S390 [liftoff]: Implement simd rounding average

Change-Id: I709253796b8493cf365f21247c367974ffbb3106
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3440435Reviewed-by: Junliang Yan <junyan@redhat.com>
Commit-Queue: Milad Farazmand <mfarazma@redhat.com>
Cr-Commit-Position: refs/heads/main@{#78954}

src/codegen/s390/macro-assembler-s390.cc

@@ -5372,12 +5372,14 @@ SIMD_BINOP_LIST_VRR_B(EMIT_SIMD_BINOP_VRR_B)
   V(I16x8MinU, vmnl, 0, 0, 1)              \
   V(I16x8MaxS, vmx, 0, 0, 1)               \
   V(I16x8MaxU, vmxl, 0, 0, 1)              \
+  V(I16x8RoundingAverageU, vavgl, 0, 0, 1) \
   V(I8x16Add, va, 0, 0, 0)                 \
   V(I8x16Sub, vs, 0, 0, 0)                 \
   V(I8x16MinS, vmn, 0, 0, 0)               \
   V(I8x16MinU, vmnl, 0, 0, 0)              \
   V(I8x16MaxS, vmx, 0, 0, 0)               \
   V(I8x16MaxU, vmxl, 0, 0, 0)              \
+  V(I8x16RoundingAverageU, vavgl, 0, 0, 0) \
   V(S128And, vn, 0, 0, 0)                  \
   V(S128Or, vo, 0, 0, 0)                   \
   V(S128Xor, vx, 0, 0, 0)                  \

src/codegen/s390/macro-assembler-s390.h

@@ -1256,6 +1256,7 @@ class V8_EXPORT_PRIVATE TurboAssembler : public TurboAssemblerBase {
   V(I16x8Shl, const Operand&)               \
   V(I16x8ShrS, const Operand&)              \
   V(I16x8ShrU, const Operand&)              \
+  V(I16x8RoundingAverageU, Simd128Register) \
   V(I8x16Add, Simd128Register)              \
   V(I8x16Sub, Simd128Register)              \
   V(I8x16Eq, Simd128Register)               \
@@ -1274,6 +1275,7 @@ class V8_EXPORT_PRIVATE TurboAssembler : public TurboAssemblerBase {
   V(I8x16Shl, const Operand&)               \
   V(I8x16ShrS, const Operand&)              \
   V(I8x16ShrU, const Operand&)              \
+  V(I8x16RoundingAverageU, Simd128Register) \
   V(S128And, Simd128Register)               \
   V(S128Or, Simd128Register)                \
   V(S128Xor, Simd128Register)               \

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

@@ -2621,6 +2621,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
   V(I16x8Shl, Register)                     \
   V(I16x8ShrS, Register)                    \
   V(I16x8ShrU, Register)                    \
+  V(I16x8RoundingAverageU, Simd128Register) \
   V(I8x16Add, Simd128Register)              \
   V(I8x16Sub, Simd128Register)              \
   V(I8x16Eq, Simd128Register)               \
@@ -2636,6 +2637,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
   V(I8x16Shl, Register)                     \
   V(I8x16ShrS, Register)                    \
   V(I8x16ShrU, Register)                    \
+  V(I8x16RoundingAverageU, Simd128Register) \
   V(S128And, Simd128Register)               \
   V(S128Or, Simd128Register)                \
   V(S128Xor, Simd128Register)               \
@@ -2796,19 +2798,6 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
 #undef EMIT_SIMD_QFM
 #undef SIMD_QFM_LIST
 
-      // vector binops
-    case kS390_I16x8RoundingAverageU: {
-      __ vavgl(i.OutputSimd128Register(), i.InputSimd128Register(0),
-               i.InputSimd128Register(1), Condition(0), Condition(0),
-               Condition(1));
-      break;
-    }
-    case kS390_I8x16RoundingAverageU: {
-      __ vavgl(i.OutputSimd128Register(), i.InputSimd128Register(0),
-               i.InputSimd128Register(1), Condition(0), Condition(0),
-               Condition(0));
-      break;
-    }
     // vector unary ops
     case kS390_F32x4RecipApprox: {
       __ mov(kScratchReg, Operand(1));

src/wasm/baseline/s390/liftoff-assembler-s390.h

@@ -2323,6 +2323,7 @@ void LiftoffAssembler::emit_smi_check(Register obj, Label* target,
   V(i16x8_shl, I16x8Shl, gp)                             \
   V(i16x8_shr_s, I16x8ShrS, gp)                          \
   V(i16x8_shr_u, I16x8ShrU, gp)                          \
+  V(i16x8_rounding_average_u, I16x8RoundingAverageU, fp) \
   V(i8x16_add, I8x16Add, fp)                             \
   V(i8x16_sub, I8x16Sub, fp)                             \
   V(i8x16_eq, I8x16Eq, fp)                               \
@@ -2338,6 +2339,7 @@ void LiftoffAssembler::emit_smi_check(Register obj, Label* target,
   V(i8x16_shl, I8x16Shl, gp)                             \
   V(i8x16_shr_s, I8x16ShrS, gp)                          \
   V(i8x16_shr_u, I8x16ShrU, gp)                          \
+  V(i8x16_rounding_average_u, I8x16RoundingAverageU, fp) \
   V(s128_and, S128And, fp)                               \
   V(s128_or, S128Or, fp)                                 \
   V(s128_xor, S128Xor, fp)                               \
@@ -2735,18 +2737,6 @@ void LiftoffAssembler::emit_i32x4_trunc_sat_f64x2_u_zero(LiftoffRegister dst,
   bailout(kSimd, "i32x4.trunc_sat_f64x2_u_zero");
 }
 
-void LiftoffAssembler::emit_i8x16_rounding_average_u(LiftoffRegister dst,
-                                                     LiftoffRegister lhs,
-                                                     LiftoffRegister rhs) {
-  bailout(kUnsupportedArchitecture, "emit_i8x16_rounding_average_u");
-}
-
-void LiftoffAssembler::emit_i16x8_rounding_average_u(LiftoffRegister dst,
-                                                     LiftoffRegister lhs,
-                                                     LiftoffRegister rhs) {
-  bailout(kUnsupportedArchitecture, "emit_i16x8_rounding_average_u");
-}
-
 void LiftoffAssembler::StackCheck(Label* ool_code, Register limit_address) {
   LoadU64(limit_address, MemOperand(limit_address));
   CmpU64(sp, limit_address);