International Business Machines Corporation

Mahmut K Ebcioglu

Stanley Everett Schuster

Terry I Chappell

Barbara A Chappell

Manoj N Kumar

David Arnold Luick

Jaime Humberto Moreno

Randall Dean Groves

Philip Braun Winterfield

Eric P Kronstadt

Gabriel Mauricio Silberman

An instruction cache design which converts a sequential instruction stream into a compound format in the instruction cache. The conversion from sequential instructions to compound instructions is performed by an instruction stream interpreter unit (ISU), which is placed between the instruction cache and main memory. The conversion process is performed when an instruction cache miss occurs. Each line in the instruction cache contains a single compound instruction. The format of this compound instruction is transparent to programmers and will vary depending on the number of execution units which are to be supported.

Method and apparatus for dynamic conversion of computer instructions

A method and apparatus for reordering memory operations in superscalar or very long instruction word (VLIW) processors is described, incorporating a mechanism that allows for arbitrary distance between reading from memory and using data loaded out-of-order, and that allows for moving load operations earlier in the execution stream. This mechanism tolerates ambiguous memory references. The mechanism executes only one additional instruction for disambiguation purposes, thus producing good performance, and integrates memory disambiguation with speculative execution of instructions. The overhead introduced is only one instruction, and the load operation can be arbitrarily moved earlier in the instruction stream. The mechanism can cope with conflicts that occur as a result of an unexpected combination of store/load instructions, can be used in a coherent multiprocessor context, and combines speculative execution with reordering of memory operations in a way which requires simple hardware support.

Method and apparatus for reordering memory operations in a superscalar

The invention provides for improved performance of out of sequence load operations. The system has an improved compiler, with an optimizer, an improved CPU with four new instructions in its instruction set, and an address compare unit (ACU). During compilation, the improved compiler identifies load operations that can be move out of sequence ahead of associated store operations and moves those load operations out of sequence and flags them as such. The associated store operations are also flagged. During processor execution of a compiled and optimized program, the address of operands fetched by the out of sequence load operations are saved to the new associative memory. On request, the ACU compares the addresses saved to the addresses generated by the associated store operations. If a comparison results in an identity between the address of a store operation and an address of the out of sequence load operation, a recovery code is run to correct the problem, if not the system continues to execute the program in its compiled order. The system also has the ability to work in a multiprogramming or multitasking environment.

Method and apparatus for improving performance of out of sequence load

A virtual multi-port RAM (VMPRAM) structure has automatic port sequencing and single-port array density and speed. VMPRAM employs input-triggered, self-resetting macros in a pipelined architecture to provide multiple self-timed on-chip cycles during one machine cycle. The VMPRAM incorporates an SRAM segmented into many input triggered, self-resetting, fast cycling blocks. A timing signal is derived from a selected SRAM block for releasing the next select signals and data to the SRAM blocks. The SRAM block inputs are only the data input bus and the decoded signals needed to select a wordline and a bitline pair, and the SRAM block cycle is only the time needed to provide adequate pulse width for word lines and bitlines. Each SRAM block, and all the circuit blocks in the path to access the SRAM blocks, are input-triggered and self-resetting. The multiple address and data input latches are multiplexed at the driver to the true and complement buses to the SRAM segments, and those buses are self-resetting. Similarly, the selected SRAM block reads data out onto a self-resetting bus, and address and data inputs are latched in blocks that are set up for the release signal by the release of the adjacent block, and these blocks are all self-resetting.

Virtual multi-port RAM employing multiple accesses during single machine

Virtual multi-port RAM

Growing community of inventors

Somers, NY, United States of America

Mahmut K Ebcioglu