S390: [wasm-simd] Implement extended multiply

PPC will be implemented separately.

Change-Id: I358f5a73275a40f8083e9c07b7028d162969f836
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2597578Reviewed-by: Junliang Yan <junyan@redhat.com>
Commit-Queue: Milad Fa <mfarazma@redhat.com>
Cr-Commit-Position: refs/heads/master@{#71834}

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

@@ -2458,6 +2458,43 @@ void InstructionSelector::VisitS128Const(Node* node) {
   }
 }
 
+void InstructionSelector::VisitI64x2ExtMulLowI32x4S(Node* node) {
+  UNIMPLEMENTED();
+}
+void InstructionSelector::VisitI64x2ExtMulHighI32x4S(Node* node) {
+  UNIMPLEMENTED();
+}
+void InstructionSelector::VisitI64x2ExtMulLowI32x4U(Node* node) {
+  UNIMPLEMENTED();
+}
+void InstructionSelector::VisitI64x2ExtMulHighI32x4U(Node* node) {
+  UNIMPLEMENTED();
+}
+void InstructionSelector::VisitI32x4ExtMulLowI16x8S(Node* node) {
+  UNIMPLEMENTED();
+}
+void InstructionSelector::VisitI32x4ExtMulHighI16x8S(Node* node) {
+  UNIMPLEMENTED();
+}
+void InstructionSelector::VisitI32x4ExtMulLowI16x8U(Node* node) {
+  UNIMPLEMENTED();
+}
+void InstructionSelector::VisitI32x4ExtMulHighI16x8U(Node* node) {
+  UNIMPLEMENTED();
+}
+void InstructionSelector::VisitI16x8ExtMulLowI8x16S(Node* node) {
+  UNIMPLEMENTED();
+}
+void InstructionSelector::VisitI16x8ExtMulHighI8x16S(Node* node) {
+  UNIMPLEMENTED();
+}
+void InstructionSelector::VisitI16x8ExtMulLowI8x16U(Node* node) {
+  UNIMPLEMENTED();
+}
+void InstructionSelector::VisitI16x8ExtMulHighI8x16U(Node* node) {
+  UNIMPLEMENTED();
+}
+
 void InstructionSelector::EmitPrepareResults(
     ZoneVector<PushParameter>* results, const CallDescriptor* call_descriptor,
     Node* node) {

