US 5,740,469 AGrant
Apparatus for Dynamically Reading/Writing Multiple Object File Formats Through Use of Object Code Readers/Writers Interfacing With Generalized Object File Format Interface and Applications Programmers' Interface
Issue Date:1998-04-14
•8 Claims
•5 Drawing Sheets
Abstract
An apparatus for allowing a single software Tool (136) to read and write multiple Object File Formats utilize dynamically configurable and loadable Object File Readers (131) and Writers (139). A separate Reader (131) and Writer (139) can be implemented for each different Object File Format and variations thereof. Tools (136) communicate with the Readers (131) and Writers (139) using a Generalized Object File Program Interface (124). This Interface (124) utilizes Data Structures implementing a Generalized Object File Internal Representation (122) and an Applications Programmers Interface (120).
Metadata
Assignee
- Motorola Inc.
Inventors
- Weiping Yin
- Paul Hancock
- Alan Weiner
Application Information
Application Number:US 4268628
Filing Date:1995-04-24
Priority Date:1995-04-24
Art Unit:272
Classifications
IPC:
G06F 500
Field of Search:
395382700;275;600;500;601;650;885;704;442;5201
Patent Drawings (5 sheets)
Description
FIELD OF THE INVENTION
The present invention generally relates to a data processing interface, and more specifically to dynamically reading and writing different object file formats.
BACKGROUND OF THE INVENTION
There is often a need for compatibility between tools from different vendors. One vendor may manufacture a superior compiler, but its debugger may not be as good as that from some other vendor. Likewise, another vendor may excel at debuggers. It would be helpful for a purchaser of software tools to be able to mix and match tools, instead of being locked into a specific vendors line of tools.
A number of approaches have been tried to achieve this compatibility. One approach has been to translate source code files. Unfortunately, this translation is rarely 100% effective. Also, it is only practical for one-time translations.
Another approach has been to directly read multiple object file formats. This has not been successful for a number of reasons. First, many vendors keep their object fie formats confidential. Secondly, adding or modifying formats requires relinking of the tool. Similar problems arise when object files are translated from one format to another.
Another approach has been to require conformance to a standardized object code format. This has not worked well for a number of reasons. First, it is difficult to impose standards upon established markets. Secondly, standards are invariably written to the lowest common denominator. This results in features lost in vendor product lines.
Several vendors, such as Intermetrics, have developed static converters for converting from their proprietary object code format to other object code formats.
SUMMARY OF THE INVENTION
In accordance with the invention, an apparatus is provided for allowing a single software Tool to read and write multiple Object File Formats utilizing dynamically configurable and loadable Object File Readers and Writers. A separate Reader and Writer can be implemented for each different Object File Format and variations thereof. Tools communicate with the Readers and Writers using a Generalized Object File Program Interface. This Interface utilizes Data Structures implementing a Generalized Object File Internal Representation and an Applications Programmers Interface (API).
These and other features, and advantages, will be more clearly understood from the following detailed description taken in conjunction with the accompanying drawings. It is important to point out that there may be other embodiments of the present invention which are not specifically illustrated.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a block diagram showing the hardware components of the present invention;
FIG. 2 is a general system overview that illustrates the problem solved with the present invention and the Prior Art solutions thereof;
FIG. 3 is a block diagram showing how the present invention differs from the Prior Art as shown in FIG. 2;
FIG. 4 is a block diagram showing how to read and write different Object File Formats in accordance with this invention; and
FIG. 5 is a block diagram showing how multiple Object Files can be combined using a single Tool in accordance with this invention.
DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT
Building a program that can be loaded into a computer memory and executed by a computer processor is a process known and utilized by all computer programmers. High level languages, such as C, C++, FORTRAN, COBOL, PL/1, Basic, Pascal, etc. are encoded by programmers in source files. These files contain instructions to the appropriate compilers, which differ from language to language, and indeed from language derivative to derivative. Source code files are read by compilers, which generate either assembler source code or Relocation Level Object Files. Assemblers are programs that translate symbolic encodings of machine level code into Relocation Level Object Files. Relocation Level Object Files are combined by linkers into Execution Level Object Files. In may cases, these Execution Level Object Files are directly loadable and executable by computer processors. In other cases, further transformations must be performed on the Execution Level Object Files before they are ready for execution.
