[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 1ad6cc116..db94f3d36 100644
--- a/src/CUDAKernels.jl
+++ b/src/CUDAKernels.jl
@@ -160,29 +160,29 @@ 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
@@ -197,7 +197,7 @@ 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 051ecc11e..d238bb8a2 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 4865baa36..f7740b61d 100644
--- a/test/runtests.jl
+++ b/test/runtests.jl
@@ -1,6 +1,6 @@
 @static if VERSION < v"1.11"
     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

Benchmark suite	Current: 180f4a5	Previous: 1d1be49	Ratio
latency/precompile	65360103745 ns	56764175729.5 ns	1.15
latency/ttfp	8508210510.5 ns	8338765591 ns	1.02
latency/import	4685973409 ns	4497026184 ns	1.04
integration/volumerhs	9609320.5 ns	9623403 ns	1.00
integration/byval/slices=1	146644 ns	147327 ns	1.00
integration/byval/slices=3	425680 ns	426202 ns	1.00
integration/byval/reference	145025.5 ns	145288 ns	1.00
integration/byval/slices=2	286085 ns	286646 ns	1.00
integration/cudadevrt	103528 ns	103542 ns	1.00
kernel/indexing	14258 ns	14224 ns	1.00
kernel/indexing_checked	14929.5 ns	15126 ns	0.99
kernel/occupancy	670.1582278481013 ns	670.1301369863014 ns	1.00
kernel/launch	2136.4 ns	2124.4 ns	1.01
kernel/rand	14792 ns	14865 ns	1.00
array/reverse/1d	19790 ns	19923 ns	0.99
array/reverse/2dL_inplace	66720 ns	66890 ns	1.00
array/reverse/1dL	69874 ns	70173 ns	1.00
array/reverse/2d	21844 ns	21888 ns	1.00
array/reverse/1d_inplace	11348 ns	9715 ns	1.17
array/reverse/2d_inplace	13262 ns	13402 ns	0.99
array/reverse/2dL	73796.5 ns	73837 ns	1.00
array/reverse/1dL_inplace	66709 ns	66895 ns	1.00
array/copy	20939 ns	20939 ns	1
array/iteration/findall/int	159242 ns	158678 ns	1.00
array/iteration/findall/bool	141881 ns	140617 ns	1.01
array/iteration/findfirst/int	161153 ns	162792 ns	0.99
array/iteration/findfirst/bool	161312.5 ns	163326 ns	0.99
array/iteration/scalar	70538 ns	73757.5 ns	0.96
array/iteration/logical	217032.5 ns	218431.5 ns	0.99
array/iteration/findmin/1d	49861.5 ns	51927 ns	0.96
array/iteration/findmin/2d	97019 ns	96989 ns	1.00
array/reductions/reduce/Int64/1d	48000 ns	44257 ns	1.08
array/reductions/reduce/Int64/dims=1	49800 ns	50794 ns	0.98
array/reductions/reduce/Int64/dims=2	68982 ns	61630 ns	1.12
array/reductions/reduce/Int64/dims=1L	88990 ns	89077 ns	1.00
array/reductions/reduce/Int64/dims=2L	90220 ns	88486.5 ns	1.02
array/reductions/reduce/Float32/1d	36714 ns	37397.5 ns	0.98
array/reductions/reduce/Float32/dims=1	47222.5 ns	41826.5 ns	1.13
array/reductions/reduce/Float32/dims=2	64405 ns	59832 ns	1.08
array/reductions/reduce/Float32/dims=1L	53545.5 ns	52417 ns	1.02
array/reductions/reduce/Float32/dims=2L	73889.5 ns	72316 ns	1.02
array/reductions/mapreduce/Int64/1d	46259.5 ns	43845 ns	1.06
array/reductions/mapreduce/Int64/dims=1	49456.5 ns	45366.5 ns	1.09
array/reductions/mapreduce/Int64/dims=2	68910 ns	61499 ns	1.12
array/reductions/mapreduce/Int64/dims=1L	88985.5 ns	88928 ns	1.00
array/reductions/mapreduce/Int64/dims=2L	90437 ns	88227 ns	1.03
array/reductions/mapreduce/Float32/1d	36765 ns	37393.5 ns	0.98
array/reductions/mapreduce/Float32/dims=1	48192 ns	50536 ns	0.95
array/reductions/mapreduce/Float32/dims=2	62947 ns	59948 ns	1.05
array/reductions/mapreduce/Float32/dims=1L	53879 ns	52655 ns	1.02
array/reductions/mapreduce/Float32/dims=2L	73500 ns	72123.5 ns	1.02
array/broadcast	19952 ns	20094 ns	0.99
array/copyto!/gpu_to_gpu	11454 ns	12939 ns	0.89
array/copyto!/cpu_to_gpu	215164 ns	218766.5 ns	0.98
array/copyto!/gpu_to_cpu	282301 ns	286506 ns	0.99
array/accumulate/Int64/1d	127397 ns	125248 ns	1.02
array/accumulate/Int64/dims=1	85196 ns	83707 ns	1.02
array/accumulate/Int64/dims=2	158769 ns	157937.5 ns	1.01
array/accumulate/Int64/dims=1L	1795083 ns	1710672 ns	1.05
array/accumulate/Int64/dims=2L	973567 ns	967046.5 ns	1.01
array/accumulate/Float32/1d	111567 ns	109958 ns	1.01
array/accumulate/Float32/dims=1	81756 ns	80681 ns	1.01
array/accumulate/Float32/dims=2	149503.5 ns	147571.5 ns	1.01
array/accumulate/Float32/dims=1L	1713669 ns	1619100 ns	1.06
array/accumulate/Float32/dims=2L	715021 ns	698632 ns	1.02
array/construct	1247.6 ns	1292.8 ns	0.97
array/random/randn/Float32	45613.5 ns	45085.5 ns	1.01
array/random/randn!/Float32	25126 ns	25182 ns	1.00
array/random/rand!/Int64	27413 ns	27387 ns	1.00
array/random/rand!/Float32	8706 ns	8796.333333333334 ns	0.99
array/random/rand/Int64	30653.5 ns	29906 ns	1.02
array/random/rand/Float32	12880 ns	13172 ns	0.98
array/permutedims/4d	52750 ns	55451 ns	0.95
array/permutedims/2d	54303.5 ns	54366.5 ns	1.00
array/permutedims/3d	54265 ns	55262 ns	0.98
array/sorting/1d	2759316.5 ns	2758900.5 ns	1.00
array/sorting/by	3368651.5 ns	3345477 ns	1.01
array/sorting/2d	1088246 ns	1082617 ns	1.01
cuda/synchronization/stream/auto	1020.5 ns	1052 ns	0.97
cuda/synchronization/stream/nonblocking	7056.8 ns	7628.6 ns	0.93
cuda/synchronization/stream/blocking	817.49 ns	835.3333333333334 ns	0.98
cuda/synchronization/context/auto	1158.9 ns	1173.3 ns	0.99
cuda/synchronization/context/nonblocking	8459.8 ns	8290.1 ns	1.02
cuda/synchronization/context/blocking	887.6607142857143 ns	927.025 ns	0.96

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