s390: [wasm-simd] Implement VisitSimd128ReverseBytes

LoadReverseSimd128 and StoreReverseSimd128 are implemented
to support the above instruction selection.

Change-Id: I5dcb30ce68b3478c69668b7589e77a52e77d9388
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1846460
Commit-Queue: Milad Farazmand <miladfar@ca.ibm.com>
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Reviewed-by: Junliang Yan <jyan@ca.ibm.com>
Reviewed-by: Joran Siu <joransiu@ca.ibm.com>
Cr-Commit-Position: refs/heads/master@{#64350}

src/codegen/cpu-features.h

@@ -62,6 +62,7 @@ enum CpuFeature {
   FLOATING_POINT_EXT,
   VECTOR_FACILITY,
   VECTOR_ENHANCE_FACILITY_1,
+  VECTOR_ENHANCE_FACILITY_2,
   MISC_INSTR_EXT2,
 #endif
 

src/codegen/s390/assembler-s390.cc

@@ -218,6 +218,11 @@ void CpuFeatures::ProbeImpl(bool cross_compile) {
         supportsCPUFeature("vx")) {
       supported_ |= (1u << VECTOR_ENHANCE_FACILITY_1);
     }
+    // Test for Vector Enhancement Facility 2 - Bit 148
+    if (facilities[2] & (one << (63 - (148 - 128))) &&
+        supportsCPUFeature("vx")) {
+      supported_ |= (1u << VECTOR_ENHANCE_FACILITY_2);
+    }
     // Test for Miscellaneous Instruction Extension Facility - Bit 58
     if (facilities[0] & (1lu << (63 - 58))) {
       supported_ |= (1u << MISC_INSTR_EXT2);

src/codegen/s390/assembler-s390.h

@@ -1146,6 +1146,19 @@ class V8_EXPORT_PRIVATE Assembler : public AssemblerBase {
   S390_VRR_E_OPCODE_LIST(DECLARE_VRR_E_INSTRUCTIONS)
 #undef DECLARE_VRR_E_INSTRUCTIONS
 
+#define DECLARE_VRR_F_INSTRUCTIONS(name, opcode_name, opcode_value)        \
+  void name(DoubleRegister v1, Register r1, Register r2) {                 \
+    uint64_t code = (static_cast<uint64_t>(opcode_value & 0xFF00)) * B32 | \
+                    (static_cast<uint64_t>(v1.code())) * B36 |             \
+                    (static_cast<uint64_t>(r1.code())) * B32 |             \
+                    (static_cast<uint64_t>(r2.code())) * B28 |             \
+                    (static_cast<uint64_t>(0)) * B8 |                      \
+                    (static_cast<uint64_t>(opcode_value & 0x00FF));        \
+    emit6bytes(code);                                                      \
+  }
+  S390_VRR_F_OPCODE_LIST(DECLARE_VRR_F_INSTRUCTIONS)
+#undef DECLARE_VRR_E_INSTRUCTIONS
+
 #define DECLARE_VRX_INSTRUCTIONS(name, opcode_name, opcode_value)       \
   void name(DoubleRegister v1, const MemOperand& opnd, Condition m3) {  \
     uint64_t code =                                                     \

src/codegen/s390/constants-s390.h

@@ -1549,21 +1549,24 @@ using SixByteInstr = uint64_t;
   V(risbhg, RISBHG,                                                            \
     0xEC5D) /* type = RIE_F ROTATE THEN INSERT SELECTED BITS HIGH (64)  */
 
