Cygnus Papers

From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 13109 invoked by alias); 11 Nov 2012 23:39:11 -0000 Received: (qmail 13084 invoked by uid 22791); 11 Nov 2012 23:39:10 -0000 X-SWARE-Spam-Status: No, hits=1.0 required=5.0 tests=BAYES_50,KHOP_THREADED,SARE_OBFU_PART_CIA X-Spam-Check-By: sourceware.org Received: from ainaz.pair.com (HELO ainaz.pair.com) (209.68.2.66) by sourceware.org (qpsmtpd/0.43rc1) with ESMTP; Sun, 11 Nov 2012 23:38:55 +0000 Received: from [172.26.27.191] (65-114-101-150.dia.static.qwest.net [65.114.101.150]) by ainaz.pair.com (Postfix) with ESMTPSA id 1F7B43F40F; Sun, 11 Nov 2012 18:38:50 -0500 (EST) Date: Sun, 11 Nov 2012 23:39:00 -0000 From: Gerald Pfeifer To: gcc-patches@gcc.gnu.org, java-patches@gcc.gnu.org Subject: Re: [wwwdocs,Java] Fix up java/papers/compiling.html In-Reply-To: Message-ID: References: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Mailing-List: contact java-patches-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Subscribe: List-Archive: List-Post: List-Help: , Sender: java-patches-owner@gcc.gnu.org X-SW-Source: 2012-q4/txt/msg00029.txt.bz2 On Sun, 4 Nov 2012, Gerald Pfeifer wrote: > Looking at that page, I noticed a lot of broken navigation (broken > images,...). This removes that navigation and also address the worst > in terms of broken markup. > > Still bin/preprocess excludes this file from the regular treatment > that our general web pages get; that'll be next, but it'll require > a fair number of additional cleanups. Done thusly, last weekend actually, but only committed/senthere now. Gerald 2012-11-11 Gerald Pfeifer * papers/compiling.html: Transform into XHTML 1.0 Transitional. Index: papers/compiling.html =================================================================== RCS file: /cvs/gcc/wwwdocs/htdocs/java/papers/compiling.html,v retrieving revision 1.2 diff -u -3 -p -r1.2 compiling.html --- papers/compiling.html 3 Nov 2012 15:15:21 -0000 1.2 +++ papers/compiling.html 11 Nov 2012 09:23:36 -0000 @@ -1,20 +1,25 @@ - - + + + + + + Cygnus Papers - - - + + - +

Compiling Java for Embedded Systems

By Per Bothner (bothner@cygnus.com)

- +

While a major factor in Java's success is its use of portable bytecodes, we believe it cannot become a mainstream programming language without mainstream implementation @@ -27,9 +32,8 @@ technology. In this paper, we discuss is using traditional compiler, linker, and debugging technology; particular emphasis is given to using Java in embedded and limited memory environments. - +

1. Java implementation
2. Issues with embedded Java @@ -42,7 +46,7 @@ memory environments.
3. Compiling Java
- 3.1 Transforming bytecodes
- 3.1 Transforming bytecodes
- 3.2 Class meta-data
- 3.3 Static references
- 3.4 Linking
- 4.2 Profiling
5 Status -
Acknowledgements -
Bibliography +
5 Status
Acknowledgements
Bibliography

1 Java implementation

Java (see Bibliography for JavaSpec) has +

Java (see Bibliography for JavaSpec) has become popular because it is a decent programming language, is buzzword-compliant (object-oriented and web-enabled), and because it is implemented by compiling to portable bytecodes (see Bibliography -for JavaVMSpec). The traditional Java +for JavaVMSpec). The traditional Java implementation model is to compile Java source code into .class files containing machine-independent byte-code instructions. These .class files are downloaded and interpreted by a browser or some other Java "Virtual -Machine" (VM). See Figure 1 for a -model of the traditional Java implementation.

+Machine" (VM). See Figure 1 for a +model of the traditional Java implementation.

Figure 1: Typical Java implementation

However, interpreting bytecodes makes Java program many times slower than comparable C or C++ programs. One approach to improving this situation is "Just-In-Time" (JIT) compilers. These dynamically translate bytecodes to machine code just before a method -is first executed.

