Improve testing on ParallelTestRunner

README.md

@@ -45,18 +45,18 @@ To ensure that everything is working you can run tests for the package with
 pkg> test AMDGPU
 ```
 
-Or specifying a subset of tests to run:
+Or specifying a subset of tests to run on 4 workers in parallel:
 
 ```julia
 julia> using Pkg
 
-julia> Pkg.test("AMDGPU"; test_args=["core", "kernelabstractions"])
+julia> Pkg.test("AMDGPU"; test_args=`--jobs=4 core kernelabstractions`)
 ```
 
-Full list of tests to run can be obtained with `--list` argument:
+Full list of tests to run can be obtained with `--list` argument (refer to [ParallelTestRunner.jl's doc](https://juliatesting.github.io/ParallelTestRunner.jl/dev/#Command-Line-Options) for the full list of available command line arguments):
 
 ```julia
-julia> Pkg.test("AMDGPU"; test_args=["--list"])
+julia> Pkg.test("AMDGPU"; test_args=`--list`])
 ```
 
 ## Questions and Contributions

docs/make.jl

@@ -35,6 +35,7 @@ function main()
                 "Performance Tips" => "tutorials/perf.md",
                 "Profiling" => "tutorials/profiling.md",
                 "Installation Tips" => "install_tips.md",
+                "Testing" => "testing.md",
             ],
             "API" => [
                 "Devices" => "api/devices.md",

docs/src/index.md

@@ -28,6 +28,7 @@ Simply add the AMDGPU.jl package to your Julia environment:
 
 ```julia
 using Pkg
+
 Pkg.add("AMDGPU")
 ```
 
@@ -62,6 +63,7 @@ To ensure that everything works, you can run the test suite:
 ```julia
 using AMDGPU
 using Pkg
+
 Pkg.test("AMDGPU")
 ```
 

docs/src/testing.md

@@ -0,0 +1,42 @@
+# Testing
+
+To ensure that everything is working you can run tests for the package with:
+
+```julia
+pkg> test AMDGPU
+```
+
+## Advanced testing options
+
+AMDGPU tests use [ParallelTestRunner.jl]((https://juliatesting.github.io/ParallelTestRunner.jl) which allow for [running tests with various (command line) options](https://juliatesting.github.io/ParallelTestRunner.jl/dev/#Running-Tests).
+
+To, e.g., launch a subset of tests `core` and `kernelabstractions` on 4 runners in parallel:
+
+```julia
+julia> using Pkg
+
+julia> Pkg.test("AMDGPU"; test_args=`--jobs=4 core kernelabstractions`)
+```
+
+The full list of tests to run can be obtained with `--list` argument:
+
+```julia
+julia> Pkg.test("AMDGPU"; test_args=`--list`])
+```
+
+## Testing categories
+
+Although tests can be run in a custom fashion upon exploring the output of listing (using the `--list` test argument), tests are organised such that following grouping by categories is possible:
+
+```
+core device hip external gpuarrays kernelabstractions wmma enzyme
+```
+
+which allows to, e.g., run all `gpuarrays` related tests as
+
+```
+julia> Pkg.test("AMDGPU"; test_args=`gpuarrays`)
+```
+
+!!! warning "Large memory tests"
+    Some tests HIP and GPUArrays tests may use > 20GB of host RAM. It is recommended to use fewer workers (<= 4) on machines that have < 32Gb of host RAM in case running tests would result in out of memory errors.

test/rocarray/broadcast.jl → test/core/rocarray_broadcast.jl

@@ -2,10 +2,6 @@ using Test
 using AMDGPU
 using AMDGPU: ROCArray
 
-import GPUArrays
-include(joinpath(pkgdir(GPUArrays), "test", "testsuite.jl"))
-testf(f, xs...; kwargs...) = TestSuite.compare(f, AMDGPU.ROCArray, xs...; kwargs...)
-
 @testset "broadcast" begin
     @test testf((x)       -> fill!(x, 1),  rand(3,3))
     @test testf((x, y)    -> map(+, x, y), rand(2, 3), rand(2, 3))

test/rocarray/blas.jl → test/hip_rocarray/blas.jl

@@ -3,10 +3,6 @@ using AMDGPU
 using AMDGPU: ROCArray, ROCVector, ROCMatrix
 using LinearAlgebra
 
-import GPUArrays
-include(joinpath(pkgdir(GPUArrays), "test", "testsuite.jl"))
-testf(f, xs...; kwargs...) = TestSuite.compare(f, AMDGPU.ROCArray, xs...; kwargs...)
-
 @assert AMDGPU.functional(:rocblas)
 
 @testset "BLAS" begin
@@ -132,7 +128,7 @@ end
             @test testf(
                 (y, a, b) -> mul!(y, Hermitian(a), b), rand(T, 5),
                 rand(T, 5, 5), rand(T, 5))
-            
+
             A_ = rand(T, m, m)
             A = A_ + A_'
             x = rand(T, m)
@@ -423,7 +419,7 @@ end
             dC = rocBLAS.trsm('L', 'U', 'N', 'N', one(T), dA, dB)
             @test collect(dC) ≈ triu(view(A,1:m,1:m)) \ view(B,1:m,1:m)
         end
-            
+
         @testset "trsm_batched" begin
             batch_count = 3
             A  = [rand(T, m, m) for ix in 1:batch_count]
@@ -661,7 +657,7 @@ end
         @test C ≈ h_C
     end
     @testset "herk T=$T" for T in (ComplexF32, ComplexF64)
-        T1 = T 
+        T1 = T
         T2 = real(T)
         # generate parameters
         α = rand(T2)
@@ -689,7 +685,7 @@ end
         @test C ≈ h_C
     end
     @testset "her2k T=$T" for T in (ComplexF32, ComplexF64)
-        T1 = T 
+        T1 = T
         T2 = real(T)
         # generate parameters
         α = rand(T1)