There are currently dozens, if not hundred, of Object File Formats in use by development tools vendors on many platforms. For example, developers employed by the original assignee of this invention, Motorola, Inc., utilize at a minimum three major families of object file formats: COFF (Common Object File Format, developed by AT&T), ELF (Executable and Linkable Format, developed by the Tools Interface Standards Committee), and IEEE695 (an IEEE standard originally developed by Hewlett Packard and Microtec Research, Inc.) Most of these families contain a number of derivatives. Many proprietary Object File Formats belonging to these families are currently in use.
Finding a way to make Tools using different Object File Formats has presented a major obstacle for Tool vendors. Previous solutions to this problem have involved providing a plethora of translators from one Object File Format to the next, or creating universal standards which all vendors must follow. Both of these solutions involve considerable effort, and have not lived up to the expectations of their proponents.
This invention addresses the problem by providing a mechanism for making Tools using various Object File Formats work together at a level never before achieved. Each vendor of a Tool or proprietary Object File Format need only provide a single Reader and a single Writer for its Object File Format. All Tools using this invention could then read and write those object files. Note that though reference is made here to a single Reader and a single Writer, it is envisioned that some vendors may elect to combine both functions into a single Reader/Writer.
Many vendors regard their Object File Formats as trade secrets which should never be disclosed. These often offer what that vendor considers to be enhancements to a more standardized Object File Format, such as COFF, ELF, or IEEE695. This invention allows vendors to maintain their trade secrets, while making the vendors Tools interoperable with Tools from other vendors.
FIG. 1 is a block diagram showing the hardware components of this invention. A general purpose computer 10 contains a computer Processor ("CPU") 12 usually connected via a Bus 16 to a Memory 14. Also connected to the Bus 16 are one or more Disk drives 20, a CRT monitor 22, a keyboard 24, and a printer 26.
FIG. 2 is a general system overview that illustrates the problem solved with this invention and the prior art solutions thereof. A.C 30 and B.C 32 are files located on disk 20 containing C language code. While there is an ANSI standard for C language compilers, most compiler vendors implement non-ANSI standard differences in their compilers. These differences are usually of two different types: either limitations, or extensions. Limitations often come about because a vendor does not think implementation of some obscure features makes economic sense. Extensions on the other hand are additions to a language. They can be features a vendor thinks helpful. Note that the usage of the C language herein is by example only. Slightly different source code language specifications have plagued compiler writers since early FORTRAN and COBOL, and continue with C++.
The A.C 30 source is compiled with C compiler A 31 into assembly source code A.S 34. Likewise, B.C 32 source code is compiled with C compiler B 33 into assembly source code B.S 36. Often, the A.C 30 source code will not be completely compilable with the B compiler 33, and the B.C 32 source code will not be completely compilable with the A compiler 31. Note that many modern compilers do not actually generate symbolic assembler code, but rather utilize for efficiency an internal assembler format.
Four different assembly language source files are shown, A.S 34, B.S 36, C.S 38, and D.S 40. Assembly language formats are even more likely to vary than higher languages because they are less likely to be subject to some international standard. Though these assembler source file formats often generate code for the same physical architectures, they do it differently. Instruction layout may differ. More importantly, the assembler pseudo-operations used to direct the assemblers vary widely.
The A.S 34 assembler source code is assembled by the A Assembler 35, while the B.S 36 assembler source is assembled by the B assembler 37, the C.S 38 assembler source is assembled by the C assembler 39, and the D.S 40 assembler source is assembled by the D assembler 41. As with compilers, assembler source code is often not compatible across different assemblers. For example, the B assembler 37 will often be unable to assemble C.S 38 assembler code created to be assembled with the C assembler 39.
Each of the four assemblers shown above will probably generate differently formatted Relocation Level Object Files. For example, assembler A 35 will produce Relocation Level Object File A.O 42, assembler B 37 will produce Relocation Level Object File B.O 44, assembler C 39 will produce Relocation Level Object File C.O 46, and assembler D 41 will produce Relocation Level Object File D.O 48.
