| |
Hypnosis
Registered: Mar 2015
Posts: 36 |
Assembler with dynamic symbol creation
I have read some documentation for an assembler that allowed labels or symbols to be created dynamically, basically by concatenating strings. Which assembler was that? |
|
| |
doynax
Account closed
Registered: Oct 2004
Posts: 212 |
CA65 supports this, as do a few others I expect. If push comes to shove you can always do ad-hoc meta-programming or preprocessing by adding external tools to the build process.
In CA65 the syntax is something along these lines: .mac smc_store name
ident(.concat(.string(name), "_value")) := *+1
lda #$00
ident(.concat(.string(name), "_addr")) := *+1
sta a:$0000
.endmac |
| |
Hypnosis
Registered: Mar 2015
Posts: 36 |
Ah, there it is. Thanks! |
| |
Compyx
Registered: Jan 2005
Posts: 631 |
Why would you want to do that? Usually when I run into this, my "you're doing it wrong" alarm goes off.
There might be a valid use for this, but I'm not seeing it. |
| |
Oswald
Registered: Apr 2002
Posts: 5022 |
probaby some complicated speedcode, where writing native generator would take less time than macroing the shit out of it. |
| |
MagerValp
Registered: Dec 2001
Posts: 1058 |
For example if you want a small DSL for writing display lists, like this one here: https://twitter.com/MagerValp/status/741681367516340224 |
| |
doynax
Account closed
Registered: Oct 2004
Posts: 212 |
Quoting CompyxWhy would you want to do that? Usually when I run into this, my "you're doing it wrong" alarm goes off. The main place I am using it as a handy way of forwarding metadata and type information to external tools, sort of like C++ name mangling. For instance to encode breakpoints/watchpoints for VICE by running a post-processing script on the label list to transcribe magic labels into monitor commands.
Quoting Oswaldprobaby some complicated speedcode, where writing native generator would take less time than macroing the shit out of it. Indeed but also note that CA65 has scopes which often can achieve the same ends in a cleaner way, at least when operating within a single compilation unit. |
| |
Compyx
Registered: Jan 2005
Posts: 631 |
Quoting doynax
Quoting CompyxWhy would you want to do that? Usually when I run into this, my "you're doing it wrong" alarm goes off. The main place I am using it as a handy way of forwarding metadata and type information to external tools, sort of like C++ name mangling. For instance to encode breakpoints/watchpoints for VICE by running a post-processing script on the label list to transcribe magic labels into monitor commands.[/quote[
I stand corrected, that does indeed sound like a valid use case.
Quoting doynax
Quoting Oswaldprobaby some complicated speedcode, where writing native generator would take less time than macroing the shit out of it. Indeed but also note that CA65 has scopes which often can achieve the same ends in a cleaner way, at least when operating within a single compilation unit.
Speedcode was my first guess, but then again, when generating speedcode, you already all you need to know about entry points or whatever and just 'poke' that in other (speed)code.
I'm kind of surprised CA65 doesn't handle scopes spread over various translation units. But since it's part of a C compiler, it makes sense, C having only a couple of namespaces/scopes. |
| |
Hypnosis
Registered: Mar 2015
Posts: 36 |
Quoting CompyxWhy would you want to do that? Usually when I run into this, my "you're doing it wrong" alarm goes off.
I think the general case is a meta programming loop with some code that has to match and reference things in another loop. I can see if I can dig up my latest exact use case. |
| |
Trash
Registered: Jan 2002
Posts: 122 |
Sometimes you have speedcode that havn't got a fixed size (and the size-changes may not be cyclic), then it is nice to be able to generate labels when generating it in order to access them from another set of speedcode. |
| |
Oswald
Registered: Apr 2002
Posts: 5022 |
Quote: Sometimes you have speedcode that havn't got a fixed size (and the size-changes may not be cyclic), then it is nice to be able to generate labels when generating it in order to access them from another set of speedcode.
just store the "labels" into a table while generating speedcode A, then read the table while generating speedcode B that is referencing A. I do this all the time, and not much code in native 6510 either. |
| |
Count Zero
Registered: Jan 2003
Posts: 1825 |
AFAIR Dreamass supports that as well. |
| |
Slammer
Registered: Feb 2004
Posts: 416 |
Quoting doynaxIndeed but also note that CA65 has scopes which often can achieve the same ends in a cleaner way, at least when operating within a single compilation unit. Doynax got a good point. Doing things with scopes often gives cleaner solutions. |
| |
Hypnosis
Registered: Mar 2015
Posts: 36 |
Quoting doynaxIndeed but also note that CA65 has scopes which often can achieve the same ends in a cleaner way, at least when operating within a single compilation unit.
