s390: [wasm-simd] Implement simd shuffle

Change-Id: I3da840cdabf6d0ed8c4d823855acb999a0167167
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2073206Reviewed-by: Junliang Yan <jyan@ca.ibm.com>
Commit-Queue: Milad Farazmand <miladfar@ca.ibm.com>
Cr-Commit-Position: refs/heads/master@{#66467}

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

@@ -3689,6 +3689,21 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       break;
     }
 #undef BINOP_EXTRACT
+    case kS390_S8x16Shuffle: {
+      Simd128Register dst = i.OutputSimd128Register(),
+                      src0 = i.InputSimd128Register(0),
+                      src1 = i.InputSimd128Register(1);
+      int32_t k8x16_indices[] = {i.InputInt32(2), i.InputInt32(3),
+                                 i.InputInt32(4), i.InputInt32(5)};
+      // create 2 * 8 byte inputs indicating new indices
+      for (int i = 0, j = 0; i < 2; i++, j = +2) {
+        __ lgfi(i < 1 ? ip : r0, Operand(k8x16_indices[j + 1]));
+        __ aih(i < 1 ? ip : r0, Operand(k8x16_indices[j]));
+      }
+      __ vlvgp(kScratchDoubleReg, ip, r0);
+      __ vperm(dst, src0, src1, kScratchDoubleReg, Condition(0), Condition(0));
+      break;
+    }
     default:
       UNREACHABLE();
   }

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

@@ -300,6 +300,7 @@ namespace compiler {
   V(S390_I8x16SubSaturateS)                 \
   V(S390_I8x16AddSaturateU)                 \
   V(S390_I8x16SubSaturateU)                 \
+  V(S390_S8x16Shuffle)                      \
   V(S390_S1x4AnyTrue)                       \
   V(S390_S1x8AnyTrue)                       \
   V(S390_S1x16AnyTrue)                      \

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

@@ -246,6 +246,7 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kS390_I8x16SubSaturateS:
     case kS390_I8x16AddSaturateU:
     case kS390_I8x16SubSaturateU:
+    case kS390_S8x16Shuffle:
     case kS390_S1x4AnyTrue:
     case kS390_S1x8AnyTrue:
     case kS390_S1x16AnyTrue:

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

@@ -2718,6 +2718,34 @@ SIMD_CONVERSION_LIST(SIMD_VISIT_CONVERSION)
 #undef SIMD_CONVERSION_LIST
 #undef SIMD_TYPES
 
+void InstructionSelector::VisitS8x16Shuffle(Node* node) {
+  uint8_t shuffle[kSimd128Size];
+  bool is_swizzle;
+  CanonicalizeShuffle(node, shuffle, &is_swizzle);
+  S390OperandGenerator g(this);
+  Node* input0 = node->InputAt(0);
+  Node* input1 = node->InputAt(1);
+  // input registers are each in reverse order, we will have to remap the
+  // shuffle indices
+  int max_index = 15;
+  int total_lane_count = 2 * kSimd128Size;
+  uint8_t shuffle_remapped[kSimd128Size];
+  for (int i = 0; i < kSimd128Size; i++) {
+    uint8_t current_index = shuffle[i];
+    shuffle_remapped[i] = (current_index <= max_index
+                               ? max_index - current_index
+                               : total_lane_count - current_index + max_index);
+  }
+  Emit(kS390_S8x16Shuffle, g.DefineAsRegister(node), g.UseRegister(input0),
+       g.UseRegister(input1),
+       // Pack4Lanes reverses the bytes, therefore we will need to pass it in
+       // reverse
+       g.UseImmediate(Pack4Lanes(shuffle_remapped + 12)),
+       g.UseImmediate(Pack4Lanes(shuffle_remapped + 8)),
+       g.UseImmediate(Pack4Lanes(shuffle_remapped + 4)),
+       g.UseImmediate(Pack4Lanes(shuffle_remapped)));
+}
+
 void InstructionSelector::VisitS128Zero(Node* node) {
   S390OperandGenerator g(this);
   Emit(kS390_S128Zero, g.DefineAsRegister(node), g.DefineAsRegister(node));
@@ -2771,8 +2799,6 @@ void InstructionSelector::VisitF32x4Min(Node* node) { UNIMPLEMENTED(); }
 
 void InstructionSelector::VisitF32x4Max(Node* node) { UNIMPLEMENTED(); }
 
-void InstructionSelector::VisitS8x16Shuffle(Node* node) { UNIMPLEMENTED(); }
-
 void InstructionSelector::VisitS8x16Swizzle(Node* node) { UNIMPLEMENTED(); }
 
 void InstructionSelector::VisitF64x2Splat(Node* node) { UNIMPLEMENTED(); }