Connectix Corporation

Chad Perry Walters

Jorg Anthony Brown

Jonathan Forrest Garber

In a computer system, a cross-compiler converts non-native code into native code immediately prior to execution of that code. The system also includes a code cache for storing cross-compiled code and a hash table for locating code blocks in the code cache. In a preferred embodiment, the system also includes an interpreter for emulating certain non-native instructions that are not converted into native code by the cross-compiler. While executing any non-native application, if the next instruction is not one of the predefined set of non-native instructions to be handled by interpretation or a special purpose procedure, then the next instruction is considered to be an 'entry point' instruction, and the cross-compiler looks up the address of the entry point instruction in the hash table to see if a corresponding native code block is already stored in the code cache. If so, the native code block in the code cache is executed until an exit instruction in the native code block is encountered. Otherwise, the cross-compiler cross-compiles all code that is reachable from the entry point instruction during execution of the program without going outside the compilation window. During compilation the cross-compiler determines the non-native condition codes generated by a non-native instruction that will not be used by any successors of the non-native instruction. The native code instructions generated by the cross-compiler do not include instructions for processing non-native condition codes generated by the non-native instruction that will not be used by any successors of the qualifying non-native instruction.

Dynamic cross-compilation system and method

A virtual memory system and method enable a computer system to use a virtual memory address space larger than the size of physical primary memory while swapping few, if any, pages out to secondary memory. Primary memory is divided into a work space, used for storing uncompressed pages in current use, and a 'Compression Heap'. A MappedOut storage space, which includes the Compression Heap and a portion of secondary memory, is used to store all pages swapped out of the work space. A virtual memory manager dynamically determines the number of pages of primary memory which need to be included in the work space, and moves pages of primary memory into the work space from the Compression Heap as needed. Pages are selected to be swapped out of the work space to the MappedOut storage space on the basis of memory usage data. A data compressor compresses at least some virtual memory pages prior to their storage in the MappedOut storage space, and a data decompressor decompresses the compressed virtual memory pages when a fault occurs on the those pages. The virtual memory manager includes logic for preferentially compressing and storing swapped out virtual memory pages in the Compression Heap so long as the MappedOut storage page includes a sufficient number of pages of primary memory to store those virtual memory pages, and for storing ones of the swapped out virtual memory pages in secondary memory when insufficient space is available in the Compression Heap.

Virtual memory management system and method using data compression

A lossless data compression system and method compresses a set of M data words stored in a computer memory. A first table stores data representing last occurrence positions among those of the M data words already processed for all distinct word values. A second table stores for each data word an entry indicating the position, if any, of a most recent prior occurrence of another data word with the same word value. A dictionary index indicates how many distinct data word values have been encountered during processing of M data words. The M data words are processed in in said M data words previously processed, identifying a longest sequence of data words starting with said each data word referencing said second table to identify data words earlier in said M data words than said each data word that match said each data word's value, identifying a longest sequence of data words earlier in said M data words than said data word which match an equal number of data words starting with said each data word, (B1) when said longest sequence's length is greater than one data word, outputting a run length code that indicates said longest sequence's length and position, and updating said first and second tables for the equal number of data words starting said each data word, and (B2) when said longest sequence's length is equal to one data word, outputting a code that indicates the position of said identified longest sequence, and updating said first and second tables for said each data word.

Growing community of inventors

Palo Alto, CA, United States of America

Chad Perry Walters