allow tuples

Author: mcabbott

src/sorting.jl

@@ -37,6 +37,13 @@ using ..CUDA: i32
     (eq && a′ == b′) || lt(a′, b′)
 end
 
+# To allow sorting tuples of numbers:
+@inline _zero(x) = Base.zero(x)
+@inline _zero(::Type{T}) where {T<:Tuple{Vararg{Any,N}}} where {N} = ntuple(i -> zero(T.parameters[i]), N)
+
+@inline _one(x) = Base.one(x)
+@inline _one(::Type{T}) where {T<:Tuple{Vararg{Any,N}}} where {N} = ntuple(i -> one(T.parameters[i]), N)
+
 
 # Batch partitioning
 """
@@ -73,7 +80,7 @@ Uses block y index to decide which values to operate on.
     sync_threads()
     blockIdx_yz = (blockIdx().z - 1i32) * gridDim().y + blockIdx().y
     idx0 = lo + (blockIdx_yz - 1i32) * blockDim().x + threadIdx().x
-    val = idx0 <= hi ? values[idx0] : one(eltype(values))
+    val = idx0 <= hi ? values[idx0] : _one(eltype(values))
     comparison = flex_lt(pivot, val, parity, lt, by)
 
     @inbounds if idx0 <= hi
@@ -183,7 +190,7 @@ Must only run on 1 SM.
         swap = if threadIdx().x <= to_move
             vals[lo + a + threadIdx().x]
         else
-            zero(eltype(vals))  # unused value
+            _zero(eltype(vals))  # unused value
         end
         sync_threads()
         if threadIdx().x <= to_move
@@ -215,7 +222,7 @@ function bitonic_median(vals :: AbstractArray{T}, swap, lo, L, stride, lt::F1, b
 
     @inbounds swap[threadIdx().x] = vals[lo + threadIdx().x * stride]
     sync_threads()
-    old_val = zero(eltype(swap))
+    old_val = _zero(eltype(swap))
 
     log_blockDim = begin
         out = 0
@@ -272,7 +279,7 @@ elements spaced by `stride`. Good for sampling pivot values as well as short sor
             buddy_val = if 1 <= buddy <= L && threadIdx().x <= L
                  swap[buddy]
             else
-                zero(eltype(swap)) # unused value
+                _zero(eltype(swap)) # unused value
             end
             sync_threads()
             if 1 <= buddy <= L && threadIdx().x <= L