[RFA] Fix bytecode interpreter bitrot

[RFA] Fix bytecode interpreter bitrot

[Keith Seitz]

[-- Attachment #1: Type: text/plain, Size: 987 bytes --]

Hi,

I don't know if anybody actually uses the bytecode interpreter anymore, 
but since we do have all the code lurking in the sources somewhere, it 
would be nice to keep it working.

It seems that over the past months, several bugs have found their way 
into the BC interpreter. I believe this patch fixes those problems.

Keith

ChangeLog
2006-09-28  Keith Seitz  <keiths@redhat.com>

         * include/java-interp.h (prepared): Change type to pc_t.
         (insn_index): Define for both DIRECT_THREADED and bytecode 
interpreters.
         * interpret.cc [!DIRECT_THREADED] (POKEI): Fix typo.
         (insn_index): Implement for bytecode interpreter.
         * interpret-run.cc [!DIRECT_THREADED] (AVAL1U): Add _Jv_Linker 
class
         qualifier to resolve_pool_entry.
         [!DIRECT_THREADED] (AVAL2U): Likewise.
         [!DIRECT_THREADED] bytecode() cannot be called without an object.
         Changed all typos.
         [!DIRECT_THREADED] Likewise for defining_class.

[-- Attachment #2: bc-bitrot.patch --]
[-- Type: text/x-patch, Size: 4182 bytes --]

Index: include/java-interp.h
===================================================================
--- include/java-interp.h	(revision 117093)
+++ include/java-interp.h	(working copy)
@@ -144,7 +144,7 @@
   int line_table_len;  
   _Jv_LineTableEntry *line_table;
 
-  void *prepared;
+  pc_t prepared;
   int number_insn_slots;
 
   unsigned char* bytecode () 
@@ -191,13 +191,9 @@
   // number info is unavailable.
   int get_source_line(pc_t mpc);
 
-
-
-#ifdef DIRECT_THREADED
   // Convenience function for indexing bytecode PC/insn slots in
   // line tables for JDWP
   jlong insn_index (pc_t pc);
-#endif
   
    public:
    
Index: interpret.cc
===================================================================
--- interpret.cc	(revision 117093)
+++ interpret.cc	(working copy)
@@ -200,7 +200,7 @@
 #define PEEKA(I)  (locals+(I))->o
 
 #define POKEI(I,V)  	\
-DEBUG_LOCALS_INSN(I,i)	\
+DEBUG_LOCALS_INSN(I,'i'); \
 ((locals+(I))->i = (V))
 
 
@@ -1307,23 +1307,27 @@
   return self->ncode;
 }
 
-#ifdef DIRECT_THREADED
 /* Find the index of the given insn in the array of insn slots
    for this method. Returns -1 if not found. */
 jlong
 _Jv_InterpMethod::insn_index (pc_t pc)
 {
   jlong left = 0;
+#ifdef DIRECT_THREADED
   jlong right = number_insn_slots;
-  insn_slot* slots = reinterpret_cast<insn_slot*> (prepared);
+  pc_t insns = prepared;
+#else
+  jlong right = code_length;
+  pc_t insns = bytecode ();
+#endif
 
   while (right >= 0)
     {
       jlong mid = (left + right) / 2;
-      if (&slots[mid] == pc)
+      if (&insns[mid] == pc)
 	return mid;
 
-      if (pc < &slots[mid])
+      if (pc < &insns[mid])
 	right = mid - 1;
       else
         left = mid + 1;
@@ -1331,7 +1335,6 @@
 
   return -1;
 }
-#endif // DIRECT_THREADED
 
 void
 _Jv_InterpMethod::get_line_table (jlong& start, jlong& end,
Index: interpret-run.cc
===================================================================
--- interpret-run.cc	(revision 117093)
+++ interpret-run.cc	(working copy)
@@ -291,10 +291,10 @@
   // class'.
 #define AVAL1U()						\
   ({ int index = get1u (pc++);					\
-      resolve_pool_entry (meth->defining_class, index).o; })
+    _Jv_Linker::resolve_pool_entry (meth->defining_class, index).o; })
 #define AVAL2U()						\
   ({ int index = get2u (pc); pc += 2;				\
-      resolve_pool_entry (meth->defining_class, index).o; })
+    _Jv_Linker::resolve_pool_entry (meth->defining_class, index).o; })
   // Note that we don't need to resolve the pool entry here as class
   // constants are never wide.
 #define AVAL2UP() ({ int index = get2u (pc); pc += 2; &pool_data[index]; })
@@ -303,7 +303,7 @@
 #define PCVAL(unionval) unionval.i
 #define AMPAMP(label) NULL
 