Relocation Level Object Files are combined by a linker to form Execution Level Object Files. In many computer architectures, these Execution Level Object Files may be directly loaded into Memory and executed. Usually it is necessary to link Relocation Level Object Files created with one vendor's assembler or compilers with that same vendor's linker. Thus, Relocation Level Object Files A.O 42 is linked with the A linker 43 into Execution Level Object File A.OUT 50, while Relocation Level Object File B.O 44 is linked with the B linker 45 into Execution Level Object File B.OUT 52, Relocation Level Object File C.O 46 is linked with the C linker 47 into Execution Level Object File C.OUT 54, and Relocation Level Object File D.O 48 is linked with the D linker 49 into Execution Level Object File D.OUT 56.
As with Relocation Level Object File, often Execution Level Object Files formats differ from one vendor to another, and from one linker to another. This is especially true, where the Execution Level Object Files are not directly executable, but rather must be further translated before usage. One example of this is the use of an S-Record Generator 84 (shown in FIG. 3) to generate loadable S-Records (App. HEX) 86. In such a case, each linker vendor would usually create its own S-Record Generator 84, recognizing its own Execution Level Object File Format.
FIG. 3 is a block diagram showing how this invention differs from the prior art as shown in FIG. 2. Again, we have two different source code formats compiled by their respective compilers: C source ABC.C 60 is compiled with ABC compiler 61 which generates ABC.S 66 assembler source, while C source code DEF.C 62 is compiled with DEF compiler 63 which generates DEF.S assembler source code 68. Likewise, assembler ABC 67 assembles assembler source format ABC.S 66 into Relocation Level Object File Format ABC.O 74, while assembler DEF 69 assembles assembler source format DEF.S 68 into Relocation Level Object File Format DEF.O 76 and assembler XYZ 71 assembles assembler source format XYZ.S 70 into Relocation Level Object File Format XYZ.O 78.
This is the point at which this invention becomes readily distinguishable from the prior art. All three Relocation Level Object File Formats ABC.O 74, DEF.O 76, and XYZ.O 78 can be linked using the same linker 80 to produce an Execution Level Object File APP.OUT 82. Additionally, should different linkers 80 be used (as was shown in FIG. 2), and different Execution Level Object File Format files are generated, a common S-Record Generator 84 can be used to translate the different Executable Level Object Files APP.OUT 82 into S-Records APP.HEX 86.
FIG. 4 is a block diagram showing how this invention is able to read and write different Object File Formats. Three different Object File Formats are shown: COFF 100, ELF 102, and IEEE695 104. COFF 100 has several COFF derivatives 106, and ELF 102 has several ELF derivatives 108. For each COFF derivative 106, ELF derivative 108, and for IEEE695 104, a Reader/Writer 110 is written. The Reader portion of the Reader/Writer 110 is used to read the Object File, while the Writer portion of the Reader/Writer 110 is used to write an Object File in the specified Object File Format.
On the other side, each Tool 112, 114, 116, 118, would have a Generalized Object File Program Interface 124. The Generalized Object File Program Interface 124 comprises essentially two parts: an Application Programmers' Interface ("API") 120, and an Generalized Object File Internal Representation 122.
Some of the Tools 112, 114, 116, 118 that can benefit from this Generalized Object File Program Interface 124 are compilers 64, assemblers 72, linkers 80, loaders (not shown), debuggers (not shown), simulators (not shown), emulators (not shown), disassemblers (not shown), and S-Record Generators 84 (see FIG. 3). Compilers 64 and assemblers 71 can utilize Writers to write Relocation Level Object Files. Linkers 80 and debuggers can utilize Readers to read Relocation Level Object Files. Linkers 80 can utilize Writers to write Execution Level Object Files. Linkers 80, loaders, debuggers, simulators, emulators, disassemblers, and S-Record Generators 84 (see FIG. 3) can utilize Readers to read Execution Level Object Files. Note that the above Reader and Writer users are suggestive only. The Generalized Object File Program Interface 124 is useful to any tools that either read or write object files.
FIG. 5 is a block diagram showing how multiple Object Files can be combined using a single Tool. Input File #1 130 has a corresponding Reader #1 131, while Input File #2 132 has corresponding Reader #2 133, and Input File #3 134 has corresponding Reader #3 135. Each of these Object Files is located on a Disk Drive 10 (see FIG. 1) and may be a different Object File Format. For example, Input File #1 130 may be encoded with a COFF derivative 106, Input File #2 131 with an ELF derivative 108, and Input File #3 134 with IEEE695 104. In that case, Reader #1 131 would read that COFF derivative 106, Reader #2 133 would read that ELF derivative 108, and Reader #3 135 would read IEEE695 104 formatted Object Files.
