[Do not merge] Test KernelIntrinsics

[christiangnrd]

Not a draft to also run benchmarks

[github-actions]

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

Click here to view the suggested changes.

diff --git a/src/MetalKernels.jl b/src/MetalKernels.jl
index 4e856194..7573c5e1 100644
--- a/src/MetalKernels.jl
+++ b/src/MetalKernels.jl
@@ -136,26 +136,26 @@ end
 
 KI.argconvert(::MetalBackend, arg) = mtlconvert(arg)
 
-function KI.kernel_function(::MetalBackend, f::F, tt::TT=Tuple{}; name=nothing, kwargs...) where {F,TT}
+function KI.kernel_function(::MetalBackend, f::F, tt::TT = Tuple{}; name = nothing, kwargs...) where {F, TT}
     kern = mtlfunction(f, tt; name, kwargs...)
-    KI.Kernel{MetalBackend, typeof(kern)}(MetalBackend(), kern)
+    return KI.Kernel{MetalBackend, typeof(kern)}(MetalBackend(), kern)
 end
 
-function (obj::KI.Kernel{MetalBackend})(args...; numworkgroups=1, workgroupsize=1)
+function (obj::KI.Kernel{MetalBackend})(args...; numworkgroups = 1, workgroupsize = 1)
     KI.check_launch_args(numworkgroups, workgroupsize)
 
-    obj.kern(args...; threads=workgroupsize, groups=numworkgroups)
+    return obj.kern(args...; threads = workgroupsize, groups = numworkgroups)
 end
 
 
-function KI.kernel_max_work_group_size(kikern::KI.Kernel{<:MetalBackend}; max_work_items::Int=typemax(Int))::Int
-    Int(min(kikern.kern.pipeline.maxTotalThreadsPerThreadgroup, max_work_items))
+function KI.kernel_max_work_group_size(kikern::KI.Kernel{<:MetalBackend}; max_work_items::Int = typemax(Int))::Int
+    return Int(min(kikern.kern.pipeline.maxTotalThreadsPerThreadgroup, max_work_items))
 end
 function KI.max_work_group_size(::MetalBackend)::Int
-    Int(device().maxThreadsPerThreadgroup.width)
+    return Int(device().maxThreadsPerThreadgroup.width)
 end
 function KI.multiprocessor_count(::MetalBackend)::Int
-    Metal.num_gpu_cores()
+    return Metal.num_gpu_cores()
 end
 
 
diff --git a/src/broadcast.jl b/src/broadcast.jl
index 5d107ec2..3455fad2 100644
--- a/src/broadcast.jl
+++ b/src/broadcast.jl
@@ -66,9 +66,9 @@ end
     if _broadcast_shapes[Is] > BROADCAST_SPECIALIZATION_THRESHOLD
         ## COV_EXCL_START
         function broadcast_cartesian_static(dest, bc, Is)
-             i = KI.get_global_id().x
-             stride = KI.get_global_size().x
-             while 1 <= i <= length(dest)
+            i = KI.get_global_id().x
+            stride = KI.get_global_size().x
+            while 1 <= i <= length(dest)
                 I = @inbounds Is[i]
                 @inbounds dest[I] = bc[I]
                 i += stride
@@ -91,13 +91,13 @@ end
        (isa(IndexStyle(dest), IndexLinear) && isa(IndexStyle(bc), IndexLinear))
         ## COV_EXCL_START
         function broadcast_linear(dest, bc)
-             i = KI.get_global_id().x
-             stride = KI.get_global_size().x
-             while 1 <= i <= length(dest)
-                 @inbounds dest[i] = bc[i]
-                 i += stride
-             end
-             return
+            i = KI.get_global_id().x
+            stride = KI.get_global_size().x
+            while 1 <= i <= length(dest)
+                @inbounds dest[i] = bc[i]
+                i += stride
+            end
+            return
         end
         ## COV_EXCL_STOP
 
@@ -168,9 +168,9 @@ end
     else
         ## COV_EXCL_START
         function broadcast_cartesian(dest, bc)
-             i = KI.get_global_id().x
-             stride = KI.get_global_size().x
-             while 1 <= i <= length(dest)
+            i = KI.get_global_id().x
+            stride = KI.get_global_size().x
+            while 1 <= i <= length(dest)
                 I = @inbounds CartesianIndices(dest)[i]
                 @inbounds dest[I] = bc[I]
                 i += stride
diff --git a/src/device/random.jl b/src/device/random.jl
index 12b053a2..edc999cd 100644
--- a/src/device/random.jl
+++ b/src/device/random.jl
@@ -89,8 +89,8 @@ end
         @inbounds global_random_counters()[simdgroupId]
     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 7be5ef43..a737e8d0 100644
--- a/src/mapreduce.jl
+++ b/src/mapreduce.jl
@@ -224,7 +224,8 @@ function GPUArrays.mapreducedim!(f::F, op::OP, R::WrappedMtlArray{T},
     # we might not be able to launch all those threads to reduce each slice in one go.
     # that's why each threads also loops across their inputs, processing multiple values
     # so that we can span the entire reduction dimension using a single item group.
-    kernel = KI.@kernel backend launch = false partial_mapreduce_device(f, op, init, Val(maxthreads), Val(Rreduce), Val(Rother),
+    kernel = KI.@kernel backend launch = false partial_mapreduce_device(
+        f, op, init, Val(maxthreads), Val(Rreduce), Val(Rother),
                                                           Val(UInt64(length(Rother))), Val(grain), Val(shuffle), R, A)
 
