S390: Push the full simd register before lazy compile

If a lazy compilation happens in between 2 Wasm calls, we need to save
the full Simd register, since we can have live v128 values.

Port: 3b302d5c

Bug: chromium:1161555
Change-Id: Id79c609cc01e896f48aff39fdcbf4aa76ae6996e
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2649260Reviewed-by: Junliang Yan <junyan@redhat.com>
Commit-Queue: Milad Fa <mfarazma@redhat.com>
Cr-Commit-Position: refs/heads/master@{#72349}

src/builtins/s390/builtins-s390.cc

@@ -2449,7 +2449,7 @@ void Builtins::Generate_WasmCompileLazy(MacroAssembler* masm) {
     constexpr RegList fp_regs = DoubleRegister::ListOf(d0, d2);
 #endif
     __ MultiPush(gp_regs);
-    __ MultiPushDoubles(fp_regs);
+    __ MultiPushV128(fp_regs);
 
     // Pass instance and function index as explicit arguments to the runtime
     // function.
@@ -2462,7 +2462,7 @@ void Builtins::Generate_WasmCompileLazy(MacroAssembler* masm) {
     __ mov(ip, r2);
 
     // Restore registers.
-    __ MultiPopDoubles(fp_regs);
+    __ MultiPopV128(fp_regs);
     __ MultiPop(gp_regs);
   }
   // Finally, jump to the entrypoint.

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

@@ -611,6 +611,20 @@ void TurboAssembler::MultiPushDoubles(RegList dregs, Register location) {
   }
 }
 
+void TurboAssembler::MultiPushV128(RegList dregs, Register location) {
+  int16_t num_to_push = base::bits::CountPopulation(dregs);
+  int16_t stack_offset = num_to_push * kSimd128Size;
+
+  SubS64(location, location, Operand(stack_offset));
+  for (int16_t i = Simd128Register::kNumRegisters - 1; i >= 0; i--) {
+    if ((dregs & (1 << i)) != 0) {
+      Simd128Register dreg = Simd128Register::from_code(i);
+      stack_offset -= kSimd128Size;
+      StoreV128(dreg, MemOperand(location, stack_offset), r0);
+    }
+  }
+}
+
 void TurboAssembler::MultiPopDoubles(RegList dregs, Register location) {
   int16_t stack_offset = 0;
 
@@ -624,6 +638,19 @@ void TurboAssembler::MultiPopDoubles(RegList dregs, Register location) {
   AddS64(location, location, Operand(stack_offset));
 }
 
+void TurboAssembler::MultiPopV128(RegList dregs, Register location) {
+  int16_t stack_offset = 0;
+
+  for (int16_t i = 0; i < Simd128Register::kNumRegisters; i++) {
+    if ((dregs & (1 << i)) != 0) {
+      Simd128Register dreg = Simd128Register::from_code(i);
+      LoadV128(dreg, MemOperand(location, stack_offset), r0);
+      stack_offset += kSimd128Size;
+    }
+  }
+  AddS64(location, location, Operand(stack_offset));
+}
+
 void TurboAssembler::LoadRoot(Register destination, RootIndex index,
                               Condition) {
   LoadU64(destination,

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

@@ -147,6 +147,9 @@ class V8_EXPORT_PRIVATE TurboAssembler : public TurboAssemblerBase {
   void MultiPushDoubles(RegList dregs, Register location = sp);
   void MultiPopDoubles(RegList dregs, Register location = sp);
 
+  void MultiPushV128(RegList dregs, Register location = sp);
+  void MultiPopV128(RegList dregs, Register location = sp);
+
   // Calculate how much stack space (in bytes) are required to store caller
   // registers excluding those specified in the arguments.
   int RequiredStackSizeForCallerSaved(SaveFPRegsMode fp_mode,