Asmusr Posted April 14 Share Posted April 14 (edited) This is my port of the ZX Spectrum game 'Through the Trap Door' to the TI-99/4A using partially automated Z80 to TMS9900 conversion. Unfortunately it wasn't possible to obtain the same frame rate as on the Spectrum, so shortly into the video I switch the emulator to 2x speed. I'm not sure I actually enjoy the game - it's very hard and sometimes frustrating - but here we go: The source code is quite advanced with its own byte code graphics language similar to GPL, and it's using a double buffer system in CPU RAM. In each frame: It clears the active screen buffer in CPU RAM. It drawn all the elements in the scene to the active screen buffer (also walls and floors) in two layers using the GPL-like language mentioned above. It iterates through the active screen buffer and compares with the passive screen buffer, sending only changed characters to the VDP memory. That sounded like a good scheme for the 9918A because the CPU to VDP transfer would be minimized, but in the game the screen buffer is 3*768 bytes long and actually both point 2 and 3 take about 500,000 clock cycles, which results in about 3 frame per second. To really bring the game up to playable speed, I would have to cut that down to half. After the automatic conversion of the original Z80 code to TMS9900 code, I had to fill up all memory on the TI, also the >4000 and the >6000 regions, to be able to run the code under emulation. That's a problem with automatic conversion: everything grows bigger, it's only by lengthy peep optimization that the code shrinks again. But, as one of the last step, I split the code into game levels and added ROM banks for each level and ended up with a little bit of RAM free. Source code: https://github.com/Rasmus-M/trapdoor trapdoor8.bin Edited April 15 by Asmusr 22 3 Quote Link to comment Share on other sites More sharing options...
Elia Spallanzani fdt Posted April 14 Share Posted April 14 The video is private 3 Quote Link to comment Share on other sites More sharing options...
jrhodes Posted April 14 Share Posted April 14 2 hours ago, Elia Spallanzani fdt said: The video is private Ditto. Glad to know its not just me getting that. 2 Quote Link to comment Share on other sites More sharing options...
RickyDean Posted April 15 Share Posted April 15 6 hours ago, Asmusr said: This is my port of the ZX Spectrum game 'Through the Trap Door' to the TI-99/4A using partially automated Z80 to TMS9900 conversion. Unfortunately it wasn't possible to obtain the same frame rate as on the Spectrum, so shortly into the video I switch the emulator to 2x speed. I'm not sure I actually enjoy the game - it's very hard and sometimes frustrating - but here we go: https://youtu.be/Yz_NefvU2Gg The source code is quite advanced with its own byte code graphics language similar to GPL, and it's using a double buffer system in CPU RAM. In each frame: It clears the active screen buffer in CPU RAM. It drawn all the elements in the scene to the active screen buffer (also walls and floors) in two layers using the GPL-like language mentioned above. It iterates through the active screen buffer and compares with the passive screen buffer, sending only changed characters to the VDP memory. That sounded like a good scheme for the 9918A because the CPU to VDP transfer would be minimized, but in the game the screen buffer is 3*768 bytes long and actually both point 2 and 3 take about 500,000 clock cycles, which results in about 3 frame per second. To really bring the game up to playable speed, I would have to cut that down to half. After the automatic conversion of the original Z80 code to TMS9900 code, I had to fill up all memory on the TI, also the >4000 and the >6000 regions, to be able to run the code under emulation. That's a problem with automatic conversion: everything grows bigger, it's only by lengthy peep optimization that the code shrinks again. But, as one of the last step, I split the code into game levels and added ROM banks for each level and ended up with a little bit of RAM free. Source code: https://github.com/Rasmus-M/trapdoor trapdoor8.bin 64 kB · 16 downloads 1 Quote Link to comment Share on other sites More sharing options...
Asmusr Posted April 15 Author Share Posted April 15 I have fixed the video. 1 Quote Link to comment Share on other sites More sharing options...
tmop69 Posted April 15 Share Posted April 15 @Asmusr the approach used for the conversion is really interesting and intriguing. This should significantly reduce the time for 1:1 conversions from Spectrum to TI99. 🙂 Are you thinking to try to apply this method to MSX1 or Colecovision porting (same Speccy's CPU)? Do you think that, in these cases, the use of the same GPU of the TI could improve the final frame rate? The result of this specific game conversion is a really slow game that does not fit too well with the game category (action/platform). But it should not impact games like Lords of Midnight and The Hobbit, that surely will be very appreciated by the 99ers. Maybe with games with less gfx on screen it will be possible to obtain a more decent frame rate. I'm thinking of 16K games like Arcadia, Horace Goes Skiing, Pssst, etc. Will you try with some other game to see the results? Could you, finally, explain more the technique used? Is the conversion executed at runtime on the TI or on the PC, etc. Thanks. Quote Link to comment Share on other sites More sharing options...
sometimes99er Posted April 15 Share Posted April 15 Nice technical achievement ! 👍 I remember seeing that game BITD. Nice graphics. I thought it was the Cookie Monster from Sesame Street, but apparently Berk is based on a British television show of the same name. I think Cookie Monster came first.https://en.wikipedia.org/wiki/The_Trap_Door_(video_game)https://en.wikipedia.org/wiki/The_Trap_Doorhttps://en.wikipedia.org/wiki/Cookie_Monster Quote Link to comment Share on other sites More sharing options...
