[simd] Implement conversion simd lowering

Implement lowering for simd operations I32x4ConvertI16x8 and
I16x8ConvertI8x16. Also, remove skip tests from status files that
were overriden when tests were renamed.

TEST=cctest/test-run-wasm-simd/RunWasm_I16x8ConvertI8x16_turbofan

Change-Id: If428f5039a32995c8ee64294c936419173a87aa7
Reviewed-on: https://chromium-review.googlesource.com/1069007Reviewed-by: Aseem Garg <aseemgarg@chromium.org>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/master@{#53343}

src/compiler/simd-scalar-lowering.cc

@@ -678,6 +678,39 @@ void SimdScalarLowering::LowerConvertFromFloat(Node* node, bool is_signed) {
   ReplaceNode(node, rep_node, kNumLanes32);
 }
 
+void SimdScalarLowering::LowerConvertFromInt(Node* node,
+                                             SimdType input_rep_type,
+                                             SimdType output_rep_type,
+                                             bool is_signed) {
+  DCHECK_EQ(1, node->InputCount());
+  Node** rep = GetReplacementsWithType(node->InputAt(0), input_rep_type);
+
+  int32_t shift_val = 0;
+  if (input_rep_type == SimdType::kInt16x8) {
+    DCHECK_EQ(output_rep_type, SimdType::kInt32x4);
+    shift_val = kShift16;
+  } else {
+    DCHECK_EQ(output_rep_type, SimdType::kInt16x8);
+    DCHECK_EQ(input_rep_type, SimdType::kInt8x16);
+    shift_val = kShift8;
+  }
+
+  int num_lanes = NumLanes(output_rep_type);
+  Node** rep_node = zone()->NewArray<Node*>(num_lanes);
+  for (int i = 0; i < num_lanes; ++i) {
+    rep_node[i] = rep[i];
+    if (!is_signed) {
+      rep_node[i] =
+          graph()->NewNode(machine()->Word32Shr(),
+                           graph()->NewNode(machine()->Word32Shl(), rep_node[i],
+                                            mcgraph_->Int32Constant(shift_val)),
+                           mcgraph_->Int32Constant(shift_val));
+    }
+  }
+
+  ReplaceNode(node, rep_node, num_lanes);
+}
+
 void SimdScalarLowering::LowerPack(Node* node, SimdType input_rep_type,
                                    SimdType output_rep_type, bool is_signed) {
   DCHECK_EQ(2, node->InputCount());
@@ -1068,6 +1101,26 @@ void SimdScalarLowering::LowerNode(Node* node) {
       LowerConvertFromFloat(node, false);
       break;
     }
+    case IrOpcode::kI32x4SConvertI16x8Low:
+    case IrOpcode::kI32x4SConvertI16x8High: {
+      LowerConvertFromInt(node, SimdType::kInt16x8, SimdType::kInt32x4, true);
+      break;
+    }
+    case IrOpcode::kI32x4UConvertI16x8Low:
+    case IrOpcode::kI32x4UConvertI16x8High: {
+      LowerConvertFromInt(node, SimdType::kInt16x8, SimdType::kInt32x4, false);
+      break;
+    }
+    case IrOpcode::kI16x8SConvertI8x16Low:
+    case IrOpcode::kI16x8SConvertI8x16High: {
+      LowerConvertFromInt(node, SimdType::kInt8x16, SimdType::kInt16x8, true);
+      break;
+    }
+    case IrOpcode::kI16x8UConvertI8x16Low:
+    case IrOpcode::kI16x8UConvertI8x16High: {
+      LowerConvertFromInt(node, SimdType::kInt8x16, SimdType::kInt16x8, false);
+      break;
+    }
     case IrOpcode::kI16x8SConvertI32x4: {
       LowerPack(node, SimdType::kInt32x4, SimdType::kInt16x8, true);
       break;

src/compiler/simd-scalar-lowering.h

@@ -91,6 +91,8 @@ class SimdScalarLowering {
   void LowerIntMinMax(Node* node, const Operator* op, bool is_max,
                       SimdType type);
   void LowerConvertFromFloat(Node* node, bool is_signed);
+  void LowerConvertFromInt(Node* node, SimdType input_rep_type,
+                           SimdType output_rep_type, bool is_signed);
   void LowerPack(Node* node, SimdType input_rep_type, SimdType output_rep_type,
                  bool is_signed);
   void LowerShiftOp(Node* node, SimdType type);

test/cctest/cctest.status

@@ -346,10 +346,6 @@
 ['(arch == mipsel or arch == mips64el or arch == mips or arch == mips64) and not simd_mips', {
   # Skip tests that fail on MIPS architectures which don't support SIMD,
   # because lowering mechanism doesn't work properly
-  'test-run-wasm-simd/RunWasm_I32x4ConvertI16x8_compiled': [SKIP],
-  'test-run-wasm-simd/RunWasm_I16x8ConvertI8x16_compiled': [SKIP],
-  'test-run-wasm-simd/RunWasm_I16x8ConvertI32x4_compiled': [SKIP],
-  'test-run-wasm-simd/RunWasm_I8x16ConvertI16x8_compiled': [SKIP],
   'test-run-wasm-simd/RunWasm_ReductionTest4_compiled': [SKIP],
   'test-run-wasm-simd/RunWasm_ReductionTest8_compiled': [SKIP],
   'test-run-wasm-simd/RunWasm_ReductionTest16_compiled': [SKIP],