At 06:20 AM 3/29/2000, Robin Davies wrote:
>A question about the rate of ftlen, ftloop, ftend (5.9.6.1- 5.9.6.3).
>
>The standard specifies ftlen as
>
> opcode ftlen(table t);
>
>Note, not kopcode or iopcode, but opcode, which makes it an x-rate opcode.
>
>Although the standard does not specify how to treat table parameters for
>intrinsic opcodes, I'm going to assume that table parameters are skipped
>when determining the rate of an x-rate opcode. (Is that correct?)
I hope so :-))
>If so, then by default rules, the opcode should evaluate at k-rate.
This is in 5.8.7.7.2, but it refers to core opcodes, not to supposed
calls to core opcodes inside other ones...
>However, the following should also be true. Given
>
> iopcode itlen(table t) {
> return (ftlen(t));
> }
>
>itlen should evaluate at ftlen at I-rate since iopcode places a rate
>restriction on the x-rate opcode ftlen. Is that correct?
Again reading 5.8.7.7.2 this is correct.
>This is an important question given that any serious attempt at a
>roll-your-own version of oscil or loscil is going to have to perform some
>pretty expensive divides at k-rate if ftlen, ftsr and ftbasecps can't be
>evaluated at i-rate.
I don't see your exact point in relating ftlen to loscil optimization...
In SAINT the e.g. loscil core opcodes is decomposed exactly as you
mention, i.e. parameters related with tables are stored in intermediate
i-registers and never calculated again during the opcode life.
I suppose that this definition of ftlen et al. has something to do with
the table score line in SASL; on the other hand nothing in loscil et al.
have a direct reference to ftlen, you just know you have to do
the same thing and calculate the table length. I made here the supposition
that since parameters in loscil are ivars, one cannot change the table
on which it works on without using a different loscil. But imho the
formal specification of loscil/oscil has nothing to do with ftlen: one
thing is the language, another your own implementation of the decoding
process. Can you agree on this ?
Best regards,
Giorgio
__________________________________________________________________
Giorgio ZOIA
Integrated Systems Laboratory - DE/LSI - EPFL
CH-1015 Lausanne - SWITZERLAND
Phone: + 41 21 693 69 79 E-mail: Giorgio.Zoia@epfl.ch
Fax: +41 21 693 46 63
__________________________________________________________________
This archive was generated by hypermail 2b29 : Mon Jan 28 2002 - 12:03:53 EST