|
|||
Encore Computer 32/8705, 32/8750, 32/8780
|
General
The CONCEPT 32/87 product line is ideal for engineering, scientific
and industrial
System Architecture The CONCEPT 32/87 computer is built around a high-speed synchronous bus called the SeIBUS that has a transfer rate of 26.67 million bytes per second. The CPU, PU, main memory and the input/output subsystem reside on the SeIBUS as shown in Figure 1. CPU/IPU The CPU and IPU are implemented with the MC 10,000 series Emitter Coupled Logic (ECL) which provides extremely fast gate switching times. Functional units within the CPU and IPU are connected by internal busses which have 75-nanosecond bus cycle time. During each 75-nanosecond cycle, four macroinstructions are in various phases of execution. This pipelined method of operation increases the effective throughput of the CPU and IPU and, hence, the system Interrupts/Traps Assigned to devices external to the CPU, are 96 interrupt levels that report real-time events to the software. The notification of these events are prioritized, scheduled and in some cases, deferrable. Interrupts must contend for recognition by the CPU as only the highest priority interrupt is recognized and executed by the CPU. Interrupt levels are assigned to individual SeIBUS devices (i.e I/0 channels, Real-Time Option Modules (RTOM) and the Input/Output Processor (lOP). The real-time clock and the interval timer interrupts use one level each Users requiring more external interrupt lines can add one or more RTOMs. Each provides 16 external interrupt levels. Traps are error conditions that are identified and reported by the CPU/IPU. All traps have the same priority with the exception of the power fail trap, which overrides all other traps and interrupts Input/Output Processor (lOP) The lOP, a powerful multiplexing channel for I/O operations, resides on the SeIBUS and operates independently of and in parallel with the CPU. The result is increased CPU availability and improved system performance. Under the control of the lOP the Multipurpose (MP) Bus, a medium speed asynchronous I/O bus, transfers data at a rate of up to 1.5 MB per second. Up to 16 device controllers supporting up to 124 devices can be handled on a single MP Bus by a single lOP. Larger systems can use multiple lOPs to implement additional MP Buses. Diagnostic Processor The Diagnostic Processor has four main functions: • Loading the CPU/1PU control store at power-up At power-up, the Diagnostic Processor loads the CPU and the IPU microdiagnostics from a dedicated floppy disc to the control store RAM. It then performs CPU and the IPU tests before loading the CPU/IPU microcode and system initialization. The non-critical modules, such as the MACC, the cache banks or the Shadow Memory, are placed off-line if a failure is detected at power-up. In a dual-processor (CPU/IPU} configuration, the failing processor is placed off-line and the remaining processor is reconfigured as the CPU. Control Console The Control Console combines the operator’s console function the control panel functions. It is connected to the Diagnostic Processor Control Console Port, which, with modems, permits remote control of the system. The Control Console has three modes; Operators Console mode, Control Panel mode, and Diagnostic mode. The operator can switch between the three modes at any time using the key switch. Operator’s Console mode is the normal mode for the control Console when the system is running. Job status, user status, and operator information are displayed. The operator can monitor and control operation of the system. In the Operator’s Console mode, the CRT also functions as a user terminal. Control Panel mode displays status and allows the operator to display and/or modify Machine State, General-Purpose Register contents memory locations. The system is in Diagnostic Mode when first powered up. .CPU microdiagnostics can be run, single-stepped, and halted from Control Console, which displays relevant information on the status of each diagnostic. Line Printer/Floppy Controller The Line Printer/Floppy Controller provides the user an interface up to two line printers and two floppy disc drives. The floppy discs double-sided, double-density discs. All Gould CSD line printer are supported by this controller. The Line Printer/Floppy Controller is an MP Bus device and is based on a microprocessor. The microprocessor communicates through MP Bus interface to the IOP and controls the line printer and the floppy disc drive through their respective interfaces. Data buffers in each of the interfaces allow high throughput with minimum of lOP intervention. Each printer port has a 256 byte buffer. The floppy disc port has a 2 KB buffer which holds up to eight sectors of data. Memory Airthierachical memory system increases system performance by g that needed data is available for CPU/IPU usage in a very fast memory. Each level in figure 2 represents a level in the memory arrachy. Each level is successively faster than the one below, with CPU/IPU exchanging data with the system cache. Main Memory Main Memory in the CONCEPT 32/87 interfaces directly to the SelBUS. The Integrated Memory Module (IMM) is a high-density memory subsystem implemented on a single board It includes up to 1 dynamic MOS RAM, memory controller, error correction logic, fresh circuitry. The IMM can overlap two read cycles, allowing reads to be initiated 300 nanoseconds. A write can be initiated every 300 nanoseconds to a single IMM. When multiple IMMs are used, they can or 4-way interleaved. Writes can then be initiated every 150 seconds. The MOS Memory is organized as 39-bit words: 32 bits of data and 7 Error Correction Code bits (ECC). During the write operations, the ECC bits are generated based on a modified Hamming code algorithm stored in memory. When a read memory operation takes place, the ECC bits are checked. A 1-bit error is detected, the