[Libre-soc-bugs] [Bug 238] POWER Compressed Formal Standard writeup

[bugzilla-daemon at libre-soc.org]

https://bugs.libre-soc.org/show_bug.cgi?id=238

--- Comment #32 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to Jacob Lifshay from comment #29)
> (In reply to Luke Kenneth Casson Leighton from comment #23)
> > | 0 | 1  | 2 3 4 | | 567.8 | 9ab  | c d e | f |
> > | 1 | o2 |  RT   | | 010.0 | RB|0 | offs  | 1 | addi.
> > | 1 | o2 |  RT   | | 010.1 | RB|0 | offs  | 1 | addis.
> > 
> > these i'm reluctant to go entirely immediate in bits 2-4 because when
> > RB=0 it encodes "li RT, #imm" even though the immediate is only in
> > the range 0-15 for addi and -8 to 7 for addis.
> 
> I'd recommend not supporting addis for 16-bit mode since most of the time
> addis is part of a addi addis sequence to add a 32-bit constant (usually
> some variable/function's address). That constant would almost always not fit
> in whatever reduced-range immediate you have available in 16-bit mode.

i added a note that the idea is, a 2nd opcode is dedicated to a reduced-shift
range *variant* of addis, which maps into *addi*.  what do you think?


> > (actually a better encoding there would be just addis, why has nobody
> > noticed a massive overlap between addis and addi before in OpenPOWER
> > ISA v3.0B????)
> 
> addis adds bits right above the bits you can specify with standard 32-bit
> addi. together they allow you to add an arbitrary 32-bit constant.
> 
> This is kinda like lui/addi on RISC-V.

yeah there's a sequence for getting 64-bit immediates without using LD
in PowerISA, it involves *six* instructions.

https://github.com/antonblanchard/microwatt/blob/master/hello_world/head.S#L34

/* Load an immediate 64-bit value into a register */
#define LOAD_IMM64(r, e)                        \
        lis     r,(e)@highest;                  \
        ori     r,r,(e)@higher;                 \
        rldicr  r,r, 32, 31;                    \
        oris    r,r, (e)@h;                     \
        ori     r,r, (e)@l;

this is one of the reasons i'd like to add that Data-Pointer concept
(the above would be one instruction using DP)

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