-  pc = bytecode ();
+  pc = meth->bytecode ();
 
 #endif /* DIRECT_THREADED */
 
@@ -1545,7 +1545,7 @@
 	pc_t base_pc = pc - 1;
 	int index = POPI ();
 
-	pc_t base = (pc_t) bytecode ();
+	pc_t base = (pc_t) meth->bytecode ();
 	while ((pc - base) % 4 != 0)
 	  ++pc;
 
@@ -1601,7 +1601,7 @@
 	unsigned char *base_pc = pc-1;
 	int index = POPI();
 
-	unsigned char* base = bytecode ();
+	unsigned char* base = meth->bytecode ();
 	while ((pc-base) % 4 != 0)
 	  ++pc;
 
@@ -2469,7 +2469,7 @@
 #ifdef DIRECT_THREADED
       void *logical_pc = (void *) ((insn_slot *) pc - 1);
 #else
-      int logical_pc = pc - 1 - bytecode ();
+      int logical_pc = pc - 1 - meth->bytecode ();
 #endif
       _Jv_InterpException *exc = meth->exceptions ();
       jclass exc_class = ex->getClass ();
@@ -2484,8 +2484,8 @@
 #else
 	      jclass handler = NULL;
 	      if (exc[i].handler_type.i != 0)
-		handler = (_Jv_Linker::resolve_pool_entry (defining_class,
-							     exc[i].handler_type.i)).clazz;
+		handler = (_Jv_Linker::resolve_pool_entry (meth->defining_class,
+							   exc[i].handler_type.i)).clazz;
 #endif /* DIRECT_THREADED */
 
 	      if (handler == NULL || handler->isAssignableFrom (exc_class))
@@ -2494,7 +2494,7 @@
 #ifdef DIRECT_THREADED
 		  pc = (insn_slot *) exc[i].handler_pc.p;
 #else
-		  pc = bytecode () + exc[i].handler_pc.i;
+		  pc = meth->bytecode () + exc[i].handler_pc.i;
 #endif /* DIRECT_THREADED */
 		  sp = stack;
 		  sp++->o = ex; // Push exception.

Re: [RFA] Fix bytecode interpreter bitrot

[Tom Tromey]

>>>>> "Keith" == Keith Seitz <keiths@redhat.com> writes:

Keith> I don't know if anybody actually uses the bytecode interpreter
Keith> anymore

Me neither.

Keith> It seems that over the past months, several bugs have found their way
Keith> into the BC interpreter. I believe this patch fixes those problems.

This is ok, thanks.

Tom

Re: [RFA] Fix bytecode interpreter bitrot

[David Daney]

Tom Tromey wrote:
>>>>>>"Keith" == Keith Seitz <keiths@redhat.com> writes:
> 
> 
> Keith> I don't know if anybody actually uses the bytecode interpreter
> Keith> anymore
> 
> Me neither.
> 

Correct me if I am wrong, but don't most of the gij test cases use it?

Also it seems that it would be very useful for loading bytecode at 
runtime (for rmi and jini code for example).

David Daney

Re: [RFA] Fix bytecode interpreter bitrot

[Casey Marshall]

David Daney wrote:
> Tom Tromey wrote:
>>>>>>> "Keith" == Keith Seitz <keiths@redhat.com> writes:
>>
>>
>> Keith> I don't know if anybody actually uses the bytecode interpreter
>> Keith> anymore
>>
>> Me neither.
>>
> 
> Correct me if I am wrong, but don't most of the gij test cases use it?
> 
> Also it seems that it would be very useful for loading bytecode at
> runtime (for rmi and jini code for example).
> 

Also applets.

Re: [RFA] Fix bytecode interpreter bitrot

[Tom Tromey]

David> Correct me if I am wrong, but don't most of the gij test cases use it?

Keith meant the old "pure bytecode" interpreter.  You have to build
libgcj specially to get this -- the default is to use the direct
threaded interpreter, which is faster but uses more memory.

Tom

Re: [RFA] Fix bytecode interpreter bitrot

[Keith Seitz]

Tom Tromey wrote:

> Keith> It seems that over the past months, several bugs have found their way
> Keith> into the BC interpreter. I believe this patch fixes those problems.
> 
> This is ok, thanks.

Thanks!

Keith