[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: 7f653a1	Previous: ac0202d	Ratio
array/accumulate/Float32/1d	406916 ns	407375 ns	1.00
array/accumulate/Float32/dims=1	377959 ns	351791 ns	1.07
array/accumulate/Float32/dims=1L	8804750 ns	8796584 ns	1.00
array/accumulate/Float32/dims=2	451000 ns	439792 ns	1.03
array/accumulate/Float32/dims=2L	2686083 ns	2603208 ns	1.03
array/accumulate/Int64/1d	913375 ns	842333 ns	1.08
array/accumulate/Int64/dims=1	1060375 ns	944125 ns	1.12
array/accumulate/Int64/dims=1L	9665667 ns	9521000 ns	1.02
array/accumulate/Int64/dims=2	1035667 ns	1247208 ns	0.83
array/accumulate/Int64/dims=2L	6523042 ns	6499125 ns	1.00
array/broadcast	230167 ns	222000 ns	1.04
array/construct	2333 ns	2417 ns	0.97
array/permutedims/2d	525334 ns	449458 ns	1.17
array/permutedims/3d	1127917 ns	1011000 ns	1.12
array/permutedims/4d	1479042 ns	1275416 ns	1.16
array/private/copy	239167 ns	229791 ns	1.04
array/private/copyto!/cpu_to_gpu	199417 ns	195833 ns	1.02
array/private/copyto!/gpu_to_cpu	196875 ns	194167 ns	1.01
array/private/copyto!/gpu_to_gpu	212042 ns	173708 ns	1.22
array/private/iteration/findall/bool	1146458 ns	1057250 ns	1.08
array/private/iteration/findall/int	1290083 ns	1228125 ns	1.05
array/private/iteration/findfirst/bool	1162291 ns	1111542 ns	1.05
array/private/iteration/findfirst/int	1185208 ns	1117792 ns	1.06
array/private/iteration/findmin/1d	1283458 ns	1229625 ns	1.04
array/private/iteration/findmin/2d	785458 ns	769709 ns	1.02
array/private/iteration/logical	1815791 ns	1688959 ns	1.08
array/private/iteration/scalar	1250250 ns	1216958 ns	1.03
array/random/rand/Float32	398916 ns	426125 ns	0.94
array/random/rand/Int64	519625 ns	513417 ns	1.01
array/random/rand!/Float32	381833 ns	313042 ns	1.22
array/random/rand!/Int64	409917 ns	406791 ns	1.01
array/random/randn/Float32	340875 ns	391833 ns	0.87
array/random/randn!/Float32	356375 ns	340208 ns	1.05
array/reductions/mapreduce/Float32/1d	439500 ns	383792 ns	1.15
array/reductions/mapreduce/Float32/dims=1	269083 ns	263000 ns	1.02
array/reductions/mapreduce/Float32/dims=1L	625417 ns	621791 ns	1.01
array/reductions/mapreduce/Float32/dims=2	362417 ns	275708 ns	1.31
array/reductions/mapreduce/Float32/dims=2L	1050208 ns	841125 ns	1.25
array/reductions/mapreduce/Int64/1d	614750 ns	621125 ns	0.99
array/reductions/mapreduce/Int64/dims=1	440208 ns	450958 ns	0.98
array/reductions/mapreduce/Int64/dims=1L	1031625 ns	1026500 ns	1.00
array/reductions/mapreduce/Int64/dims=2	570334 ns	778292 ns	0.73
array/reductions/mapreduce/Int64/dims=2L	2246167 ns	2179166 ns	1.03
array/reductions/reduce/Float32/1d	437750 ns	412916 ns	1.06
array/reductions/reduce/Float32/dims=1	287917 ns	345084 ns	0.83
array/reductions/reduce/Float32/dims=1L	632167 ns	635167 ns	1.00
array/reductions/reduce/Float32/dims=2	231458 ns	228750 ns	1.01
array/reductions/reduce/Float32/dims=2L	450125 ns	448708 ns	1.00
array/reductions/reduce/Int64/1d	593875 ns	576542 ns	1.03
array/reductions/reduce/Int64/dims=1	613375 ns	614541 ns	1.00
array/reductions/reduce/Int64/dims=1L	1024666 ns	1030125 ns	0.99
array/reductions/reduce/Int64/dims=2	244459 ns	240250 ns	1.02
array/reductions/reduce/Int64/dims=2L	655375 ns	644833 ns	1.02
array/shared/copy	136625 ns	136583 ns	1.00
array/shared/copyto!/cpu_to_gpu	37708 ns	38292 ns	0.98
array/shared/copyto!/gpu_to_cpu	38000 ns	38167 ns	1.00
array/shared/copyto!/gpu_to_gpu	38292 ns	38917 ns	0.98
array/shared/iteration/findall/bool	1142708 ns	1076292 ns	1.06
array/shared/iteration/findall/int	1304458 ns	1220458 ns	1.07
array/shared/iteration/findfirst/bool	1002791 ns	956459 ns	1.05
array/shared/iteration/findfirst/int	848000 ns	650041 ns	1.30
array/shared/iteration/findmin/1d	1152333 ns	1089541 ns	1.06
array/shared/iteration/findmin/2d	1088958 ns	1075500 ns	1.01
array/shared/iteration/logical	1446500 ns	1553459 ns	0.93
array/shared/iteration/scalar	4160.714285714285 ns	4023.714285714286 ns	1.03
array/sorting/1d	1603541 ns	1786083 ns	0.90
array/sorting/2d	8302125 ns	8305833 ns	1.00
integration/byval/reference	1111167 ns	1119042 ns	0.99
integration/byval/slices=1	1104792 ns	1109583 ns	1.00
integration/byval/slices=2	2022583 ns	2021291 ns	1.00
integration/byval/slices=3	7723750 ns	6610875 ns	1.17
integration/metaldevrt	387458 ns	379833 ns	1.02
kernel/indexing	207125 ns	208708 ns	0.99
kernel/indexing_checked	374583 ns	370667 ns	1.01
kernel/launch	1891.6 ns	1883.3 ns	1.00
kernel/rand	393750 ns	393792 ns	1.00
latency/import	2208580083 ns	2038510291 ns	1.08
latency/precompile	44965841000 ns	38792921333 ns	1.16
latency/ttfp	2539808458 ns	2354706125 ns	1.08
metal/synchronization/context	583.5635359116022 ns	583.7912087912088 ns	1.00
metal/synchronization/stream	331.4663677130045 ns	332.58744394618833 ns	1.00

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

[codecov]

Codecov Report

❌ Patch coverage is 96.66667% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 86.20%. Comparing base (b70608b) to head (7f653a1).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/MetalKernels.jl	94.11%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #688      +/-   ##
==========================================
- Coverage   86.21%   86.20%   -0.01%     
==========================================
  Files          76       76              
  Lines        5246     5264      +18     
==========================================
+ Hits         4523     4538      +15     
- Misses        723      726       +3     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.