README.jmd

## CuCountmap

`cucountmap` is a faster `countmap` equivalent utilizing CUDA.jl  for `Vector{T}` where `isbits(T)` and `sizeof(T) <= 2`.

### Usage

```julia
using CuCountMap

v = rand(Int16, 1_000_000)

cucountmap(v) # converts v to cu(v) and then run countmap

using CUDA: cu

cuv = cu(v)
countmap(cuv) # StatsBase.countmap is overloaded for CuArrays
```


### Example & Benchmarks

```julia
using CUDA
using CuCountMap
using StatsBase: countmap

v = rand(Int16, 10_000_000);

using BenchmarkTools

cpu_to_gpu_benchmark = @benchmark gpu_countmap = cucountmap($v)
```

```julia
cpu_to_cpu_benchmark = @benchmark cpu_countmap = countmap($v)
```

```julia
cuv = CUDA.cu(v)
gpu_to_gpu_benchmark = @benchmark gpu_countmap2 = countmap(cuv)
```

#### Benchmark Plot

```julia
using Plots
using Statistics: mean

cpu_to_gpu = mean(cpu_to_gpu_benchmark.times)/1000/1000
gpu_to_gpu = mean(gpu_to_gpu_benchmark.times)/1000/1000
cpu_to_cpu = mean(cpu_to_cpu_benchmark.times)/1000/1000

plot(
["CPU Array on CPU \n countmap(v)", "convert CPU Array to GPU array on GPU \n cucountmap(cu(v))", "GPU array on GPU \n cucountmap(cuv)"],
[cpu_to_cpu, cpu_to_gpu, gpu_to_gpu],
seriestypes = :bar, title="CuCountMap.cucountmap vs StatsBase.countmap", label="ms",
legendtitle="Mean time")
```