Re: [PATCH] Add MIPS32r2/MT ASE pause instruction

[Andrew Pinski <andrew dot pinski at caviumnetworks dot com>]

On Thu, Dec 8, 2011 at 11:52 AM, Richard Sandiford
<rdsandiford@googlemail.com> wrote:
> "Maciej W. Rozycki" <macro@linux-mips.org> writes:
>>> Index: opcodes/mips-opc.c
>>> ===================================================================
>>> RCS file: /cvs/src/src/opcodes/mips-opc.c,v
>>> retrieving revision 1.89
>>> diff -u -p -r1.89 mips-opc.c
>>> --- opcodes/mips-opc.c    29 Nov 2011 20:28:55 -0000   Â1.89
>>> +++ opcodes/mips-opc.c    7 Dec 2011 22:39:50 -0000
>>> @@ -1169,6 +1169,7 @@ const struct mips_opcode mips_builtin_op
>>> Â{"ori", Â Â "t,r,i", Â Â Â Â0x34000000, 0xfc000000, WR_t|RD_s, Â Â Â Â Â Â Â0, Â Â Â Â Â Â ÂI1 Â Â Â},
>>> Â{"pabsdiff.ob", "X,Y,Q",0x78000009, 0xfc20003f, Â Â WR_D|RD_S|RD_T|FP_D, Â Â0, Â Â Â Â Â Â ÂSB1 Â Â },
>>> Â{"pabsdiffc.ob", "Y,Q", Â Â 0x78000035, 0xfc2007ff, RD_S|RD_T|FP_D, Â Â Â Â WR_MACC, Â Â Â ÂSB1 Â Â },
>>> +{"pause", Â "", Â Â Â Â Â Â 0x00000140, 0xffffffff, TRAP, Â Â Â Â Â Â Â Â Â 0, Â Â Â Â Â Â ÂI33|MT32},
>>> Â{"pavg.ob", "X,Y,Q", Â Â Â Â0x78000008, 0xfc20003f, WR_D|RD_S|RD_T|FP_D, Â Â0, Â Â Â Â Â Â ÂSB1 Â Â },
>>> Â{"pickf.ob", "X,Y,Q", Â Â Â 0x78000002, 0xfc20003f, WR_D|RD_S|RD_T|FP_D, Â Â0, Â Â Â Â Â Â ÂMX|SB1 Â},
>>> Â{"pickf.ob", "D,S,T", Â Â Â 0x4ac00002, 0xffe0003f, WR_D|RD_S|RD_T, Â Â Â Â 0, Â Â Â Â Â Â ÂN54 Â Â },
>>
>> ÂI think it's redundant to say |MT32 here as MT32 already implies I33 --
>> the MT ASE requires rev. 2 of the architecture.
>>
>> ÂAlso please use NODS rather than TRAP -- although both macros expand to
>> the same numerical value, they have been semantically split so that the
>> former refers to instructions that must not be scheduled into a delay slot
>> for architectural reasons (undefined behaviour) and the latter is for
>> instructions that are meant to generate a trap and we chose not to
>> schedule these instructions into a delay slot merely to make life of other
>> software easier (delay-slot execution is just fine).
>>
>> ÂYes, I know most MT instructions use the TRAP flag, but that's because
>> there's really no reason I know of they should have this flag at all in
>> the first place, so they have been left as classified "for someone's
>> convenience" (some can legitimately trap indeed, i.e. FORK when run out of
>> thread contexts). ÂPAUSE's delay-slot operation is on the other hand
>> undefined (which actually brings a concern about forward compatibility of
>> old software, hmm...).
>>
>> ÂI can't approve the change, but it looks good to me otherwise.
>
> OK with Maciej's changes (i.e. MT32 rather than "I33|MT32" and
> NODS rather than TRAP).

It is part of the MIPS32R2 instruction set as of v2.6 (and not just
MT32).  So should it be just I33 then?

Thanks,
Andrew Pinski