s390: [wasm-simd] Implementing simd comparisons

Change-Id: I60e839b0272a7dc13852549f543c9fa724f7fd36
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1994821Reviewed-by: Joran Siu <joransiu@ca.ibm.com>
Commit-Queue: Milad Farazmand <miladfar@ca.ibm.com>
Cr-Commit-Position: refs/heads/master@{#65712}

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

@@ -3107,6 +3107,229 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
              Condition(0));
       break;
     }
+    // vector comparisons
+    case kS390_I32x4MinS: {
+      __ vmn(i.OutputSimd128Register(), i.InputSimd128Register(0),
+             i.InputSimd128Register(1), Condition(0), Condition(0),
+             Condition(2));
+      break;
+    }
+    case kS390_I32x4MinU: {
+      __ vmnl(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(0),
+              Condition(2));
+      break;
+    }
+    case kS390_I16x8MinS: {
+      __ vmn(i.OutputSimd128Register(), i.InputSimd128Register(0),
+             i.InputSimd128Register(1), Condition(0), Condition(0),
+             Condition(1));
+      break;
+    }
+    case kS390_I16x8MinU: {
+      __ vmnl(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(0),
+              Condition(1));
+      break;
+    }
+    case kS390_I8x16MinS: {
+      __ vmn(i.OutputSimd128Register(), i.InputSimd128Register(0),
+             i.InputSimd128Register(1), Condition(0), Condition(0),
+             Condition(0));
+      break;
+    }
+    case kS390_I8x16MinU: {
+      __ vmnl(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(0),
+              Condition(0));
+      break;
+    }
+    case kS390_I32x4MaxS: {
+      __ vmx(i.OutputSimd128Register(), i.InputSimd128Register(0),
+             i.InputSimd128Register(1), Condition(0), Condition(0),
+             Condition(2));
+      break;
+    }
+    case kS390_I32x4MaxU: {
+      __ vmxl(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(0),
+              Condition(2));
+      break;
+    }
+    case kS390_I16x8MaxS: {
+      __ vmx(i.OutputSimd128Register(), i.InputSimd128Register(0),
+             i.InputSimd128Register(1), Condition(0), Condition(0),
+             Condition(1));
+      break;
+    }
+    case kS390_I16x8MaxU: {
+      __ vmxl(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(0),
+              Condition(1));
+      break;
+    }
+    case kS390_I8x16MaxS: {
+      __ vmx(i.OutputSimd128Register(), i.InputSimd128Register(0),
+             i.InputSimd128Register(1), Condition(0), Condition(0),
+             Condition(0));
+      break;
+    }
+    case kS390_I8x16MaxU: {
+      __ vmxl(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(0),
+              Condition(0));
+      break;
+    }
+    case kS390_F32x4Eq: {
+      __ vfce(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(0),
+              Condition(2));
+      break;
+    }
+    case kS390_I32x4Eq: {
+      __ vceq(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(2));
+      break;
+    }
+    case kS390_I16x8Eq: {
+      __ vceq(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(1));
+      break;
+    }
+    case kS390_I8x16Eq: {
+      __ vceq(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(0));
+      break;
+    }
+    case kS390_F32x4Ne: {
+      __ vfce(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(0),
+              Condition(2));
+      __ vno(i.OutputSimd128Register(), i.OutputSimd128Register(),
+             i.OutputSimd128Register(), Condition(0), Condition(0),
+             Condition(2));
+      break;
+    }
+    case kS390_I32x4Ne: {
+      __ vceq(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(2));
+      __ vno(i.OutputSimd128Register(), i.OutputSimd128Register(),
+             i.OutputSimd128Register(), Condition(0), Condition(0),
+             Condition(2));
+      break;
+    }
+    case kS390_I16x8Ne: {
+      __ vceq(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(1));
+      __ vno(i.OutputSimd128Register(), i.OutputSimd128Register(),
+             i.OutputSimd128Register(), Condition(0), Condition(0),
+             Condition(1));
