Xargon/Supermodel
1
0
Fork 0
mirror of https://github.com/RetroDECK/Supermodel.git synced 2024-11-24 06:35:41 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity
1088 commits 1 branch 0 tags 78 MiB
Commit graph

52 commits

Author SHA1 Message Date
gm-matthew 3cf8a1fa94 Implementing texture NP values 
For some reason Model 3 uses vertex coordinates rather than texel coordinates to calculate mipmap levels
Revised microtexture implementation; results are very close to real hardware when running at native resolution with supersampling disabled
 2024-07-05 15:57:25 +01:00
Ian Curtis edb11dc223 Performance improvements 
The old texture code was being bottle necked by the texture reads. We mirrored the real3d texture memory directly, including the mipmaps in a single large texture. I *think* most h/w has some sort of texture cache for a 2x2 or 4x4 block of pixels for a texture. What we were doing was reading the base texture, then reading the mipmap data from a totally separate part of the same texture which I can only assume flushed this cache. What I did was to create mipmap chains for the texture sheet, then copy the mipmap data there. Doing this basically doubles performance.
 2024-06-13 13:36:30 +01:00
Ian Curtis c039d08c03 Add supersampling anti-aliasing 
Late christmas present. Due to the way alpha works on the model3 adding regular anti-aliasing doesn't really work. Supersampling is very much a brute force solution, render the scene at a higher resolution and mipmap it.

It's enabled via command line with the -ss option, for example -ss=4 for 4x supersampling or by adding Supersampling = 4 in the config file.

Note non power of two values work as well, so 3 gives a very good balance between speed and quality. 8 will make your GPU bleed, since it is essentially rendering 64 pixels for every visible pixel on the screen.
 2023-12-26 18:25:03 +00:00
gm-matthew 9f66fcaac7 Floating point reversed z-buffer and new clipping code 
Also always draw nodes with culling disabled even if they test as being outside the visible frustum
 2023-12-22 02:00:47 +00:00
Ian Curtis a214c6dae8 Prior scroll fog logic was nearly correct. For scroll fogging to draw it needs either a start value or a fog density value, but these can come from a different viewport if they have the same colour fog set. This fixes the credits in vf3 which sets scroll fog, but it never draws on the original hardware. 
For a long time we've had bug reports that in vf3 the background in the Dural levels was the wrong colour, it should have been much darker. We thought it was again missing scroll fog, but gm_mathew worked out it was actually coming from the fogAmbient paramater. Not only does it darken the fog, but it also has an effect on the 2d background from the tilegen, similar to scroll fog. This also fixes the sky in lemans24.
 2023-10-22 21:07:35 +01:00
Ian Curtis 6595b9320e Rewrite the renderer a bit to spit out the finished graphics from the 3d chip on separate buffers. One buffer is for opaque pixels, and 2 more for translucent pixels. Before the frame was composited on the back buffer, which meant the tilegen had to have been drawn first. This way the images are now totally independant of the tilegen chip so can be drawn as soon as the register write 0xC is written to the tilegen. 
Some games update the tilegen after the ping_ping bit has flipped at 66% of the frame, so we need to split the tilegen drawing up into two stages to get some effects to work. So having the tilegen draw independantly of the 3d chip can make this happen.
 2023-10-14 20:05:00 +01:00
Ian Curtis d656643087 gm_mathew worked out that the camera flashes in sega rally 2 have the 'reset-matrix' attribute set inside the culling nodes, which was unsupported until now because we never found any games to have used this bit. The reset matrix is a bit strange, I would assume it would just reset the matrix back to identity, instead it seems to just reset the rotation back to identity whilst preserving the scale/position. 2023-09-27 16:33:53 +01:00
Ian Curtis 40c8259130 Rewrite the whole project for GL4+. I figured if we removed the limitation of a legacy rendering API we could improve things a bit. With GL4+ we can do unsigned integer math in the shaders. This allows us to upload a direct copy of the real3d texture sheet, and texture directly from this memory given the x/y pos and type. This massively simplifies the binding and invalidation code. Also the crazy corner cases will work because it essentially works the same way as the original hardware. 
The standard triangle render requires gl 4.1 core, so should work on mac. The quad renderer runs on 4.5 core. The legacy renderer should still work, and when enabled a regular opengl context will be created, which allows functions marked depreciated in the core profiles to still work. This will only work in windows/linux I think. Apple doesn't support this.