Asmusr Posted April 15 Author Share Posted April 15 4 hours ago, tmop69 said: @Asmusr the approach used for the conversion is really interesting and intriguing. This should significantly reduce the time for 1:1 conversions from Spectrum to TI99. 🙂 Are you thinking to try to apply this method to MSX1 or Colecovision porting (same Speccy's CPU)? Do you think that, in these cases, the use of the same GPU of the TI could improve the final frame rate? The result of this specific game conversion is a really slow game that does not fit too well with the game category (action/platform). But it should not impact games like Lords of Midnight and The Hobbit, that surely will be very appreciated by the 99ers. Maybe with games with less gfx on screen it will be possible to obtain a more decent frame rate. I'm thinking of 16K games like Arcadia, Horace Goes Skiing, Pssst, etc. Will you try with some other game to see the results? Could you, finally, explain more the technique used? Is the conversion executed at runtime on the TI or on the PC, etc. Thanks. The conversion takes place on a PC, and to be honest there is a lot of work to do afterwards. In particular, everything to do with graphics, sound and input hardware have to be rewritten. Still, it would take a lot longer to develop a game like Through the Trap Door from scratch. Commented source code, or more likely a commented disassembly, is a requirement, and on the Spectrum there are many examples using SkoolKit. I'm not aware of anything similar for the MSX1 or Colecovision. PSSST, The Hobbit, and Lords of Midnight all have disassemblies for the Spectrum, but I'm picking another game for my next project. Something that should run fast enough, and a genre I think the TI community would enjoy. There is even a chance of improved graphics, music and speech. 7 3 Quote Link to comment Share on other sites More sharing options...
+Retrospect Posted April 16 Share Posted April 16 One of the "Dizzy" games might have been a nice game to try and convert. In some of those games, hardly anything moves on the screen apart from the Dizzy character itself , repeatedly flapping it's arms up and down. Mind, all of the Dizzy games ran on a 48K Spectrum. Quote Link to comment Share on other sites More sharing options...
Asmusr Posted April 16 Author Share Posted April 16 7 hours ago, Retrospect said: One of the "Dizzy" games might have been a nice game to try and convert. In some of those games, hardly anything moves on the screen apart from the Dizzy character itself , repeatedly flapping it's arms up and down. Mind, all of the Dizzy games ran on a 48K Spectrum. Yes it looks like a good game to port, but no source code, I'm afraid. 1 Quote Link to comment Share on other sites More sharing options...
OX. Posted April 21 Share Posted April 21 Nice conversion Rasmus, keep them coming! 1 Quote Link to comment Share on other sites More sharing options...
+arcadeshopper Posted April 22 Share Posted April 22 very interesting game! have you considered porting msx stuff? Quote Link to comment Share on other sites More sharing options...
Asmusr Posted April 22 Author Share Posted April 22 3 hours ago, arcadeshopper said: very interesting game! have you considered porting msx stuff? Not without source code or commented disassembly. Quote Link to comment Share on other sites More sharing options...
Tursi Posted April 23 Share Posted April 23 I wonder if we can analyze the final resulting code, and maybe come up with a handful of peephole optimizations that could be automatically applied to improve performance? 3 Quote Link to comment Share on other sites More sharing options...
Asmusr Posted April 27 Author Share Posted April 27 On 4/23/2023 at 9:36 PM, Tursi said: I wonder if we can analyze the final resulting code, and maybe come up with a handful of peephole optimizations that could be automatically applied to improve performance? Here are a few examples for consideration: movb *ix,a ; LD A,(IX+$00) ; inc ix ; INC IX ; movb @timer,a ; LD A,($DA14) ; sb one,a ; DEC A ; movb a,@timer ; LD ($DA14),A ; .push hl ; PUSH HL ; movb @hit,a ; LD A,($D300) ; li hl,frame_ref ; LD HL,$CE1A ; cb a,*hl ; CP (HL) ; .pop hl ; POP HL ; dec de ; DEC DE ; movb d,a ; LD A,D ; socb @e,a ; OR E ; jne start_loop ; JR NZ,$C4FD ; a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; inc hl ; INC HL ; inc hl ; INC HL ; inc hl ; INC HL ; inc hl ; INC HL ; cb a,*hl ; CP (HL) ; dec hl ; DEC HL ; dec hl ; DEC HL ; dec hl ; DEC HL ; dec hl ; DEC HL ; li de,prep_open ; LD DE,$D241 ; Take care of little/big endian movb *ix,@l ; LD L,(IX+$00) ; movb @1(ix),h ; LD H,(IX+$01) ; movb @e,*hl ; LD (HL),E ; inc hl ; INC HL ; movb d,*hl ; LD (HL),D ; li hl,routine_1 ; LD HL,$CE14 ; movb *hl,a ; LD A,(HL) ; inc hl ; INC HL ; socb *hl,a ; OR (HL) ; .push af ; PUSH AF ; Does not affect flags on Z80 jne somewhere ; CALL Z,$CE21 ; Take care of flags 1 Quote Link to comment Share on other sites More sharing options...
