![]() |
|
non-paged supervisor code calling paged application code - Printable Version +- Boriel Basic Forum (https://forum.boriel.com) +-- Forum: Compilers and Computer Languages (https://forum.boriel.com/forumdisplay.php?fid=12) +--- Forum: ZX Basic Compiler (https://forum.boriel.com/forumdisplay.php?fid=11) +---- Forum: Help & Support (https://forum.boriel.com/forumdisplay.php?fid=16) +---- Thread: non-paged supervisor code calling paged application code (/showthread.php?tid=2651) |
non-paged supervisor code calling paged application code - sdo303 - 2026-02-15 Firstly, a proper thankyou to Boriel for the compiler. And secondly, also a proper thankyou to Duefectu for the beginner's guide book (I bought a hard copy), which I have read cover to cover (but I did skip the sound/audio sections ). And I bought a hard copy of Simon G's NextTech book too, again I have read it cover to cover (but skipped the sound/audio parts).Using Boriel ZXBasic and as little Z80 machine code as I can get away with (until I need to performance tune)... my project is that I am trying to write a quasi "character cell" based application, like being on an old VT100, whereby the application uses all 64KB of the Z80 for itself, and thus no ZX Spectrum ROM, and which internally leverages a poor man's virtual memory framework, whereby I have a supervisor layer of non-paged code running in MMU0/1, which boots from $0000 and jumps over IM1 at $0038 to begin my main code properly at say $0050 where the application begins properly... and so I have my core non-paged system variables mapped in MMU2 and then I have my paged code to run from MMU3 - and then various different aspects of paged application data paging in and out MMU4/5/6, with finally MMU7 again being static non-paged to host the 5KB of screen tile map for an 80x32 screen, with the final 3KB of MMU7 being available for the stack pointer from $FFFE downwards. phew... I know my code will never work first time, I get that... but I feel fairly sure that I have the basic (excuse the pun) initial structure of what I want to achieve, with all of my system variables (so far identified) defined at static addresses in MMU2, and my main supervisor MMU paging and also my screen updating routines and keyboard reading routines written (to be permanently resident in MMU0/1) and the first of my paged blocks of code to run from MMU3, and I have gotten the source code clean enough pass through zxbc without the compiler spitting any hard errors... (just lots and lots of warnings)... but I've now hit a roadblock... I get how I can have many separate pages of 8KB deposited across the ZX Next vast 200+ pages of 8KB across 1.75 MB of RAM, and I get how I can have my code paged in at MMU3 for my application to execute from $6000, and I can get my non-paged supervisor layer code to make calls such as uw_dummy = USR $6000 to call my paged code in MMU3, and thus wait for a return... but what I don't understand is how I can have many different paged code blocks where each is aware of *ALL* of the available core supervisor routines and their parameter structures and thus able to makes calls back in to the supervisor to run any of the many core system routines... i.e. how can I make zxbc firstly compile the application core without dropping routines that zxbc thinks are unused, and yet also somehow get the compilations of many different "paged" application code to be aware of the core supervisor routines and their parameters and any return value data types (of functions). e.g. I get this from my compilation of my "core.bas": core.bas:115: warning: [W170] Function 's_crash' is never called and has been ignored core.bas:154: warning: [W170] Function 's_fc_ram_copy' is never called and has been ignored core.bas:165: warning: [W170] Function 's_clear_screen' is never called and has been ignored core.bas:126: warning: [W170] Function 's_print_char_at' is never called and has been ignored core.bas:143: warning: [W170] Function 's_print_string_at' is never called and has been ignored core.bas:217: warning: [W170] Function 's_display_monitor' is never called and has been ignored core.bas:283: warning: [W170] Function 