s390: [wasm-simd] Add to simd conversion operations

Change-Id: Id3d6566dc7b8a4cc10574abe79246ebf1398f016
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2043024
Reviewed-by: Junliang Yan <jyan@ca.ibm.com>
Commit-Queue: Milad Farazmand <miladfar@ca.ibm.com>
Cr-Commit-Position: refs/heads/master@{#66185}

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

@@ -3544,6 +3544,71 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       break;
     }
 #undef CONVERT_INT32_TO_FLOAT
+#define VECTOR_UNPACK(op, mode)                                             \
+  __ op(i.OutputSimd128Register(), i.InputSimd128Register(0), Condition(0), \
+        Condition(0), Condition(mode));
+    case kS390_I32x4SConvertI16x8Low: {
+      VECTOR_UNPACK(vupl, 1);
+      break;
+    }
+    case kS390_I32x4SConvertI16x8High: {
+      VECTOR_UNPACK(vuph, 1);
+      break;
+    }
+    case kS390_I32x4UConvertI16x8Low: {
+      VECTOR_UNPACK(vupll, 1);
+      break;
+    }
+    case kS390_I32x4UConvertI16x8High: {
+      VECTOR_UNPACK(vuplh, 1);
+      break;
+    }
+    case kS390_I16x8SConvertI8x16Low: {
+      VECTOR_UNPACK(vupl, 0);
+      break;
+    }
+    case kS390_I16x8SConvertI8x16High: {
+      VECTOR_UNPACK(vuph, 0);
+      break;
+    }
+    case kS390_I16x8UConvertI8x16Low: {
+      VECTOR_UNPACK(vupll, 0);
+      break;
+    }
+    case kS390_I16x8UConvertI8x16High: {
+      VECTOR_UNPACK(vuplh, 0);
+      break;
+    }
+#undef VECTOR_UNPACK
+    case kS390_I16x8SConvertI32x4:
+      __ vpks(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(2));
+      break;
+    case kS390_I8x16SConvertI16x8:
+      __ vpks(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(1));
+      break;
+#define VECTOR_PACK_UNSIGNED(mode)                                             \
+  Simd128Register tempFPReg = i.ToSimd128Register(instr->TempAt(0));           \
+  __ vx(kScratchDoubleReg, kScratchDoubleReg, kScratchDoubleReg, Condition(0), \
+        Condition(0), Condition(mode));                                        \
+  __ vmx(tempFPReg, i.InputSimd128Register(0), kScratchDoubleReg,              \
+         Condition(0), Condition(0), Condition(mode));                         \
+  __ vmx(kScratchDoubleReg, i.InputSimd128Register(1), kScratchDoubleReg,      \
+         Condition(0), Condition(0), Condition(mode));                         \
+  __ vpkls(i.OutputSimd128Register(), tempFPReg, kScratchDoubleReg,            \
+           Condition(0), Condition(mode));
+    case kS390_I16x8UConvertI32x4: {
+      // treat inputs as signed, and saturate to unsigned (negative to 0)
+      VECTOR_PACK_UNSIGNED(2)
+      break;
+    }
+    case kS390_I8x16UConvertI16x8: {
+      // treat inputs as signed, and saturate to unsigned (negative to 0)
+      VECTOR_PACK_UNSIGNED(1)
+      break;
+    }
+#undef VECTOR_PACK_UNSIGNED
     default:
       UNREACHABLE();
   }

