82596 Co-Processor Performance - Intel Embedded Intel486 Hardware Reference Manual

Embedded intel486 processor
Table of Contents

Advertisement

The CA, 596RESET, and PORT# signals are generated according to the equations shown in
ure
7-26. The M/IO# and D/C# signals are also generated by the glue logic. When both HOLD
and HLDA are asserted, indicating that the coprocessor has requested and been granted the bus,
M/IO# and D/C# must be driven high.
HLDA
ADDRESS
ADS#
D0
RESET
D/C#, M/IO#
CLK
CA = (CA Address Decode) & ADS# & D0
596RESET = (596RESET Address Decode) & D0 I RESET
PORT# = (Port Access Address Decode) & ADS#
Figure 7-26. 596RESET, CA, and PORT# Equations
Caching of the coprocessor memory structures in the Intel486 processor internal cache may be
disadvantageous, because these memory structures are not directly executable by the processor.
Typically, most coprocessor bus activity consists of receiving and transmitting frames, managing
the receive frame area, and prefetching descriptor pointers. The system control block is typically
accessed only once by the processor for every update of this area made by the coprocessor. The
processor gains no advantage from caching locations which are used only once. Also, each time
a cached memory location is written to by the coprocessor, a cache invalidation cycle must be
performed.
For systems in which caching is obligatory, external logic must monitor ADS# and W/R# and
drive the EADS# cache invalidation input to the processor.
7.6.1.8

82596 Co-processor Performance

With a 25-MHz clock, the 82596 coprocessor can transfer data at up to 80 Mbyte/second in burst
cycles, or 50 Mbytes/second in non-burst cycles. With a 33-MHz clock, the rates are
106 Mbytes/second for burst and 66 Mbytes/second for non-burst. Most transfers in a Intel486
85C220
PLD
596RESET
PORT#
PERIPHERAL SUBSYSTEM
HOLD
CA
82596 LAN
Controller
Registered
Registered
Combinatorial
Fig-
7-49

Advertisement

Table of Contents
loading

Table of Contents