Ok, I've uploaded my current snapshot to
http://hg.genunix.org/libdisasm.hg for all that wish to take a look.
I've also included copies of libdisasm*.[ch] to simplify the build
(obviously these will be redundant once/if this is integrated).
Currently, there is an environment variable that can be set,
_LIBDISASM_DEBUG, that can take a combination of the following values:
none Do no special formatting (i.e. no synthetic
instructions, no elimination of redundant values
compat Output as compatible as possible (more below) with the
current closed source disassembler (based on trial and error) --
Default.
synth-all Output all supported synthetic instructions
format Wriite on stderr the individual field values of the instructions
binary Output the instruction value in binary on stderr
Values can be combined with a , (i.e. export _LIBDISASM_DEBUG=compat,format)
The following table shows what is done when:
For addresses,
[ %rXX + %g0 ] -> [ %rXX ] compat, synth-all
[ %rXX + 0x0 ] -> [ %rXX] compat, synth-all
[ %g0 + %rXX ] -> [%rXX] compat, synth-all
Synthetic Instructions:
subcc %rs1, reg_or_imm, %g0 -> cmp %rs1, reg_or_imm compat,synth-all
jmpl address, %g0 -> jmp address
compat
jmpl address, %o7 -> call address
compat
bn.a.pt %xcc, label -> iprefetch label
orcc %g0, %rs2, %g0 -> tst %rs2
compat, synth-all
jmpl %i7+8, %g0 -> ret
compat, synth-all
jmpl %o7+8, %g0 -> retl
compat, synth-all
restore %g0, %g0, %g0 -> restore
compat,synth-all
save %g0, %g0, %g0 -> save
compat,synth-all
sra %rs1, %g0, %rd -> signx %rs1, %rd
sra %rd, $g0, %rd -> signx %rd
xnor %rs1, %g0, %rd -> not %rs1, %rd
compat,synth-all
xnor %rd, %g0, %rd -> not %rd
compat,synth-all
sub %g0, %rs2, %rd ->neg %rs2, %rd
compat,synth-all
sub %g0, %rd, %rd -> neg %rd
compat,synth-all
casa [%rs1]#ASI_P, %rs2, %rd -> cas [%rs1], %rs2, %rd synth-all
casa [%rs1]#ASI_L_P, %rs2, %rd -> casl [%rs1], %rs2, %rd synth-all
casxa [%rs1]#ASI_P, %rs2, %rd -> casx [%rs1], %rs2, %rd synth-all
casxa [%rs1]#ASI_L_P, %rs2, %rd -> casxa [%rs1], %rs2, %rd synth-all
add %rd, 1, %rd -> inc %rd
synth-all
add %rd, const13, %rd -> inc const13, %rd
synth-all
addcc %rd, 1, %rd -> incc %rd
synth-all
addcc %rd, const13, %rd -> incc const13, %rd
synth-all
sub %rd, 1, %rd -> dec %rd
synth-all
sub %rd, const13, %rd -> dec const13, %rd
synth-all
subcc %rd, 1, %rd -> deccc %rd
synth-all
subcc %rd, const13, %rd -> deccc const13, %rd
synth-all
andcc %rs1, reg_or_imm, %g0 -> btst reg_or_imm, %rs1
compat,synth-all
or %rd, reg_or_imm, %rd -> bset reg_or_imm, %rd
synth-all
andn %rd, reg_or_imm, %rd -> bclr reg_or_imm, %rd
xor %rd,reg_or_imm, %rd -> btog reg_or_imm, %rd
or %g0, %g0,%rd -> clr %rd
compat,synth-all
stb %g0, [addr] -> clrb [addr]
compat, synth-all
sth %g0, [addr] -> clrh [addr]
compat, synth-all
stw %g0, [addr] -> clr [addr]
compat, synth-all
stx %g0, [addr] -> clrx [addr]
compat, synth-all
srl %rs1, %g0, %rd -> clruw %rs1, %rd
srl %rd, %g0, %rd -> clruw %rd
or %g0, reg_or_imm, %rd -> mov reg_or_imm, %rd
compat, synth-all
rd %y, %rd -> mov %y, %rd
rd %asrN, %rd -> mov %asrN, %rd
wr %g0, reg_or_imm, %y -> mov reg_or_imm, %y
compat, synth-all
wr %g0, reg_or_imm, %asrN -> mov reg_or_imm, %asrN
compat, synth-all
The current closed source disassembler does not seem to use the
correct floating point register names for double and quad precision
floats -- it's always %fNN (not %dNN or %qNN). While in some cases
%fNN is an alias, as far as I can tell, that is not the case for all
of the double and quad registers (%q48 or %d50 for example). My
preference would be to use the %dNN or %qNN names as I suspect it
makes the intent more obvious.
The other deficiency with compat mode is that for certain load and
store variants, the closed source library will still output '[%rs1 +
0x0]' but not in others. I'm not convinced it'd be a good use of time
to try to figure out when it does and when it does not print the '+
0x0', and instead just never print '+ 0x0' except when
_LIBDISASM_DEBUG=none
All of the settings are easily tweaked if there's a strong desire to
have or not have certain synthetic instructions be used.
I've tested it (using the bundled compare.pl script) against /usr/bin,
/bin, /usr/sbin, /sbin, /kernel/drv/sparcv9,
/platform/sun4u/kernel/sparcv9/{gen}unix, and so far everything
appears to check out. I'm planning to add a little more
sophistication to the compare.pl script to recognize some of the
differences I mentioned, and ignore those, that way the noise is cut
down further.
So, the next question I'd have is what the next step would be to get
things closer towards integration. Or rather, what things I can do now
to help make the process go smooth when things are ready.
_______________________________________________
tools-discuss mailing list
tools-discuss@opensolaris.org