+is first executed.

This can provide substantial speed-up, but it is still slower than +

This can provide substantial speed-up, but it is still slower than C or C++. There are three main drawbacks with the JIT approach -compared to conventional compilers:

+compared to conventional compilers:

The compilation is done every time the application is executed, which means start-up times are much worse than pre-compiled code.
The JIT compiler must remain in (virtual) memory while the application is executing. This memory may be quite costly in an embedded application.

While JIT compilers have an important place in a Java system, for +

While JIT compilers have an important place in a Java system, for frequently used applications it is better to use a more traditional "ahead-of-time" or batch compiler. While many think of Java as primarily an internet/web language, others are interested in using Java as an alternative to traditional languages such as C++, provided performance is acceptable. For embedded applications, it makes much more sense to pre-compile the Java program, especially if the program -is to be in ROM.

+is to be in ROM.

Cygnus is building a Java programming environment that is based on +

Cygnus is building a Java programming environment that is based on a conventional compiler, linker, and debugger, using Java-enhanced versions of the existing GNU programming tools. These have been ported to just about every chip in use (including many chips only used in embedded systems), and so we will have a uniquely portable Java -system. See Figure 2 for a model of the -compilation process.

+system. See Figure 2 for a model of the +compilation process.

- +

Figure 2: Compiled Java

Top of Page

2 Issues with embedded Java

Sun's motto for Java is "Write once, run anywhere". As +

Sun's motto for Java is "Write once, run anywhere". As part of that, Sun has been pushing Java as also suitable for embedded systems, announcing specifications for "Embedded Java" and "Personal Java" specifications. The latter (just published at the time of writing) is primarily a restricted library that a Java -application can expect.

+application can expect.

"Embedded systems" covers a large spectrum from 4-bit +

"Embedded systems" covers a large spectrum from 4-bit chips with tens of bytes of RAM, to 32- and 64-bit systems with megabytes of RAM. I believe it will be very difficult to squeeze a reasonably complete implementation of Java into less than one @@ -150,25 +154,25 @@ general, there is less memory available environment where Java usually runs. Those designers that have been leery of C++ because of performance concerns (perceived or real) will not embrace Java. On the other hand, those that have been leery of C++ -because of its complexity will find Java easier to master.

+because of its complexity will find Java easier to master.

Top of Page

2.1 Advantages of Java

Java has a number of advantages for embedded systems. Using classes +

Java has a number of advantages for embedded systems. Using classes to organize the code enforces modularity, and encourages data abstraction. Java has many useful and standardized classes (for graphics, networking, simple math and containers, internationalization, files, and much more). This means a designer can count on having these libraries on any (full) implementation of -Java.

+Java.

Java programs are generally more portable than C or C++ programs:

The size of integer, floats, and character is defined by the language.
Initial values of fields are defined, and the languages require that local variables be set before use in a way that the compiler can check.

In fact, the only major non-determinacy in Java is due to +

In fact, the only major non-determinacy in Java is due to time-dependencies between interacting threads. Safety-critical -applications will find the following features very useful:

+applications will find the following features very useful:

More disciplined use of pointers, called "references" @@ -198,24 +202,24 @@ case from error cases, and to handle the manner.

The portability of Java (including a portable binary format) means +

The portability of Java (including a portable binary format) means that an application can run on many hardware platforms, with no -porting effort (at least that's the theory).

+porting effort (at least that's the theory).

Top of Page

2.2 Code compactness

It has been argued that even for ROM-based embedded systems (where +

It has been argued that even for ROM-based embedded systems (where portability is not an issue), it still makes sense to use a bytecode-based implementation of Java, since the bytecodes are supposedly more compact than native code. However, it is not at all clear if that is really the case. The actual bytecode instructions of the Fib class (a program to calculate Fibonacci numbers, which is -discussed in the section, 5. Status) only -take 134 bytes, while the compiled instructions for i86 and Sparc take +discussed in the section, 5. Status) only +take 134 bytes, while the compiled instructions for i86 and SPARC take 373 and 484 bytes respectively. (This is if we assume external classes are also pre-compiled; otherwise, 417 and 540 bytes are needed, respectively.) However, there is quite a bit of symbol table @@ -226,21 +230,21 @@ information is represented - but it does space. (Pre-compiled code also needs space for the reflective data structure, about 520 bytes for this example.) Our tentative conclusion is that the space advantage of bytecodes is minor at best, -whereas the speed penalty is major.