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

@@ -212,6 +212,8 @@ namespace compiler {
   V(S390_F32x4Neg)                          \
   V(S390_F32x4RecipApprox)                  \
   V(S390_F32x4RecipSqrtApprox)              \
+  V(S390_F32x4SConvertI32x4)                \
+  V(S390_F32x4UConvertI32x4)                \
   V(S390_I32x4Splat)                        \
   V(S390_I32x4ExtractLane)                  \
   V(S390_I32x4ReplaceLane)                  \
@@ -236,8 +238,10 @@ namespace compiler {
   V(S390_I32x4ShrU)                         \
   V(S390_I32x4SConvertF32x4)                \
   V(S390_I32x4UConvertF32x4)                \
-  V(S390_F32x4SConvertI32x4)                \
+  V(S390_I32x4SConvertI16x8Low)             \
-  V(S390_F32x4UConvertI32x4)                \
+  V(S390_I32x4SConvertI16x8High)            \
+  V(S390_I32x4UConvertI16x8Low)             \
+  V(S390_I32x4UConvertI16x8High)            \
   V(S390_I16x8ExtractLaneU)                 \
   V(S390_I16x8ExtractLaneS)                 \
   V(S390_I16x8ReplaceLane)                  \
@@ -259,6 +263,12 @@ namespace compiler {
   V(S390_I16x8ShrS)                         \
   V(S390_I16x8ShrU)                         \
   V(S390_I16x8Neg)                          \
+  V(S390_I16x8SConvertI32x4)                \
+  V(S390_I16x8UConvertI32x4)                \
+  V(S390_I16x8SConvertI8x16Low)             \
+  V(S390_I16x8SConvertI8x16High)            \
+  V(S390_I16x8UConvertI8x16Low)             \
+  V(S390_I16x8UConvertI8x16High)            \
   V(S390_I8x16Splat)                        \
   V(S390_I8x16ExtractLaneU)                 \
   V(S390_I8x16ExtractLaneS)                 \
@@ -280,6 +290,8 @@ namespace compiler {
   V(S390_I8x16ShrS)                         \
   V(S390_I8x16ShrU)                         \
   V(S390_I8x16Neg)                          \
+  V(S390_I8x16SConvertI16x8)                \
+  V(S390_I8x16UConvertI16x8)                \
   V(S390_S1x4AnyTrue)                       \
   V(S390_S1x8AnyTrue)                       \
   V(S390_S1x16AnyTrue)                      \

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

@@ -158,6 +158,8 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_F32x4Neg:
     case kS390_F32x4RecipApprox:
     case kS390_F32x4RecipSqrtApprox:
+    case kS390_F32x4SConvertI32x4:
+    case kS390_F32x4UConvertI32x4:
     case kS390_I32x4Splat:
     case kS390_I32x4ExtractLane:
     case kS390_I32x4ReplaceLane:
@@ -181,8 +183,10 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_I32x4Neg:
     case kS390_I32x4SConvertF32x4:
     case kS390_I32x4UConvertF32x4:
-    case kS390_F32x4SConvertI32x4:
+    case kS390_I32x4SConvertI16x8Low:
-    case kS390_F32x4UConvertI32x4:
+    case kS390_I32x4SConvertI16x8High:
+    case kS390_I32x4UConvertI16x8Low:
+    case kS390_I32x4UConvertI16x8High:
     case kS390_I16x8Splat:
     case kS390_I16x8ExtractLaneU:
     case kS390_I16x8ExtractLaneS:
@@ -205,6 +209,12 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_I16x8ShrS:
     case kS390_I16x8ShrU:
     case kS390_I16x8Neg:
+    case kS390_I16x8SConvertI32x4:
+    case kS390_I16x8UConvertI32x4:
+    case kS390_I16x8SConvertI8x16Low:
+    case kS390_I16x8SConvertI8x16High:
+    case kS390_I16x8UConvertI8x16Low:
+    case kS390_I16x8UConvertI8x16High:
     case kS390_I8x16Splat:
     case kS390_I8x16ExtractLaneU:
     case kS390_I8x16ExtractLaneS:
@@ -226,6 +236,8 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_I8x16ShrS:
     case kS390_I8x16ShrU:
     case kS390_I8x16Neg:
+    case kS390_I8x16SConvertI16x8:
+    case kS390_I8x16UConvertI16x8:
     case kS390_S1x4AnyTrue:
     case kS390_S1x8AnyTrue:
     case kS390_S1x16AnyTrue:

