[wasm-simd] Optimize S16x8Dup and S8x16Dup

Use punpcklqdq/punpckhqdq instead of pshufd.

Bug: v8:12075
Change-Id: I0260136d3727fde7bae4359df015d2b0090e8b29
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3092554Reviewed-by: Deepti Gandluri <gdeepti@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#76272}

src/compiler/backend/ia32/code-generator-ia32.cc

@@ -3539,10 +3539,10 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       uint8_t half_dup = lane4 | (lane4 << 2) | (lane4 << 4) | (lane4 << 6);
       if (lane < 4) {
         __ Pshuflw(dst, src, half_dup);
-        __ Pshufd(dst, dst, uint8_t{0});
+        __ Punpcklqdq(dst, dst);
       } else {
         __ Pshufhw(dst, src, half_dup);
-        __ Pshufd(dst, dst, uint8_t{0xaa});
+        __ Punpckhqdq(dst, dst);
       }
       break;
     }
@@ -3570,10 +3570,10 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       uint8_t half_dup = lane4 | (lane4 << 2) | (lane4 << 4) | (lane4 << 6);
       if (lane < 4) {
         __ Pshuflw(dst, dst, half_dup);
-        __ Pshufd(dst, dst, uint8_t{0});
+        __ Punpcklqdq(dst, dst);
       } else {
         __ Pshufhw(dst, dst, half_dup);
-        __ Pshufd(dst, dst, uint8_t{0xaa});
+        __ Punpckhqdq(dst, dst);
       }
       break;
     }

src/compiler/backend/x64/code-generator-x64.cc

@@ -3903,10 +3903,10 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       uint8_t half_dup = lane4 | (lane4 << 2) | (lane4 << 4) | (lane4 << 6);
       if (lane < 4) {
         ASSEMBLE_SIMD_IMM_INSTR(Pshuflw, dst, 0, half_dup);
-        __ Pshufd(dst, dst, uint8_t{0});
+        __ Punpcklqdq(dst, dst);
       } else {
         ASSEMBLE_SIMD_IMM_INSTR(Pshufhw, dst, 0, half_dup);
-        __ Pshufd(dst, dst, uint8_t{0xaa});
+        __ Punpckhqdq(dst, dst);
       }
       break;
     }
@@ -3924,10 +3924,10 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       uint8_t half_dup = lane4 | (lane4 << 2) | (lane4 << 4) | (lane4 << 6);
       if (lane < 4) {
         __ Pshuflw(dst, dst, half_dup);
-        __ Pshufd(dst, dst, uint8_t{0});
+        __ Punpcklqdq(dst, dst);
       } else {
         __ Pshufhw(dst, dst, half_dup);
-        __ Pshufd(dst, dst, uint8_t{0xaa});
+        __ Punpckhqdq(dst, dst);
       }
       break;
     }