A single Tool 136, such as a linker, could thus read each of the Object File Formats. Indeed, as shown here, these different Object File Formats can be mixed. One example is that relocatable Object Files from assemblers and compilers from different vendors can be combined into a single executable object file. Vendors can maintain secrecy as to their proprietary Object Code Format even while delivering Reader/Writer black boxes to their customers.
The Tool 136 reads each of the input files 130, 132, 134 through the corresponding Readers 131, 133, 135. It can then write an Output File 138 of a specified Object File Format through use of the appropriate Writer 139. The Tool 136 communicates with the Readers 131, 133, 135, and the Writer 139 through the Generalized Object File Program Interface 124. The Generalized Object File Program Interface 124 has essentially two parts: an Application Programmers' Interface ("API") 120, and a Generalized Object File Internal Representation 122.
Preferably, Readers and Writers for the various Object Code Formats are written as Dynamically Loadable Modules ("DLM"s). For example, under Microsoft Windows, Dynamic Linkable Libraries ("DLL"s) may be used to implement Readers and Writers. There are usually separate DLM Reader and Writer modules for each Object File Format, and derivatives thereof. In the preferred embodiment, each of the Tools 136, Readers 131, and Writers 139, when activated, is located in a separate partition, segment, or address space in Memory 14.
Included in each Tool 136 is an Application Programmers' Interface ("API") 120, where selection of which Readers and Writers can be dynamically performed based on Object File Format. The API combined with the use of DLMs allows addition and modification of Object File Readers 131 and Writers 139 without the necessity of relinking Tools 136. This flexibility can be extended by using a configuration file (not shown) stored on Disk 20 (see FIG. 1) to dynamically select which DLM Reader and Writer modules to load and access for which Object File Formats.
The Tools 136 and the Readers 131 and Writers 139 communicate through sharing an Generalized Object File Internal Representation 122. This Generalized Object File Internal Representation 122 primarily comprises Data Structures stored in Memory 14 (see FIG. 1). In the preferred embodiment, an object-oriented language such as C++ is used to specify classes and subclasses of objects. These objects include Data Structures used to pass information between the Tools 136 and the Readers 131 and Writers 139. The Generalized Object File Internal Representation 122 is a generalized or virtualized Object File Format. It usually is a super-set of each of the Object File Formats, forming a union of the capabilities available in the various Object File Formats supported by the Generalized Object File Internal Representation 122. A partial example of a C++ implementation supporting COFF, ELF, and IEEE695 is included herein as an Appendix.
A Tool 136 communicates with Readers 131 and Writers 139 by first inserting commands and data into Generalized Object File Internal Representation 122 Data Structures stored in Memory 14. One of the Readers 131 or Writers 139 is selected based on Object File Format. The API is then invoked to transfer the Data Structures from the Tool 136 to the selected Reader 131 or Writer 139. The selected Reader 131 or Writer 139 executes the command, reading or writing (as appropriate) from/to the appropriate Object File. The selected Reader 131 or Writer 139 will then insert a result and possibly data into a Data Structure, which is returned through the API to the Tool 136. This is repeated as necessary.
This invention is applicable in all known general purpose computer environments. However, the capability for Dynamically Linking modules at execution time is helpful, since it obviates the need to relink Tools when Readers or Writers are modified or added. Also note that though the Appendix shows a partial example using an object-oriented language such as C++, this invention is also usable with less object-oriented languages such as the C language.
Those skilled in the art will recognize that modifications and variations can be made without departing from the spirit of the invention. Therefore, it is intended that this invention encompass all such variations and modifications as fall within the scope of the appended claims.
THE DESIGN OF THE OBJECT FILE READERS AND WRITERS
The following paper describes the API of the Dynamically Loadable Object File Readers and Writers. It also describes how to construct an object file reader or writer which will work with this API.
The Object File Readers and writers were written using object oriented design paradigms. All object file readers and writers are based upon the same objects which defines the interface used to call them. The base objects also handle all file I/O operations for the derived objects.
The Object File Readers have been implemented in C++ and make use of C++'s classes and inheritance features. However, the importance of C++ is minimal, the Object File Readers could have been written in C, Smalltalk, or any of a wide variety of other languages.
