| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
got rid of the bullshit dance between XRGB32 and RGBA32, now
all rasters are RGBA32 and premultiply alpha before loading
up an image and drawing over the destination. this lets the
user compose their own scenes with correct transparency.
no more mediocre clearcolor.
as a consequence drawing got slower, but if i get the turbo
drawing pool working properly that should go away.
textures also take alpha pre-multiplied channels into account,
dividing them when sampling, so loading transparent textures
will show the correct colors.
|
|
|
|
|
|
|
|
|
|
| |
this commit also includes the turbo drawing pool.
the experiment was successful in getting reasonable
drawing times to a fullhd image, but the process
causes glitches when moving the objects around,
which is unacceptable.
it's been commented out for now.
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
got a bit tired of having to add a ulong *buf every time
i needed to get some data out of the shaders, so i made
it possible to create as many extra buffers as needed,
which then can be addressed from the fragment shader
and drawn using the same drawing routines that were used
for the color buffer.
these so called Rasters are very similar to the OpenGL
FBOs, in case you are familiar, but we address them by
strings instead of numbers. and they can also be used as
textures if you create one (see alloctexture) and then
memdraw into it after every shot.
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
| |
very simple, without anti-aliasing, made for getting
OIT (order independent transparency) rendering of
arbitrary objects.
also added switches for blending, depth testing and the
A-buffer to the camera.
|
| |
|
| |
|
|
|
|
|
|
|
| |
vertex attributes were not being updated after clipping
with their corresponding work rectangles, nor when
reordering the points before rasterization, which caused
serious artifacts.
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
| |
it's possible for interactive programs to keep updating
the geometry of the scene during rendering, which caused
primitives to look deformed or shattered in the final
image.
to avoid this we take a snapshot of the current state of
things (scene and camera), and render based on that copy.
|
| |
|
|
|
|
|
|
| |
we are still doing alpha blending, so we use the
X chan internally, but we don't care about it
when (mem)drawing to the screen.
|
|
|
|
|
|
|
|
| |
colors are now properly processed in linear RGB space for
lighting, shading and blending. sRGB is assumed for any
texture sampled and the destination framebuffer.
it's not perfect, but it does the job for now.
|
| |
|
|
|
|
| |
added a normals buffer for debugging.
|
| |
|
|
|
|
|
|
| |
among these other things are clamping the color
channels to [0,1] internally, and adding a
modulation function for mixing colors/points.
|
| |
|
| |
|
|
|
|
| |
this provides fair scheduling, minimizing contention.
|
| |
|
| |
|
| |
|
|
|
|
|
| |
mixed up CLIP[TB] and the slope was being computed as
an integer division, which caused artifacts.
|
|
|
|
|
|
|
|
| |
also got rid of the dependency on OBJ for the entire renderer,
instead letting the user load a Model from any given OBJ. this
modularity will allow for other formats to be used in the same
way, relying on a single, internal representation for the
entire pipeline.
|
| |
|
|
|
|
|
|
| |
this fixes a bug i introduced with the fully concurrent pipeline,
where instead of testing whether the triangle *crossed* any of the
tiles, i checked whether any of the points *were inside*.
|
| |
|
|
|
|
|
| |
this is only an structural replacement and doesn't add
support for the other primitive types.
|
|
|
|
|
| |
this makes more sense memory-wise, and it also fixes a
clipping glitch that was occurring at tile boundaries.
|
| |
|
|
|
|
|
| |
- got rid of the z-buffer lock to avoid contention.
- little improvements to fb.c
|
|
|
|
|
|
|
|
|
|
| |
> They made two mistakes. they hanged the wrong man and they didn't finish the job.
so Clint Eastwood came back to kick my ass. the mistakes in
question were that the Viewport shouldn't know about double
buffering, conceptually it has a framebuffer and that's it.
the second one was passing it to the renderer, when the
renderer couldn't care less about what a viewport is.
|
| |
|
|
|
|
|
| |
do the frame buffer clearing and swapping as part
of the rendering process, not within shootcamera.
|
| |
|
| |
|