+whereas the speed penalty is major.

Top of Page

2.3 Space for standard run-time

In addition to the space needed for the user code, there is also a +

In addition to the space needed for the user code, there is also a large chunk of fixed code for the run-time environment. This includes code for the standard libraries (such as java.lang), code for loading new classes, the garbage collector, and an interpreter or -just-in-time compiler.

+just-in-time compiler.

In a memory-tight environment, it is desirable to be able to leave +

In a memory-tight environment, it is desirable to be able to leave out some of this support code. For example, if there is no need for loading new classes at run-time, we can leave out the code for reading class files, and interpreting (or JIT-compiling) bytecodes. (If you @@ -248,15 +252,15 @@ have a dynamic loader, you could still d compile them off-line.) Similarly, some embedded applications might not need support for Java's Abstract Windowing Toolkit, or networking, or decryption, while another might need some or all of -these.

+these.

Depending on a conventional (static) linker to select only the code +

Depending on a conventional (static) linker to select only the code that is actually needed does not work, since a Java class can be referenced using a run-time String expression passed to the Class.forName method. If that feature is not used, then a -static linker can be used.

+static linker can be used.

The Java run-time needs to have access to the name and type of +

The Java run-time needs to have access to the name and type of every field and method of every class that is loaded. This is not needed for normal operation (especially not when using precompiled methods); however, a program can examine this information using the @@ -267,9 +271,9 @@ works by looking up fields and methods b JNI thus requires extra run-time memory for field and method information. Since the JNI is also quite slow, an embedded application may prefer to use a lower-level (but less portable) Native -Interface.

+Interface.

Because applications and resources vary so widely, it is important +

Because applications and resources vary so widely, it is important to have a Java VM/run-time than can be easily configured. Some applications may need access to dynamically loaded classes, the Java Native Interface, reflection, and a large class library. Other @@ -279,24 +283,24 @@ different algorithms for garbage collect implementations. This implies the need for a configuration utility, so one can select the features one wants in a VM, much like one might need to configure kernel options before building an operating -system.

+system.

Top of Page

2.4 Garbage collection

Programmers used to traditional malloc/free-style heap management +

Programmers used to traditional malloc/free-style heap management tend to be skeptical about the efficiency of garbage collection. It is true that garbage collection usually takes a significant toll on execution time, and can lead to large unpredictable pauses. However, it is important to remember that is also an issue for manual heap allocations using malloc and free. There are many very poorly written malloc/free implementations in common use, -just as there are inefficient implementations of garbage collection.

+just as there are inefficient implementations of garbage collection.

There are a number of incremental, parallel, or generational +

There are a number of incremental, parallel, or generational garbage collection algorithms that provide performance as good or better than malloc/free. What is difficult, however, is ensuring that pause times are @@ -306,58 +310,58 @@ triggered. The solution is to make sure garbage collection on each allocation. Unfortunately, the only known algorithms that can handle hard-real time either require hardware assistance or are very inefficient. However, "soft" -real-time can be implemented at reasonable cost.

+real-time can be implemented at reasonable cost.

Top of Page

3 Compiling Java

The core tool of Cygnus's Java implementation is the +

The core tool of Cygnus's Java implementation is the compiler. This is jc1, a new GCC front-end (see -Bibliography for GCC). This has similar structure +Bibliography for GCC). This has similar structure as existing front-ends (such as cc1plus, the C++ front-end), and shares most of the code with them. The most unusual aspect of jc1 is that its "parser" reads -either Java source files or Java bytecode files. (The first +either Java source files or Java bytecode files. (The first release will only directly support bytecodes; parsing Java source will be done by invoking Sun's javac. A future version will provide an integrated Java parser, largely for the sake of compilation speed.) In any case, it is important that jc1 can read -bytecodes, for the following reasons:

+bytecodes, for the following reasons:

It is the natural way to get declarations of external classes; in this respect, a Java bytecode file is like a C++ pre-compiled header file.
It is needed so we can support code produced from other tools that produce Java bytecodes (such as the Kawa Scheme-to-Java-bytecode -compiler; see Bibliography for Kawa).

