- weighted cuda compressor.

- faster and better BC3n compressor
- add normal map flag to DDS files that store normal maps.
- stop using RXGB fourcc code.
- move tools to tools/
- add temptative config dialog for UI based tool. 
- add experimental normal map mipmap generation.
- start adding support for input dds files in nvcompress.
- many other small fixes and cleanup.

src/nvimage/nvtt/cuda/CudaMath.h

@@ -111,7 +111,7 @@ inline __device__ __host__ float3 clamp(float3 v, float3 a, float3 b)
 
 inline __device__ __host__ float3 normalize(float3 v)
 {
-    float len = 1.0f / dot(v, v);
+    float len = 1.0f / sqrtf(dot(v, v));
     return make_float3(v.x * len, v.y * len, v.z * len);
 }