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

@@ -4113,6 +4113,80 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
             Condition(0), Condition(0), Condition(2));
       break;
     }
+#define ASSEMBLE_SIMD_I64X2_EXT_MUL(UNPACK_INSTR)                              \
+  __ UNPACK_INSTR(kScratchDoubleReg, i.InputSimd128Register(0), Condition(0),  \
+                  Condition(0), Condition(2));                                 \
+  __ UNPACK_INSTR(i.OutputSimd128Register(), i.InputSimd128Register(1),        \
+                  Condition(0), Condition(0), Condition(2));                   \
+  Register scratch_0 = r0;                                                     \
+  Register scratch_1 = r1;                                                     \
+  for (int lane = 0; lane < 2; lane++) {                                       \
+    __ vlgv(scratch_0, kScratchDoubleReg, MemOperand(r0, lane), Condition(3)); \
+    __ vlgv(scratch_1, i.OutputSimd128Register(), MemOperand(r0, lane),        \
+            Condition(3));                                                     \
+    __ Mul64(scratch_0, scratch_1);                                            \
+    scratch_0 = r1;                                                            \
+    scratch_1 = ip;                                                            \
+  }                                                                            \
+  __ vlvgp(i.OutputSimd128Register(), r0, r1);
+    case kS390_I64x2ExtMulLowI32x4S: {
+      ASSEMBLE_SIMD_I64X2_EXT_MUL(vupl)
+      break;
+    }
+    case kS390_I64x2ExtMulHighI32x4S: {
+      ASSEMBLE_SIMD_I64X2_EXT_MUL(vuph)
+      break;
+    }
+    case kS390_I64x2ExtMulLowI32x4U: {
+      ASSEMBLE_SIMD_I64X2_EXT_MUL(vupll)
+      break;
+    }
+    case kS390_I64x2ExtMulHighI32x4U: {
+      ASSEMBLE_SIMD_I64X2_EXT_MUL(vuplh)
+      break;
+    }
+#undef ASSEMBLE_SIMD_I64X2_EXT_MUL
+#define ASSEMBLE_SIMD_I32X4_I16X8_EXT_MUL(UNPACK_INSTR, MODE)                 \
+  __ UNPACK_INSTR(kScratchDoubleReg, i.InputSimd128Register(0), Condition(0), \
+                  Condition(0), Condition(MODE));                             \
+  __ UNPACK_INSTR(i.OutputSimd128Register(), i.InputSimd128Register(1),       \
+                  Condition(0), Condition(0), Condition(MODE));               \
+  __ vml(i.OutputSimd128Register(), kScratchDoubleReg,                        \
+         i.OutputSimd128Register(), Condition(0), Condition(0),               \
+         Condition(MODE + 1));
+    case kS390_I32x4ExtMulLowI16x8S: {
+      ASSEMBLE_SIMD_I32X4_I16X8_EXT_MUL(vupl, 1)
+      break;
+    }
+    case kS390_I32x4ExtMulHighI16x8S: {
+      ASSEMBLE_SIMD_I32X4_I16X8_EXT_MUL(vuph, 1)
+      break;
+    }
+    case kS390_I32x4ExtMulLowI16x8U: {
+      ASSEMBLE_SIMD_I32X4_I16X8_EXT_MUL(vupll, 1)
+      break;
+    }
+    case kS390_I32x4ExtMulHighI16x8U: {
+      ASSEMBLE_SIMD_I32X4_I16X8_EXT_MUL(vuplh, 1)
+      break;
+    }
+    case kS390_I16x8ExtMulLowI8x16S: {
+      ASSEMBLE_SIMD_I32X4_I16X8_EXT_MUL(vupl, 0)
+      break;
+    }
+    case kS390_I16x8ExtMulHighI8x16S: {
+      ASSEMBLE_SIMD_I32X4_I16X8_EXT_MUL(vuph, 0)
+      break;
+    }
+    case kS390_I16x8ExtMulLowI8x16U: {
+      ASSEMBLE_SIMD_I32X4_I16X8_EXT_MUL(vupll, 0)
+      break;
+    }
+    case kS390_I16x8ExtMulHighI8x16U: {
+      ASSEMBLE_SIMD_I32X4_I16X8_EXT_MUL(vuplh, 0)
+      break;
+    }
+#undef ASSEMBLE_SIMD_I32X4_I16X8_EXT_MUL
     case kS390_StoreCompressTagged: {
       CHECK(!instr->HasOutput());
       size_t index = 0;

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

@@ -256,6 +256,10 @@ namespace compiler {
   V(S390_I64x2ExtractLane)                  \
   V(S390_I64x2Eq)                           \
   V(S390_I64x2BitMask)                      \
+  V(S390_I64x2ExtMulLowI32x4S)              \
+  V(S390_I64x2ExtMulHighI32x4S)             \
+  V(S390_I64x2ExtMulLowI32x4U)              \
+  V(S390_I64x2ExtMulHighI32x4U)             \
   V(S390_I32x4Splat)                        \
   V(S390_I32x4ExtractLane)                  \
   V(S390_I32x4ReplaceLane)                  \
@@ -286,6 +290,10 @@ namespace compiler {
   V(S390_I32x4Abs)                          \
   V(S390_I32x4BitMask)                      \
   V(S390_I32x4DotI16x8S)                    \
+  V(S390_I32x4ExtMulLowI16x8S)              \
+  V(S390_I32x4ExtMulHighI16x8S)             \
+  V(S390_I32x4ExtMulLowI16x8U)              \
+  V(S390_I32x4ExtMulHighI16x8U)             \
   V(S390_I16x8Splat)                        \
   V(S390_I16x8ExtractLaneU)                 \
   V(S390_I16x8ExtractLaneS)                 \
@@ -321,6 +329,10 @@ namespace compiler {
   V(S390_I16x8RoundingAverageU)             \
   V(S390_I16x8Abs)                          \
   V(S390_I16x8BitMask)                      \
+  V(S390_I16x8ExtMulLowI8x16S)              \
+  V(S390_I16x8ExtMulHighI8x16S)             \
+  V(S390_I16x8ExtMulLowI8x16U)              \
+  V(S390_I16x8ExtMulHighI8x16U)             \
   V(S390_I8x16Splat)                        \
   V(S390_I8x16ExtractLaneU)                 \
   V(S390_I8x16ExtractLaneS)                 \