A GL 4.1 GPU is now the min required spec. Sorry if you have an OLDER gpu. GL 4.1 is over 12 years old now.

This is a big update so I apologise in advance if I accidently broke something :]
 2022-11-07 21:33:01 +00:00
toxieainc 22ffb5cc19 Harmless math warning fixes. map->unordered_map in CNew3D::CacheModel. Avoid general aliasing problems (using bit_cast or the recommended/optimized-away memcpy). 2022-10-03 17:59:59 -07:00
Ian Curtis 0ef90899e1 Double buffer the line of sight values. Should allow the correct values to be returned if GPU threading is enabled. 2022-01-02 12:48:09 +00:00
Ian Curtis ba917aca6d Update glew version 2020-07-31 19:18:51 +00:00
Ian Curtis 06688de5fc Really should cull against 5 frustum planes. 2019-11-15 20:00:25 +00:00
Ian Curtis bf4d725970 Smooth texture repeat makes no sense for alpha/contour textures with pixel dilate. HW seems to treat them as non smooth anyway. Also implement some line of sight stuff, used by Scud. Really need to check the threading/synchronisation of the register reads but it basically works. 2018-10-19 20:59:46 +00:00
Ian Curtis 3eeec35e7c Correctly emulate the real3d pro-1000 texture modes in our shader. 2018-10-13 13:29:45 +00:00
Ian Curtis b5f9ad9651 Quad rendering engine. Set QuadRendering = 1 in the ini file to use, or -quad-rendering at the end of the command line to use. 2018-09-13 12:50:34 +00:00
Ian Curtis e4f5f0bcaf Sega bass fishing is starting some off the meshes with a shared vertex. Without having a previous vertex to share with, the values were just uninitialised and junk. The hardware itself seems to hold the value from the previous mesh, and sega bass fishing relies on this undefined behaviour to function correctly. Storing the cached vertices as member variables instead of locally to the function means it holds the values from the last mesh, and uses those values instead of random garbage. 2018-09-02 21:36:24 +00:00
Ian Curtis c7ffd0a808 Draw transparent polys to separate layers and composite at the end. This solves a tonne of transparency errors we had been battling with for a long time. The model3 is strange in the fact it only supports a max of two translucent overlapped polys. They are not blended into the frame normally. Doing this means the topmost translucent polys only are visible in the scene, the equivalent of doing a depth pass first, but without the added cost. 2018-06-16 21:31:29 +00:00
Bart Trzynadlowski 3f9b1b76e5 Added a stretch mode (-stretch) and modified new renderer's viewport code (hopefully correctly) to handle aspect ratio correction and widening the way the legacy one does. 2018-05-03 03:46:44 +00:00
Ian Curtis 8266eecabb Parse out line of sight position. Only known to be used by scud. 2018-01-15 21:27:21 +00:00
Ian Curtis eb798ed15e Star wars is the only game to pass unsigned fixed shaded values (per vertex brightness) to the renderer. Originally we thought that the specular flag would turn on unsigned values since it's the only game to set specular with these polys, but this logic turned out to be incorrect. The JTAG interface seems to config the GPU to turn on this functionality. (Harry Tuttle) 2017-10-05 19:15:00 +00:00
Ian Curtis 2786d95795 add interface for sun clamp 2017-09-02 14:54:16 +00:00
Ian Curtis 0efd4dac39 Finish fixed shading for 2.0 hardware. I'm pretty sure it actually works identically to step 1.5 hardware. The oddball is LA machine guns where the viewport ambient doesn't seem to effect the brightness. But the ambient works differently in this game because it uses the unclamped light model. Still need to investigate if the diffuse factor effects fixed shading. 2017-08-15 23:21:57 +00:00
Ian Curtis c12e4c3215 Fixed shading (per vertex poly colours) on step 1.5 hardware have the viewport ambient value added to them. This fixes various shading on scud. To do this had to switch the maths to the vertex shader. 2017-08-14 09:14:06 +00:00
