kurt@fluke.UUCP (Kurt Guntheroth) (05/29/85)
I frankly don't know what Doug Pardee's problem is. He seems to understand
that machine architecture is only important to compiler writers since users only
deal with the machine at a high level. Why is it not obvious to him that
the machine language should therefore be designed for the compiler writers,
no matter how few of them there are?
He understands that good compiler writers are few and far between. Why
doesn't he understand that a mediocre compiler writer is more likely to do
an acceptable job if the machine is easy to compile for?
He notices that there are many compilers for the 68000, but very few good
ones. Why doesn't he infer from this fact that the 68000 is not a very easy
instruction set to compile for? (And the iAPX-86 architecture is much worse).
He believes a 'dozen' top notch compiler writers should be sufficient to
meet the world's compiler needs. Can't he understand that market forces
would not create so many compiler-writing jobs if this were true.
*** THE ONE TRUE MEASURE OF INSTRUCTION SET QUALITY ***
An instruction set can be judged as 'good' if and only if it produces
optimal usage of computer resources on user jobs. If the user programs
in a higl level language, the instruction set balances hardware throughput
against the ability of the compiler to get around any obstacles put into the
architecture to achieve the additional machine throughput. Any part of the
hardware architecture not heavily used by user jobs is wasted and it
behooves hardware designers to insure they do not waste their time designing
parts of the computer that cannot effectively be used.
--
Kurt Guntheroth
John Fluke Mfg. Co., Inc.
{uw-beaver,decvax!microsof,ucbvax!lbl-csam,allegra,ssc-vax}!fluke!kurt