+      break;
+    }
+    case kS390_I8x16Ne: {
+      __ vceq(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(0));
+      __ vno(i.OutputSimd128Register(), i.OutputSimd128Register(),
+             i.OutputSimd128Register(), Condition(0), Condition(0),
+             Condition(0));
+      break;
+    }
+    case kS390_F32x4Lt: {
+      __ vfch(i.OutputSimd128Register(), i.InputSimd128Register(1),
+              i.InputSimd128Register(0), Condition(0), Condition(0),
+              Condition(2));
+      break;
+    }
+    case kS390_F32x4Le: {
+      __ vfche(i.OutputSimd128Register(), i.InputSimd128Register(1),
+               i.InputSimd128Register(0), Condition(0), Condition(0),
+               Condition(2));
+      break;
+    }
+    case kS390_I32x4GtS: {
+      __ vch(i.OutputSimd128Register(), i.InputSimd128Register(0),
+             i.InputSimd128Register(1), Condition(0), Condition(2));
+      break;
+    }
+    case kS390_I32x4GeS: {
+      __ vceq(kScratchDoubleReg, i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(2));
+      __ vch(i.OutputSimd128Register(), i.InputSimd128Register(0),
+             i.InputSimd128Register(1), Condition(0), Condition(2));
+      __ vo(i.OutputSimd128Register(), i.OutputSimd128Register(),
+            kScratchDoubleReg, Condition(0), Condition(0), Condition(2));
+      break;
+    }
+    case kS390_I32x4GtU: {
+      __ vchl(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(2));
+      break;
+    }
+    case kS390_I32x4GeU: {
+      __ vceq(kScratchDoubleReg, i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(2));
+      __ vchl(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(2));
+      __ vo(i.OutputSimd128Register(), i.OutputSimd128Register(),
+            kScratchDoubleReg, Condition(0), Condition(0), Condition(2));
+      break;
+    }
+    case kS390_I16x8GtS: {
+      __ vch(i.OutputSimd128Register(), i.InputSimd128Register(0),
+             i.InputSimd128Register(1), Condition(0), Condition(1));
+      break;
+    }
+    case kS390_I16x8GeS: {
+      __ vceq(kScratchDoubleReg, i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(1));
+      __ vch(i.OutputSimd128Register(), i.InputSimd128Register(0),
+             i.InputSimd128Register(1), Condition(0), Condition(1));
+      __ vo(i.OutputSimd128Register(), i.OutputSimd128Register(),
+            kScratchDoubleReg, Condition(0), Condition(0), Condition(1));
+      break;
+    }
+    case kS390_I16x8GtU: {
+      __ vchl(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(1));
+      break;
+    }
+    case kS390_I16x8GeU: {
+      __ vceq(kScratchDoubleReg, i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(1));
+      __ vchl(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(1));
+      __ vo(i.OutputSimd128Register(), i.OutputSimd128Register(),
+            kScratchDoubleReg, Condition(0), Condition(0), Condition(1));
+      break;
+    }
+    case kS390_I8x16GtS: {
+      __ vch(i.OutputSimd128Register(), i.InputSimd128Register(0),
+             i.InputSimd128Register(1), Condition(0), Condition(0));
+      break;
+    }
+    case kS390_I8x16GeS: {
+      __ vceq(kScratchDoubleReg, i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(0));
+      __ vch(i.OutputSimd128Register(), i.InputSimd128Register(0),
+             i.InputSimd128Register(1), Condition(0), Condition(0));
+      __ vo(i.OutputSimd128Register(), i.OutputSimd128Register(),
+            kScratchDoubleReg, Condition(0), Condition(0), Condition(0));
+      break;
+    }
+    case kS390_I8x16GtU: {
+      __ vchl(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(0));
+      break;
+    }
+    case kS390_I8x16GeU: {
+      __ vceq(kScratchDoubleReg, i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(0));
+      __ vchl(i.OutputSimd128Register(), i.InputSimd128Register(0),
+              i.InputSimd128Register(1), Condition(0), Condition(0));
+      __ vo(i.OutputSimd128Register(), i.OutputSimd128Register(),
+            kScratchDoubleReg, Condition(0), Condition(0), Condition(0));
+      break;
+    }
     default:
       UNREACHABLE();
   }