Bart Trzynadlowski a315627401 Added support for loading shaders from files for new engine and tile renderer. Changed help text to reflect new engine being default. 2017-08-11 00:41:10 +00:00
Ian Curtis b00315b072 Revert previous fixed shading changes. Implement new logic for fixed shading on step 1.5 hardware based upon Harry's findings. 2017-06-20 20:13:42 +00:00
Ian Curtis 9b226ed478 parse out some culling node values 2017-04-17 10:40:07 +00:00
Bart Trzynadlowski f34e25dfc7 Massive internal change: removed CConfig object and replaced it with a hierarchical config object system (Util::Config::Node). Games are now defined in an XML file. Hopefully I didn't break too many things :/ 2017-03-27 03:19:15 +00:00
Ian Curtis ff624afb4d The mipmap data in some games seems to being sent behind the base textures, and when the base textures are bound and created the mipmap chain is not yet valid in memory. To fix this, we invalidate the base textures, when the mipmap chain is updated. 2017-03-25 00:06:24 +00:00
Ian Curtis 5b9741bd5d Support high priority polygons. They are treated as a separate overlay plane by the h/w. Fixes some missing fx in harley. 2017-02-20 17:22:32 +00:00
Ian Curtis 24cbeed526 Dynamically calculate near/far frustum planes for each viewport priority layer by clipping the overlapping meshes with the 4 frustum planes. A bit brute forcy, but the results are great. 2017-02-07 14:05:03 +00:00
Ian Curtis c82dc55485 cosmetics 2016-12-22 22:11:19 +00:00
Ian Curtis 72dd0dbf98 only need to cull against 4 clipping planes 2016-12-22 22:09:13 +00:00
Ian Curtis da8b63b44b The model3 shares vertices between polygons. But it also seems to do this for polygons with different textures. This means the texture coordinate for potentially different sizes textures is shared. Currently this was broken leading to some corrupted looking textures. The fix is to simply cache the old texture coordinates, and recalculate the normalised coordinates in the new texture coordinate space. 2016-12-06 14:39:46 +00:00
Ian Curtis 10a5d2c0e1 refactor 2016-12-06 12:25:34 +00:00
Ian Curtis fe60d115d2 Revert changes .. if we clip against the real3d clipping planes, our wide screen hack won't work anymore. 2016-11-26 01:19:51 +00:00
Ian Curtis 665fa90425 Get clipping planes directly from the viewport data. Also no need to clip against a far plane 2016-11-26 01:12:08 +00:00
Ian Curtis f0a15febd1 implement scroll fog, fixes a tonne of missing fx 2016-10-09 16:34:12 +00:00
Ian Curtis 85c4d85562 refactor 2016-06-19 21:43:09 +00:00
Ian Curtis c2b1db11f8 Cull geometry based on the culling distances in the nodes. Should give some sort of speedup on lower end pcs where the CPU is predominantly the bottle neck. 2016-06-16 20:05:29 +00:00
Ian Curtis 5081ee99dc Add code to convert the 16bit float format used by the hardware back into float 754 format. 16bit floats are used for the culling radiuses in the nodes. 2016-06-03 00:01:34 +00:00
Ian Curtis 36074d9fd5 Y texture offsets are just 5 bits. Texture coordinate wrap around happens in the same texture sheet, not into the next one. Fixes various texturing bugs in sega rally. 2016-05-30 13:25:00 +00:00
Ian Curtis ff8d311dcc support colour table pointer 2016-05-18 23:06:41 +00:00
Ian Curtis 43add90d2f Fix bad texture seems 2016-05-09 16:26:34 +00:00
Ian Curtis ded3168060 Split face colour from per vertex poly colour. Fixes colour bleeding between connected polys in harley. 2016-05-07 15:28:04 +00:00
Ian Curtis 58d187e59a parse viewports in normal order 2016-04-19 22:05:12 +00:00
Ian Curtis 057ee94210 simplify 2016-04-18 14:06:10 +00:00
Ian Curtis aee89390ba add LOD table data. Maybe we can get clipping planes out of these 2016-03-28 20:11:46 +00:00
Ian Curtis 4992e59673 Cache rom models, for better performance. 2016-03-26 22:44:26 +00:00
Ian Curtis 0327db246f WIP 2016-03-24 13:17:17 +00:00
Ian Curtis f031e5d095 finish front/back face culling code based on matrix determinant. Optimised opengl to avoid redundant state changes. 2016-03-22 23:39:59 +00:00