Finish porting the object barrier for riscv

Author: caizixian

openjdk/cpu/riscv/mmtkObjectBarrierSetAssembler_riscv.cpp

@@ -4,8 +4,73 @@
 
 #define __ masm->
 
+// extern void object_reference_write_slow_call(void* obj, void* dst, uint64_t val);
+
+// void write_post(char* SIDE_METADATA_BASE_ADDRESS, void* obj, uint64_t val) {
+//     char tmp2 = *(SIDE_METADATA_BASE_ADDRESS + ((uint64_t)obj >> 6));
+//     char tmp3 = ((uint64_t)obj >> 3) & 7;
+//     tmp2 = tmp2 >> tmp3;
+//     if ((tmp2 & 1) == 1) {
+//         object_reference_write_slow_call(obj, 0, val);
+//     }
+// }
+
+// write_post(char*, void*, unsigned long):
+//         srli    a5,a1,6
+//         add     a0,a0,a5
+//         lbu     a5,0(a0)
+//         srli    a3,a1,3
+//         andi    a3,a3,7
+//         sraw    a5,a5,a3
+//         andi    a5,a5,1
+//         mv      a4,a1
+//         bne     a5,zero,.L4
+//         ret
+// .L4:
+//         li      a1,0
+//         mv      a0,a4
+//         tail    _Z32object_reference_write_slow_callPvS_m
 void MMTkObjectBarrierSetAssembler::object_reference_write_post(MacroAssembler* masm, DecoratorSet decorators, Address dst, Register val, Register tmp1, Register tmp2) const {
-  assert(false, "Not implemented");
+  // tmp1 and tmp2 is from MacroAssembler::access_store_at
+  // See tmplateTable_riscv, we don't actually get any temporary register.
+  // printf("object_reference_write_post\n");
+  if (can_remove_barrier(decorators, val, /* skip_const_null */ true)) return;
+#if MMTK_ENABLE_BARRIER_FASTPATH
+  Label done;
+
+  Register obj = dst.base();
+  assert(obj->is_valid(), "dst must be an offset from a base register");
+  assert_different_registers(obj, t0, t1);
+  assert_different_registers(val, t0, t1);
+
+  // t0 = load-byte (SIDE_METADATA_BASE_ADDRESS + (obj >> 6));
+  __ mv(t0, obj);
+  __ srli(t0, t0, 6); // t0 = obj >> 6;
+  __ li(t1, SIDE_METADATA_BASE_ADDRESS);
+  __ add(t0, t0, t1); // t0 = SIDE_METADATA_BASE_ADDRESS + (obj >> 6);
+  __ lbu(t0, Address(t0, 0));
+  // t1 = (obj >> 3) & 7
+  __ mv(t1, obj);
+  __ srli(t1, t1, 3);
+  __ andi(t1, t1, 7);
+  // t0 = t0 >> t1
+  __ sraw(t0, t0, t1);
+  // if ((t0 & 1) == 1) fall through to slowpath;
+  __ andi(t0, t0, 1);
+  __ bnez(t0, done); // (t0 & 1) == 1 is equivalent to (t0 & 1) != 0
+  // setup calling convention
+  __ mv(c_rarg0, obj);
+  __ la(c_rarg1, dst);
+  __ mv(c_rarg2, val == noreg ? zr : val);
+  __ call_VM_leaf_base(FN_ADDR(MMTkBarrierSetRuntime::object_reference_write_slow_call), 3);
+
+  __ bind(done);
+#else
+  __ mv(c_rarg0, obj);
+  __ la(c_rarg1, dst);
+  __ mv(c_rarg2, val == noreg ? zr : val);
+  __ call_VM_leaf_base(FN_ADDR(MMTkBarrierSetRuntime::object_reference_write_post_call), 3);
+#endif
 }
 
 void MMTkObjectBarrierSetAssembler::arraycopy_epilogue(MacroAssembler* masm, DecoratorSet decorators, bool is_oop,