[ia32][wasm] Add I32x4AddHoriz and I16x8AddHoriz

Add phaddd, phaddw and AVX version
Add vmovdqu and Movdqu macro

Bug: 
Change-Id: I4f5c0cf96ab481fc18f0a0d554101a996a16c954
Reviewed-on: https://chromium-review.googlesource.com/715677
Commit-Queue: Jing Bao <jing.bao@intel.com>
Reviewed-by: Bill Budge <bbudge@chromium.org>
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#48621}

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

@@ -1863,6 +1863,15 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
         __ movss(operand, i.InputDoubleRegister(index));
       }
       break;
+    case kIA32Movdqu:
+      if (instr->HasOutput()) {
+        __ Movdqu(i.OutputSimd128Register(), i.MemoryOperand());
+      } else {
+        size_t index = 0;
+        Operand operand = i.MemoryOperand(&index);
+        __ Movdqu(operand, i.InputSimd128Register(index));
+      }
+      break;
     case kIA32BitcastFI:
       if (instr->InputAt(0)->IsFPStackSlot()) {
         __ mov(i.OutputRegister(), i.InputOperand(0));
@@ -2040,6 +2049,17 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
                 i.InputOperand(1));
       break;
     }
+    case kSSEI32x4AddHoriz: {
+      CpuFeatureScope sse_scope(tasm(), SSSE3);
+      __ phaddd(i.OutputSimd128Register(), i.InputOperand(1));
+      break;
+    }
+    case kAVXI32x4AddHoriz: {
+      CpuFeatureScope avx_scope(tasm(), AVX);
+      __ vphaddd(i.OutputSimd128Register(), i.InputSimd128Register(0),
+                 i.InputOperand(1));
+      break;
+    }
     case kSSEI32x4Sub: {
       __ psubd(i.OutputSimd128Register(), i.InputOperand(1));
       break;
@@ -2226,6 +2246,17 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
                  i.InputOperand(2), i.InputInt8(1));
       break;
     }
+    case kSSEI16x8AddHoriz: {
+      CpuFeatureScope sse_scope(tasm(), SSSE3);
+      __ phaddw(i.OutputSimd128Register(), i.InputOperand(1));
+      break;
+    }
+    case kAVXI16x8AddHoriz: {
+      CpuFeatureScope avx_scope(tasm(), AVX);
+      __ vphaddw(i.OutputSimd128Register(), i.InputSimd128Register(0),
+                 i.InputOperand(1));
+      break;
+    }
     case kIA32I8x16Splat: {
       XMMRegister dst = i.OutputSimd128Register();
       __ Movd(dst, i.InputOperand(0));

src/compiler/ia32/instruction-codes-ia32.h

@@ -103,6 +103,7 @@ namespace compiler {
   V(IA32Movl)                      \
   V(IA32Movss)                     \
   V(IA32Movsd)                     \
+  V(IA32Movdqu)                    \
   V(IA32BitcastFI)                 \
   V(IA32BitcastIF)                 \
   V(IA32Lea)                       \
@@ -122,6 +123,8 @@ namespace compiler {
   V(AVXI32x4ShrS)                  \
   V(SSEI32x4Add)                   \
   V(AVXI32x4Add)                   \
+  V(SSEI32x4AddHoriz)              \
+  V(AVXI32x4AddHoriz)              \
   V(SSEI32x4Sub)                   \
   V(AVXI32x4Sub)                   \
   V(SSEI32x4Mul)                   \
@@ -152,6 +155,8 @@ namespace compiler {
   V(IA32I16x8ExtractLane)          \
   V(SSEI16x8ReplaceLane)           \
   V(AVXI16x8ReplaceLane)           \
+  V(SSEI16x8AddHoriz)              \
+  V(AVXI16x8AddHoriz)              \
   V(IA32I8x16Splat)                \
   V(IA32I8x16ExtractLane)          \
   V(SSEI8x16ReplaceLane)           \

src/compiler/ia32/instruction-scheduler-ia32.cc

@@ -108,6 +108,8 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kAVXI32x4ShrS:
     case kSSEI32x4Add:
     case kAVXI32x4Add:
+    case kSSEI32x4AddHoriz:
+    case kAVXI32x4AddHoriz:
     case kSSEI32x4Sub:
     case kAVXI32x4Sub:
     case kSSEI32x4Mul:
@@ -138,6 +140,8 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kIA32I16x8ExtractLane:
     case kSSEI16x8ReplaceLane:
     case kAVXI16x8ReplaceLane:
+    case kSSEI16x8AddHoriz:
+    case kAVXI16x8AddHoriz:
     case kIA32I8x16Splat:
     case kIA32I8x16ExtractLane:
     case kSSEI8x16ReplaceLane:
@@ -161,6 +165,7 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kIA32Movl:
     case kIA32Movss:
     case kIA32Movsd:
+    case kIA32Movdqu:
       // Moves are used for memory load/store operations.
       return instr->HasOutput() ? kIsLoadOperation : kHasSideEffect;
 

