Shadow mapping for point lights demo
- Thread starter Humus
- Start date
It's not Luth's (Luleå University of Technology) server, it's me and my roommates own server, the only connection to luth is the domain name as I'm on the campus network.
Are you sure you're using right address? It's currently on a temporary address (http://humus2.campus.luth.se/~humus/) due to some maintainance, when we get the HD we're waiting on it'll be put back on the old address (http://esprit.campus.luth.se/~humus/).
Are you sure you're using right address? It's currently on a temporary address (http://humus2.campus.luth.se/~humus/) due to some maintainance, when we get the HD we're waiting on it'll be put back on the old address (http://esprit.campus.luth.se/~humus/).
I get 130fps on a Radeon9700 PRO in default.
I'm actually a little disappointed in that figure. I only scanned Humus's code, but I suspect there's some optimizations to be done using buffered primitives/indices, range extensions, and stripification. I think the demo only uses procedural glVertex* calls for drawing. Huddy hammered us over and over at the last Mojo saying that sending 1 triangle at a time to the Radeon 9700 will destroy performance and cause the card to achieve only a fraction of its true performance.
I'm not criticizing Humus tho, it's clearly just a demonstration demo, and not supposed to be optimized. The code is written to be easy to read and understand. It's a very nice demo, and I suggest anyone look at Humus's code to learn GL programming.
-DC
p.s. now that I thought about it, it's not so disappointing. It's rendering 12 render targets (1 for each light, 1 for each cube face) plus the main rendertarget. So multiply that FPS by 13 to get the true number of "frames" being rendered.
I'm actually a little disappointed in that figure. I only scanned Humus's code, but I suspect there's some optimizations to be done using buffered primitives/indices, range extensions, and stripification. I think the demo only uses procedural glVertex* calls for drawing. Huddy hammered us over and over at the last Mojo saying that sending 1 triangle at a time to the Radeon 9700 will destroy performance and cause the card to achieve only a fraction of its true performance.
I'm not criticizing Humus tho, it's clearly just a demonstration demo, and not supposed to be optimized. The code is written to be easy to read and understand. It's a very nice demo, and I suggest anyone look at Humus's code to learn GL programming.
-DC
p.s. now that I thought about it, it's not so disappointing. It's rendering 12 render targets (1 for each light, 1 for each cube face) plus the main rendertarget. So multiply that FPS by 13 to get the true number of "frames" being rendered.
DemoCoder, it runs ok on your Radeon 9700? That's great, I've had several reports of people not being able to run it on the R9700, getting serious artefacts. This is the first report of it working (well, I assume it does as you didn't complain about artefacts 
)
Yes, I only use immediate mode in this demo, but then I have the extreme scene complexity of 20 quads ... so I don't think it matters. I even used immediate mode in my GameEngine demo from early this year, which had a decent number of polygons. But that's about where one should have moved over to vertex array + VAO/VAR. I'm working on another larger project where I plan to use this technique for shadows, it's designed around vertex arrays + VAO from the start. It really matters in that demo, there a large performance difference even between using VAO and not.
)Yes, I only use immediate mode in this demo, but then I have the extreme scene complexity of 20 quads
... so I don't think it matters. I even used immediate mode in my GameEngine demo from early this year, which had a decent number of polygons. But that's about where one should have moved over to vertex array + VAO/VAR. I'm working on another larger project where I plan to use this technique for shadows, it's designed around vertex arrays + VAO from the start. It really matters in that demo, there a large performance difference even between using VAO and not.