[Do not merge] Test KernelIntrinsics

[christiangnrd]

[only tests]
[only benchmarks]

[github-actions]

Your PR requires formatting changes to meet the project's style guidelines.
Please consider running Runic (git runic master) to apply these changes.

Click here to view the suggested changes.

diff --git a/src/CUDAKernels.jl b/src/CUDAKernels.jl
index 5e39ab68e..7f8ec7ad4 100644
--- a/src/CUDAKernels.jl
+++ b/src/CUDAKernels.jl
@@ -162,29 +162,29 @@ end
 
 KI.argconvert(::CUDABackend, arg) = cudaconvert(arg)
 
-function KI.kernel_function(::CUDABackend, f::F, tt::TT=Tuple{}; name=nothing, kwargs...) where {F,TT}
+function KI.kernel_function(::CUDABackend, f::F, tt::TT = Tuple{}; name = nothing, kwargs...) where {F, TT}
     kern = cufunction(f, tt; name, kwargs...)
-    KI.Kernel{CUDABackend, typeof(kern)}(CUDABackend(), kern)
+    return KI.Kernel{CUDABackend, typeof(kern)}(CUDABackend(), kern)
 end
 
 function (obj::KI.Kernel{CUDABackend})(args...; numworkgroups = 1, workgroupsize = 1)
     KI.check_launch_args(numworkgroups, workgroupsize)
 
-    obj.kern(args...; threads=workgroupsize, blocks=numworkgroups)
+    obj.kern(args...; threads = workgroupsize, blocks = numworkgroups)
     return nothing
 end
 
 
-function KI.kernel_max_work_group_size(kernel::KI.Kernel{<:CUDABackend}; max_work_items::Int=typemax(Int))::Int
+function KI.kernel_max_work_group_size(kernel::KI.Kernel{<:CUDABackend}; max_work_items::Int = typemax(Int))::Int
     kernel_config = launch_configuration(kernel.kern.fun)
 
-    Int(min(kernel_config.threads, max_work_items))
+    return Int(min(kernel_config.threads, max_work_items))
 end
 function KI.max_work_group_size(::CUDABackend)::Int
-    Int(attribute(device(), CUDA.DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK))
+    return Int(attribute(device(), CUDA.DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK))
 end
 function KI.multiprocessor_count(::CUDABackend)::Int
-    Int(attribute(device(), CUDA.DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT))
+    return Int(attribute(device(), CUDA.DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT))
 end
 
 ## indexing
@@ -199,7 +199,7 @@ end
 end
 
 @device_override @inline function KI.get_global_id()
-    return (; x = Int((blockIdx().x-1)*blockDim().x + threadIdx().x), y = Int((blockIdx().y-1)*blockDim().y + threadIdx().y), z = Int((blockIdx().z-1)*blockDim().z + threadIdx().z))
+    return (; x = Int((blockIdx().x - 1) * blockDim().x + threadIdx().x), y = Int((blockIdx().y - 1) * blockDim().y + threadIdx().y), z = Int((blockIdx().z - 1) * blockDim().z + threadIdx().z))
 end
 
 @device_override @inline function KI.get_local_size()
diff --git a/src/accumulate.jl b/src/accumulate.jl
index 54fab2119..e0631d387 100644
--- a/src/accumulate.jl
+++ b/src/accumulate.jl
@@ -22,9 +22,9 @@ function partial_scan(op::Function, output::AbstractArray{T}, input::AbstractArr
     temp = CuDynamicSharedArray(T, (2*threads,))
 
     # iterate the main dimension using threads and the first block dimension
-    i = (KI.get_group_id().x-1i32) * KI.get_local_size().x + KI.get_local_id().x
+    i = (KI.get_group_id().x - 1i32) * KI.get_local_size().x + KI.get_local_id().x
     # iterate the other dimensions using the remaining block dimensions
-    j = (KI.get_group_id().z-1i32) * KI.get_num_groups().y + KI.get_group_id().y
+    j = (KI.get_group_id().z - 1i32) * KI.get_num_groups().y + KI.get_group_id().y
 
     if j > length(Rother)
         return
@@ -105,9 +105,9 @@ function aggregate_partial_scan(op::Function, output::AbstractArray,
     block = KI.get_group_id().x
 
     # iterate the main dimension using threads and the first block dimension
-    i = (KI.get_group_id().x-1i32) * KI.get_local_size().x + KI.get_local_id().x
+    i = (KI.get_group_id().x - 1i32) * KI.get_local_size().x + KI.get_local_id().x
     # iterate the other dimensions using the remaining block dimensions
-    j = (KI.get_group_id().z-1i32) * KI.get_num_groups().y + KI.get_group_id().y
+    j = (KI.get_group_id().z - 1i32) * KI.get_num_groups().y + KI.get_group_id().y
 
