Remove some comptime stuff that wasn't really necessary for perf in the

rchol.

src/rcholesky.jl

@@ -37,16 +37,15 @@ end
 # TODO (cg 2022/05/30 11:05): Continually look to squeeze allocations out of
 # here. Maybe I can pre-allocate things for the BLAS calls, even?
 function rchol_instantiate!(strbuf::RCholesky{T}, V::VecchiaConfig{H,D,F},
-                            params::AbstractVector{T}, 
-                            ::Val{Z}, ::Val{N}) where{H,D,F,T,Z,N}
+                            params::AbstractVector{T}, ::Val{Z}) where{H,D,F,T,Z}
   checkthreads()
   @assert !strbuf.is_instantiated[] RCHOL_INSTANTIATE_ERROR
   strbuf.is_instantiated[] = true
   kernel  = V.kernel
   cpts_sz = V.chunksize*V.blockrank
   pts_sz  = V.chunksize
   # allocate three buffers:
-  bufs = allocate_crchol_bufs(Val(N), Val(D), Val(Z), cpts_sz, pts_sz)
+  bufs = allocate_crchol_bufs(Threads.nthreads(), Val(D), Val(Z), cpts_sz, pts_sz)
   # do the main loop:
   m = cld(length(V.condix), Threads.nthreads())
   @sync for (i, chunk) in enumerate(Iterators.partition(1:length(V.condix), m))
@@ -115,7 +114,7 @@ function rchol(V::VecchiaConfig{H,D,F}, params::AbstractVector{T};
   # compute the out type and the number of threads to pass in as vals:
   Z   = promote_type(H, T)
   N   = Threads.nthreads()
-  rchol_instantiate!(out, V, params, Val(Z), Val(N))
+  rchol_instantiate!(out, V, params, Val(Z))
   out
 end
 

src/utils.jl

@@ -205,6 +205,14 @@ end
   end
 end
 
+# I think I went a little overboard with the compile-time stuff above. This
+# allocation happens once in a function call that will take a long time for real
+# problems, so it doesn't seem worth the compiler stress.
+function allocate_crchol_bufs(n::Int64, ::Val{D}, ::Val{Z}, 
+                              cpts_sz, pts_sz) where{N,D,Z}
+  [crcholbuf(Val(D), Val(Z), cpts_sz, pts_sz) for _ in 1:n]
+end
+
 function pretty_print_number(x)
   if x < zero(x)
     @printf "-%06.3f " abs(x)