S390 [liftoff]: Implement FP promote and demote

Change-Id: Ie5eac61238442bedcb5f9d61a67cc47de261655a
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3447692Reviewed-by: Junliang Yan <junyan@redhat.com>
Commit-Queue: Milad Farazmand <mfarazma@redhat.com>
Cr-Commit-Position: refs/heads/main@{#79002}

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

@@ -5876,6 +5876,40 @@ void TurboAssembler::I8x16SubSatU(Simd128Register dst, Simd128Register src1,
 }
 #undef BINOP_EXTRACT
 
+void TurboAssembler::F64x2PromoteLowF32x4(Simd128Register dst,
+                                          Simd128Register src,
+                                          Simd128Register scratch1,
+                                          Register scratch2, Register scratch3,
+                                          Register scratch4) {
+  Register holder = scratch3;
+  for (int index = 0; index < 2; ++index) {
+    vlgv(scratch2, src, MemOperand(scratch2, index + 2), Condition(2));
+    MovIntToFloat(scratch1, scratch2);
+    ldebr(scratch1, scratch1);
+    MovDoubleToInt64(holder, scratch1);
+    holder = scratch4;
+  }
+  vlvgp(dst, scratch3, scratch4);
+}
+
+void TurboAssembler::F32x4DemoteF64x2Zero(Simd128Register dst,
+                                          Simd128Register src,
+                                          Simd128Register scratch1,
+                                          Register scratch2, Register scratch3,
+                                          Register scratch4) {
+  Register holder = scratch3;
+  for (int index = 0; index < 2; ++index) {
+    vlgv(scratch2, src, MemOperand(r0, index), Condition(3));
+    MovInt64ToDouble(scratch1, scratch2);
+    ledbr(scratch1, scratch1);
+    MovFloatToInt(holder, scratch1);
+    holder = scratch4;
+  }
+  vx(dst, dst, dst, Condition(0), Condition(0), Condition(2));
+  vlvg(dst, scratch3, MemOperand(r0, 2), Condition(2));
+  vlvg(dst, scratch4, MemOperand(r0, 3), Condition(2));
+}
+
 // Vector LE Load and Transform instructions.
 #ifdef V8_TARGET_BIG_ENDIAN
 #define IS_BIG_ENDIAN true

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

@@ -1134,6 +1134,12 @@ class V8_EXPORT_PRIVATE TurboAssembler : public TurboAssemblerBase {
                           Simd128Register src2, Simd128Register scratch);
   void I8x16UConvertI16x8(Simd128Register dst, Simd128Register src1,
                           Simd128Register src2, Simd128Register scratch);
+  void F64x2PromoteLowF32x4(Simd128Register dst, Simd128Register src,
+                            Simd128Register scratch1, Register scratch2,
+                            Register scratch3, Register scratch4);
+  void F32x4DemoteF64x2Zero(Simd128Register dst, Simd128Register src,
+                            Simd128Register scratch1, Register scratch2,
+                            Register scratch3, Register scratch4);
   void S128Select(Simd128Register dst, Simd128Register src1,
                   Simd128Register src2, Simd128Register mask);
 

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

@@ -3064,32 +3064,15 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       break;
     }
     case kS390_F64x2PromoteLowF32x4: {
-      Register holder = r1;
-      for (int index = 0; index < 2; ++index) {
-        __ vlgv(r0, i.InputSimd128Register(0), MemOperand(r0, index + 2),
-                Condition(2));
-        __ MovIntToFloat(kScratchDoubleReg, r0);
-        __ ldebr(kScratchDoubleReg, kScratchDoubleReg);
-        __ MovDoubleToInt64(holder, kScratchDoubleReg);
-        holder = ip;
-      }
-      __ vlvgp(i.OutputSimd128Register(), r1, ip);
+      __ F64x2PromoteLowF32x4(i.OutputSimd128Register(),
+                              i.InputSimd128Register(0), kScratchDoubleReg, r0,
+                              r1, ip);
       break;
     }
     case kS390_F32x4DemoteF64x2Zero: {
-      Simd128Register dst = i.OutputSimd128Register();
-      Register holder = r1;
-      for (int index = 0; index < 2; ++index) {
-        __ vlgv(r0, i.InputSimd128Register(0), MemOperand(r0, index),
-                Condition(3));
-        __ MovInt64ToDouble(kScratchDoubleReg, r0);
-        __ ledbr(kScratchDoubleReg, kScratchDoubleReg);
-        __ MovFloatToInt(holder, kScratchDoubleReg);
-        holder = ip;
-      }
-      __ vx(dst, dst, dst, Condition(0), Condition(0), Condition(2));
-      __ vlvg(dst, r1, MemOperand(r0, 2), Condition(2));
-      __ vlvg(dst, ip, MemOperand(r0, 3), Condition(2));
+      __ F32x4DemoteF64x2Zero(i.OutputSimd128Register(),
+                              i.InputSimd128Register(0), kScratchDoubleReg, r0,
+                              r1, ip);
       break;
     }
     case kS390_I32x4TruncSatF64x2SZero: {

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

@@ -2581,7 +2581,7 @@ void LiftoffAssembler::emit_f64x2_convert_low_i32x4_u(LiftoffRegister dst,
 
 void LiftoffAssembler::emit_f64x2_promote_low_f32x4(LiftoffRegister dst,
                                                     LiftoffRegister src) {
-  bailout(kSimd, "f64x2.promote_low_f32x4");
+  F64x2PromoteLowF32x4(dst.fp(), src.fp(), kScratchDoubleReg, r0, r1, ip);
 }
 
 void LiftoffAssembler::emit_i64x2_bitmask(LiftoffRegister dst,
@@ -2688,7 +2688,7 @@ void LiftoffAssembler::emit_f32x4_uconvert_i32x4(LiftoffRegister dst,
 
 void LiftoffAssembler::emit_f32x4_demote_f64x2_zero(LiftoffRegister dst,
                                                     LiftoffRegister src) {
-  bailout(kSimd, "f32x4.demote_f64x2_zero");
+  F32x4DemoteF64x2Zero(dst.fp(), src.fp(), kScratchDoubleReg, r0, r1, ip);
 }
 
 void LiftoffAssembler::emit_i8x16_sconvert_i16x8(LiftoffRegister dst,