Split out MTTKRP benchmarks

benchmark/benchmarks.jl

@@ -1,5 +1,7 @@
 using BenchmarkTools
 
+BenchmarkTools.DEFAULT_PARAMETERS.seconds = 1
+
 const SUITE = BenchmarkGroup()
 
 module BenchmarkGCP

benchmark/gcp.jl

@@ -6,9 +6,7 @@ using Distributions
 
 const SUITE = BenchmarkGroup()
 
-BenchmarkTools.DEFAULT_PARAMETERS.seconds = 1
-
-bench_gcp = SUITE["gcp"] = BenchmarkGroup()
+bench_gcp = SUITE
 
 # Benchmark least squares loss 
 for sz in [(15, 20, 25), (30, 40, 50)], r in 1:2
@@ -42,19 +40,3 @@ for sz in [(15, 20, 25), (30, 40, 50)], r in 1:2
     X = [rand(Bernoulli(M[I]/(M[I] + 1))) for I in CartesianIndices(size(M))]
     bench_gcp["bernoulliOdds-size(X)=$sz, rank(X)=$r"] = @benchmarkable gcp($X, $r, BernoulliOddsLoss())
 end
-
-
-# MTTKRP benchmarks
-bench_mttkrp = SUITE["mttkrp"] = BenchmarkGroup()
-
-szs = [10,30,50,80,120,200]
-shapes = [(sz, sz, sz) for sz in szs]
-n = 1
-rs = 20:20:200
-
-for sz in shapes, r in rs
-    Random.seed!(0)
-    X = randn(sz)
-    U = [randn(Ik,r) for Ik in sz]
-    bench_mttkrp["mttkrp-size(X)=$sz, rank(X)=$r"] = @benchmarkable GCPDecompositions.mttkrp($X, $U, $n)
-end

benchmark/mttkrp.jl

@@ -0,0 +1,20 @@
+using BenchmarkTools
+
+using Random
+using GCPDecompositions
+
+const SUITE = BenchmarkGroup()
+
+bench_mttkrp = SUITE
+
+szs = [10,30,50,80,120,200]
+shapes = [(sz, sz, sz) for sz in szs]
+n = 1
+rs = 20:20:200
+
+for sz in shapes, r in rs
+    Random.seed!(0)
+    X = randn(sz)
+    U = [randn(Ik,r) for Ik in sz]
+    bench_mttkrp["mttkrp-size(X)=$sz, rank(X)=$r"] = @benchmarkable GCPDecompositions.mttkrp($X, $U, $n)
+end