More words to flesh out the docs

docs/src/index.md

@@ -1,6 +1,11 @@
 # CUSPARSE.jl
 
 ### Container types
+
+Analogous to `CUDArt.jl` and its `CudaArray` types, you can use
+a variety of sparse matrix containers similar to those in Julia
+base.
+
 ```@docs
 CudaSparseVector
 CudaSparseMatrixCSC
@@ -12,6 +17,11 @@ CUSPARSE.CompressedSparse
 ```
 
 ### Utility types
+
+These types are used to give CUSPARSE internals information
+about the matrices they are working on. Most are opaque pointers
+or `enums`.
+
 ```@docs
 CUSPARSE.cusparseMatDescr_t
 CUSPARSE.cusparseDirection_t
@@ -43,6 +53,13 @@ CUSPARSE.cusparsetype
 ```
 
 ### Type Conversions
+
+These functions allow you to switch between different sparse
+matrix formats. This is particularly useful because Julia
+uses a compressed sparse **column** representation internally,
+while most CUSPARSE routines work with compressed sparse **row**
+matrices (or variants of them, such as the `BSR` and `HYB` formats).
+
 ```@docs
 switch2csc
 switch2csr
@@ -86,6 +103,15 @@ gemm
 ```
 
 ### Preconditioners
+
+These are in-place incomplete preconditioners for a linear solve.
+They are in-place in the sense that doing a full Cholesky/LU solve
+would require more memory than the starting matrix uses, and this
+may not be possible on the restricted memory of the GPU. Instead,
+an incomplete solution is performed, where the structure of the
+matrix is untouched (so it occupies the same amount of memory)
+but its values are updated. 
+
 ```@docs
 ic0!
 ic02!