     @inbounds if i <= length(Rdim) && j <= length(Rother)
         I = Rother[j]
diff --git a/src/device/random.jl b/src/device/random.jl
index 7d72d90a1..063c736ed 100644
--- a/src/device/random.jl
+++ b/src/device/random.jl
@@ -73,8 +73,8 @@ end
         @inbounds global_random_counters()[warpId]
     elseif field === :ctr2
         globalId = KI.get_global_id().x +
-                   (KI.get_global_id().y - 1i32) * KI.get_global_size().x +
-                   (KI.get_global_id().z - 1i32) * KI.get_global_size().x * KI.get_global_size().y
+            (KI.get_global_id().y - 1i32) * KI.get_global_size().x +
+            (KI.get_global_id().z - 1i32) * KI.get_global_size().x * KI.get_global_size().y
         globalId%UInt32
     end::UInt32
 end
diff --git a/src/mapreduce.jl b/src/mapreduce.jl
index 97a4176b4..6fccff91e 100644
--- a/src/mapreduce.jl
+++ b/src/mapreduce.jl
@@ -294,8 +294,9 @@ function GPUArrays.mapreducedim!(f::F, op::OP, R::AnyCuArray{T},
         end
 
         partial_kernel(f, op, init, Rreduce, Rother, Val(shuffle), partial, A;
-                    threads=partial_threads, blocks=partial_blocks, shmem=partial_shmem)
-                    # workgroupsize=partial_threads, numworkgroups=partial_blocks, shmem=partial_shmem)
+            threads = partial_threads, blocks = partial_blocks, shmem = partial_shmem
+        )
+        # workgroupsize=partial_threads, numworkgroups=partial_blocks, shmem=partial_shmem)
 
         GPUArrays.mapreducedim!(identity, op, R, partial; init)
     end
diff --git a/test/base/kernelabstractions.jl b/test/base/kernelabstractions.jl
index 2f2c4300b..1e674d3be 100644
--- a/test/base/kernelabstractions.jl
+++ b/test/base/kernelabstractions.jl
@@ -4,9 +4,14 @@ using SparseArrays
 
 include(joinpath(dirname(pathof(KernelAbstractions)), "..", "test", "testsuite.jl"))
 
-Testsuite.testsuite(()->CUDABackend(false, false), "CUDA", CUDA, CuArray, CuDeviceArray; skip_tests=Set([
-    "CPU synchronization",
-    "fallback test: callable types",]))
+Testsuite.testsuite(
+    () -> CUDABackend(false, false), "CUDA", CUDA, CuArray, CuDeviceArray; skip_tests = Set(
+        [
+            "CPU synchronization",
+            "fallback test: callable types",
+        ]
+    )
+)
 for (PreferBlocks, AlwaysInline) in Iterators.product((true, false), (true, false))
     Testsuite.unittest_testsuite(()->CUDABackend(PreferBlocks, AlwaysInline), "CUDA", CUDA, CuDeviceArray)
 end
diff --git a/test/runtests.jl b/test/runtests.jl
index 802c832e5..11584ca92 100644
--- a/test/runtests.jl
+++ b/test/runtests.jl
@@ -1,6 +1,6 @@
 @static if VERSION < v"1.11" && get(ENV, "BUILDKITE_PIPELINE_NAME", "CUDA.jl") == "CUDA.jl"
     using Pkg
-    Pkg.add(url="https://github.com/JuliaGPU/KernelAbstractions.jl", rev="main")
+    Pkg.add(url = "https://github.com/JuliaGPU/KernelAbstractions.jl", rev = "main")
 end
 
 using Distributed

[github-actions]