'fub_key_get_minutes' is never called and has been ignored ...and I get this for my first stage of my application: stage-include.bas:37: warning: [W170] Function 's_next_reg' is never called and has been ignored stage-include.bas:38: warning: [W170] Function 's_return_to_zxos' is never called and has been ignored stage-include.bas:39: warning: [W170] Function 's_crash' is never called and has been ignored stage-include.bas:40: warning: [W170] Function 's_fc_ram_copy' is never called and has been ignored stage-include.bas:42: warning: [W170] Function 's_clear_screen' is never called and has been ignored stage-include.bas:43: warning: [W170] Function 's_print_char_at' is never called and has been ignored stage-include.bas:44: warning: [W170] Function 's_print_string_at' is never called and has been ignored stage-include.bas:45: warning: [W170] Function 's_display_monitor' is never called and has been ignored stage-include.bas:47: warning: [W170] Function 'fub_key_get_minutes' is never called and has been ignored stage-include.bas:48: warning: [W170] Function 's_key_wait_for_vblank' is never called and has been ignored stage-include.bas:49: warning: [W170] Function 's_get_key' is never called and has been ignored ... stage-init.bas:8: warning: [W170] Function 's_stage_init' is never called and has been ignored stage-init.bas:20: warning: [W170] Function 's_init_keyboard_table_ZX_keys' is never called and has been ignored stage-init.bas:21: warning: [W170] Function 's_init_keyboard_table_ZX_symkeys' is never called and has been ignored stage-init.bas:18: warning: [W170] Function 's_init_load_palette' is never called and has been ignored stage-init.bas:186: warning: [W170] Function 's_init_load_palette_entry' is never called and has been ignored stage-init.bas:223: warning: [W170] Function 's_init_clear_screen' is never called and has been ignored stage-init.bas:26: warning: [W170] Function 's_init_title_screen' is never called and has been ignored ...where stage-include.bas was my attempt to make each "stage" aware of a jump-table of locations of core routines: Asm ; fixed jump table addresses for the linker _s_main equ $0050 _s_next_reg equ $0053 _s_return_to_zxos equ $0056 _s_crash equ $0059 _s_fc_ram_copy equ $005C _s_clear_screen equ $005F _s_print_char_at equ $0062 _s_print_string_at equ $0065 _s_display_monitor equ $0068 _fub_key_get_minutes equ $006B _s_key_wait_for_vblank equ $006E _s_get_key equ $0071 End Asm ' implementation blocks to stop "not implemented" errors ' these contain no code; the equate above handles the address Sub s_main(): End Sub Sub s_next_reg(reg As UByte, value As UByte): End Sub Sub s_return_to_zxos(): End Sub Sub s_crash(error_code As UByte, error_text As String): End Sub Sub Fastcall s_fc_ram_copy(src_address As UInteger, dest_address As UInteger, byte_count As UInteger): End Sub Sub s_clear_screen(): End Sub Sub s_print_char_at(x As UByte, y As UByte, char As UByte, colour As UByte): End Sub Sub s_print_string_at(x As UByte, y As UByte, text As String, colour As UByte): End Sub Sub s_display_monitor(): End Sub Function fub_key_get_minutes() As UByte: return 0: End Function Sub s_key_wait_for_vblank(): End Sub Sub s_get_key(): End Sub But clearly there is something I'm not getting, and I've reached the limits of capability of AI helper tools. I had rationalised to myself that my build script should only call to zxbc to produce "-f asm" ASM files... as I expect to have to then somehow "link" the static core ASM and the many paged stage ASM together at some point... and I also seem to have rationalised that I may perhaps need to use sjasmplus to assemble my ".nex" package. Does the above sound achievable? Or am I asking too much of the "Next" as a platform? i.e. is the "Next" only any good for paging graphics data, and no good for paging application code? RE: non-paged supervisor code calling paged application code - sdo303 - 2026-02-15 ok - so I have managed to get the compilation (to ASM) of my "core.bas" supervisor to no longer drop the routines (as uncalled) by getting my s_main() routine to call this dummy "do nothing" routine: Sub s_super_keeper() Dim p, q, swap As