s390: [wasm-simd] Implement simd conversion operations

Change-Id: I6f7d3a5f123edea8674c0f9217b03760f3af016f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2028451Reviewed-by: Junliang Yan <jyan@ca.ibm.com>
Commit-Queue: Milad Farazmand <miladfar@ca.ibm.com>
Cr-Commit-Position: refs/heads/master@{#66059}

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

@@ -3493,6 +3493,57 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       __ vsel(dst, src1, src2, mask, Condition(0), Condition(0));
       break;
     }
+    // vector conversions
+#define CONVERT_FLOAT_TO_INT32(convert)                                       \
+  for (int index = 0; index < 4; index++) {                                   \
+    __ vlgv(kScratchReg, kScratchDoubleReg, MemOperand(r0, index),            \
+            Condition(2));                                                    \
+    __ vlvg(kScratchDoubleReg, kScratchReg, MemOperand(r0, 0), Condition(2)); \
+    __ convert(kScratchReg, kScratchDoubleReg, kRoundToZero);                 \
+    __ vlvg(dst, kScratchReg, MemOperand(r0, index), Condition(2));           \
+  }
+    case kS390_I32x4SConvertF32x4: {
+      Simd128Register src = i.InputSimd128Register(0);
+      Simd128Register dst = i.OutputSimd128Register();
+      // NaN to 0
+      __ vlr(kScratchDoubleReg, src, Condition(0), Condition(0), Condition(0));
+      __ vfce(kScratchDoubleReg, kScratchDoubleReg, kScratchDoubleReg,
+              Condition(0), Condition(0), Condition(2));
+      __ vn(kScratchDoubleReg, src, kScratchDoubleReg, Condition(0),
+            Condition(0), Condition(0));
+      CONVERT_FLOAT_TO_INT32(ConvertFloat32ToInt32)
+      break;
+    }
+    case kS390_I32x4UConvertF32x4: {
+      Simd128Register src = i.InputSimd128Register(0);
+      Simd128Register dst = i.OutputSimd128Register();
+      // NaN to 0, negative to 0
+      __ vx(kScratchDoubleReg, kScratchDoubleReg, kScratchDoubleReg,
+            Condition(0), Condition(0), Condition(0));
+      __ vfmax(kScratchDoubleReg, src, kScratchDoubleReg, Condition(1),
+               Condition(0), Condition(2));
+      CONVERT_FLOAT_TO_INT32(ConvertFloat32ToUnsignedInt32)
+      break;
+    }
+#undef CONVERT_FLOAT_TO_INT32
+#define CONVERT_INT32_TO_FLOAT(convert)                                       \
+  Simd128Register src = i.InputSimd128Register(0);                            \
+  Simd128Register dst = i.OutputSimd128Register();                            \
+  for (int index = 0; index < 4; index++) {                                   \
+    __ vlgv(kScratchReg, src, MemOperand(r0, index), Condition(2));           \
+    __ convert(kScratchDoubleReg, kScratchReg);                               \
+    __ vlgv(kScratchReg, kScratchDoubleReg, MemOperand(r0, 0), Condition(2)); \
+    __ vlvg(dst, kScratchReg, MemOperand(r0, index), Condition(2));           \
+  }
+    case kS390_F32x4SConvertI32x4: {
+      CONVERT_INT32_TO_FLOAT(ConvertIntToFloat)
+      break;
+    }
+    case kS390_F32x4UConvertI32x4: {
+      CONVERT_INT32_TO_FLOAT(ConvertUnsignedIntToFloat)
+      break;
+    }
+#undef CONVERT_INT32_TO_FLOAT
     default:
       UNREACHABLE();
   }

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

@@ -234,6 +234,10 @@ namespace compiler {
   V(S390_I32x4Shl)                          \
   V(S390_I32x4ShrS)                         \
   V(S390_I32x4ShrU)                         \
+  V(S390_I32x4SConvertF32x4)                \
+  V(S390_I32x4UConvertF32x4)                \
+  V(S390_F32x4SConvertI32x4)                \
+  V(S390_F32x4UConvertI32x4)                \
   V(S390_I16x8ExtractLaneU)                 \
   V(S390_I16x8ExtractLaneS)                 \
   V(S390_I16x8ReplaceLane)                  \

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

@@ -179,6 +179,10 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_I32x4ShrS:
     case kS390_I32x4ShrU:
     case kS390_I32x4Neg:
+    case kS390_I32x4SConvertF32x4:
+    case kS390_I32x4UConvertF32x4:
+    case kS390_F32x4SConvertI32x4:
+    case kS390_F32x4UConvertI32x4:
     case kS390_I16x8Splat:
     case kS390_I16x8ExtractLaneU:
     case kS390_I16x8ExtractLaneS:

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

@@ -2581,15 +2581,15 @@ void InstructionSelector::VisitWord64AtomicStore(Node* node) {
   V(I8x16Neg)             \
   V(S128Not)
 
-#define SIMD_SHIFT_OPCODES(V) \
-  V(I32x4Shl)                 \
-  V(I32x4ShrS)                \
-  V(I32x4ShrU)                \
-  V(I16x8Shl)                 \
-  V(I16x8ShrS)                \
-  V(I16x8ShrU)                \
-  V(I8x16Shl)                 \
-  V(I8x16ShrS)                \
+#define SIMD_SHIFT_LIST(V) \
+  V(I32x4Shl)              \
+  V(I32x4ShrS)             \
+  V(I32x4ShrU)             \
+  V(I16x8Shl)              \
+  V(I16x8ShrS)             \
+  V(I16x8ShrU)             \
+  V(I8x16Shl)              \
+  V(I8x16ShrS)             \
   V(I8x16ShrU)
 
 #define SIMD_BOOL_LIST(V) \
@@ -2600,6 +2600,12 @@ void InstructionSelector::VisitWord64AtomicStore(Node* node) {
   V(S1x8AllTrue)          \
   V(S1x16AllTrue)
 
+#define SIMD_CONVERSION_LIST(V) \
+  V(I32x4SConvertF32x4)         \
+  V(I32x4UConvertF32x4)         \
+  V(F32x4SConvertI32x4)         \
+  V(F32x4UConvertI32x4)
+
 #define SIMD_VISIT_SPLAT(Type)                               \
   void InstructionSelector::Visit##Type##Splat(Node* node) { \
     S390OperandGenerator g(this);                            \
@@ -2666,9 +2672,9 @@ SIMD_UNOP_LIST(SIMD_VISIT_UNOP)
          g.UseUniqueRegister(node->InputAt(0)),         \
          g.UseUniqueRegister(node->InputAt(1)));        \
   }
-SIMD_SHIFT_OPCODES(SIMD_VISIT_SHIFT)
+SIMD_SHIFT_LIST(SIMD_VISIT_SHIFT)
 #undef SIMD_VISIT_SHIFT
-#undef SIMD_SHIFT_OPCODES
+#undef SIMD_SHIFT_LIST
 
 #define SIMD_VISIT_BOOL(Opcode)                                           \
   void InstructionSelector::Visit##Opcode(Node* node) {                   \
@@ -2679,6 +2685,17 @@ SIMD_SHIFT_OPCODES(SIMD_VISIT_SHIFT)
   }
 SIMD_BOOL_LIST(SIMD_VISIT_BOOL)
 #undef SIMD_VISIT_BOOL
+#undef SIMD_BOOL_LIST
+
+#define SIMD_VISIT_CONVERSION(Opcode)                   \
+  void InstructionSelector::Visit##Opcode(Node* node) { \
+    S390OperandGenerator g(this);                       \
+    Emit(kS390_##Opcode, g.DefineAsRegister(node),      \
+         g.UseRegister(node->InputAt(0)));              \
+  }
+SIMD_CONVERSION_LIST(SIMD_VISIT_CONVERSION)
+#undef SIMD_VISIT_CONVERSION
+#undef SIMD_CONVERSION_LIST
 #undef SIMD_TYPES
 
 void InstructionSelector::VisitS128Zero(Node* node) {
@@ -2766,22 +2783,6 @@ void InstructionSelector::VisitF32x4Min(Node* node) { UNIMPLEMENTED(); }
 
 void InstructionSelector::VisitF32x4Max(Node* node) { UNIMPLEMENTED(); }
 
-void InstructionSelector::VisitF32x4SConvertI32x4(Node* node) {
-  UNIMPLEMENTED();
-}
-
-void InstructionSelector::VisitF32x4UConvertI32x4(Node* node) {
-  UNIMPLEMENTED();
-}
-
-void InstructionSelector::VisitI32x4SConvertF32x4(Node* node) {
-  UNIMPLEMENTED();
-}
-
-void InstructionSelector::VisitI32x4UConvertF32x4(Node* node) {
-  UNIMPLEMENTED();
-}
-
 void InstructionSelector::VisitI32x4SConvertI16x8Low(Node* node) {
   UNIMPLEMENTED();
 }