CUDA.jl Benchmarks

Details

Benchmark suite	Current: 3c75eb6	Previous: 9f56ee2	Ratio
latency/precompile	54337646854.5 ns	45061042309.5 ns	1.21
latency/ttfp	12957095931 ns	12778273638 ns	1.01
latency/import	3551763524 ns	3543310654 ns	1.00
integration/volumerhs	9438341.5 ns	9438152 ns	1.00
integration/byval/slices=1	145862 ns	145916 ns	1.00
integration/byval/slices=3	423025 ns	423037 ns	1.00
integration/byval/reference	143927 ns	143945 ns	1.00
integration/byval/slices=2	284696 ns	284549 ns	1.00
integration/cudadevrt	102606 ns	102670.5 ns	1.00
kernel/indexing	13705 ns	13342 ns	1.03
kernel/indexing_checked	14354 ns	14086 ns	1.02
kernel/occupancy	663.8220858895705 ns	660.9593749999999 ns	1.00
kernel/launch	2178.5555555555557 ns	2204 ns	0.99
kernel/rand	16127 ns	14561 ns	1.11
array/reverse/1d	18528 ns	18374 ns	1.01
array/reverse/2dL_inplace	66523 ns	66121 ns	1.01
array/reverse/1dL	69060 ns	68975 ns	1.00
array/reverse/2d	20665 ns	20697 ns	1.00
array/reverse/1d_inplace	8551.666666666666 ns	8587.333333333334 ns	1.00
array/reverse/2d_inplace	10725 ns	10398 ns	1.03
array/reverse/2dL	72862 ns	72780 ns	1.00
array/reverse/1dL_inplace	65960 ns	66028.5 ns	1.00
array/copy	18769 ns	18723 ns	1.00
array/iteration/findall/int	151676 ns	149865 ns	1.01
array/iteration/findall/bool	134533.5 ns	132264 ns	1.02
array/iteration/findfirst/int	83676 ns	83815 ns	1.00
array/iteration/findfirst/bool	81796 ns	81845 ns	1.00
array/iteration/scalar	69218.5 ns	67004 ns	1.03
array/iteration/logical	204523.5 ns	200939.5 ns	1.02
array/iteration/findmin/1d	86343.5 ns	88921.5 ns	0.97
array/iteration/findmin/2d	117623 ns	117673.5 ns	1.00
array/reductions/reduce/Int64/1d	44657 ns	43989 ns	1.02
array/reductions/reduce/Int64/dims=1	47418.5 ns	43251 ns	1.10
array/reductions/reduce/Int64/dims=2	66765.5 ns	59962 ns	1.11
array/reductions/reduce/Int64/dims=1L	88038.5 ns	87716 ns	1.00
array/reductions/reduce/Int64/dims=2L	87506 ns	85104 ns	1.03
array/reductions/reduce/Float32/1d	35566 ns	34799.5 ns	1.02
array/reductions/reduce/Float32/dims=1	42739 ns	49659.5 ns	0.86
array/reductions/reduce/Float32/dims=2	61141 ns	57379 ns	1.07
array/reductions/reduce/Float32/dims=1L	52574 ns	52018 ns	1.01
array/reductions/reduce/Float32/dims=2L	70620 ns	70083 ns	1.01
array/reductions/mapreduce/Int64/1d	44678 ns	43593 ns	1.02
array/reductions/mapreduce/Int64/dims=1	47434.5 ns	43176 ns	1.10
array/reductions/mapreduce/Int64/dims=2	66337 ns	60012 ns	1.11
array/reductions/mapreduce/Int64/dims=1L	87984 ns	87745 ns	1.00
array/reductions/mapreduce/Int64/dims=2L	87291 ns	85352.5 ns	1.02
array/reductions/mapreduce/Float32/1d	35288 ns	34798.5 ns	1.01
array/reductions/mapreduce/Float32/dims=1	42716 ns	49547 ns	0.86
array/reductions/mapreduce/Float32/dims=2	61166 ns	56871.5 ns	1.08
array/reductions/mapreduce/Float32/dims=1L	52387 ns	51976 ns	1.01
array/reductions/mapreduce/Float32/dims=2L	70448.5 ns	69534 ns	1.01
array/broadcast	20682 ns	20738 ns	1.00
array/copyto!/gpu_to_gpu	11274 ns	11258 ns	1.00
array/copyto!/cpu_to_gpu	215613 ns	217274 ns	0.99
array/copyto!/gpu_to_cpu	283122 ns	283488 ns	1.00
array/accumulate/Int64/1d	120812 ns	118668 ns	1.02
array/accumulate/Int64/dims=1	81479 ns	80440 ns	1.01
array/accumulate/Int64/dims=2	156886 ns	156264 ns	1.00
array/accumulate/Int64/dims=1L	1753125.5 ns	1694690 ns	1.03
array/accumulate/Int64/dims=2L	963051 ns	961398 ns	1.00
array/accumulate/Float32/1d	104207.5 ns	101158 ns	1.03
array/accumulate/Float32/dims=1	77898 ns	77281 ns	1.01
array/accumulate/Float32/dims=2	146734 ns	144244 ns	1.02
array/accumulate/Float32/dims=1L	1622772 ns	1585543 ns	1.02
array/accumulate/Float32/dims=2L	669505.5 ns	657740 ns	1.02
array/construct	1347.85 ns	1296.1 ns	1.04
array/random/randn/Float32	38901.5 ns	38521.5 ns	1.01
array/random/randn!/Float32	31520 ns	31518 ns	1.00
array/random/rand!/Int64	34377 ns	34237 ns	1.00
array/random/rand!/Float32	8547.333333333334 ns	8618.333333333334 ns	0.99
array/random/rand/Int64	36018 ns	37340 ns	0.96
array/random/rand/Float32	13130 ns	13131 ns	1.00
array/permutedims/4d	51657 ns	52877.5 ns	0.98
array/permutedims/2d	52161 ns	52537 ns	0.99
array/permutedims/3d	53114 ns	53135 ns	1.00
array/sorting/1d	2735846.5 ns	2735479 ns	1.00
array/sorting/by	3304793 ns	3304665.5 ns	1.00
array/sorting/2d	1069043 ns	1068975 ns	1.00
cuda/synchronization/stream/auto	1049.3 ns	1032.4 ns	1.02
cuda/synchronization/stream/nonblocking	7695 ns	7639.6 ns	1.01
cuda/synchronization/stream/blocking	831.7951807228916 ns	843.4102564102565 ns	0.99
cuda/synchronization/context/auto	1195.9 ns	1179 ns	1.01
cuda/synchronization/context/nonblocking	7504.3 ns	8045 ns	0.93
cuda/synchronization/context/blocking	971.4230769230769 ns	931.1724137931035 ns	1.04

This comment was automatically generated by workflow using github-action-benchmark.