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

@@ -204,6 +204,10 @@ namespace compiler {
   V(S390_F32x4AddHoriz)                     \
   V(S390_F32x4Sub)                          \
   V(S390_F32x4Mul)                          \
+  V(S390_F32x4Eq)                           \
+  V(S390_F32x4Ne)                           \
+  V(S390_F32x4Lt)                           \
+  V(S390_F32x4Le)                           \
   V(S390_I32x4Splat)                        \
   V(S390_I32x4ExtractLane)                  \
   V(S390_I32x4ReplaceLane)                  \
@@ -211,6 +215,16 @@ namespace compiler {
   V(S390_I32x4AddHoriz)                     \
   V(S390_I32x4Sub)                          \
   V(S390_I32x4Mul)                          \
+  V(S390_I32x4MinS)                         \
+  V(S390_I32x4MinU)                         \
+  V(S390_I32x4MaxS)                         \
+  V(S390_I32x4MaxU)                         \
+  V(S390_I32x4Eq)                           \
+  V(S390_I32x4Ne)                           \
+  V(S390_I32x4GtS)                          \
+  V(S390_I32x4GeS)                          \
+  V(S390_I32x4GtU)                          \
+  V(S390_I32x4GeU)                          \
   V(S390_I16x8Splat)                        \
   V(S390_I16x8ExtractLaneU)                 \
   V(S390_I16x8ExtractLaneS)                 \
@@ -219,6 +233,16 @@ namespace compiler {
   V(S390_I16x8AddHoriz)                     \
   V(S390_I16x8Sub)                          \
   V(S390_I16x8Mul)                          \
+  V(S390_I16x8MinS)                         \
+  V(S390_I16x8MinU)                         \
+  V(S390_I16x8MaxS)                         \
+  V(S390_I16x8MaxU)                         \
+  V(S390_I16x8Eq)                           \
+  V(S390_I16x8Ne)                           \
+  V(S390_I16x8GtS)                          \
+  V(S390_I16x8GeS)                          \
+  V(S390_I16x8GtU)                          \
+  V(S390_I16x8GeU)                          \
   V(S390_I8x16Splat)                        \
   V(S390_I8x16ExtractLaneU)                 \
   V(S390_I8x16ExtractLaneS)                 \
@@ -226,6 +250,16 @@ namespace compiler {
   V(S390_I8x16Add)                          \
   V(S390_I8x16Sub)                          \
   V(S390_I8x16Mul)                          \
+  V(S390_I8x16MinS)                         \
+  V(S390_I8x16MinU)                         \
+  V(S390_I8x16MaxS)                         \
+  V(S390_I8x16MaxU)                         \
+  V(S390_I8x16Eq)                           \
+  V(S390_I8x16Ne)                           \
+  V(S390_I8x16GtS)                          \
+  V(S390_I8x16GeS)                          \
+  V(S390_I8x16GtU)                          \
+  V(S390_I8x16GeU)                          \
   V(S390_StoreSimd128)                      \
   V(S390_LoadSimd128)
 

