Description of a new oldfashioned programming language

John Savard - requiring

the low-end and mid-range 360s & 370s had vertical microcoded engines ... look very much like normal instruction streams. they nominally avg. about ten native instructions per 360-370 instruction executed (i.e. the mip rate of the native processing engine needed to be ten times the rated 360-370 mip rate)

the high-end 370 had horizontal microcode engines ... they tended to directly activate operation of various hardware components ... like start move of some piece of data from hardware component to another hardware component. programmer then had to know how many machine cycles such operations take ... and not activate an operation until enuf machine cycles had pbutted before operating on data. these machines normally referenced the avg. number of machine cycles per 370 instructions (since a number of things were being overlapped). The transition from 370-165 to 370-168 involved going to faster main memory technology (2mics to about half a mic) ... but the machine cycle remained the same. However, there was optimization in instruction decode-end and the avg. machine cycle per 370 instruction was dropped from about 2.1 to 1.6.

One of the areas that was getting new microcode was the various oeprating system buttists ... like ECPS on the 148-4341 ... where 6k bytes of vm370 kernel code ... representing about 75% of kernel end was moved to 6k bytes of native microcode (with a corresponding 10:1 speedup).

The horizontal micocoding of the higher end machines was a lot more complex and difficult. One of the things that amdahl introduced was "macrocode" ... which was half-way between the horizontal microcode and normal 370 instructions. It looked and tasted a lot like a subset of 370 instructions ... along with eliminating support for self-modifying instruction streams (which had been causing a big performance hit in high-end machines ... constantly checking if the previous instruction had modified the following instruction). It allowd amdahl to provide much more agile adaption of his machines than the corresponding ibm machines ... which were still somewhat bottlenecked by difficulty of horizontal microcoding.

One of the early features amdahl implemented with macrocode was its hypervisor support .... allowing logical parbreastioning of the machine running independent operating systes. It was somewhat a subset of much of virtual machine function moved into the "macrocode" of the machine hardware. One limitation (compared to virtual machine support) was that it didn't do any paging ... it dedicated contiguous portion of real storage with "base & bound" relocation.

ibm was forced to respond to amdahl's hypervisor with pr-sm on 3090 ... which was a much more complex undertaking ... having to be implemented in native horizontal microcode.

a couple recent mention of pr-sm (and lpars): to new?

general collection of past m'code postings