1. The Design of the Data Structures Used By The Readers, Writers, And Applications
The data structures used by the object file readers and writers form half of the API. The other half consists of the function calls used to access the readers and writers. The data structures themselves were designed by taking the union of the structures found in three common object file formats: COFF (Common Object File Format), ELF (Executable Linkable Format), and IEEE695. Since the structures are a union of these object file format structures, they can be used to read all of these object file formats without loss of information. Other object file formats which contain similar structures can also use these data structures, however, they may experience some data loss in the process. All data within these structures must be accessed using function calls. This enables the storage mechanisms to be hidden and allows optimizations in some cases.
The structures are: FileHeader, Section, Reloc, Line, and Symbol.
The FileHeader structure contains the information taken from the file header (and auxiliary tie header) in the object file. Most of the information in the file header is contained as data in the reader or writer. Thus, it is most useful for diagnostic purposes since it only contains information about the object file itself and this is precisely the information which is being abstracted. The Section structure contains information taken from the section headers. It defines information about individual sections.
The Reloc structure contains information about relocation entries.
The Line structure contains information about line number entries. These are used for debugging and are only found in COFF and IEEE695 files. ELF files use DWARF debugging format, which is read and written as section data.
The Symbol structure contains information about symbols (and auxiliary symbols for COFF) in the object files. Symbols contain a lot of format specific information which must be abstracted. For instance, COFF file symbols depend upon the order of the symbols to give added meaning. This ordering information must be abstracted so that it is file format independent.
The raw section data is written out as arrays of unsigned characters.
2. The Design of the Object File Readers
The readers are based upon a generalized base class, InObj, which provides a framework and defines the interface through which clients call the readers. All format specific information must be handled through classes derived from class InObj. For instance, relocRead is declared in class InObj, but it has a trivial function body "{}." In the derived class InCOFF, relocRead will find a specific relocation entry in a COFF file and read it. The same function could be overloaded in class InIEEE to read a relocation entry from an IEEE695 file. These functions must handle all file format specific information. They must be able to find the correct location in the file for each piece of information.
Class InObj also handles all file i/o for the derived classes. It contains an object of type InEndianFile which will overload all file i/o functions to do all the byte swapping necessary to compensate for differences in endianness within the file and the host processor. Class InEndianFile overloads operator >> for several types, so that when the derived classes use these operators they will get a consistent byte ordering.
Each object file reader must have a companion function which can open a file and determine whether it is an object file of a format which that reader can recognize. If the file can be read by the reader, this function should be able to instantiate a reader for the object file and return a pointer to the reader.
With this approach, a reader may be compiled into a library and loaded at run time. Such libraries could be linked into any application with knowledge of the reader interface. Such libraries need not be known to the application when it is written, but can be linked in at a later time.
For example: suppose that one tools vendor were to write a linker with this technology. Suppose this linker runs under Microsoft Windows. Then the readers for the object files which it uses could be written as Windows DLLs. Now suppose that another vendor writes a DLL which reads its own object files. The second vendor could ship its DLL to the customers of the first vendor, and the first vendor's linker would be able to link the second vendor' s object files.
Here is how it could work. First, the first vendor would have to have a means of locating the new DLL. Next, the linker could load the DLL, and call the identification function, inObjectCreate. If the call succeeds, the DLL will create an object file reader for the second vendor's object file. Since all data structures are normalized, the second vendor's reader would be virtually indistinguishable from the first vendor's reader in most important respects. Thus, the files could be linked.
Note that although this example deals with Microsoft Windows, this technology is not limited to Windows. Other operating systems may require other mechanisms, but the same results could be obtained. For instance, interprocess communications could be used under UNIX instead of DLLs.
CLASS HIERARCHY OF THE OBJECT FILE READERS ##STR1## See iobj.hxx, iendfile.hxx, icreate.hxx, icoff.hxx, ielf.hxx, and iieee.hxx for prototypes of the functions used.
3. The Design of the Object File Writers
The design of the object file writers is very similar to that of the readers. Class OutObj is a base class from which all of the writers are derived. Class OutObj contains an object of type OutEndianFile which handles file i/o and endian byte swapping. However, the user of an object file writer is assumed to know what kind of object file is to be written, so there is no need for a function like inObjectCreate.