Kawa

Some libraries are (unfortunately) distributed as Java bytecodes without source.

Much of the work of the compiler is the same whether we are +

Much of the work of the compiler is the same whether we are compiling from source code or from byte codes. For example emitting code to handle method invocation is the same either way. When we compile from source, we need a parser, semantic analysis, and error-checking. On the other hand, when parsing from bytecodes, we need to extract higher-level information from the lower-level -bytecodes, which we will discuss in 3.1 -Transforming bytecodes.

+bytecodes, which we will discuss in 3.1 +Transforming bytecodes.

Top of Page

-
3.1 Transforming bytecodes
+
3.1 Transforming bytecodes

This section describes how jc1 works.

The executable content of a bytecode file contains a vector of +

The executable content of a bytecode file contains a vector of bytecode instructions for each (non-native) method. The bytecodes are instructions for an abstract machine with some local variable registers and a stack used for expression evaluation. (The first few @@ -377,9 +381,9 @@ and are enforced by the Java bytecode ve analysis in jc1, which lets us know for every program point, the current stack pointer, and the type of every local variable -and stack slot.

+and stack slot.

Internally GCC uses two main +

Internally GCC uses two main representations: The tree representation is at the level of an abstract syntax tree, and is used to represent high-level (fully-typed) expressions, declarations, and @@ -389,9 +393,9 @@ to represent instructions, instruction p calculations in general. Constant folding is done using trees, but otherwise most optimizations are done at the -rtl level.

+rtl level.

The basic strategy for translating Java stack-oriented bytecodes is +

The basic strategy for translating Java stack-oriented bytecodes is that we create a dummy local variable for each Java local variable or stack slot. These are mapped to GCC "virtual registers," and standard GCC register @@ -399,23 +403,24 @@ allocation later assigns each virtual register to a hard register or a stack location. This makes it easy to map each opcode into tree-nodes or rtl to manipulate the virtual -registers.

+registers.

As an example, consider how to compile iadd, which adds +

As an example, consider how to compile iadd, which adds the top two ints on the stack. For illustration, assume the stack pointer is 3, and virtual registers 50, 51, and 52 are associated with stack slots 0, 1, and 2. The code generated by jc1 is the -following:

+following:

reg51 := vreg51 + vreg52

Note that the stack exists only at compile-time. There is no stack, -stack pointer, or stack operations in the emitted code.

Note that the stack exists only at compile-time. There is no stack, +stack pointer, or stack operations in the emitted code.

This simple model has some problems, compared to conventional -compilers:

This simple model has some problems, compared to conventional +compilers:

We would like to do constant folding, which is done at the tree level. However, tree nodes are typed.
The simple-minded approach uses lots of virtual registers, and the code generated is very bad. Running the optimizer (with the -O flag) fixes the generated code, but you still get a lot -of useless stack slots. It would be nice not to have to +of useless stack slots. It would be nice not to have to run the optimizer, and if you do, not to make unnecessary work for it.

The last problem we solve by using a separate virtual register for +

The last problem we solve by using a separate virtual register for each machine mode. For example, for local variable slot 5 we might use vreg40 when it contains an integer, and vreg41 when it points an object reference. This is safe, because the Java verifier does not allow storing an integer in an object slot and later -reusing the same value as an object reference or float.

+reusing the same value as an object reference or float.

