Had some further exchanges in private with Don, but here is the bottom
line as far as I am understanding things...
The relocation of a value have nothing to do with the instruction, or
the addressing mode of the instruction. If that is being used, it is
"wrong". The addressing mode, if the expression is relocatable, will
decide what kind of relocatable record will be created, but it does not
decide if there even should be a relocatable record or not.
This should all be about the value of the evaluated expression. If the
evaluated expression is relative, then it needs to be relocated by the
linker. If it is absolute, then MACRO-11 can create the end results
directly.
A few examples:
.MAIN. MACRO V05.05 Saturday 7-MAR-20 15:42 Page 1
1 000000 .PSECT FOO,I,RW
2
3 000000 000042 .WORD 42
4 000002 000000 Y: .WORD 0
5
6 000004 005067 177772 CLR Y
7 000010 005037 000002' CLR @#Y
8 000014 005067 001234' CLR X
9 000020 005037 001234 CLR @#X
10
11 000000 .ASECT
12 001000 .=1000
13 001234 X=1234
14
15 001000 005067 000230 CLR X
16 001004 005037 001234 CLR @#X
17 001010 005067 000002' CLR Y
18 001014 005037 000002' CLR @#Y
19
20 000001 .END
.MAIN. MACRO V05.05 Saturday 7-MAR-20 15:42 Page 1-1
Symbol table
X = 001234 Y 000002R 002
. ABS. 001020 000 (RW,I,GBL,ABS,OVR)
000000 001 (RW,I,LCL,REL,CON)
FOO 000024 002 (RW,I,LCL,REL,CON)
Errors detected: 0
*** Assembler statistics
Work file reads: 0
Work file writes: 0
Size of work file: 40 Words ( 1 Pages)
Size of core pool: 9524 Words ( 36 Pages)
Operating system: RSX-11M/M-PLUS
Elapsed time: 00:00:00.01
,TT=TT
Line 6 becomes an absolute expression since it's a subtraction of the
relative address of Y from the current position, both in the same psect.
Such an operation renders an absolute value.
Line 7 becomes relocatable, since the absolute address of Y is unknown
because it's in a psect that can be moved around at link time.
Line 8 becomes relocatable because even though X is at a known address,
the address of the instruction is unknown and subject to the location of
the psect, which is only decided at link time.
Line 9 becomes absolute since it is an expression only dependent on an
absolute value.
Line 15 becomes absolute since it's an expression subtracting two
absolute values which obviously have an absolute result.
Line 16 becomes absolute since it is an expression only dependent on an
absolute value.
Line 17 becomes relocatable because the address of Y is in another
psect, and the subtraction of a relative address from an absolute
renders a relative value.
Line 18 becomes relocatable because the absolute address of Y is
unknown. MACRO-11 only have a relative value of Y.
Note that all references to Y contains 000002. This is because that is
the offset of Y within the psect. This is added to the base of the psect
(the part that comes from this object file), which is all the linker
will know about.
Johnny
On 2020-03-07 14:47, Rhialto wrote:
On Sat 07 Mar 2020 at 01:59:03 +0100, Johnny Billquist wrote:
In the end, that is a different thing. Absolute addressing is not what Bob
was thinking about here...
Here is what MACRO-11 does:
1 000000 .ASECT
2 001000 .=1000
3 001234 X=1234
4
5 001000 005067 000230 CLR X
6
Note that the addressing is still relative. But the actual offset required
can be computed by MACRO-11 at compile time.
This is the output of dumpobj for that code. It looks like indeed there
is an unneeded relocation record in there:
GSD:
GLOBAL X =1234 DEF ABS flags=110
MODNAME TEST =0 flags=0
PSECT =0 CON RW REL LCL I flags=40
PSECT . ABS.=1004 OVR RW ABS GBL I flags=104
XFER . ABS.=1 flags=10
ENDGSD
RLD
Location counter definition . ABS.+1000
TEXT ADDR=1000 LEN=4
001000: 005067 001234 7...
RLD
Internal displaced 1002=1234
ENDMOD
It looks like the 2nd RLD record is generated by the code
text_displaced_word() in object.c. That in turn must be called from this
code in assemble_aux.c, since there are no other callers to
store_displaced_word():
void mode_extension(
...
if (value->type == EX_LIT) {
if (mode->rel) /* PC-relative? */
store_displaced_word(str, tr, 2, value->data.lit);
else
store_word(str, tr, 2, value->data.lit); /* Just a
known
value. */
My guess is that this could become something like
if (value->type == EX_LIT) {
if (mode->rel) { /* PC-relative? */
if (current_pc->section->flags & PSECT_REL) {
store_displaced_word(str, tr, 2, value->data.lit);
} else {
/* I can compute this myself. */
store_word(str, tr, 2, value->data.lit - DOT - 2);
}
} else {
store_word(str, tr, 2, value->data.lit); /* Just a
known
value. */
}
It seems like this case didn't come up in any previous tests; the
regression tests show no change at all. (I have tested somewhat strictly
by hand that various RSX-Kermit-11 source files generated the same
object files as on RSX and that the listings, including macro expansion,
didn't have any unexplained differences).
Any interesting test cases I can use with this, so see if it is correct
this way?
1 .TITLE TEST RELOCATABLE
REFERENCES
2 000000 .ASECT
3 001000 .=1000
4 001234 X == 1234
5 001000 005067 000230 CLR X
. =****** X =001234 G
. ABS. 001004 000 (RW,I,GBL,ABS,OVR,NOSAV)
Johnny
-Olaf.
--
Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: b...@softjar.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol
_______________________________________________
Simh mailing list
Simh@trailing-edge.com
http://mailman.trailing-edge.com/mailman/listinfo/simh