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

@@ -202,6 +202,10 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_I64x2ExtractLane:
     case kS390_I64x2Eq:
     case kS390_I64x2BitMask:
+    case kS390_I64x2ExtMulLowI32x4S:
+    case kS390_I64x2ExtMulHighI32x4S:
+    case kS390_I64x2ExtMulLowI32x4U:
+    case kS390_I64x2ExtMulHighI32x4U:
     case kS390_I32x4Splat:
     case kS390_I32x4ExtractLane:
     case kS390_I32x4ReplaceLane:
@@ -232,6 +236,10 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_I32x4Abs:
     case kS390_I32x4BitMask:
     case kS390_I32x4DotI16x8S:
+    case kS390_I32x4ExtMulLowI16x8S:
+    case kS390_I32x4ExtMulHighI16x8S:
+    case kS390_I32x4ExtMulLowI16x8U:
+    case kS390_I32x4ExtMulHighI16x8U:
     case kS390_I16x8Splat:
     case kS390_I16x8ExtractLaneU:
     case kS390_I16x8ExtractLaneS:
@@ -267,6 +275,10 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_I16x8RoundingAverageU:
     case kS390_I16x8Abs:
     case kS390_I16x8BitMask:
+    case kS390_I16x8ExtMulLowI8x16S:
+    case kS390_I16x8ExtMulHighI8x16S:
+    case kS390_I16x8ExtMulLowI8x16U:
+    case kS390_I16x8ExtMulHighI8x16U:
     case kS390_I8x16Splat:
     case kS390_I8x16ExtractLaneU:
     case kS390_I8x16ExtractLaneS:

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

@@ -2446,6 +2446,10 @@ void InstructionSelector::VisitWord64AtomicStore(Node* node) {
   V(I64x2Sub)              \
   V(I64x2Mul)              \
   V(I64x2Eq)               \
+  V(I64x2ExtMulLowI32x4S)  \
+  V(I64x2ExtMulHighI32x4S) \
+  V(I64x2ExtMulLowI32x4U)  \
+  V(I64x2ExtMulHighI32x4U) \
   V(I32x4Add)              \
   V(I32x4AddHoriz)         \
   V(I32x4Sub)              \
@@ -2461,6 +2465,10 @@ void InstructionSelector::VisitWord64AtomicStore(Node* node) {
   V(I32x4GtU)              \
   V(I32x4GeU)              \
   V(I32x4DotI16x8S)        \
+  V(I32x4ExtMulLowI16x8S)  \
+  V(I32x4ExtMulHighI16x8S) \
+  V(I32x4ExtMulLowI16x8U)  \
+  V(I32x4ExtMulHighI16x8U) \
   V(I16x8Add)              \
   V(I16x8AddHoriz)         \
   V(I16x8Sub)              \
@@ -2482,6 +2490,10 @@ void InstructionSelector::VisitWord64AtomicStore(Node* node) {
   V(I16x8AddSatU)          \
   V(I16x8SubSatU)          \
   V(I16x8RoundingAverageU) \
+  V(I16x8ExtMulLowI8x16S)  \
+  V(I16x8ExtMulHighI8x16S) \
+  V(I16x8ExtMulLowI8x16U)  \
+  V(I16x8ExtMulHighI8x16U) \
   V(I8x16Add)              \
   V(I8x16Sub)              \
   V(I8x16Mul)              \