[libre-riscv-dev] [Bug 269] auto-conversion / parser of POWER ISA Spec v3.0B

[bugzilla-daemon at libre-riscv.org]

http://bugs.libre-riscv.org/show_bug.cgi?id=269

--- Comment #33 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to Michael Nolan from comment #32)
> I'm trying to get addpcis working (so I can test NIA and CIA without messing
> with branches), and the instruction builds the displacement from "concat(d0,
> d1, d2)". We don't have separate d0, d1, and d2 fields, but we do have a D
> field and a d0_d1_d2 field. Should this sort of thing (I expect we'll run
> into this again with shifts) be corrected in the ISA spec, or by adding the
> relevant fields in in fields.text?

they're already there, what i did however was to take d0,d1,d2 and cat them
with underscores into a compound named field.

   d0,d1,d2 (16:25,11:15,31)
        Immediate fields that are concatenated to specify a
        16-bit signed two's complement integer which is
        sign-extended to 64 bits.
        Formats: DX

what should actually happen is that 3 fields should be created.

might need to keep the compound name as well.

hmmm line 225

https://git.libre-riscv.org/?p=soc.git;a=blob;f=src/soc/decoder/power_fields.py;h=26b59d46863fcd313e6d952b8f5fab22129afef0;hb=1117d7c3cd52cc73a757f3c6f62c6602551d7ad3#l225

i apologise, i write these types of line-based parsers on a one-off basis quite
regularly, without expectation of modifying them, afterwards.

i think that one would be easier to process back in decode_instructions.

line 24

https://git.libre-riscv.org/?p=soc.git;a=blob;f=src/soc/decoder/power_fields.py;h=26b59d46863fcd313e6d952b8f5fab22129afef0;hb=1117d7c3cd52cc73a757f3c6f62c6602551d7ad3#l24

here it is just collecting d0,d1,d2 *without* splitting by commas

but in between the brackets *is* split

err nowait, line 213 is where the brackets are silently discarded (spec[1:-1])

then split by comma to recognise 16, 23:24, 31

then split by colon to recognise slices from indices

so hmmm..

suggest taking f from line 210 and splitting it by comma.  if no commas skip
this next bit.

then take s from line 213

use zip on those two lists.

then *recreate* the lines as if they were actually in fields.text with "%s:
[%s]"


 d0 : [16]
 d1 ....
 d2 ....

append those together and *re-call* decodeinstrfields, and do res.update() on
the return dict.

only do that if f from line 210 has a comma in it.

sorry it is 3:30 am if that is totally confusing i can tackle it tomorrow
morning

-- 
You are receiving this mail because:
You are on the CC list for the bug.