add some preliminary gpu test CI code

Author: sunxd3

.buildkite/pipeline.yml

@@ -0,0 +1,28 @@
+env:
+  # SECRET_CODECOV_TOKEN: "nkcRFVXdaPNAbiI0x3qK/XUG8rWjBc8fU73YEyP35SeS465XORqrIYrHUbHuJTRyeyqNRdsHaBcV1P7TBbKAaTQAjHQ1Q0KYfd0uRMSWpZSCgTBz5AwttAxVfFrX+Ky3PzTi2TfDe0uPFZtFo0Asq6sUEr1on+Oo+j+q6br2NK6CrA5yKKuTX4Q2V/UPOIK4vNXY3+zDTKSNtr+HQOlcVEeRIk/0ZQ78Cjd52flEaVw8GWo/CC4YBzLtcOZgaFdgOTEDNHMr0mw6zLE4Y6nxq4lHVSoraSjxjhkB0pXTZ1c51yHX8Jc+q6HC5s87+2Zq5YtsuQSGao+eMtkTAYwfLw==;U2FsdGVkX18z27J3+gNgxsPNnXA0ad4LvZnXeohTam7/6UPqX5+3BYI0tAiVkCho4vlJyL7dd8JEyNtk9BFXsg=="
+
+steps:
+  - label: "Julia v{{matrix.version}}, {{matrix.label}}"
+    plugins:
+      - JuliaCI/julia#v1:
+          version: "{{matrix.version}}"
+      # - JuliaCI/julia-coverage#v1:
+      #     dirs:
+      #       - src
+      #       - ext
+    command: julia --eval='include("test/run_extra.jl")'
+    agents:
+      queue: "juliagpu"
+      cuda: "*"
+    if: build.message !~ /\[skip tests\]/
+    timeout_in_minutes: 60
+    env:
+      LABEL: "{{matrix.label}}"
+      TEST_TYPE: ext
+    matrix:
+      setup:
+        version:
+          - "1"
+          - "1.10"
+        label:
+          - "cuda"

test/ext/cuda.jl

@@ -1,31 +1,31 @@
-using CUDA, Test, LinearAlgebra, Distributions
-using Flux
+using Pkg
+Pkg.activate(@__DIR__)
+Pkg.develop(; path=joinpath(@__DIR__, "..", "..", ".."))
 
-if CUDA.functional()
-    @testset "rand with CUDA" begin
-        dists = [
-            MvNormal(CUDA.zeros(2), I), MvNormal(CUDA.zeros(2), cu([1.0 0.5; 0.5 1.0]))
-        ]
+using NormalizingFlows
+using CUDA, Test, LinearAlgebra, Distributions, Flux
 
-        @testset "$dist" for dist in dists
-            x = rand_device(CUDA.default_rng(), dist)
-            xs = rand_device(CUDA.default_rng(), dist, 100)
-            @info logpdf(dist, x)
-            @test x isa CuArray
-            @test xs isa CuArray
-        end
+@testset "rand with CUDA" begin
+    dists = [MvNormal(CUDA.zeros(2), I), MvNormal(CUDA.zeros(2), cu([1.0 0.5; 0.5 1.0]))]
 
-        @testset "$dist" for dist in dists
-            CUDA.allowscalar(true)
-            ts = reduce(∘, [Bijectors.PlanarLayer(2) for _ in 1:2])
-            ts_g = gpu(ts)
-            flow = Bijectors.transformed(dist, ts_g)
+    @testset "$dist" for dist in dists
+        x = rand_device(CUDA.default_rng(), dist)
+        xs = rand_device(CUDA.default_rng(), dist, 100)
+        @info logpdf(dist, x)
+        @test x isa CuArray
+        @test xs isa CuArray
+    end
+
+    @testset "$dist" for dist in dists
+        CUDA.allowscalar(true)
+        ts = reduce(∘, [Bijectors.PlanarLayer(2) for _ in 1:2])
+        ts_g = gpu(ts)
+        flow = Bijectors.transformed(dist, ts_g)
 
-            y = rand_device(CUDA.default_rng(), flow)
-            ys = rand_device(CUDA.default_rng(), flow, 100)
-            @info logpdf(flow, y)
-            @test y isa CuArray
-            @test ys isa CuArray
-        end
+        y = rand_device(CUDA.default_rng(), flow)
+        ys = rand_device(CUDA.default_rng(), flow, 100)
+        @info logpdf(flow, y)
+        @test y isa CuArray
+        @test ys isa CuArray
     end
 end