HomeForumsWhat's newResources 
 
 
Reading from the framebuffer?
XL2 - Aug 23, 2018

 Prev  1  2  3  4  Next 

   XL2 Dec 3, 2018 
And it seems like it works!
I haven't tried on real hardware, but it should work too.
In these pictures, only the "sparkles" are using VDP2 transparency so it might be a bit hard to see.
It's also very low definition at just 160x112 so I guess it will look better with flat polygons and gouraud shading.

The only remaining issue is the scale as I just want to scale x by 0,455 and y 0.5 to have everything fit where it should go, but simply scaling my matrix will lead to issues (I just want screen space scaling).
Maybe I'll just manualy fix the draw commands or manualy generate draw commands, but it would be better to avoid always "patching" SGL and redo everything it does.

Edit : Added another picture with a basic box to better show how the transparency works on both VDP2 layers and VDP1. But the scale is still wrong, I will try to find a way to fix it this week.

   mrkotfw Dec 4, 2018 
Excellent work.

This is a bit off topic, but I wonder if it's possible to use the SCU-DSP for software translucency (similar to PlayStation blending modes).

Possibly perform it in 8x8 or 16x16 "tiles". You have two tiles in each bank, and DMA the results to VDP2.

   XL2 Dec 5, 2018 

  mrkotfw said:

You mean using software rendering? I already wrote a simple software renderer using lines for transparency before trying the current technique.
The line technique is looking quite good but it creates too many draw commands.
The SCU could be useful to create the edges, but it has too limited memory for anything ambitious.
You can of course use different transparency ratios on the NBG0 or NBG1 layer, but it will apply on the whole layer.
I never tried per pixel color ratios, but that could be another solution using the BR technique for different blending values.
The only negative points are the sorting and the low definition.
For the speed, it seems to be quite fast with little impact on the framerate, but I need to use custom draw commands as it doesn't play nice with SGL.

Edit : With custom draw commands and proper scaling. I could probably not scale up that much and accept some lost pixels on the side of the screen to avoid such large edges, but I don't know if I should.

   XL2 Dec 7, 2018 
Tested on real hardware and it works fine!
The framerate is mostly at 30 FPS, but it slows down more often to 20 than before (some optimizations would help improve that).

On a side note, could anyone give me a hand transforming some SGL functions?
It would be faster that using my own code since I could just use the whole SGL pipeline.
I just need to scale at screen space (again, x * 0.454545 and y * 0.5) after the transformation.
I know almost nothing about assembly atm, so could anyone point me where to look at?
Thanks!

Code:
  