     # how many threads do we want?
@@ -260,7 +261,8 @@ function GPUArrays.mapreducedim!(f::F, op::OP, R::WrappedMtlArray{T},
         # we can cover the dimensions to reduce using a single group
         kernel(f, op, init, Val(maxthreads), Val(Rreduce), Val(Rother),
                Val(UInt64(length(Rother))), Val(grain), Val(shuffle), R, A;
-               workgroupsize = threads, numworkgroups = groups)
+            workgroupsize = threads, numworkgroups = groups
+        )
     else
         # temporary empty array whose type will match the final partial array
 	    partial = similar(R, ntuple(_ -> 0, Val(ndims(R)+1)))
@@ -287,7 +289,8 @@ function GPUArrays.mapreducedim!(f::F, op::OP, R::WrappedMtlArray{T},
         partial_kernel(f, op, init, Val(threads), Val(Rreduce),
                         Val(Rother), Val(UInt64(length(Rother))),
                         Val(grain), Val(shuffle), partial, A;
-                        numworkgroups = partial_groups, workgroupsize = partial_threads)
+            numworkgroups = partial_groups, workgroupsize = partial_threads
+        )
 
         GPUArrays.mapreducedim!(identity, op, R, partial; init=init)
     end
diff --git a/test/kernelabstractions.jl b/test/kernelabstractions.jl
index cda5b249..339fcbc8 100644
--- a/test/kernelabstractions.jl
+++ b/test/kernelabstractions.jl
@@ -7,6 +7,6 @@ Testsuite.testsuite(MetalBackend, "Metal", Metal, MtlArray, Metal.MtlDeviceArray
     "Convert",           # depends on https://github.com/JuliaGPU/Metal.jl/issues/69
     "SpecialFunctions",  # gamma and erfc not currently supported on Metal.jl
     "sparse",            # not supported yet
-    "CPU synchronization",
-    "fallback test: callable types",
+            "CPU synchronization",
+            "fallback test: callable types",
 ]))
diff --git a/test/runtests.jl b/test/runtests.jl
index 32b45c8c..14fcfb93 100644
--- a/test/runtests.jl
+++ b/test/runtests.jl
@@ -1,6 +1,6 @@
 @static if VERSION < v"1.11" && get(ENV, "BUILDKITE_PIPELINE_NAME", "Metal.jl") == "Metal.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 Metal

[github-actions]