UInteger p = @s_next_reg p = @s_return_to_zxos p = @s_crash p = @s_fc_ram_copy p = @s_clear_screen p = @s_print_char_at p = @s_print_string_at p = @s_display_monitor p = @fub_sys_get_minutes p = @s_wait_for_vblank p = @s_get_key swap = p p = q q = swap End Sub ...with a final variable swap to stop the compiler from noticing that 'p' does nothing genuinely useful... ...now back to trying to determine how to make calls back to routines in the supervisor.... hmmm. RE: non-paged supervisor code calling paged application code - boriel - 2026-02-16 A trick (for the moment) to mark a function / sub as "used", is to do something like: Code: DIM dummy_var As UInteger: REM will be dropped by the optimizerRE: non-paged supervisor code calling paged application code - sdo303 - 2026-02-16 That worked nicely - as a method for tricking the compiler optimization to not discard what appears to be (as yet) unused routines in the core supervisor. Thank you for that. And I've been dreaming/thinking of a way of using shims for the "core" subs and functions, in the paged code setions, and I'll use the same trick to keep all the shims present in the paged code sections, and to quieten down the compiler optimisation discard warnings. When I've got a decent demo working of core performing multiple (at least two) MMU page swaps for paged code sections, and the core calling those paged sections, and then having those paged code sections make calls back to general purpose routines in the core... when I have this working then I will share all the code and build scripts and config in case anyone else ever wants to try anything similar. RE: non-paged supervisor code calling paged application code - boriel - 2026-02-17 This is something also being worked on... for the moment you can call "submodules" compiled in paged memory. @Duefectu has documented how to do it. RE: non-paged supervisor code calling paged application code - sdo303 - 2026-02-17 (2026-02-17, 08:28 AM)boriel Wrote: This is something also being worked on... for the moment you can call "submodules" compiled in paged memory. @Duefectu has documented how to do it. Thank you. I've gotten my code to compile to asm... Code: PS D:\zxnext\projects\demo> .\build.ps1...using thisPowerShell: Code: $ErrorActionPreference = 'Stop'So, I have four new ".asm" files... ...does anyone have any tips on how I can assemble them join them together in to some kind of '.nex' package ? Please thank you!
RE: non-paged supervisor code calling paged application code - sdo303 - 2026-02-18 so, now my header of "core.bas" is: Code: #PRAGMA push(case_insensitive)...and my sjasmplus linker is : Code: DEVICE ZXSPECTRUMNEXTbut for some reason I still see the normal Spectrum ROM in zesarux : Code: ¤ Debug CPU (step)I've tried for literally hours in a loop with Gemini to unmap the Spectrum ROM and get my code at page 50 to map to MMU0. Could anyone please help me? Thank you! RE: non-paged supervisor code calling paged application code - sdo303 - 2026-02-19 well I changed the sjasmplus linker to attempt to run a short piece of ASM from $8000 and I set the SP to $BEEF and I can see in zesarux that the SP in indeed $BEEF so.... we know that my project.nex is making it to be started but for some reason zesarux is stubbornly doing two things... 1) holding on to speccy ROM 2) but more tellingly zesarux is not executing my code at $8000 whose job is this : named: load.bas Code: Asmhere's my build script named "build.ps1" : Code: $ErrorActionPreference = 'Stop'here's my sjasmplus linker named "project.cfg" Code: DEVICE ZXSPECTRUMNEXThere's my zesarux.cfg file : Code: --machine tbblueso I still cannot fathom a way to get the speccy ROM out of the way :/ RE: non-paged supervisor code calling paged application code - sdo303 - 2026-02-20 well, I managed to get my bas to compile to bin, and to get sjasmplus to create a nex file, and it seems to be ok in the "nex" viewer extension in vscode... but can I get my nex file to take over the machine... no way... many hours... and I'm still stuck with zesarux refusing to let me take over MMU0..5 the best I can get is : MMU=80008001000a000b0004000500360037 in the zesarux debug. |