pec@necntc.nec.com (Paul Cohen) (05/31/90)
In Article 1570 of comp.sys.ibm.pc.programmer, jhuang@sci.ccny.cuny.edu (Jian Huang) comments: > NEC 25 is actually a microcontrollor which has some useful > features such as macro service (kind of DMA) and bank > switch. The V25 has eight macro-service channels in addition to two independent DMA channels. Macro-service is like DMA except that it services internal peripherals whereas DMA is for external accesses. The internal peripherals of the V25/35 include two timers, two full duplex UART's, a programmable interrupt controller and a wait state generator. As mentioned, the parts support register bank switching. They provide 24 I/O lines and eight analog comparator inputs as well. These processors have a prefetch pointer which is separate from the instruction pointer to support its six byte instruction prefetch queue and to reduce branch penalty. > After using it in our PC-based motion controllor > card we found that its main problem is the same as one in > intel 8088 --- " branch penalty " which got improved in > 80188. So if you don't care speed then it is a good chip > for compact motherboard design. The manuals for these parts show that the execution time for a far branch for the 80188 is 14 cycles, in sharp contrast to the 15 cycles for the V-Series parts and the 8088. Branch penalty is only one of many factors that affect performance. Another is memory fetch time - two cycles per byte for the V25 and four for the 8088 and 80188. The most widely accepted method for judging performance is with benchmarks so I submit the following data from NEC's V-Series benchmark report (Stock Number 501270) which shows execution time in seconds: 0 Wait States 2 Wait States V25 80188 V25 80188 EDN A 18 16.5 26 22.375 EDN E 220 280.5 269 337 EDN F 240 274 354 395 EDN H 290 377.25 456 532.5 EDN I 51961 57856 74635 78204 EDN K 1094 1391 1636 1991.5 IEEE Digital Filter 318 363.875 447 537 Sieve 307236 328403 509934 534263 Fibonacci 2804014 2700922 4363825 4474487 The V25 measurements shown above are for the ROMless 70320 which is slightly slower than the 70322 ROM part. For these nine benchmarks the execution time for the V25 at zero wait states ranges from 76.8% to 109.1% of the execution time for the 80188 with the average being 89.5%. At two wait states the range is from 79.8% to 116%, the average being 91.7%. These benchmarks were measured using exactly the same code for the two processors. 8 MHz. versions of each part were measured. The V-Series parts have some additional instructions which speed up some of the benchmarks, as shown below. EDN F 167 247 EDN K 718 1120