[Cfp-interest] more background: why we don't have presubstitution

[David Hough CFP]

I left this out of the historical material I sent yesterday:

One can imagine the original idea being implemented by having all
hardware trap enable bits on all the time.      There would be a table
of actions to be taken for all floating-point traps....

Now the foregoing sounds very inefficient - taking a hardware trap on every
rounding error?      But there's an obvious optimization available -
once the detail record has been created/flag has been raised, there's no
need to take a trap to implement the default exception handling - just
compute the default result and continue.    And this is what almost all
hardware does almost all the time.

The idea of a table of actions on exceptions lends itself to presubstitution.
After all, if the default action were to return a value that is
just an entry in the (user space)
table, how hard can it be to substitute a different value?
But the obvious optimization mentioned above, hardware handling exceptions by
returning a default value without trapping, interferes with  this idea.
The hardware does not have a table of default actions; it synthesizes the
infs and NaNs it needs when it needs them.    To get to a table of
substituted values in user space, that might be thread-specific or worse
global to many threads and hence subject to read/write race conditions
or deadlocks... is going to be at least as expensive as taking a trap to
user mode.     So while attaching a presubstituted value to an exceptional
operation might be a clean syntactic way to express the intended
program flow, the actual implementation is
going to have to at least involve tests and branches if not traps.