error is corrected. Two bit errors and multiple bit errors with an even number of bits wrong cause an error dection signal. Cache Cache memory is a very high-speed memory. It is integral to the CPU and IPU, and logically resides between the processor and main memory. (See Figure 1). Cache memory reduces the effective cycle time of the memory system and is transparent to the user. It provides a double word, which contains two word instructions or four halfword instructions or an operand, to the CPU/IPU in 75 nanoseconds. The 32 KB cache memory is structured with four banks of 8 KB each. The 64 KB cache memory contains eight banks of 8KB each. They use respectively 4-way and 8-way set associative mapping. A modified Least Recently Used (LRU) algorithm is used to allocate data to a bank. A cache control unit fetches data from main memory, monitors SeIBUS activity, and monitors the relevance of the contents of cache memory. It also loads the cache memory with requested data from main memory (see Figure 2) and copies the memory address for that block of data into the cache index. The cache system performs write throughs to main memory and monitors writes on the SelBUS to insure that the cache contains an accurate copy of main memory. Shadow Memory The optional Shadow Memory provides a high-speed copy of a region of main memory. Like cache, access time to this RAM is 75 ns. Unlike cache, the same addresses are always represented in Shadow Memory. The addresses contained in cache are a function of the recent history of the processor utilization. The addresses contained in the Shadow Memory are hardware assigned. Each Shadow Memory module maps a single 128 KB block of main memory. The CONCEPT 32/87 can support up to 256 KB of Shadow Memory in both CPU and IPU. Shadow Memory is initialized either by writing to the shadowed address range during system initialization or by the processors microcode during a warm start (main memory backed up by battery power). Shadow Memory is transparent to the user. I/O reads from the shadowed addresses occur from the main memory. Writes to the shadowed address range are updated into the Shadow Memory directly from the SeIBUS. The Shadow Memory can be used with a region of shared memory. This permits 75 ns access to shared data structures. Input/Output Subsystem The dual bus architecture of the CONCEPT 32/87 provides a choice of I/O performance levels. High performance I/O resides on the SeIBUS, while medium performance I/O resides on the MP Bus. This two-bus I/O structure allows great flexibility in system integration. The basic I/O subsystem of the CONCEPT 32/87 consists of the: • Input/Output (lOP) Data Types Instruction Repertoire Instructions are classified as either halfword instructions (16-bits) or word instructions (32-bits). The word instructions refer to memory locations; the halfword instructions deal with register operands. The functional classification and corresponding number of instructions for the CONCEPT 32/87 are as follows:
Privileged/Unprivileged Operations The CPU is capable of both privileged and unprivileged operation. Privileged operation allows the CPU to perform control functions and I/O instructions. Unprivileged operation is the normal user program mode of the CPU. In this mode, memory protection is in effect and all privileged operations are prohibited. This prevents unprivileged users from interfering with each other or the system. The memory protect system provides write protection by individual protection granules. A protection granule consists of 2 KB. Up to 16 MB can be protected at a time The memory protect registers can be changed by executing privileged instruction. Addressing The CONCEPT 32/87 provides two memory addressing environments: mapped and nonmapped. It also supports two addressing modes: the base register mode and the non-base register mode. When in the non-base register mode, two options are available under each environment: extended and non-extended. The user controls the selection of the options under each environment. These options determine the rules for logical address generation. In the base register mode, the CPU uses eight base registers to extend the addressing range of a single task. This mode allows 16 MB addressing for both code and data. In the mapped environment, the CPU uses 2048 MAP registers to translate logical address into physical address. The map registers for a user program are loaded on an as needed basis. There are two banks so one may be cleared "off line" to minimize context switch time. The memory is partitioned in 8 KB blocks and the write protect on 2KB boundaries. The addressing range for both environments and modes are summarized below.
Floating-Point Processor The Floating-Point Processor is a high-speed ECL-based single and double-precision computational processor with 64-bit wide busses to achieve high-speed floating-point operations. The processor is an integral pan of the CPU as well as of the IPU. All the floating-point related instructions process either words or doublewords. They are: • Add Multiply Accelerator (MACC) To enhance computational performance, the Multiply Accelerator (MACC) can be added to the CPU and the IPU of the CONCEPT 32/87 The MACC computes products for the nine integer and floating-point instructions. The execution time of each instruction varies with the value of the operands and whether it is an integer or floating-point instruction. As the MACC is transparent to the software, its presence is only visible by the decrease in execution time for multiply computations. Computational benchmarks such as Whetstone II (Double Precision show) performance improvement of up to 20%. Individual multiply instruction execution times average approximately one third of the time of the CONCEPT 32/87 without a MACC.
|
|
||||||||||||||||||||||||||
Send mail to tmghouston@aol.com
|
|||||||||||||||||||||||||||