The other two problems we solve by modeling the stack as a stack of +

The other two problems we solve by modeling the stack as a stack of tree nodes, and not storing the results in their "home" virtual registers unless we have to. Thus jc1actually -does the following for iadd:

+does the following for iadd:

- tree arg1 = pop_value (int_type_node); - tree arg2 = pop_value (int_type_node); +
+ tree arg1 = pop_value (int_type_node); + tree arg2 = pop_value (int_type_node); push_value (fold (build (PLUS_EXPR, int_type_node, arg1, -arg2))); +arg2)));
-
The build function is the standard GCC function for +
The build function is the standard GCC function for creating tree-nodes, while fold is the standard function for doing constant folding. The functions pop_value and push_value are specific to jc1, and keep track of which stack location corresponds to which tree node. No -code is actually generated yet.
+code is actually generated yet.
-
This works for straight-line code (i.e. within a basic block). When +
This works for straight-line code (i.e. within a basic block). When we come to a branch or a branch target, we have to flush the stack of tree nodes, and make sure the value of each stack slot gets saved in its "home" virtual register. The stack is usually empty at branches and branch targets, so this does not happen very often. Otherwise, we only emit actual rtl instructions to evaluate the expressions when we get to a side-effecting operation -(such as a store or a method call).
+(such as a store or a method call).
-
Since we only allocate a virtual register when we need one, we are +
Since we only allocate a virtual register when we need one, we are now using fewer virtual registers, which leads to better code. We also get the benefits of GCC constant folding, plus the existing machinery for selecting the right instructions for addition and other -expressions.
+expressions.
-
The result is that we end up generating code using the same +
The result is that we end up generating code using the same mechanisms that the GCC C and C++ front-ends do, and -therefore we can expect similar code quality.
+therefore we can expect similar code quality.
-
Top of Page
+
Top of Page

3.2 Class meta-data

-
Compiling the executable content is only part of the problem. The +
Compiling the executable content is only part of the problem. The Java run-time also needs an extensive data structure that describes each class with its fields and methods. This is the "meta-data" or "reflective" data for the class. The compiler has to somehow make it possible for the run-time system to correctly initialize the data structures before the compiled -classes can be used.
+classes can be used.
-
If we are compiling from bytecodes, the compiler can just pass +
If we are compiling from bytecodes, the compiler can just pass through the meta-data as they are encoded in the class file. (If the run-time supports dynamically loading new classes, it already knows -how to read meta-data from a class file.)
+how to read meta-data from a class file.)
-
It is inconvenient if the meta-data and the compiled code are in +
It is inconvenient if the meta-data and the compiled code are in different files. The run-time should be able to create its representation of the meta-data without having to go beyond its address space. For example reading in the meta-data from an external file may cause consistency problems, and it may not even be possible -for embedded systems.
+for embedded systems.
-
A possible solution is to emit something like:
+
A possible solution is to emit something like:

- static const char FooClassData[] = "\xCa\xfe\xba\xbe..."; - static { - LoadClassFromByteArray(FooClassData); - Patch up method to point to code; + static const char FooClassData[] = "\xCa\xfe\xba\xbe..."; + static { + LoadClassFromByteArray(FooClassData); + Patch up method to point to code;
-
The code marked static is compiled into + +
The code marked static is compiled into a dummy function that is executed at program startup. This can be handled using whatever mechanism is used to execute C++ static initializers. This @@ -521,9 +527,9 @@ dummy function reads the meta-data in ex FooClassData, creates the internal representation, and enters the class into the class table. It then patches up the method -descriptors so that they point to the compiled code.
+descriptors so that they point to the compiled code.
-
This works, but it is rather wasteful in terms of memory and +
This works, but it is rather wasteful in terms of memory and startup time. We need space for both the external representation (in FooClassData) and the internal representation, and we have to spend time creating the latter from the former. It is much better @@ -532,54 +538,54 @@ representation. If we use initialized st compiler statically allocate and initialize the internal data structures. This means the actual initialization needed at run is very little Ð most of it is just entering the meta-data into a global -symbol table.
+symbol table.
-
Consider the following example class:
+
Consider the following example class:

- public class Foo extends Bar { - public int a; - public int f(int j) { return a+j; } - }; + public class Foo extends Bar { + public int a; + public int f(int j) { return a+j; } + };
-
That class compiles into something like the following:
+
That class compiles into something like the following:

-int Foo_f (Foo* this, int j) -{ return this->a + j; } - -struct Method Foo_methods[1] = {{ - /* name: */ "f"; - /* code: */ (Code) &Foo_f; - /* access: */ PUBLIC, - /* etc */ ... -}}; - -struct Class Foo_class = { - /* name: */ "Foo", - /* num_methods: */ 1, - /* methods: */ Foo_methods, - /* super: */ &Bar_class, - /* num_fields: */ 1, - /* etc */ ... -}; - -static { - RegisterClass (&Foo_class, "Foo"); +int Foo_f (Foo* this, int j) +{ return this->a + j; } + +struct Method Foo_methods[1] = {{ + /* name: */ "f"; + /* code: */ (Code) &Foo_f; + /* access: */ PUBLIC, + /* etc */ ... +}}; + +struct Class Foo_class = { + /* name: */ "Foo", + /* num_methods: */ 1, + /* methods: */ Foo_methods, + /* super: */ &Bar_class, + /* num_fields: */ 1, + /* etc */ ... +}; + +static { + RegisterClass (&Foo_class, "Foo"); }
-Thus startup is fast, and does not require any dynamic allocation. +Thus startup is fast, and does not require any dynamic allocation. -Top of Page + Top of Page 3.3 Static references -A class may make references to static + A class may make references to static fields and methods of another class. If we can assume that the other class will also be jc1-compiled, then jc1 can emit a direct reference to the @@ -587,39 +593,39 @@ external static field or me a C++ compiler would). That is, a call to a static method can be compiled as a direct function call. If you want to make a static call from a pre-compiled -class to a known class that you do not know is pre-compiled, +class to a known class that you do not know is pre-compiled, you can try using extra indirection and a "trampoline" stub that jumps to the correct method. (Not that this feature is very useful: it makes little sense to pre-compile a class without also -pre-compiling the other classes that it statically depends on.) +pre-compiling the other classes that it statically depends on.) -A related problem has to do with string constants. The language + A related problem has to do with string constants. The language specification requires that string literals that are equal will compile into the same String object at run-time. This complicates separate compilation, and it makes it difficult to statically allocate the strings (as in done for C), unless you have a truly unusual linker. So, we compile references to string literals as indirect references to a pointer that gets initialized at class initialization -time. +time. -Top of Page + Top of Page 3.4 Linking -The jc1 program creates standard + The jc1 program creates standard assembler files that are processed by the standard unmodified GNU assembler. The resulting object files can be placed in a dynamic or static library, or linked (together with the run-time system) into an executable, using a standard linker. The only linker extension we really need is support for static initializers, but we already have -that (since GCC supports C++). +that (since GCC supports C++). -While we do not need any linker modification, there are some -that may be desirable. Here are some ideas. +While we do not need any linker modification, there are some +that may be desirable. Here are some ideas. -Java needs a lot of names at run-time for classes, + Java needs a lot of names at run-time for classes, files, local variables, methods, type signatures, and source files. These are represented in the constant pool of .class files as @@ -630,20 +636,20 @@ meta-data. Two different class files may same name. It may be desirable to combine them to save space (and to speed up equality tests). That requires linker support. The compiler could place these names in a special section of the object file, and -the linker could then combine identical names. +the linker could then combine identical names. -The run-time maintains a global table of all loaded classes, so it + The run-time maintains a global table of all loaded classes, so it can find a class given its name. When most of the classes are statically compiled and allocated, it is reasonable to pre-compute (the static part of) the global class table. One idea is that the linker could build a perfect hash table of the classes in a library or -program. +program. -Top of Page + Top of Page 4. Run-time -Running a compiled Java program will need a suitable Java run-time + Running a compiled Java program will need a suitable Java run-time environment. This is in principle no different from C++ (which requires an extensive library, as well as support for memory allocation, exceptions, and so on). However, Java requires more @@ -652,9 +658,9 @@ support for threads, garbage collection, and all the primitive Java methods. Full Java support also means being able to dynamically load new bytecoded classes, though this may not be needed in some embedded environments. Basically, the appropriate Java -run-time environment is a Java Virtual Machine. +run-time