-#define S390_VRX_OPCODE_LIST(V)                                             \
-  V(vleb, VLEB, 0xE700) /* type = VRX   VECTOR LOAD ELEMENT (8)  */         \
-  V(vleh, VLEH, 0xE701) /* type = VRX   VECTOR LOAD ELEMENT (16)  */        \
-  V(vleg, VLEG, 0xE702) /* type = VRX   VECTOR LOAD ELEMENT (64)  */        \
-  V(vlef, VLEF, 0xE703) /* type = VRX   VECTOR LOAD ELEMENT (32)  */        \
-  V(vllez, VLLEZ,                                                           \
-    0xE704) /* type = VRX   VECTOR LOAD LOGICAL ELEMENT AND ZERO  */        \
-  V(vlrep, VLREP, 0xE705) /* type = VRX   VECTOR LOAD AND REPLICATE  */     \
-  V(vl, VL, 0xE706)       /* type = VRX   VECTOR LOAD  */                   \
-  V(vlbb, VLBB, 0xE707)   /* type = VRX   VECTOR LOAD TO BLOCK BOUNDARY  */ \
-  V(vsteb, VSTEB, 0xE708) /* type = VRX   VECTOR STORE ELEMENT (8)  */      \
-  V(vsteh, VSTEH, 0xE709) /* type = VRX   VECTOR STORE ELEMENT (16)  */     \
-  V(vsteg, VSTEG, 0xE70A) /* type = VRX   VECTOR STORE ELEMENT (64)  */     \
-  V(vstef, VSTEF, 0xE70B) /* type = VRX   VECTOR STORE ELEMENT (32)  */     \
-  V(vst, VST, 0xE70E)     /* type = VRX   VECTOR STORE  */
+#define S390_VRX_OPCODE_LIST(V)                                               \
+  V(vleb, VLEB, 0xE700) /* type = VRX   VECTOR LOAD ELEMENT (8)  */           \
+  V(vleh, VLEH, 0xE701) /* type = VRX   VECTOR LOAD ELEMENT (16)  */          \
+  V(vleg, VLEG, 0xE702) /* type = VRX   VECTOR LOAD ELEMENT (64)  */          \
+  V(vlef, VLEF, 0xE703) /* type = VRX   VECTOR LOAD ELEMENT (32)  */          \
+  V(vllez, VLLEZ,                                                             \
+    0xE704) /* type = VRX   VECTOR LOAD LOGICAL ELEMENT AND ZERO  */          \
+  V(vlrep, VLREP, 0xE705) /* type = VRX   VECTOR LOAD AND REPLICATE  */       \
+  V(vl, VL, 0xE706)       /* type = VRX   VECTOR LOAD  */                     \
+  V(vlbb, VLBB, 0xE707)   /* type = VRX   VECTOR LOAD TO BLOCK BOUNDARY  */   \
+  V(vsteb, VSTEB, 0xE708) /* type = VRX   VECTOR STORE ELEMENT (8)  */        \
+  V(vsteh, VSTEH, 0xE709) /* type = VRX   VECTOR STORE ELEMENT (16)  */       \
+  V(vsteg, VSTEG, 0xE70A) /* type = VRX   VECTOR STORE ELEMENT (64)  */       \
+  V(vstef, VSTEF, 0xE70B) /* type = VRX   VECTOR STORE ELEMENT (32)  */       \
+  V(vst, VST, 0xE70E)     /* type = VRX   VECTOR STORE  */                    \
+  V(vlbr, VLBR, 0xE606) /* type = VRX   VECTOR LOAD BYTE REVERSED ELEMENTS */ \
+  V(vstbr, VSTBR, 0xE60E) /* type = VRX   VECTOR STORE BYTE REVERSED ELEMENTS \
+                           */
 
 #define S390_RIE_G_OPCODE_LIST(V)                                             \
   V(lochi, LOCHI,                                                             \

src/compiler/backend/s390/code-generator-s390.cc

@@ -2517,6 +2517,27 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     case kS390_LoadReverse64RR:
       __ lrvgr(i.OutputRegister(), i.InputRegister(0));
       break;
+    case kS390_LoadReverseSimd128RR:
+      __ vlgv(r0, i.InputSimd128Register(0), MemOperand(r0, 0), Condition(3));
+      __ vlgv(r1, i.InputSimd128Register(0), MemOperand(r0, 1), Condition(3));
+      __ lrvgr(r0, r0);
+      __ lrvgr(r1, r1);
+      __ vlvg(i.OutputSimd128Register(), r0, MemOperand(r0, 1), Condition(3));
+      __ vlvg(i.OutputSimd128Register(), r1, MemOperand(r0, 0), Condition(3));
+      break;
+    case kS390_LoadReverseSimd128: {
+      AddressingMode mode = kMode_None;
+      MemOperand operand = i.MemoryOperand(&mode);
+      if (CpuFeatures::IsSupported(VECTOR_ENHANCE_FACILITY_2)) {
+        __ vlbr(i.OutputSimd128Register(), operand, Condition(4));
+      } else {
+        __ lrvg(r0, operand);
+        __ lrvg(r1, MemOperand(operand.rx(), operand.rb(),
+                               operand.offset() + kBitsPerByte));
+        __ vlvgp(i.OutputSimd128Register(), r1, r0);
+      }
+      break;
+    }
     case kS390_LoadWord64:
       ASSEMBLE_LOAD_INTEGER(lg);
       EmitWordLoadPoisoningIfNeeded(this, instr, i);
@@ -2558,6 +2579,23 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     case kS390_StoreReverse64:
       ASSEMBLE_STORE_INTEGER(strvg);
       break;
+    case kS390_StoreReverseSimd128: {
+      size_t index = 0;
+      AddressingMode mode = kMode_None;
+      MemOperand operand = i.MemoryOperand(&mode, &index);
+      if (CpuFeatures::IsSupported(VECTOR_ENHANCE_FACILITY_2)) {
+        __ vstbr(i.InputSimd128Register(index), operand, Condition(4));
+      } else {
+        __ vlgv(r0, i.InputSimd128Register(index), MemOperand(r0, 1),
+                Condition(3));
+        __ vlgv(r1, i.InputSimd128Register(index), MemOperand(r0, 0),
+                Condition(3));
+        __ strvg(r0, operand);
+        __ strvg(r1, MemOperand(operand.rx(), operand.rb(),
+                                operand.offset() + kBitsPerByte));
+      }
+      break;
+    }
     case kS390_StoreFloat32:
       ASSEMBLE_STORE_FLOAT32();
       break;