src/compiler/ia32/instruction-selector-ia32.cc

@@ -250,8 +250,10 @@ void InstructionSelector::VisitLoad(Node* node) {
     case MachineRepresentation::kWord32:
       opcode = kIA32Movl;
       break;
+    case MachineRepresentation::kSimd128:
+      opcode = kIA32Movdqu;
+      break;
     case MachineRepresentation::kWord64:   // Fall through.
-    case MachineRepresentation::kSimd128:  // Fall through.
     case MachineRepresentation::kNone:
       UNREACHABLE();
       return;
@@ -340,8 +342,10 @@ void InstructionSelector::VisitStore(Node* node) {
       case MachineRepresentation::kWord32:
         opcode = kIA32Movl;
         break;
+      case MachineRepresentation::kSimd128:
+        opcode = kIA32Movdqu;
+        break;
       case MachineRepresentation::kWord64:   // Fall through.
-      case MachineRepresentation::kSimd128:  // Fall through.
       case MachineRepresentation::kNone:
         UNREACHABLE();
         return;
@@ -1905,6 +1909,7 @@ VISIT_ATOMIC_BINOP(Xor)
 
 #define SIMD_BINOP_LIST(V) \
   V(I32x4Add)              \
+  V(I32x4AddHoriz)         \
   V(I32x4Sub)              \
   V(I32x4Mul)              \
   V(I32x4MinS)             \
@@ -1916,7 +1921,8 @@ VISIT_ATOMIC_BINOP(Xor)
   V(I32x4MinU)             \
   V(I32x4MaxU)             \
   V(I32x4GtU)              \
-  V(I32x4GeU)
+  V(I32x4GeU)              \
+  V(I16x8AddHoriz)
 
 #define SIMD_UNOP_LIST(V) V(I32x4Neg)
 

src/compiler/instruction-selector.cc

@@ -2150,15 +2150,11 @@ void InstructionSelector::VisitI32x4MinU(Node* node) { UNIMPLEMENTED(); }
 void InstructionSelector::VisitI32x4MaxU(Node* node) { UNIMPLEMENTED(); }
 
 void InstructionSelector::VisitI32x4ShrU(Node* node) { UNIMPLEMENTED(); }
-#endif  // !V8_TARGET_ARCH_ARM && !V8_TARGET_ARCH_ARM64 && !V8_TARGET_ARCH_X64
-        // && !V8_TARGET_ARCH_IA32 && !V8_TARGET_ARCH_MIPS
-        // && !V8_TARGET_ARCH_MIPS64
 
-#if !V8_TARGET_ARCH_ARM && !V8_TARGET_ARCH_ARM64 && !V8_TARGET_ARCH_X64 && \
-    !V8_TARGET_ARCH_MIPS && !V8_TARGET_ARCH_MIPS64
 void InstructionSelector::VisitI32x4AddHoriz(Node* node) { UNIMPLEMENTED(); }
 #endif  // !V8_TARGET_ARCH_ARM && !V8_TARGET_ARCH_ARM64 && !V8_TARGET_ARCH_X64
-        // && !V8_TARGET_ARCH_MIPS && !V8_TARGET_ARCH_MIPS64
+        // && !V8_TARGET_ARCH_IA32 && !V8_TARGET_ARCH_MIPS
+        // && !V8_TARGET_ARCH_MIPS64
 
 #if !V8_TARGET_ARCH_ARM && !V8_TARGET_ARCH_ARM64 && !V8_TARGET_ARCH_MIPS && \
     !V8_TARGET_ARCH_MIPS64
@@ -2248,10 +2244,11 @@ void InstructionSelector::VisitI16x8SubSaturateS(Node* node) {
         // && !V8_TARGET_ARCH_MIPS && !V8_TARGET_ARCH_MIPS64
 
 #if !V8_TARGET_ARCH_ARM && !V8_TARGET_ARCH_ARM64 && !V8_TARGET_ARCH_X64 && \
-    !V8_TARGET_ARCH_MIPS && !V8_TARGET_ARCH_MIPS64
+    !V8_TARGET_ARCH_IA32 && !V8_TARGET_ARCH_MIPS && !V8_TARGET_ARCH_MIPS64
 void InstructionSelector::VisitI16x8AddHoriz(Node* node) { UNIMPLEMENTED(); }
 #endif  // !V8_TARGET_ARCH_ARM && !V8_TARGET_ARCH_ARM64 && !V8_TARGET_ARCH_X64
-        // && !V8_TARGET_ARCH_MIPS && !V8_TARGET_ARCH_MIPS64
+        // && !V8_TARGET_ARCH_IA32 && !V8_TARGET_ARCH_MIPS
+        // && !V8_TARGET_ARCH_MIPS64
 