The object file writers also have some additional constraints which do not apply to the readers. The writers are responsible for calculating any file pointers within an object file. So, the writers must have information about the contents of the file before the actual data is written. For instance, the section headers contain pointers to the section data and relocation entries within the file. Thus, the writers must know how many sections there are in the file, the size of the contents of each section, the number of relocation entries, and the number of line number entries for each section before they can calculate the offsets. The object file writers ensure that they have this information by requiring that the section headers be written before any section data, relocation entries, line number entries, or symbols are written.
In most other respects, the writers are very similar in design to the readers. ##SPC1##
Claims
We claim:
1. An apparatus allowing a software Tool to read a plurality of Object File Formats comprising: a plurality of Object Code Readers, wherein: each of the plurality of Object Code Readers reads data formatted in at least one of a plurality of Object File Formats, and at least two of said plurality of Object Code Readers read data formatted in different ones of the plurality of Object File Formats; one or more Data Structures storable in a Memory, wherein: said one or more Data Structures provide a Generalized Object File Format interface; and an Applications Programmers' Interface (API) providing selectable communications allowing data stored in the one or more Data Structures to be transferred between the Tool and one or more of said plurality of Object Code Readers.
2. The apparatus claimed in claim 1 which additionally comprises: one or more Object Code Writers, wherein: each of the one or more Object Code Writers writes data formatted in an Object File Format.
3. The apparatus claimed in claim 2 wherein within the element of one or more Object Code Writers: at least two of said Object Code Writers write data formatted in different Object File Formats.
4. The apparatus claimed in claim 1 which additionally comprises: a Memory, wherein: data representing Object Files is stored in the Data Structures in the Memory.
5. The apparatus claimed in claim 1 wherein within the element of the plurality of Object Code Readers: the Object Code Readers are Dynamically Loadable Modules.
6. The apparatus claimed in claim 1 wherein within the element of the plurality of Object Code Readers: the Object Code Readers are Dynamically Linkable Library (DLL) modules.
7. The apparatus claimed in claim 1 wherein within the element of an Application Programmers' Interface (API): the selection of which Object Code Reader to communicate with depends on which Object Code Format is encountered.
8. An apparatus allowing a software Tool to read and write a plurality of Object File Formats comprising: one or more Object Code Readers, wherein: each of the one or more Object Code Readers reads data formatted in at least one of the plurality of Object File Formats, and each of the one or more Object Code Readers is a Dynamically Loadable Module; one or more Object Code Writers, wherein: each of the one or more Object Code Writers writes data formatted in at least one of the plurality of Object File Formats, and each of the one or more Object Code Writers is a Dynamically Loadable Module; one or more Data Structures storable in a Memory which provide a Generalized Object File Format interface; and an Applications Programmers' Interface (API) providing selectable communications allowing data stored in the one or more Data Structures to be transferred between the Tool and one or more of the Object Code Readers, wherein: the selection of which Object Code Reader to communicate with and which Object Code Writer to communicate with depends on which Object Code Format is encountered.
Patent Citations (16)
| Patent | Date | Inventor | Cited By |
|---|---|---|---|
| US4751740 | 1988-06-01 | Wright | |
| US5097533 | 1992-03-01 | Burger et al. | |
| US5119465 | 1992-06-01 | Jack et al. | |
| US5280613 | 1994-01-01 | Chan et al. | |
| US5313635 | 1994-05-01 | Ishizuka et al. | |
| US5339419 | 1994-08-01 | Chan et al. | |
| US5339431 | 1994-08-01 | Rupp et al. | |
| US5339433 | 1994-08-01 | Frid-Nielsen | |
| US5363487 | 1994-11-01 | Willman et al. | |
| US5371885 | 1994-12-01 | Letwin | |
| US5421001 | 1995-05-01 | Methe | |
| US5421015 | 1995-05-01 | Khoyi et al. | |
| US5561800 | 1996-10-01 | Sabatella | |
| US5574903 | 1996-11-01 | Szymanski et al. | |
| US5636357 | 1997-06-01 | Weiner | |
| EPX96106279 | 1996-12-01 |
Non-Patent Literature (2)
- Allen, et al.; "Program Loading inOSF/1"; Proceedings of the Winter 1991 USENIX Conference; pp. 145-160 (1991).
- Kuo, et al: "A Retargetable Compiler for the Generation of Control Routines of Microprocessor-Based Digital Systems"; pp. 469-476, (1990).