After an evening reflecting on this issue, I think the following might be a
rule worth considering. Sub-rule (2) causes some significant complications,
that will be discussed later.
Rules for determining x-opcode rates:
(1) If an x-opcode has a rate-determined argument, x-opcodes are assigned
the rate of their fastest rate-determined argument.
(2) If (1) fails, and if the x-opcode occurs within an expression that will
be passed as a formal argument to another opcode, *and* the formal argument
is not xsig, then the x-opcode takes on the rate of the formal argument.
(3) if (2) fails, and if the x-opcode occurs within the right-hand-side of
an assignment expression, then the rate of the x-opcode defaults to the rate
of the left-hand-side of the assignment statement.
(4) if (3) fails, the x-opcode is enclosed in the body of an if or while
statement, then the x-opcode defaults to the slowest rate permitted by all
enclosing guard conditions.
(5) If (4) fails, the x-opcode defaults to k-rate.
Note, particularly, that the enclosing opcode match has been removed.
Discussion:
The case that requires particular attention, is the case where x-opcode
results are passed to xsig arguments of enclosing x-opcode invocations.
e.g.:
opcode x1(xsig x) {
return (x);
}
opcode x2(xsig x) {
return (x);
}
instr i() {
x2(x1(aVar)); // (1)
x2(x1(ftlen(tbl)); // (2)
}
Application of the rules would procede as follows:
(1) Should produce the following: x2 has one argument: ((x1 has 1 asig
argument) -> x1 is a-rate) -> x2 is a-rate, but it doesn't as the rules are
given.
(2) Should produce the following: x2 has one argument: (x1 has 1 argument:
(ftlen has no arguments -> ftlen is xsig) -> x1 is xsig -> x2 is xsig) ->
apply rule (4) to x2: x2 is ksig. Now, rule (2) can be applied to x1: x1 is
k-rate. now, rule (2) can be applied to ftlen: ftlen is ksig. Note that the
proposed rules require two passes on the expression tree to determine the
rate of the opcodes.
I have no idea how you would describe (2) in verbiage suitable for a
standards document.
For the record, I think the existing rules would be applied as follows:
(2) x2 has one argument: (x1 has one argument: (ftlen has no arguments, and
no enclosing context -> ftlen is k-rate) -> x1 is k-rate) -> x2 is k-rate.
Only one pass on the expression tree.
Conclusion:
Sub-rule (2), if adopted, would need to be carefully formulated.
regards,
robin davies.
_________________________________________________________________________
Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com.
This archive was generated by hypermail 2b29 : Mon Jan 28 2002 - 12:11:07 EST