Processadores Bipartidos Aspectos de Projeto

Felipe M. G. França, Edil S. T. Fernandes1987 · DOI

Abstract: As motivações e contextos do aparecimento de processadores bipartidos devem ser esclarecidos. A detecção de paralelismo potencial ao nível convencional de instruções durante a execução de uma tarefa sequencial permite que se dobre o alcance máximo do desempenho. RISC e VLSI caracterizam o meio-ambiente de dois exemplos que serão vistos aqui.

Introduction

In the late 1980s, computer architecture faced two significant hurdles: the von Neumann bottleneck, which limited throughput between the CPU and memory, and the Flynn bottleneck, which restricted traditional SISD (Single Instruction Stream, Single Data Stream) architectures to executing only one instruction per clock cycle.

This paper, originally published in 1987 by researchers at COPPE/UFRJ, explores the design of "bipartite" processors. These architectures split the traditional CPU into specialized units to exploit parallelism within sequential tasks, effectively bridging the "semantic gap" between high-level languages and machine code.

Bipartite Architectures: PIPE and FILA&PILHA

The core philosophy of a bipartite processor is the decoupling of memory access from instruction execution. By allowing a dedicated unit to calculate addresses and fetch data independently, the system can hide memory latency and increase instruction throughput.

General diagram of a bipartite processor with dedicated Access and Execution units.

The PIPE Processor

Developed at the University of Wisconsin-Madison, the PIPE architecture utilizes two physically identical processors configured to operate in "Access" and "Execution" modes. The Access processor runs ahead of the Execution processor, fetching data into hardware queues (FIFOs). This smoothing of timing differences minimizes memory conflicts and allows for a 50% performance gain in scalar processing compared to supercomputers like the CRAY-1.

The FILA&PILHA Machine

A domestic project from COPPE/UFRJ, the FILA&PILHA machine also employs two identical units but adopts a stack-based approach. It separates code and data into distinct memory areas to facilitate simultaneous access. The S/CE unit manages sequencing and address calculation, while the E/A unit handles data access and execution. This design aligns with RISC principles by using implicit operands, which helps mitigate the code-bloat typically associated with early RISC architectures.

Evaluation Methodology and Results

To evaluate these architectures, the researchers employed the Wortman method, calculating the global cost of "semantic primitives" (high-level language fragments). The systems were tested against standard benchmarks of the era: Acker (recursion), Puzzle (computation), Search (branching), and Sieve (array manipulation).

Performance comparison across benchmark programs for various architectures including FILA&PILHA (Fi&Pi).

As shown in the comparative data, the FILA&PILHA (Fi&Pi) architecture demonstrated competitive performance against established hardware like the Intel 8086/80286 and the Motorola 68000. While the static code size was larger than CISC counterparts, the use of a "Local Stack" acting as a data cache significantly reduced main memory references, proving the efficiency of the bipartite organization in cushioning communication costs.

Conclusion

Bipartite processors represent a strategic approach to overcoming the fundamental limits of sequential processing. By decoupling access and execution, these designs can virtually double potential performance. For the Brazilian context of the 1980s, such RISC-influenced bipartite designs offered a viable path toward technological independence, as they were more feasible to implement with limited VLSI resources than complex CISC processors.

Summary updated

Abstract

As motivações e contextos do aparecimento de processadores bipartidos devem ser esclarecidos. A detecção de paralelismo potencial ao nível convencional de instruções durante a execução de uma tarefa sequencial permite que se dobre o alcance máximo do desempenho. RISC e VLSI caracterizam o meio-ambiente de dois exemplos que serão vistos aqui.
BibTeX
@misc{franca1987processadores,
  author = {Felipe M. G. França and Edil S. T. Fernandes},
  title = {{Processadores Bipartidos Aspectos de Projeto}},
  year = {1987},
  doi = {10.5753/sbac-pad.1987.23577},
}
Figure