 #if !V8_TARGET_ARCH_ARM && !V8_TARGET_ARCH_ARM64 && !V8_TARGET_ARCH_X64 && \
     !V8_TARGET_ARCH_MIPS && !V8_TARGET_ARCH_MIPS64

src/ia32/assembler-ia32.h

@@ -1453,6 +1453,12 @@ class Assembler : public AssemblerBase {
     vinstr(0x5B, dst, xmm0, src, kF3, k0F, kWIG);
   }
 
+  void vmovdqu(XMMRegister dst, const Operand& src) {
+    vinstr(0x6F, dst, xmm0, src, kF3, k0F, kWIG);
+  }
+  void vmovdqu(const Operand& dst, XMMRegister src) {
+    vinstr(0x7F, src, xmm0, dst, kF3, k0F, kWIG);
+  }
   void vmovd(XMMRegister dst, Register src) { vmovd(dst, Operand(src)); }
   void vmovd(XMMRegister dst, const Operand& src) {
     vinstr(0x6E, dst, xmm0, src, k66, k0F, kWIG);

src/ia32/disasm-ia32.cc

@@ -941,6 +941,15 @@ int DisassemblerIA32::AVXInstruction(byte* data) {
                        NameOfXMMRegister(vvvv));
         current += PrintRightXMMOperand(current);
         break;
+      case 0x6f:
+        AppendToBuffer("vmovdqu %s,", NameOfXMMRegister(regop));
+        current += PrintRightOperand(current);
+        break;
+      case 0x7f:
+        AppendToBuffer("vmovdqu ");
+        current += PrintRightOperand(current);
+        AppendToBuffer(",%s", NameOfXMMRegister(regop));
+        break;
       default:
         UnimplementedInstruction();
     }

src/ia32/macro-assembler-ia32.h

@@ -214,6 +214,8 @@ class TurboAssembler : public Assembler {
     }                                                           \
   }
 
+  AVX_OP2_WITH_TYPE(Movdqu, movdqu, XMMRegister, const Operand&)
+  AVX_OP2_WITH_TYPE(Movdqu, movdqu, const Operand&, XMMRegister)
   AVX_OP2_WITH_TYPE(Movd, movd, XMMRegister, Register)
   AVX_OP2_WITH_TYPE(Movd, movd, XMMRegister, const Operand&)
   AVX_OP2_WITH_TYPE(Movd, movd, Register, XMMRegister)

src/ia32/sse-instr.h

@@ -44,6 +44,8 @@
   V(pxor, 66, 0F, EF)
 
 #define SSSE3_INSTRUCTION_LIST(V) \
+  V(phaddd, 66, 0F, 38, 02)       \
+  V(phaddw, 66, 0F, 38, 01)       \
   V(pshufb, 66, 0F, 38, 00)       \
   V(psignb, 66, 0F, 38, 08)       \
   V(psignw, 66, 0F, 38, 09)       \

test/cctest/test-disasm-ia32.cc

@@ -667,6 +667,8 @@ TEST(DisasmIa320) {
       __ vcvttps2dq(xmm1, xmm0);
       __ vcvttps2dq(xmm1, Operand(ebx, ecx, times_4, 10000));
 
+      __ vmovdqu(xmm0, Operand(ebx, ecx, times_4, 10000));
+      __ vmovdqu(Operand(ebx, ecx, times_4, 10000), xmm0);
       __ vmovd(xmm0, edi);
       __ vmovd(xmm0, Operand(ebx, ecx, times_4, 10000));
       __ vmovd(eax, xmm1);

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

@@ -1610,6 +1610,8 @@ WASM_SIMD_SELECT_TEST(8x16)
 WASM_SIMD_NON_CANONICAL_SELECT_TEST(32x4)
 WASM_SIMD_NON_CANONICAL_SELECT_TEST(16x8)
 WASM_SIMD_NON_CANONICAL_SELECT_TEST(8x16)
+#endif  // V8_TARGET_ARCH_ARM || V8_TARGET_ARCH_ARM64 || V8_TARGET_ARCH_X64 ||
+        // V8_TARGET_ARCH_MIPS || V8_TARGET_ARCH_MIPS64
 
 // Test binary ops with two lane test patterns, all lanes distinct.
 template <typename T>
@@ -1653,8 +1655,6 @@ WASM_SIMD_COMPILED_TEST(I16x8AddHoriz) {
   RunBinaryLaneOpTest<int16_t>(execution_mode, kExprI16x8AddHoriz,
                                {{1, 5, 9, 13, 17, 21, 25, 29}});
 }
-#endif  // V8_TARGET_ARCH_ARM || V8_TARGET_ARCH_ARM64 || V8_TARGET_ARCH_X64 ||
-        // V8_TARGET_ARCH_MIPS || V8_TARGET_ARCH_MIPS64
 
 #if V8_TARGET_ARCH_ARM || V8_TARGET_ARCH_ARM64 || V8_TARGET_ARCH_MIPS || \
     V8_TARGET_ARCH_MIPS64