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

@@ -150,6 +150,10 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_F32x4AddHoriz:
     case kS390_F32x4Sub:
     case kS390_F32x4Mul:
+    case kS390_F32x4Eq:
+    case kS390_F32x4Ne:
+    case kS390_F32x4Lt:
+    case kS390_F32x4Le:
     case kS390_I32x4Splat:
     case kS390_I32x4ExtractLane:
     case kS390_I32x4ReplaceLane:
@@ -157,6 +161,16 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_I32x4AddHoriz:
     case kS390_I32x4Sub:
     case kS390_I32x4Mul:
+    case kS390_I32x4MinS:
+    case kS390_I32x4MinU:
+    case kS390_I32x4MaxS:
+    case kS390_I32x4MaxU:
+    case kS390_I32x4Eq:
+    case kS390_I32x4Ne:
+    case kS390_I32x4GtS:
+    case kS390_I32x4GeS:
+    case kS390_I32x4GtU:
+    case kS390_I32x4GeU:
     case kS390_I16x8Splat:
     case kS390_I16x8ExtractLaneU:
     case kS390_I16x8ExtractLaneS:
@@ -165,6 +179,16 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_I16x8AddHoriz:
     case kS390_I16x8Sub:
     case kS390_I16x8Mul:
+    case kS390_I16x8MinS:
+    case kS390_I16x8MinU:
+    case kS390_I16x8MaxS:
+    case kS390_I16x8MaxU:
+    case kS390_I16x8Eq:
+    case kS390_I16x8Ne:
+    case kS390_I16x8GtS:
+    case kS390_I16x8GeS:
+    case kS390_I16x8GtU:
+    case kS390_I16x8GeU:
     case kS390_I8x16Splat:
     case kS390_I8x16ExtractLaneU:
     case kS390_I8x16ExtractLaneS:
@@ -172,6 +196,16 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_I8x16Add:
     case kS390_I8x16Sub:
     case kS390_I8x16Mul:
+    case kS390_I8x16MinS:
+    case kS390_I8x16MinU:
+    case kS390_I8x16MaxS:
+    case kS390_I8x16MaxU:
+    case kS390_I8x16Eq:
+    case kS390_I8x16Ne:
+    case kS390_I8x16GtS:
+    case kS390_I8x16GeS:
+    case kS390_I8x16GtU:
+    case kS390_I8x16GeU:
       return kNoOpcodeFlags;
 
     case kS390_LoadWordS8:

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

@@ -2522,17 +2522,51 @@ void InstructionSelector::VisitWord64AtomicStore(Node* node) {
   V(F32x4AddHoriz)         \
   V(F32x4Sub)              \
   V(F32x4Mul)              \
+  V(F32x4Eq)               \
+  V(F32x4Ne)               \
+  V(F32x4Lt)               \
+  V(F32x4Le)               \
   V(I32x4Add)              \
   V(I32x4AddHoriz)         \
   V(I32x4Sub)              \
   V(I32x4Mul)              \
+  V(I32x4MinS)             \
+  V(I32x4MinU)             \
+  V(I32x4MaxS)             \
+  V(I32x4MaxU)             \
+  V(I32x4Eq)               \
+  V(I32x4Ne)               \
+  V(I32x4GtS)              \
+  V(I32x4GeS)              \
+  V(I32x4GtU)              \
+  V(I32x4GeU)              \
   V(I16x8Add)              \
   V(I16x8AddHoriz)         \
   V(I16x8Sub)              \
   V(I16x8Mul)              \
+  V(I16x8MinS)             \
+  V(I16x8MinU)             \
+  V(I16x8MaxS)             \
+  V(I16x8MaxU)             \
+  V(I16x8Eq)               \
+  V(I16x8Ne)               \
+  V(I16x8GtS)              \
+  V(I16x8GeS)              \
+  V(I16x8GtU)              \
+  V(I16x8GeU)              \
   V(I8x16Add)              \
   V(I8x16Sub)              \
-  V(I8x16Mul)
+  V(I8x16Mul)              \
+  V(I8x16MinS)             \
+  V(I8x16MinU)             \
+  V(I8x16MaxS)             \
+  V(I8x16MaxU)             \
+  V(I8x16Eq)               \
+  V(I8x16Ne)               \
+  V(I8x16GtS)              \
+  V(I8x16GeS)              \
+  V(I8x16GtU)              \
+  V(I8x16GeU)
 