Metal Benchmarks

Details

Benchmark suite	Current: 3be3a7c	Previous: b94fd4b	Ratio
array/accumulate/Float32/1d	1082834 ns	1098958 ns	0.99
array/accumulate/Float32/dims=1	1756458 ns	1554708 ns	1.13
array/accumulate/Float32/dims=1L	10573166 ns	9848583.5 ns	1.07
array/accumulate/Float32/dims=2	2191167 ns	1886771 ns	1.16
array/accumulate/Float32/dims=2L	7852125 ns	7256459 ns	1.08
array/accumulate/Int64/1d	1368750 ns	1261958 ns	1.08
array/accumulate/Int64/dims=1	2006875 ns	1824291.5 ns	1.10
array/accumulate/Int64/dims=1L	12374354 ns	11664208.5 ns	1.06
array/accumulate/Int64/dims=2	2391083.5 ns	2170333.5 ns	1.10
array/accumulate/Int64/dims=2L	10179771 ns	10120062.5 ns	1.01
array/broadcast	551520.5 ns	605916 ns	0.91
array/construct	6542 ns	6292 ns	1.04
array/permutedims/2d	1217833 ns	1168125 ns	1.04
array/permutedims/3d	1831458.5 ns	1673084 ns	1.09
array/permutedims/4d	2757333 ns	2365959 ns	1.17
array/private/copy	828291.5 ns	545792 ns	1.52
array/private/copyto!/cpu_to_gpu	752521 ns	802916 ns	0.94
array/private/copyto!/gpu_to_cpu	752666.5 ns	801917 ns	0.94
array/private/copyto!/gpu_to_gpu	551708 ns	634458 ns	0.87
array/private/iteration/findall/bool	1539041 ns	1402750 ns	1.10
array/private/iteration/findall/int	1665709 ns	1564021 ns	1.07
array/private/iteration/findfirst/bool	2112708 ns	2055916 ns	1.03
array/private/iteration/findfirst/int	2194250 ns	2064479.5 ns	1.06
array/private/iteration/findmin/1d	2674916 ns	2499959 ns	1.07
array/private/iteration/findmin/2d	1914958 ns	1790791 ns	1.07
array/private/iteration/logical	2782333.5 ns	2631896 ns	1.06
array/private/iteration/scalar	3324937.5 ns	5047625 ns	0.66
array/random/rand/Float32	828937.5 ns	582958 ns	1.42
array/random/rand/Int64	901000 ns	775667 ns	1.16
array/random/rand!/Float32	540833 ns	574750 ns	0.94
array/random/rand!/Int64	532833 ns	550792 ns	0.97
array/random/randn/Float32	1045791 ns	1006937.5 ns	1.04
array/random/randn!/Float32	724125 ns	755666 ns	0.96
array/reductions/mapreduce/Float32/1d	801458 ns	1029500 ns	0.78
array/reductions/mapreduce/Float32/dims=1	828708 ns	840875 ns	0.99
array/reductions/mapreduce/Float32/dims=1L	1370146 ns	1324000 ns	1.03
array/reductions/mapreduce/Float32/dims=2	880958 ns	860875 ns	1.02
array/reductions/mapreduce/Float32/dims=2L	1894104.5 ns	1799541 ns	1.05
array/reductions/mapreduce/Int64/1d	1280208.5 ns	1374875 ns	0.93
array/reductions/mapreduce/Int64/dims=1	1122020.5 ns	1097625 ns	1.02
array/reductions/mapreduce/Int64/dims=1L	1998208 ns	2002854 ns	1.00
array/reductions/mapreduce/Int64/dims=2	1357291.5 ns	1145000 ns	1.19
array/reductions/mapreduce/Int64/dims=2L	4164313 ns	3614000 ns	1.15
array/reductions/reduce/Float32/1d	810270.5 ns	1028437.5 ns	0.79
array/reductions/reduce/Float32/dims=1	819396 ns	832667 ns	0.98
array/reductions/reduce/Float32/dims=1L	1371729.5 ns	1318416.5 ns	1.04
array/reductions/reduce/Float32/dims=2	880500 ns	853041.5 ns	1.03
array/reductions/reduce/Float32/dims=2L	1885250 ns	1810250 ns	1.04
array/reductions/reduce/Int64/1d	1265750 ns	1516958 ns	0.83
array/reductions/reduce/Int64/dims=1	1113209 ns	1095375 ns	1.02
array/reductions/reduce/Int64/dims=1L	2066750 ns	2023499.5 ns	1.02
array/reductions/reduce/Int64/dims=2	1347646 ns	1240750 ns	1.09
array/reductions/reduce/Int64/dims=2L	4256479.5 ns	4233875 ns	1.01
array/shared/copy	215250 ns	252417 ns	0.85
array/shared/copyto!/cpu_to_gpu	83750 ns	80750 ns	1.04
array/shared/copyto!/gpu_to_cpu	83709 ns	80667 ns	1.04
array/shared/copyto!/gpu_to_gpu	84500 ns	83083 ns	1.02
array/shared/iteration/findall/bool	1543875 ns	1427208.5 ns	1.08
array/shared/iteration/findall/int	1661625 ns	1559875 ns	1.07
array/shared/iteration/findfirst/bool	1711792 ns	1649000 ns	1.04
array/shared/iteration/findfirst/int	1792250 ns	1672458 ns	1.07
array/shared/iteration/findmin/1d	2234874.5 ns	2115583 ns	1.06
array/shared/iteration/findmin/2d	1913416 ns	1792625 ns	1.07
array/shared/iteration/logical	2558542 ns	2292167 ns	1.12
array/shared/iteration/scalar	206083 ns	199958 ns	1.03
integration/byval/reference	1593645.5 ns	1544250 ns	1.03
integration/byval/slices=1	1600792 ns	1560229.5 ns	1.03
integration/byval/slices=2	2732166 ns	2598333.5 ns	1.05
integration/byval/slices=3	19612583 ns	8092333 ns	2.42
integration/metaldevrt	861354.5 ns	868125 ns	0.99
kernel/indexing	485750 ns	592667 ns	0.82
kernel/indexing_checked	487250 ns	598292 ns	0.81
kernel/launch	12625 ns	11791.5 ns	1.07
kernel/rand	536250 ns	570709 ns	0.94
latency/import	1692594000.5 ns	1425597062.5 ns	1.19
latency/precompile	30415191770.5 ns	25453724708 ns	1.19
latency/ttfp	2536515958.5 ns	2341177208 ns	1.08
metal/synchronization/context	19792 ns	19667 ns	1.01
metal/synchronization/stream	19292 ns	18459 ns	1.05

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