environment is a Java Virtual Machine. -It is possible to have jc1 produce code + It is possible to have jc1 produce code compatible with the Java Native Interface ABI. Such code could run under any Java VM that implements the JNI. However, the JNI has relatively high overhead, so if you are not concerned about binary @@ -664,65 +670,65 @@ portability, use .class fil plan to support the portable JNI, we will also support such a lower-level ABI. Certainly the standard Java functionality (such as that in java.lang will be compiled to the -lower-level ABI. +lower-level ABI. -A low-level ABI is inherently dependent on a specific VM. We are + A low-level ABI is inherently dependent on a specific VM. We are using Kaffe, a free Java VM, written by Tim Wilkinson -(see Bibliography for Kaffe), with help from +(see Bibliography for Kaffe), with help from volunteers around the "Net." Kaffe uses either a JIT compiler on many common platforms, or a conventional bytecode interpreter, which is quite portable (except for thread support). Using a JIT compiler makes it easy to call between pre-compiled and dynamically loaded methods (since both use the same -calling convention). +calling convention). -We are making many enhancements to make Kaffe a more suitable + We are making many enhancements to make Kaffe a more suitable target for pre-compiled code. One required change is to add a hook so that pre-compiled and pre-allocated classes can be added to the global table of loaded classes. This means implementing the -RegisterClass function. +RegisterClass function. -Other changes are not strictly needed, but are highly + Other changes are not strictly needed, but are highly desirable. The original Kaffe meta-data had some redundant data. Sometimes redundancy can increase run-time efficiency -(e.g., caching, or a method dispatch table for virtual method +(e.g.., caching, or a method dispatch table for virtual method calls). However, the gain has to be balanced against the extra complication and space. Space is especially critical for embedded applications, which is an important target for us. Therefore, we have put some effort into streamlining the Kaffe data structures, such as -replacing linked lists by arrays. +replacing linked lists by arrays. -Top of Page + Top of Page 4.1 Debugging -Our debugging strategy for Java is to enhance + Our debugging strategy for Java is to enhance GDB (the GNU debugger) so it can understand Java-compiled code. This follows from our philosophy of treating Java like other programming languages. This also makes it easier to debug -multi-language applications (C and Java). +multi-language applications (C and Java). -Adding support for Java requires adding a Java expression parser, + Adding support for Java requires adding a Java expression parser, and routines to print values and types in Java syntax. It should be easy to modify the existing C and C++ language support routines, since Java expressions and primitive types are very similar to those of -C++. +C++. -Adding support for Java objects is somewhat more work. Getting, + Adding support for Java objects is somewhat more work. Getting, setting, and printing of Object fields is basically the same as for C++. Printing an Object reference can be done using a format similar to that used by the default toString method the class followed by the address such as java.io.DataInput@cf3408. Sometimes you instead -want to print the contents of the object, rather than its +want to print the contents of the object, rather than its address (or identity). Strings should, by default, be printed using their contents, rather than their address. For other objects, GDB can invoke the toString method to get a printable representation, and print that. However, there should be different -options to get different styles of output. +options to get different styles of output. GDB can evaluate a general user-supplied expression, including a function call. For Java, this @@ -743,7 +749,7 @@ so it can get the information it needs f process. This also makes GDB start-up faster, since it makes it easier to only create type information when needed. -Potentially duplicated meta-data includes the source line + Potentially duplicated meta-data includes the source line numbers. This is because a Java program needs to be able to do a stack trace, even without an external debugger. Ideally, the stack trace should include source line numbers. Therefore, it is best to put the @@ -751,9 +757,9 @@ line numbers in a special read-only sect would be pointed to by the method meta-data, where both GDB and the internal Java stack dumper can get at it. (For embedded systems one would probably leave out line numbers in the run-time, and only keep -it in the debug section of the executable file.) +it in the debug section of the executable file.) -Extracting