 #define SIMD_VISIT_SPLAT(Type)                               \
   void InstructionSelector::Visit##Type##Splat(Node* node) { \
@@ -2587,30 +2621,10 @@ void InstructionSelector::VisitI32x4Shl(Node* node) { UNIMPLEMENTED(); }
 
 void InstructionSelector::VisitI32x4ShrS(Node* node) { UNIMPLEMENTED(); }
 
-void InstructionSelector::VisitI32x4MaxS(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI32x4MinS(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI32x4Eq(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI32x4Ne(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI32x4MinU(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI32x4MaxU(Node* node) { UNIMPLEMENTED(); }
-
 void InstructionSelector::VisitI32x4ShrU(Node* node) { UNIMPLEMENTED(); }
 
 void InstructionSelector::VisitI32x4Neg(Node* node) { UNIMPLEMENTED(); }
 
-void InstructionSelector::VisitI32x4GtS(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI32x4GeS(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI32x4GtU(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI32x4GeU(Node* node) { UNIMPLEMENTED(); }
-
 void InstructionSelector::VisitI16x8Shl(Node* node) { UNIMPLEMENTED(); }
 
 void InstructionSelector::VisitI16x8ShrS(Node* node) { UNIMPLEMENTED(); }
@@ -2625,14 +2639,6 @@ void InstructionSelector::VisitI16x8SubSaturateS(Node* node) {
   UNIMPLEMENTED();
 }
 
-void InstructionSelector::VisitI16x8MinS(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI16x8MaxS(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI16x8Eq(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI16x8Ne(Node* node) { UNIMPLEMENTED(); }
-
 void InstructionSelector::VisitI16x8AddSaturateU(Node* node) {
   UNIMPLEMENTED();
 }
@@ -2641,20 +2647,8 @@ void InstructionSelector::VisitI16x8SubSaturateU(Node* node) {
   UNIMPLEMENTED();
 }
 
-void InstructionSelector::VisitI16x8MinU(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI16x8MaxU(Node* node) { UNIMPLEMENTED(); }
-
 void InstructionSelector::VisitI16x8Neg(Node* node) { UNIMPLEMENTED(); }
 
-void InstructionSelector::VisitI16x8GtS(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI16x8GeS(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI16x8GtU(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI16x8GeU(Node* node) { UNIMPLEMENTED(); }
-
 void InstructionSelector::VisitI16x8RoundingAverageU(Node* node) {
   UNIMPLEMENTED();
 }
@@ -2673,18 +2667,6 @@ void InstructionSelector::VisitI8x16SubSaturateS(Node* node) {
   UNIMPLEMENTED();
 }
 
-void InstructionSelector::VisitI8x16MinS(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI8x16MaxS(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI8x16Eq(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI8x16Ne(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI8x16GtS(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI8x16GeS(Node* node) { UNIMPLEMENTED(); }
-
 void InstructionSelector::VisitI8x16AddSaturateU(Node* node) {
   UNIMPLEMENTED();
 }
@@ -2693,14 +2675,6 @@ void InstructionSelector::VisitI8x16SubSaturateU(Node* node) {
   UNIMPLEMENTED();
 }
 
-void InstructionSelector::VisitI8x16MinU(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI8x16MaxU(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI8x16GtU(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitI8x16GeU(Node* node) { UNIMPLEMENTED(); }
-
 void InstructionSelector::VisitS128And(Node* node) { UNIMPLEMENTED(); }
 
 void InstructionSelector::VisitS128Or(Node* node) { UNIMPLEMENTED(); }
@@ -2711,14 +2685,6 @@ void InstructionSelector::VisitS128Not(Node* node) { UNIMPLEMENTED(); }
 
 void InstructionSelector::VisitS128Zero(Node* node) { UNIMPLEMENTED(); }
 
-void InstructionSelector::VisitF32x4Eq(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitF32x4Ne(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitF32x4Lt(Node* node) { UNIMPLEMENTED(); }
-
-void InstructionSelector::VisitF32x4Le(Node* node) { UNIMPLEMENTED(); }
-
 void InstructionSelector::EmitPrepareResults(
     ZoneVector<PushParameter>* results, const CallDescriptor* call_descriptor,
     Node* node) {