sometimes99er Posted April 27 Share Posted April 27 (edited) 34 minutes ago, Asmusr said: Here are a few examples for consideration: movb @timer,a ; LD A,($DA14) ; sb one,a ; DEC A ; movb a,@timer ; LD ($DA14),A ; dec @timer ; LD A,($DA14) ; ; DEC A ; ; LD ($DA14),A ; I guess timer has to be a word then. Edited April 27 by sometimes99er Quote Link to comment Share on other sites More sharing options...
sometimes99er Posted April 27 Share Posted April 27 44 minutes ago, Asmusr said: Here are a few examples for consideration: inc hl ; INC HL ; inc hl ; INC HL ; inc hl ; INC HL ; inc hl ; INC HL ; cb a,*hl ; CP (HL) ; dec hl ; DEC HL ; dec hl ; DEC HL ; dec hl ; DEC HL ; dec hl ; DEC HL ; inct hl ; INC HL ; ; INC HL ; inct hl ; INC HL ; ; INC HL ; cb a,*hl ; CP (HL) ; dect hl ; DEC HL ; ; DEC HL ; dect hl ; DEC HL ; ; DEC HL ; 1 Quote Link to comment Share on other sites More sharing options...
sometimes99er Posted April 27 Share Posted April 27 (edited) 49 minutes ago, Asmusr said: Here are a few examples for consideration: movb *ix,a ; LD A,(IX+$00) ; inc ix ; INC IX ; movb *ix+,a ; LD A,(IX+$00) ; ; INC IX ; Edited April 27 by sometimes99er 1 Quote Link to comment Share on other sites More sharing options...
Asmusr Posted April 27 Author Share Posted April 27 51 minutes ago, Asmusr said: inc hl ; INC HL ; inc hl ; INC HL ; inc hl ; INC HL ; inc hl ; INC HL ; cb a,*hl ; CP (HL) ; dec hl ; DEC HL ; dec hl ; DEC HL ; dec hl ; DEC HL ; dec hl ; DEC HL ; cb a,@4(hl) 1 Quote Link to comment Share on other sites More sharing options...
sometimes99er Posted April 27 Share Posted April 27 4 minutes ago, Asmusr said: cb a,@4(hl) Yep, I was coming to that. 😁 Quote Link to comment Share on other sites More sharing options...
sometimes99er Posted April 27 Share Posted April 27 (edited) 2 hours ago, Asmusr said: Here are a few examples for consideration: a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; a hl,hl ; ADD HL,HL ; sla hl,9 ; ADD HL,HL ; ; ADD HL,HL ; ; ADD HL,HL ; ; ADD HL,HL ; ; ADD HL,HL ; ; ADD HL,HL ; ; ADD HL,HL ; ; ADD HL,HL ; ; ADD HL,HL ; ADD HL,HL is 2x HL, which is bit-shifting one to the left. And this is done 9 times. Edited April 27 by sometimes99er 1 Quote Link to comment Share on other sites More sharing options...
+TheBF Posted April 27 Share Posted April 27 Nice work @sometimes99er. It's fun to see the 9900 instruction set doing it's job. 1 Quote Link to comment Share on other sites More sharing options...
sometimes99er Posted April 27 Share Posted April 27 (edited) 9 hours ago, Asmusr said: Here are a few examples for consideration: .push hl ; PUSH HL ; movb @hit,a ; LD A,($D300) ; li hl,frame_ref ; LD HL,$CE1A ; cb a,*hl ; CP (HL) ; .pop hl ; POP HL ; cb @hit,@frame_ref ; PUSH HL ; ; LD A,($D300) ; ; LD HL,$CE1A ; ; CP (HL) ; ; POP HL ; Edited April 27 by sometimes99er 2 Quote Link to comment Share on other sites More sharing options...
+FarmerPotato Posted April 27 Share Posted April 27 On 4/14/2023 at 12:52 PM, Asmusr said: That sounded like a good scheme for the 9918A because the CPU to VDP transfer would be minimized, but in the game the screen buffer is 3*768 bytes long and actually both point 2 and 3 take about 500,000 clock cycles, which results in about 3 frame per second. To really bring the game up to playable speed, I would have to cut that down to half. 16-bit optimizations... 1. Are the buffers compared byte to byte? Using word compare could be faster. 9900 CB has to fetch both bytes anyway! It might be ok to write both bytes to VDP, regardless of which changed. It depends on the probability of changing adjacent characters. Quite likely I think. 2. I have a hunch there is an XOR optimization to be had... 2 Quote Link to comment Share on other sites More sharing options...
+TheBF Posted April 27 Share Posted April 27 1 hour ago, sometimes99er said: cb @hit,@frame_ref ; PUSH HL ; ; LD A,($D300) ; ; LD HL,$CE1A ; ; CP (HL) ; ; POP HL ; Memory to memory architecture for the win. 3 Quote Link to comment Share on other sites More sharing options...
Like us on Facebook
Follow us on Twitter
AtariAge on Instagram
Watch us on YouTube
Chat on Discord
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.