src/compiler/backend/s390/instruction-codes-s390.h

@@ -148,6 +148,8 @@ namespace compiler {
   V(S390_LoadReverse16RR)                   \
   V(S390_LoadReverse32RR)                   \
   V(S390_LoadReverse64RR)                   \
+  V(S390_LoadReverseSimd128RR)              \
+  V(S390_LoadReverseSimd128)                \
   V(S390_LoadReverse16)                     \
   V(S390_LoadReverse32)                     \
   V(S390_LoadReverse64)                     \
@@ -161,6 +163,7 @@ namespace compiler {
   V(S390_StoreReverse16)                    \
   V(S390_StoreReverse32)                    \
   V(S390_StoreReverse64)                    \
+  V(S390_StoreReverseSimd128)               \
   V(S390_StoreFloat32)                      \
   V(S390_StoreDouble)                       \
   V(S390_DecompressSigned)                  \

src/compiler/backend/s390/instruction-scheduler-s390.cc

@@ -134,6 +134,8 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_LoadReverse16RR:
     case kS390_LoadReverse32RR:
     case kS390_LoadReverse64RR:
+    case kS390_LoadReverseSimd128RR:
+    case kS390_LoadReverseSimd128:
     case kS390_LoadAndTestWord32:
     case kS390_LoadAndTestWord64:
     case kS390_LoadAndTestFloat32:
@@ -165,6 +167,7 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_StoreWord16:
     case kS390_StoreWord32:
     case kS390_StoreWord64:
+    case kS390_StoreReverseSimd128:
     case kS390_StoreReverse16:
     case kS390_StoreReverse32:
     case kS390_StoreReverse64:

src/compiler/backend/s390/instruction-selector-s390.cc

@@ -1171,9 +1171,22 @@ void InstructionSelector::VisitWord32ReverseBytes(Node* node) {
 }
 
 void InstructionSelector::VisitSimd128ReverseBytes(Node* node) {
-  // TODO(miladfar): Implement the s390 selector for reversing SIMD bytes.
-  // Check if the input node is a Load and do a Load Reverse at once.
-  UNIMPLEMENTED();
+  S390OperandGenerator g(this);
+  NodeMatcher input(node->InputAt(0));
+  if (CanCover(node, input.node()) && input.IsLoad()) {
+    LoadRepresentation load_rep = LoadRepresentationOf(input.node()->op());
+    if (load_rep.representation() == MachineRepresentation::kSimd128) {
+      Node* base = input.node()->InputAt(0);
+      Node* offset = input.node()->InputAt(1);
+      Emit(kS390_LoadReverseSimd128 | AddressingModeField::encode(kMode_MRR),
+           // TODO(miladfar): one of the base and offset can be imm.
+           g.DefineAsRegister(node), g.UseRegister(base),
+           g.UseRegister(offset));
+      return;
+    }
+  }
+  Emit(kS390_LoadReverseSimd128RR, g.DefineAsRegister(node),
+       g.UseRegister(node->InputAt(0)));
 }
 
 template <class Matcher, ArchOpcode neg_opcode>

src/diagnostics/s390/disasm-s390.cc

@@ -827,6 +827,13 @@ bool Decoder::DecodeGeneric(Instruction* instr) {
     S390_VRR_E_OPCODE_LIST(DECODE_VRR_E_INSTRUCTIONS)
 #undef DECODE_VRR_E_INSTRUCTIONS
 
+#define DECODE_VRR_F_INSTRUCTIONS(name, opcode_name, opcode_value) \
+  case opcode_name:                                                \
+    Format(instr, #name "\t'f1,'r1,'r2");                          \
+    break;
+    S390_VRR_F_OPCODE_LIST(DECODE_VRR_F_INSTRUCTIONS)
+#undef DECODE_VRR_F_INSTRUCTIONS
+
 #define DECODE_VRX_INSTRUCTIONS(name, opcode_name, opcode_value) \
   case opcode_name:                                              \
     Format(instr, #name "\t'f1,'d1('r2d,'r3),'m4");              \