How would scopes solve the problem? |
| |
Hypnosis
Registered: Mar 2015
Posts: 36 |
Creating a meta-list of addresses where the data is defined and read from the list where addresses should be used would work for my simple case. Does assemblers handle this without requiring data and code to be defined in a certain source code order? I think Kickassembler does. How about the others? |
| |
Slammer
Registered: Feb 2004
Posts: 416 |
I don't know the functionality of CA65, but in general the solutions I have seen involving scoping seem better than those where you generate numbered labels to one big global scope. Scoping is also better than putting labels into a list yourself. In KickAss the scoping solution works well for accessing labels in executed macro calls, and I'am planning to expand the functionality to loops as well. So accessing the label of one loop from another could look like this:
.const count=25
loop1: .for(var i=0; i<count; i++) {
lda #i
sta loop2[i].color+1
}
loop2: .for(var i=0; i<count; i++) {
color: lda #0
sta $d020
}
Today you have to put your labels in a list yourself, but I find the above better.
Edit: Scoping solution = you access the scope instance of a given iteration of the loop |
| |
soci
Registered: Sep 2003
Posts: 474 |
Example of doynax translated:
smc_store .segment name
\{name}_value = *+1
lda #$00
\{name}_addr = *+1
sta @w$0000
.endm
*= $1000
#smc_store test
sta test_value
But don't do it, this is evil ;)
smc_store .macro
value = *+1
lda #$00
addr = *+1
sta @w$0000
.endm
*= $1000
test #smc_store
sta test.value
Somewhat better.
Hypnosis: Later 64tass versions tolerate "wrong" definition and use ordering.
*= $1000
.1000 4c 06 10 jmp $1006 jmp l[0]
.1003 4c 07 10 jmp $1007 jmp l[1]
=[$1006,$1007] l = [b1, b2]
.1006 ea nop b1 nop
.1007 ea nop b2 nop
Unsurprisingly it has suboptimal performance compared to proper ordering.
Slammer: I was thinking about this sort of list generation some time ago for .rept when it was still hard to implement it. Now it'd be much easier, but I can't think of a compelling use for it in my sources (yet). Even if it'd be available now probably I'd still write the above example like this:
*= $1000
count = 25
.for i = 0, i < count, i = i + 1
lda #i
sta (+)+1
.section speedcode
+ lda #0
sta $d020
.send
.next
.dsection speedcode ; right after the modifier
This compiles the two code parts in tandem so an anonymous label is good enough. Results in:
.1000 a9 01 lda #$00 lda #i
.1002 8d 7e 10 sta $107e sta (+)+1
.107d a9 00 lda #$00 + lda #0
.107f 8d 20 d0 sta $d020 sta $d020
.1005 a9 01 lda #$01 lda #i
.1007 8d 83 10 sta $1083 sta (+)+1
.1082 a9 00 lda #$00 + lda #0
.1084 8d 20 d0 sta $d020 sta $d020
...
In case of more labels it'd name them properly and stick in a .block scope to avoid the symbol collisions. |
| |
Hypnosis
Registered: Mar 2015
Posts: 36 |
Interesting use of sections! |
| |
Hypnosis
Registered: Mar 2015
Posts: 36 |
Quoting SlammerScoping solution = you access the scope instance of a given iteration of the loop
Ah, that was the missing part in my puzzle. I haven't seen this before. |
| |
Slammer
Registered: Feb 2004
Posts: 416 |
Soci: I prefer the scoping solution so I will stick to that, but I guess its a matter of taste. And nobody says you can't do both. Seems like we pretty much agree on scoping for macro executions...
.macro smc_store() {
.label value =*+1
lda #$00
.label addr = *+1
sta.a $0000
}
*= $1000
test: smc_store()
sta test.value |
| |
Peacemaker
Registered: Sep 2004
Posts: 243 |
at the moment i doing it like this with kickass:
.const count=25
loop1: .for(var i=0; i<count; i++) {
lda #i
sta loop2+1+i*5
}
loop2: .for(var i=0; i<count; i++) {
color: lda #0
sta $d020
} |
| |
Oswald
Registered: Apr 2002
Posts: 5022 |
Quote: at the moment i doing it like this with kickass:
.const count=25
loop1: .for(var i=0; i<count; i++) {
lda #i
sta loop2+1+i*5
}
loop2: .for(var i=0; i<count; i++) {
color: lda #0
sta $d020
}
lda #i
ldy #$01
sta (fe),y
lda fe
clc
adc #$05
sta fe
bcc *+4
inc ff |
| |
Peacemaker
Registered: Sep 2004
Posts: 243 |
thanks oswald, but that is slower =) |
| |
Oswald
Registered: Apr 2002
Posts: 5022 |
no, thats the generator. just wanted to show how simple is it you macro nerds :) |
| |
Compyx
Registered: Jan 2005
Posts: 631 |
But what are 'fe' and 'ff', macro's? ;) |
| |
chatGPZ
Registered: Dec 2001
Posts: 11133 |
those are not macros, those are labels! |