sglC08.o: file format coff-sh Disassembly of section SLPROG: 00000000 <_slPutPolygon>: 0: e0 01 mov #1,r0 2: c0 73 mov.b r0,@(115,gbr) 4: c5 55 mov.w @(170,gbr),r0 6: 70 01 add #1,r0 8: c1 55 mov.w r0,@(170,gbr) a: c6 22 mov.l @(136,gbr),r0 c: 51 41 mov.l @(4,r4),r1 e: 63 03 mov r0,r3 10: c6 1d mov.l @(116,gbr),r0 12: 65 3d extu.w r3,r5 14: 62 0d extu.w r0,r2 16: 32 1c add r1,r2 18: 32 56 cmp/hi r5,r2 1a: 8d 46 bt.s aa <ptpl_09> 1c: 65 29 swap.w r2,r5 1e: 52 43 mov.l @(12,r4),r2 20: 43 29 shlr16 r3 22: 40 29 shlr16 r0 24: 30 2c add r2,r0 26: 30 36 cmp/hi r3,r0 28: 8d 3f bt.s aa <ptpl_09> 2a: 25 0d xtrct r0,r5 2c: d6 7b mov.l 21c <IMM_ComRdPtr>,r6 ! 260ffc44 2e: c6 12 mov.l @(72,gbr),r0 30: 63 62 mov.l @r6,r3 32: 67 03 mov r0,r7 34: 30 30 cmp/eq r3,r0 36: 8f 3a bf.s ae <ptpl_01> 38: c6 07 mov.l @(28,gbr),r0 3a: 17 11 mov.l r1,@(4,r7) 3c: 17 23 mov.l r2,@(12,r7) 3e: 66 03 mov r0,r6 40: c6 09 mov.l @(36,gbr),r0 42: 41 08 shll2 r1 44: 41 08 shll2 r1 46: 30 1c add r1,r0 48: c2 09 mov.l r0,@(36,gbr) 4a: e0 fc mov #-4,r0 4c: 24 09 and r0,r4 4e: 51 40 mov.l @(0,r4),r1 50: 52 42 mov.l @(8,r4),r2 52: 53 44 mov.l @(16,r4),r3 54: 21 09 and r0,r1 56: 22 09 and r0,r2 58: e0 fe mov #-2,r0 5a: 23 09 and r0,r3 5c: 17 12 mov.l r1,@(8,r7) 5e: 17 24 mov.l r2,@(16,r7) 60: 17 35 mov.l r3,@(20,r7) 62: e1 14 mov #20,r1 64: 17 10 mov.l r1,@(0,r7) 66: 60 53 mov r5,r0 68: c2 1d mov.l r0,@(116,gbr) 6a: 77 18 add #24,r7 6c: 50 60 mov.l @(0,r6),r0 6e: 51 61 mov.l @(4,r6),r1 70: 52 62 mov.l @(8,r6),r2 72: 53 63 mov.l @(12,r6),r3 74: 17 00 mov.l r0,@(0,r7) 76: 17 11 mov.l r1,@(4,r7) 78: 17 22 mov.l r2,@(8,r7) 7a: 17 33 mov.l r3,@(12,r7) 7c: 50 64 mov.l @(16,r6),r0 7e: 51 65 mov.l @(20,r6),r1 80: 52 66 mov.l @(24,r6),r2 82: 53 67 mov.l @(28,r6),r3 84: 17 04 mov.l r0,@(16,r7) 86: 17 15 mov.l r1,@(20,r7) 88: 17 26 mov.l r2,@(24,r7) 8a: 17 37 mov.l r3,@(28,r7) 8c: 50 68 mov.l @(32,r6),r0 8e: 51 69 mov.l @(36,r6),r1 90: 52 6a mov.l @(40,r6),r2 92: 53 6b mov.l @(44,r6),r3 94: 17 08 mov.l r0,@(32,r7) 96: 17 19 mov.l r1,@(36,r7) 98: 17 2a mov.l r2,@(40,r7) 9a: 17 3b mov.l r3,@(44,r7) 9c: e0 30 mov #48,r0 9e: 30 7c add r7,r0 a0: d1 5f mov.l 220 <IMM_FRT_INTR>,r1 ! 21000000 a2: c2 12 mov.l r0,@(72,gbr) a4: 21 01 mov.w r0,@r1 a6: 00 0b rts a8: e0 01 mov #1,r0 [...] And so on...

   XL2 Dec 20, 2018 
(very) Slowly getting everything together.
On a side note, gouraud shading on the Saturn simply looks amazing, even with OpenGL's default gouraud shading settings you can't replicate such amazing lighting.
I ditched the color bank sprites for RGB sprites only to allow better lighting.
I think it's worth it as long as I can keep up the draw distance.
While I still have a ton of things to do on the engine, hopefully next year I'll be able to add dynamic lighting.

Here are some shots in SSF.

   David Gámiz Jiménez Dec 21, 2018 
Seriously XL2.

You are amazing. Little by little, you are getting very far away! Congratulations!! Again XD

With respect to penalize drawing distance. You know what will happen, it is expensive to calculate. But you also know, that you have room to optimize. Just like you can apply a LOD by distance, for vertices and textures. You can scale the lighting detail by distances. From Gouraud with four vertices to a flat face. That affects several local lights to only one or only pre-calculated light. Finally for geometry in the far-clipping directly without shading, since with so few pixels hardly notice the effect. Adding up everything, you can compensate to have that magnificent light closer to the camera.

We already know, that all this will depend on your implementation of BSP or similar that you finally do. And that is a lot of work. Even, really, I do not know where SGL (closed) will let you do all this, to be able to optimize even more. It is clear that your creativity has no limit! Cheer up!

For anything you know that you can count, humbly, with my help.

Regards!

PD: It is difficult see in the screenshots. Is possible the Local Light are sum to main Paralell Light source in Sonic model?

   XL2 Dec 21, 2018 

  David Gmiz Jimnez said:

About possible slowdowns, I meant mainly with the gouraud shading adding something like 20 or 25% rendering time, which is quite a lot.
About the CPU, it can always be optimized, so I'm not too worried.
I will try to use the current octree implementation and add a PVS to it.
It should work as I can already build the BSP tree on the computer (I use it for lighting), so I just need to raycast to find visible nodes/leaves.
But that raycasting will be slow, so I might need to speed it up.
It won't eliminate overdraw, but it should reduce it quite a bit, even if it's useless to solve the problem with VDP2 transparent sprites.
That octree/pvs system won't be as efficient as a BSP/portal system obviously, but it's simplier for now to reuse what I have and it will work with Z-Treme's maps, so I can quickly test.
About the light on Sonic, it's just a "demo" light that turns around, I haven't implemented local lights on the Saturn side (only while converting on PC).
For the dynamic lighting, I will probably just use it on the high quality model, not on the low quality model. I want to try something like that :

https://imgur.com/Zsx9QzR



But that's something that won't happen for a long time as I have many more important things to do on this project.

