Re: [Bug gas/336] gas cannot assemble pdtlb,l instruction

[Nick Clifton]

Hi Randolph,

> so, apparently, gas thinks that:
>
> pdtlb   0(%r29)     -> valid for .level 1.1 or .level 2.0
> pdtlb   %r0(%r29)   -> ditto
> pdtlb,l 0(%r29)     -> invalid always
> pdtlb,l %r0(%r29)   -> valid iff .level 2.0
>
> i can't say that this is strictly wrong, just seems inconsistent :)

I agree.  This behaviour is explicitly coded into the sources.  if you 
have a look at the header file include/opcodes/hppa.h and search for the 
pdtlb instruction you will find four entries:

 { "pdtlb", 0x04001600, 0xfc003fdf, "cLcZx(s,b)", pa20, FLAG_STRICT},
 { "pdtlb", 0x04001600, 0xfc003fdf, "cLcZx(b)", pa20, FLAG_STRICT},
 { "pdtlb", 0x04001200, 0xfc003fdf, "cZx(s,b)", pa10, 0},
 { "pdtlb", 0x04001200, 0xfc003fdf, "cZx(b)", pa10, 0},

It is the presence of FLAG_STRICT which is forcing the assembler to 
require an explicit register name for the "x" field.  My assumption is 
that this is because the HPPA 2.0 ISA specification tightened up the 
syntax for the pdtlb instruction, ruling out register name 
abbreviations.  Presumably the "0" abbreviation is still supported for 
level 1.1 since that might be used in already existing code.  I cannot 
be definitive about this since I did not write the code and I am not an 
expert on the HPPA ISA.

Cheers
  Nick