gnu@sun.uucp (John Gilmore) (10/16/84)
Curiously enough, MVCL is such a complicated instruction that it is almost always faster to just run the old 360 MVC loop. I don't know if this has changed in newer 370's and Amdahls but it was true of the 145, 148, 155, 158, and 470V/6. PS: It is "MVCL" not "MVL". There's also "CLCL" (compare logical characters long), which was similarly slow but had the advantage that it would stop with the registers pointing to the mismatch, a feature not available in CLC. PPS: In many ways the 370 tried to innovate but just came up with worse instructions than the 360 way. Example: I doubt there was a single 360 assembler programmer who never wanted the "LT" (load from memory and test) instruction -- you had to do "L" then "LTR" (load then load&test-register). They "fixed" this in the 370 with the "ICM" instruction (insert characters under mask), which took a 4-bit mask and loaded 0 to 4 bytes into a register, setting the condition code the same as LTR. However, it's more than twice as slow as "L/LTR" and only accepts a subset of the possible addressing modes. Years ago I read an introduction to System/370 that talked about the new instructions and why they were created and in hindsight the design was clearly done by a committee. PPPS: In popular parlance the ICM and STCM instructions were called "Ick-em" and "Stick-em". Just thought you might want to know.
herbie@watdcsu.UUCP (Herb Chong, Computing Services) (10/25/84)
The newest S/370 machines still have the same slowness of execution
with the "interruptible" instructions MVCL and CLCL. This varies
from model to model, but not much has changed. Incidentally, the 370/XA
architecture is changed in the way addressing is handled (31 bits), but
runs slower in S/370 emulation mode than 370/XA mode because the S/370
instructions are interpreted differently (I think it's using a different
set of microcode). The 370/XA Principles of Operation would be the ultimate
reference for these things. Of course, even in the slow S/370 mode, the top
of the line models are still faster than any of the older S/370 machines.
Herb...
I'm user-friendly -- I don't byte, I nybble....
UUCP: {decvax|utzoo|ihnp4|allegra|clyde}!watmath!watdcsu!herbie
CSNET: herbie%watdcsu@waterloo.csnet
ARPA: herbie%watdcsu%waterloo.csnet@csnet-relay.arpa
BITNET: herbie at watdcs,herbie at watdcsu