expected[i]);
+      }
+    }
+  }
 }
 
 // Combine 3 shuffles a, b, and c by applying both a and b and then applying c