DeadlyNinja
Veteran
That's not a fair comparison since the pic on the right isn't a direct feed picture. Of course the game looks amazing.
That's not a fair comparison since the pic on the right isn't a direct feed picture. Of course the game looks amazing.
On a console it's pretty doable. Wii/GC/PS2 can do regular per channel multiplication for sure if not a dot product. All you have to do is a full screen pass with the CPU where you add R+G+B and write it into A. With only 256K pixels, that's a piece of cake on a CPU, and they can access video memory very efficiently. Of course, you have to draw everything again to use the value, and there's no normalization.you can not do normal map on the cpu.That is a per pixel instruction.
You can do better then that - if you run deferred shading numbers on PS2, @30fps you get something like 32instructions/pixel(single VU), which allows for a lot more then just dot product.Mintmaster said:All you have to do is a full screen pass with the CPU where you add R+G+B and write it into A. With only 256K pixels, that's a piece of cake on a CPU, and they can access video memory very efficiently.
No dot-product, dependant lookup with a hard-wired 3x2 matrix transform.Still, it's pretty likely that Wii/GC has a dot-product operation on the GPU.
Oh I know (about doing better than just a dot3), I just wanted to point out to bomlat that it was doable.You can do better then that - if you run deferred shading numbers on PS2, @30fps you get something like 32instructions/pixel(single VU), which allows for a lot more then just dot product.
Basically you get you SM3.0+, with dynamic branching @ 1pixel granularity, and no texture lookups. Unfortunately there are other issues that make this less then practical, but cpu setup is up to it.
What exactly can you put into that matrix? A vertex attribute? Only a constant? Can the vector being transformed come from a texture?No dot-product, dependant lookup with a hard-wired 3x2 matrix transform.