7009231: C1: Incorrect CAS code for longs on SPARC 32bit

Summary: Fix CAS of longs on SPARC 32bit and cmove on SPARC 64bit.
Reviewed-by: kvn

src/cpu/sparc/vm/c1_LIRAssembler_sparc.cpp

@@ -1705,8 +1705,7 @@ void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Op
 }
 
 
-void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result) {
-
+void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) {
   Assembler::Condition acond;
   switch (condition) {
     case lir_cond_equal:        acond = Assembler::equal;        break;
@@ -1737,7 +1736,12 @@ void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, L
     ShouldNotReachHere();
   }
   Label skip;
-  __ br(acond, false, Assembler::pt, skip);
+#ifdef _LP64
+    if  (type == T_INT) {
+      __ br(acond, false, Assembler::pt, skip);
+    } else
+#endif
+      __ brx(acond, false, Assembler::pt, skip); // checks icc on 32bit and xcc on 64bit
   if (opr1->is_constant() && opr1->type() == T_INT) {
     Register dest = result->as_register();
     if (Assembler::is_simm13(opr1->as_jint())) {
@@ -2688,6 +2692,11 @@ void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
 #ifdef _LP64
     __ mov(cmp_value_lo, t1);
     __ mov(new_value_lo, t2);
+    // perform the compare and swap operation
+    __ casx(addr, t1, t2);
+    // generate condition code - if the swap succeeded, t2 ("new value" reg) was
+    // overwritten with the original value in "addr" and will be equal to t1.
+    __ cmp(t1, t2);
 #else
     // move high and low halves of long values into single registers
     __ sllx(cmp_value_hi, 32, t1);         // shift high half into temp reg
@@ -2696,13 +2705,15 @@ void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
     __ sllx(new_value_hi, 32, t2);
     __ srl(new_value_lo, 0, new_value_lo);
     __ or3(t2, new_value_lo, t2);          // t2 holds 64-bit value to swap
-#endif
     // perform the compare and swap operation
     __ casx(addr, t1, t2);
     // generate condition code - if the swap succeeded, t2 ("new value" reg) was
     // overwritten with the original value in "addr" and will be equal to t1.
-    __ cmp(t1, t2);
-
+    // Produce icc flag for 32bit.
+    __ sub(t1, t2, t2);
+    __ srlx(t2, 32, t1);
+    __ orcc(t2, t1, G0);
+#endif
   } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj) {
     Register addr = op->addr()->as_pointer_register();
     Register cmp_value = op->cmp_value()->as_register();

src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp

@@ -662,7 +662,7 @@ void LIRGenerator::do_AttemptUpdate(Intrinsic* x) {
 
   // generate conditional move of boolean result
   LIR_Opr result = rlock_result(x);
-  __ cmove(lir_cond_equal, LIR_OprFact::intConst(1), LIR_OprFact::intConst(0), result);
+  __ cmove(lir_cond_equal, LIR_OprFact::intConst(1), LIR_OprFact::intConst(0), result, T_LONG);
 }
 
 
@@ -699,10 +699,10 @@ void LIRGenerator::do_CompareAndSwap(Intrinsic* x, ValueType* type) {
   else {
     ShouldNotReachHere();
   }
-
   // generate conditional move of boolean result
   LIR_Opr result = rlock_result(x);
-  __ cmove(lir_cond_equal, LIR_OprFact::intConst(1), LIR_OprFact::intConst(0), result);
+  __ cmove(lir_cond_equal, LIR_OprFact::intConst(1), LIR_OprFact::intConst(0),
+           result, as_BasicType(type));
   if (type == objectType) {  // Write-barrier needed for Object fields.
     // Precise card mark since could either be object or array
     post_barrier(addr, val.result());

src/cpu/x86/vm/c1_LIRAssembler_x86.cpp

@@ -2036,7 +2036,7 @@ void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
   }
 }
 