symbolic information from the process rather than from + Extracting symbolic information from the process rather than from the executable file is also more flexible, since it makes it easier to also support new classes that are loaded in at run-time. While the first releases will concentrate on debugging pre-compiled Java code, @@ -762,30 +768,30 @@ the VM. This problem is eased if the VM since in that case the representation of dynamically-(JIT-)compiled Java code is the same as pre-compiled code. However, we still need to provide hooks so that GDB knows when a new -class is loaded into the VM. +class is loaded into the VM. Long-term, it might be useful to download Java code into GDB itself (so we can extend GDB using Java), but that requires integrating a Java evaluator into GDB. -Top of Page + Top of Page 4.2 Profiling -One problem with Java is the lack of profiling tools. This makes it + One problem with Java is the lack of profiling tools. This makes it difficult to track down the "hot-spots" in an application. Using GCC to compile Java to native code lets us use existing profiling tools, such as gprof, -and the gcov coverage analyzer. +and the gcov coverage analyzer. -Top of Page + Top of Page -5 Status +5 Status -As of early July 1997, jc1 was able to + As of early July 1997, jc1 was able to compile a simple test program, which calculates the Fibonacci numbers (up to 36), both iteratively and recursively (which is slow!), and prints the results. No manual tinkering was needed with the assembly @@ -794,9 +800,9 @@ code generated, which was assembled and run-time engine. On a SparcStation 5 running Solaris2, it took 16 seconds to execute. In comparison, the same program dynamically compiled by Kaffe's JIT takes 26 seconds, and Sun's JDK 1.1 -takes 88 seconds to run the same program. +takes 88 seconds to run the same program. -These numbers are encouraging, but they need some context. Start-up + These numbers are encouraging, but they need some context. Start-up times (for class loading and just-in-time compilation) should in all cases be fairly minor, since the execution time is dominated by recursive calls. In the jc1-compiled case, @@ -809,61 +815,59 @@ be some slight extra speed up if all the simple C-like features that are easy to compile: integer arithmetic, simple control structures, and direct recursion. The results cannot be directly generalized to larger programs that use object-oriented -features more heavily. +features more heavily. -The basic structure of the compiler works, but there is still quite + The basic structure of the compiler works, but there is still quite a bit of work to do. Many of the byte-codes are not supported yet, and neither is exception handling. Only some very simple Java-specific optimizations are implemented. (Of course, jc1 benefits from the existing -language-independent optimizations in GCC.) +language-independent optimizations in GCC.) -The Kaffe VM works on most Java code. We have enhanced it in + The Kaffe VM works on most Java code. We have enhanced it in various ways, and modified the data structures to be simpler and more -suitable for being emitted by the compiler. +suitable for being emitted by the compiler. -The Java support in the GDB debugger is + The Java support in the GDB debugger is partially written, but there is still quite a bit to go before it is user-friendly. We have not started work on our own library -implementation or source-code compiler. +implementation or source-code compiler. -Top of Page + Top of Page - Acknowledgements -This paper is partly based on a document -called "A GCC-based Java Implementation" (Ó 1997, IEEE), which was presented at IEEE Compcon, -Spring 1997. +This paper is partly based on a document +called "A GCC-based Java Implementation" (Ó 1997, IEEE), which was presented at IEEE Compcon, +Spring 1997. Bibliography -[GCC] Using GNU CC -Richard Stallman -Free Software -Foundation, 1996. - -[JavaSpec] The Java Language -Specification -James Gosling, Bill Joy, Guy Steele -Addison-Wesley, 1996. - -[JavaVMSpec] The Java Virtual Machine -Specification -Tim Lindholm, Frank Yellin -Addison-Wesley, 1996. - -[Kaffe] Kaffe, a virtual machine to run -Java code -Tim Wilkinson -http://www.kaffe.org/ - -[Kawa] Kawa, the Java-based Scheme System -Per Bothner -http://www.cygnus.com/~bothner/kawa.html +[GCC] Using GNU CC +Richard Stallman +Free Software +Foundation, 1996. + +[JavaSpec] The Java Language +Specification +James Gosling, Bill Joy, Guy Steele +Addison-Wesley, 1996. + +[JavaVMSpec] The Java Virtual Machine +Specification +Tim Lindholm, Frank Yellin +Addison-Wesley, 1996. + +[Kaffe] Kaffe, a virtual machine to run +Java code +Tim Wilkinson +http://www.kaffe.org/ + +[Kawa] Kawa, the Java-based Scheme System +Per Bothner +http://www.cygnus.com/~bothner/kawa.html -Top of Page + Top of Page - +