EDIT : I did a quick test on real hardware (gouraud shading on everything + BR-style transparency) and the game still ran at 20-30 fps. There is room for improvements, hopefully I can reach a stable 30 fps will all these effects.

   luisoft Jan 4, 2019 
Impressive work!! Hey , you are working in framebuffer effects right? Its hard do a similar effect like Tomb Raider save state from ps1?
https://youtu.be/BltNSs9Vzdg

1:25:50 This effect copy the framebuffer to a polygon and put a purple pallete.

   XL2 Jan 4, 2019 

  luisoft said:


Yes and no. The problem is the lack of texture coordinates, which limits what you could do.

   mrkotfw Jan 7, 2019 
Sonic R did the above entirely in software.

   David Gámiz Jiménez Jan 7, 2019 
Sonic R not make exactly the same. The texture reflection is static.
In other hand. Tomb Raider in PS1 not use a UV refraction mapping, only a planar proyection of the previous frame buffer how texture plus inked and light whit gouraud source light.
All this is possible in SS, in fact actually the SS version had the inked Gouraud sourced part. The miss part is the frame buffer texture, possible like XL2 made it. And last the UV planar protection, RockinB made it a long time.
All is possible in SS, tricky but possible.

   luisoft Feb 20, 2019 
News in your engine? Can i download the last build? I really like this project Burning Rangers transparency is really cool!
Do you test multiplicative lighting to see if run better than additive?

   XL2 Feb 20, 2019 

  luisoft said:

Nothing new, I didn't have much time recently, but I put back collision detection and removed the framebuffer clearing to gain a bit more rendering speed.
I'm not sure I'll keep the BR style transparency because of limited ram and vram, and it's also slower.
About multiplicative lighting, it's one option right now for smoother shading, I know it can probably be done using lookup tables, but normal additive lighting isn't bad either.
The problem is that it would kill vdp1 half transparency, which was my backup plan instead of BR-style transparency (by using lines to rasterize polygons).

   XL2 Mar 29, 2019 
Still no name nor any gameplay to show, but here is some footage from last month's version.
The textures aren't very good and they are too dark and I didn't put enough lights so it's hard to see.
I didn't have much time recently so I haven't made much progress, but I am currently working on a Quake .map converter, which should allow me to make maps faster and help with moving entities/switches/doors.
Anyway, it runs at 30 fps, but I still have visual glitches that I will try to fix once I am done with the converter.

https://youtu.be/fsaj5g7p3fQ


   luisoft Mar 29, 2019 
I see this video! Really impressive!!! The texture mapping is really nice in this demo! And running at solid 30fps show the power of your engine. This small glitches is not really a big problem the demo is really nice!! I love one day get your code and study hard this game and the awesome Sonic Z. Using bsp tree speed up the Sonic Z game? Not clearing the framebuffer cause visual glitches?

   XL2 Mar 29, 2019 

  luisoft said:

Not clearing the framebuffer doesn't cause glitches if you always overwrite all the pixels, which is what I do.
The glitches must be related to dma or cache, I will work on that later.
As for Sonic Z-Treme, the maps wouldn't work with a bsp tree and I don't see how I could ever make it work, so this project is on hold, most likely cancelled.
The bsp helps mainly for visibility calculation (pvs, portals), which is very nice for indoor maps, but not so much for outdoor maps.

   luisoft Mar 30, 2019 
All times i turn on my saturn i NEED play Sonic Z many times! Its a dream play a sonic3d into saturn hardware. "this project is on hold, most likely cancelled" man not give up your project is really nice. I understand its hard create big levels because of saturn ram size, Sonic Xtreme team in past not have saturn in mind when create the xtreme. If you put Sonic Z as a Scroll Plane game will be much better to extract power of the saturn.
https://youtu.be/IUXuZlexwi4

A similar game can be done with saturn using 2 scroll planes one for water and other for the level and can use your nice sonic 3d model.

   XL2 Apr 16, 2019 
Here is a demo for lighting using lightmaps and alpha blending on the VDP2. It's still the same framebuffer trick I mentionned before. The edges are sadly a bit too rough, so they don't fully cover the cube, but if it was an interior map with the screen filled with polygons it would be OK.

   luisoft Apr 16, 2019 
Impressive work again! This texture mapping is software? Your engine is really nice and get better and better all days. Nice Work man!

   XL2 Apr 16, 2019 
It's all made by the hardware.


 Prev  1  2  3  4  Next