[wasm-simd][scalar-lowering] Implement i32x4_dot_i16x8_s

Bug: v8:10993
Change-Id: I9b3cd1499cc9ebb93690e4940e9d94c5f445e315
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2477432Reviewed-by: Bill Budge <bbudge@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70587}

src/compiler/simd-scalar-lowering.cc

@@ -141,6 +141,7 @@ void SimdScalarLowering::LowerGraph() {
   V(I32x4ShrU)                    \
   V(I32x4MinU)                    \
   V(I32x4MaxU)                    \
+  V(I32x4DotI16x8S)               \
   V(I32x4Eq)                      \
   V(I32x4Ne)                      \
   V(I32x4LtS)                     \
@@ -1620,6 +1621,25 @@ void SimdScalarLowering::LowerNode(Node* node) {
       LowerIntMinMax(node, machine()->Uint32LessThan(), false, rep_type);
       break;
     }
+    case IrOpcode::kI32x4DotI16x8S: {
+      // i32x4.dot_i16x8_s wants the inputs to be i16x8, but outputs to i32x4.
+      DCHECK_EQ(2, node->InputCount());
+      Node** rep_left =
+          GetReplacementsWithType(node->InputAt(0), SimdType::kInt16x8);
+      Node** rep_right =
+          GetReplacementsWithType(node->InputAt(1), SimdType::kInt16x8);
+      int num_lanes = NumLanes(rep_type);
+      Node** rep_node = zone()->NewArray<Node*>(num_lanes);
+      for (int i = 0; i < num_lanes; ++i) {
+        Node* lo = graph()->NewNode(machine()->Int32Mul(), rep_left[i * 2],
+                                    rep_right[i * 2]);
+        Node* hi = graph()->NewNode(machine()->Int32Mul(), rep_left[i * 2 + 1],
+                                    rep_right[i * 2 + 1]);
+        rep_node[i] = graph()->NewNode(machine()->Int32Add(), lo, hi);
+      }
+      ReplaceNode(node, rep_node, num_lanes);
+      break;
+    }
     case IrOpcode::kI64x2Neg: {
       DCHECK_EQ(1, node->InputCount());
       Node** rep = GetReplacementsWithType(node->InputAt(0), rep_type);

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

@@ -2229,7 +2229,7 @@ WASM_SIMD_TEST_NO_LOWERING(I16x8Q15MulRSatS) {
 }
 #endif  // V8_TARGET_ARCH_ARM64
 
-WASM_SIMD_TEST_NO_LOWERING(I32x4DotI16x8S) {
+WASM_SIMD_TEST(I32x4DotI16x8S) {
   WasmRunner<int32_t, int16_t, int16_t> r(execution_tier, lower_simd);
   int32_t* g = r.builder().template AddGlobal<int32_t>(kWasmS128);
   byte value1 = 0, value2 = 1;