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

@@ -2554,6 +2554,8 @@ void InstructionSelector::VisitWord64AtomicStore(Node* node) {
   V(I16x8GeS)              \
   V(I16x8GtU)              \
   V(I16x8GeU)              \
+  V(I16x8SConvertI32x4)    \
+  V(I16x8UConvertI32x4)    \
   V(I8x16Add)              \
   V(I8x16Sub)              \
   V(I8x16Mul)              \
@@ -2567,18 +2569,28 @@ void InstructionSelector::VisitWord64AtomicStore(Node* node) {
   V(I8x16GeS)              \
   V(I8x16GtU)              \
   V(I8x16GeU)              \
+  V(I8x16SConvertI16x8)    \
+  V(I8x16UConvertI16x8)    \
   V(S128And)               \
   V(S128Or)                \
   V(S128Xor)
 
-#define SIMD_UNOP_LIST(V) \
+#define SIMD_UNOP_LIST(V)   \
-  V(F32x4Abs)             \
+  V(F32x4Abs)               \
-  V(F32x4Neg)             \
+  V(F32x4Neg)               \
-  V(F32x4RecipApprox)     \
+  V(F32x4RecipApprox)       \
-  V(F32x4RecipSqrtApprox) \
+  V(F32x4RecipSqrtApprox)   \
-  V(I32x4Neg)             \
+  V(I32x4Neg)               \
-  V(I16x8Neg)             \
+  V(I32x4SConvertI16x8Low)  \
-  V(I8x16Neg)             \
+  V(I32x4SConvertI16x8High) \
+  V(I32x4UConvertI16x8Low)  \
+  V(I32x4UConvertI16x8High) \
+  V(I16x8Neg)               \
+  V(I16x8SConvertI8x16Low)  \
+  V(I16x8SConvertI8x16High) \
+  V(I16x8UConvertI8x16Low)  \
+  V(I16x8UConvertI8x16High) \
+  V(I8x16Neg)               \
   V(S128Not)
 
 #define SIMD_SHIFT_LIST(V) \
@@ -2783,53 +2795,6 @@ void InstructionSelector::VisitF32x4Min(Node* node) { UNIMPLEMENTED(); }
 
 void InstructionSelector::VisitF32x4Max(Node* node) { UNIMPLEMENTED(); }
 
-void InstructionSelector::VisitI32x4SConvertI16x8Low(Node* node) {
-  UNIMPLEMENTED();
-}
-
-void InstructionSelector::VisitI32x4SConvertI16x8High(Node* node) {
-  UNIMPLEMENTED();
-}
-
-void InstructionSelector::VisitI32x4UConvertI16x8Low(Node* node) {
-  UNIMPLEMENTED();
-}
-
-void InstructionSelector::VisitI32x4UConvertI16x8High(Node* node) {
-  UNIMPLEMENTED();
-}
-
-void InstructionSelector::VisitI16x8SConvertI8x16Low(Node* node) {
-  UNIMPLEMENTED();
-}
-
-void InstructionSelector::VisitI16x8SConvertI8x16High(Node* node) {
-  UNIMPLEMENTED();
-}
-
-void InstructionSelector::VisitI16x8UConvertI8x16Low(Node* node) {
-  UNIMPLEMENTED();
-}
-
-void InstructionSelector::VisitI16x8UConvertI8x16High(Node* node) {
-  UNIMPLEMENTED();
-}
-
-void InstructionSelector::VisitI16x8SConvertI32x4(Node* node) {
-  UNIMPLEMENTED();
-}
-void InstructionSelector::VisitI16x8UConvertI32x4(Node* node) {
-  UNIMPLEMENTED();
-}
-
-void InstructionSelector::VisitI8x16SConvertI16x8(Node* node) {
-  UNIMPLEMENTED();
-}
-
-void InstructionSelector::VisitI8x16UConvertI16x8(Node* node) {
-  UNIMPLEMENTED();
-}
-
 void InstructionSelector::VisitS8x16Shuffle(Node* node) { UNIMPLEMENTED(); }
 
 void InstructionSelector::VisitS8x16Swizzle(Node* node) { UNIMPLEMENTED(); }