-void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result) {
+void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) {
   Assembler::Condition acond, ncond;
   switch (condition) {
     case lir_cond_equal:        acond = Assembler::equal;        ncond = Assembler::notEqual;     break;

src/cpu/x86/vm/c1_LIRGenerator_x86.cpp

@@ -741,7 +741,7 @@ void LIRGenerator::do_AttemptUpdate(Intrinsic* x) {
 
   // generate conditional move of boolean result
   LIR_Opr result = rlock_result(x);
-  __ cmove(lir_cond_equal, LIR_OprFact::intConst(1), LIR_OprFact::intConst(0), result);
+  __ cmove(lir_cond_equal, LIR_OprFact::intConst(1), LIR_OprFact::intConst(0), result, T_LONG);
 }
 
 
@@ -810,7 +810,8 @@ void LIRGenerator::do_CompareAndSwap(Intrinsic* x, ValueType* type) {
 
   // generate conditional move of boolean result
   LIR_Opr result = rlock_result(x);
-  __ cmove(lir_cond_equal, LIR_OprFact::intConst(1), LIR_OprFact::intConst(0), result);
+  __ cmove(lir_cond_equal, LIR_OprFact::intConst(1), LIR_OprFact::intConst(0),
+           result, as_BasicType(type));
   if (type == objectType) {   // Write-barrier needed for Object fields.
     // Seems to be precise
     post_barrier(addr, val.result());

src/share/vm/c1/c1_LIR.hpp

@@ -1568,15 +1568,16 @@ class LIR_Op2: public LIR_Op {
     assert(code == lir_cmp, "code check");
   }
 
-  LIR_Op2(LIR_Code code, LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result)
+  LIR_Op2(LIR_Code code, LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type)
     : LIR_Op(code, result, NULL)
     , _opr1(opr1)
     , _opr2(opr2)
-    , _type(T_ILLEGAL)
+    , _type(type)
     , _condition(condition)
     , _fpu_stack_size(0)
     , _tmp(LIR_OprFact::illegalOpr) {
     assert(code == lir_cmove, "code check");
+    assert(type != T_ILLEGAL, "cmove should have type");
   }
 
   LIR_Op2(LIR_Code code, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result = LIR_OprFact::illegalOpr,
@@ -1993,8 +1994,8 @@ class LIR_List: public CompilationResourceObj {
   void cmp_mem_int(LIR_Condition condition, LIR_Opr base, int disp, int c, CodeEmitInfo* info);
   void cmp_reg_mem(LIR_Condition condition, LIR_Opr reg, LIR_Address* addr, CodeEmitInfo* info);
 
-  void cmove(LIR_Condition condition, LIR_Opr src1, LIR_Opr src2, LIR_Opr dst) {
-    append(new LIR_Op2(lir_cmove, condition, src1, src2, dst));
+  void cmove(LIR_Condition condition, LIR_Opr src1, LIR_Opr src2, LIR_Opr dst, BasicType type) {
+    append(new LIR_Op2(lir_cmove, condition, src1, src2, dst, type));
   }
 
   void cas_long(LIR_Opr addr, LIR_Opr cmp_value, LIR_Opr new_value,

src/share/vm/c1/c1_LIRAssembler.cpp

@@ -685,7 +685,7 @@ void LIR_Assembler::emit_op2(LIR_Op2* op) {
       break;
 
     case lir_cmove:
-      cmove(op->condition(), op->in_opr1(), op->in_opr2(), op->result_opr());
+      cmove(op->condition(), op->in_opr1(), op->in_opr2(), op->result_opr(), op->type());
       break;
 
     case lir_shl:

src/share/vm/c1/c1_LIRAssembler.hpp

@@ -217,7 +217,7 @@ class LIR_Assembler: public CompilationResourceObj {
   void volatile_move_op(LIR_Opr src, LIR_Opr result, BasicType type, CodeEmitInfo* info);
   void comp_mem_op(LIR_Opr src, LIR_Opr result, BasicType type, CodeEmitInfo* info);  // info set for null exceptions
   void comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr result, LIR_Op2* op);
-  void cmove(LIR_Condition code, LIR_Opr left, LIR_Opr right, LIR_Opr result);
+  void cmove(LIR_Condition code, LIR_Opr left, LIR_Opr right, LIR_Opr result, BasicType type);
 
   void call(        LIR_OpJavaCall* op, relocInfo::relocType rtype);
   void ic_call(     LIR_OpJavaCall* op);

src/share/vm/c1/c1_LIRGenerator.cpp

@@ -856,7 +856,7 @@ void LIRGenerator::profile_branch(If* if_instr, If::Condition cond) {
     __ cmove(lir_cond(cond),
              LIR_OprFact::intptrConst(taken_count_offset),
              LIR_OprFact::intptrConst(not_taken_count_offset),
-             data_offset_reg);
+             data_offset_reg, as_BasicType(if_instr->x()->type()));
 
     // MDO cells are intptr_t, so the data_reg width is arch-dependent.
     LIR_Opr data_reg = new_pointer_register();
@@ -2591,7 +2591,7 @@ void LIRGenerator::do_IfOp(IfOp* x) {
   LIR_Opr reg = rlock_result(x);
 
   __ cmp(lir_cond(x->cond()), left.result(), right.result());
-  __ cmove(lir_cond(x->cond()), t_val.result(), f_val.result(), reg);
+  __ cmove(lir_cond(x->cond()), t_val.result(), f_val.result(), reg, as_BasicType(x->x()->type()));
 }
 
 

test/compiler/7009231/Test7009231.java

+/*
+ * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+/**
+ * @test
+ * @bug 7009231
+ * @summary C1: Incorrect CAS code for longs on SPARC 32bit
+ *
+ * @run main/othervm -Xbatch Test7009231
+ *
+ */
+
+import java.util.Random;
+import java.util.concurrent.atomic.AtomicLong;
+
+
+public class Test7009231 {
+    public static void main(String[] args) throws InterruptedException {
+        doTest(8);
+    }
+
+    private static void doTest(int nThreads) throws InterruptedException {
+        Thread[]         aThreads = new Thread[nThreads];
+        final AtomicLong atl      = new AtomicLong();
+
+        for (int i = 0; i < nThreads; i++) {
+          aThreads[i] = new RunnerThread(atl, 1L << (8 * i));
+        }
+
+        for (int i = 0; i < nThreads; i++) {
+          aThreads[i].start();
+        }
+
+        for (int i = 0; i < nThreads; i++) {
+          aThreads[i].join();
+        }
+    }
+
+    public static class RunnerThread extends Thread {
+        public RunnerThread(AtomicLong atomic, long lMask) {
+            m_lMask  = lMask;
+            m_atomic = atomic;
+        }
+
+        public void run() {
+            AtomicLong atomic = m_atomic;
+            long       lMask  = m_lMask;
+            for (int i = 0; i < 100000; i++) {
+                setBit(atomic, lMask);
+                clearBit(atomic, lMask);
+            }
+        }
+
+        protected void setBit(AtomicLong atomic, long lMask) {
+            long lWord;
+            do {
+                lWord = atomic.get();
+            } while (!atomic.compareAndSet(lWord, lWord | lMask));
+
+            if ((atomic.get() & lMask) == 0L) {
+                throw new InternalError();
+            }
+        }
+
+        protected void clearBit(AtomicLong atomic, long lMask) {
+            long lWord;
+            do {
+                lWord = atomic.get();
+            } while (!atomic.compareAndSet(lWord, lWord & ~lMask));
+
+            if ((atomic.get() & lMask) != 0L) {
+                throw new InternalError();
+            }
+        }
+
+        private long m_lMask;
+        private AtomicLong m_atomic;
+    }
+}