From b3efa6af67cc7fed7cbba3b0e2dbc50f6010f99d Mon Sep 17 00:00:00 2001 From: Tom Lin Date: Thu, 10 Jun 2021 04:20:40 +0100 Subject: [PATCH] Initial Julia implementation --- .github/workflows/main.yaml | 17 + JuliaStream.jl/.JuliaFormatter.toml | 2 + JuliaStream.jl/.gitignore | 5 + JuliaStream.jl/Manifest.toml | 411 ++++++++++++++++++++++++ JuliaStream.jl/Project.toml | 14 + JuliaStream.jl/README.md | 30 ++ JuliaStream.jl/src/AMDGPUStream.jl | 178 ++++++++++ JuliaStream.jl/src/CUDAStream.jl | 146 +++++++++ JuliaStream.jl/src/DistributedStream.jl | 84 +++++ JuliaStream.jl/src/JuliaStream.jl | 4 + JuliaStream.jl/src/PlainStream.jl | 64 ++++ JuliaStream.jl/src/Stream.jl | 292 +++++++++++++++++ JuliaStream.jl/src/StreamData.jl | 8 + JuliaStream.jl/src/ThreadedStream.jl | 72 +++++ 14 files changed, 1327 insertions(+) create mode 100644 JuliaStream.jl/.JuliaFormatter.toml create mode 100644 JuliaStream.jl/.gitignore create mode 100644 JuliaStream.jl/Manifest.toml create mode 100644 JuliaStream.jl/Project.toml create mode 100644 JuliaStream.jl/README.md create mode 100644 JuliaStream.jl/src/AMDGPUStream.jl create mode 100644 JuliaStream.jl/src/CUDAStream.jl create mode 100644 JuliaStream.jl/src/DistributedStream.jl create mode 100644 JuliaStream.jl/src/JuliaStream.jl create mode 100644 JuliaStream.jl/src/PlainStream.jl create mode 100644 JuliaStream.jl/src/Stream.jl create mode 100644 JuliaStream.jl/src/StreamData.jl create mode 100644 JuliaStream.jl/src/ThreadedStream.jl diff --git a/.github/workflows/main.yaml b/.github/workflows/main.yaml index 20e1034..83bcd9d 100644 --- a/.github/workflows/main.yaml +++ b/.github/workflows/main.yaml @@ -3,6 +3,23 @@ on: [push, pull_request] jobs: + test-julia: + runs-on: ubuntu-18.04 + steps: + - uses: actions/checkout@v2 + - name: Setup project + run: julia --project -e 'import Pkg; Pkg.instantiate()' + - name: Test run PlainStream.jl + run: julia --project src/PlainStream.jl --arraysize 100 + - name: Test run ThreadedStream.jl + run: julia --threads 2 --project src/ThreadedStream.jl --arraysize 100 + - name: Test run DistributedStream.jl + run: julia -p2 --project src/DistributedStream.jl --arraysize 100 + - name: Test run CUDAStream.jl + run: julia --project src/CUDAStream.jl --list + - name: Test run AMDGPUStream.jl + run: julia --project src/AMDGPUStream.jl --list + test: runs-on: ubuntu-18.04 steps: diff --git a/JuliaStream.jl/.JuliaFormatter.toml b/JuliaStream.jl/.JuliaFormatter.toml new file mode 100644 index 0000000..ac95ddd --- /dev/null +++ b/JuliaStream.jl/.JuliaFormatter.toml @@ -0,0 +1,2 @@ +indent = 2 +margin = 100 \ No newline at end of file diff --git a/JuliaStream.jl/.gitignore b/JuliaStream.jl/.gitignore new file mode 100644 index 0000000..12b143b --- /dev/null +++ b/JuliaStream.jl/.gitignore @@ -0,0 +1,5 @@ +*.jl.cov +*.jl.*.cov +*.jl.mem +/docs/build/ +/docs/Manifest.toml \ No newline at end of file diff --git a/JuliaStream.jl/Manifest.toml b/JuliaStream.jl/Manifest.toml new file mode 100644 index 0000000..c60d77f --- /dev/null +++ b/JuliaStream.jl/Manifest.toml @@ -0,0 +1,411 @@ +# This file is machine-generated - editing it directly is not advised + +[[AMDGPU]] +deps = ["AbstractFFTs", "Adapt", "BinaryProvider", "CEnum", "GPUArrays", "GPUCompiler", "LLVM", "Libdl", "LinearAlgebra", "MacroTools", "Printf", "Random", "Requires", "Setfield", "hsa_rocr_jll", "hsakmt_roct_jll"] +git-tree-sha1 = "04fdb3923ac6f55fa7347dce0f0f6f10e321e2e9" +uuid = "21141c5a-9bdb-4563-92ae-f87d6854732e" +version = "0.2.7" + +[[AbstractFFTs]] +deps = ["LinearAlgebra"] +git-tree-sha1 = "485ee0867925449198280d4af84bdb46a2a404d0" +uuid = "621f4979-c628-5d54-868e-fcf4e3e8185c" +version = "1.0.1" + +[[Adapt]] +deps = ["LinearAlgebra"] +git-tree-sha1 = "84918055d15b3114ede17ac6a7182f68870c16f7" +uuid = "79e6a3ab-5dfb-504d-930d-738a2a938a0e" +version = "3.3.1" + +[[ArgParse]] +deps = ["Logging", "TextWrap"] +git-tree-sha1 = "3102bce13da501c9104df33549f511cd25264d7d" +uuid = "c7e460c6-2fb9-53a9-8c5b-16f535851c63" +version = "1.1.4" + +[[ArgTools]] +uuid = "0dad84c5-d112-42e6-8d28-ef12dabb789f" + +[[Artifacts]] +uuid = "56f22d72-fd6d-98f1-02f0-08ddc0907c33" + +[[BFloat16s]] +deps = ["LinearAlgebra", "Test"] +git-tree-sha1 = "4af69e205efc343068dc8722b8dfec1ade89254a" +uuid = "ab4f0b2a-ad5b-11e8-123f-65d77653426b" +version = "0.1.0" + +[[Base64]] +uuid = "2a0f44e3-6c83-55bd-87e4-b1978d98bd5f" + +[[BinaryProvider]] +deps = ["Libdl", "Logging", "SHA"] +git-tree-sha1 = "ecdec412a9abc8db54c0efc5548c64dfce072058" +uuid = "b99e7846-7c00-51b0-8f62-c81ae34c0232" +version = "0.5.10" + +[[Bzip2_jll]] +deps = ["Artifacts", "JLLWrappers", "Libdl", "Pkg"] +git-tree-sha1 = "c3598e525718abcc440f69cc6d5f60dda0a1b61e" +uuid = "6e34b625-4abd-537c-b88f-471c36dfa7a0" +version = "1.0.6+5" + +[[CEnum]] +git-tree-sha1 = "215a9aa4a1f23fbd05b92769fdd62559488d70e9" +uuid = "fa961155-64e5-5f13-b03f-caf6b980ea82" +version = "0.4.1" + +[[CUDA]] +deps = ["AbstractFFTs", "Adapt", "BFloat16s", "CEnum", "CompilerSupportLibraries_jll", "DataStructures", "ExprTools", "GPUArrays", "GPUCompiler", "LLVM", "LazyArtifacts", "Libdl", "LinearAlgebra", "Logging", "MacroTools", "Memoize", "Printf", "Random", "Random123", "RandomNumbers", "Reexport", "Requires", "SparseArrays", "SpecialFunctions", "TimerOutputs"] +git-tree-sha1 = "364179416eabc34c9ca32126a6bdb431680c3bad" +uuid = "052768ef-5323-5732-b1bb-66c8b64840ba" +version = "3.2.1" + +[[ChainRulesCore]] +deps = ["Compat", "LinearAlgebra", "SparseArrays"] +git-tree-sha1 = "8b31cc69cbc38c5c826aaa1c890c694be3622d99" +uuid = "d360d2e6-b24c-11e9-a2a3-2a2ae2dbcce4" +version = "0.10.3" + +[[Compat]] +deps = ["Base64", "Dates", "DelimitedFiles", "Distributed", "InteractiveUtils", "LibGit2", "Libdl", "LinearAlgebra", "Markdown", "Mmap", "Pkg", "Printf", "REPL", "Random", "SHA", "Serialization", "SharedArrays", "Sockets", "SparseArrays", "Statistics", "Test", "UUIDs", "Unicode"] +git-tree-sha1 = "e4e2b39db08f967cc1360951f01e8a75ec441cab" +uuid = "34da2185-b29b-5c13-b0c7-acf172513d20" +version = "3.30.0" + +[[CompilerSupportLibraries_jll]] +deps = ["Artifacts", "Libdl"] +uuid = "e66e0078-7015-5450-92f7-15fbd957f2ae" + +[[ConstructionBase]] +deps = ["LinearAlgebra"] +git-tree-sha1 = "1dc43957fb9a1574fa1b7a449e101bd1fd3a9fb7" +uuid = "187b0558-2788-49d3-abe0-74a17ed4e7c9" +version = "1.2.1" + +[[DataStructures]] +deps = ["Compat", "InteractiveUtils", "OrderedCollections"] +git-tree-sha1 = "4437b64df1e0adccc3e5d1adbc3ac741095e4677" +uuid = "864edb3b-99cc-5e75-8d2d-829cb0a9cfe8" +version = "0.18.9" + +[[Dates]] +deps = ["Printf"] +uuid = "ade2ca70-3891-5945-98fb-dc099432e06a" + +[[DelimitedFiles]] +deps = ["Mmap"] +uuid = "8bb1440f-4735-579b-a4ab-409b98df4dab" + +[[Distributed]] +deps = ["Random", "Serialization", "Sockets"] +uuid = "8ba89e20-285c-5b6f-9357-94700520ee1b" + +[[DocStringExtensions]] +deps = ["LibGit2"] +git-tree-sha1 = "a32185f5428d3986f47c2ab78b1f216d5e6cc96f" +uuid = "ffbed154-4ef7-542d-bbb7-c09d3a79fcae" +version = "0.8.5" + +[[Downloads]] +deps = ["ArgTools", "LibCURL", "NetworkOptions"] +uuid = "f43a241f-c20a-4ad4-852c-f6b1247861c6" + +[[Elfutils_jll]] +deps = ["Artifacts", "Bzip2_jll", "JLLWrappers", "Libdl", "Pkg", "XZ_jll", "Zlib_jll", "argp_standalone_jll", "fts_jll", "obstack_jll"] +git-tree-sha1 = "76cbf1134983cfb371ad77117bb2659600ed64d6" +uuid = "ab5a07f8-06af-567f-a878-e8bb879eba5a" +version = "0.179.0+0" + +[[ExprTools]] +git-tree-sha1 = "10407a39b87f29d47ebaca8edbc75d7c302ff93e" +uuid = "e2ba6199-217a-4e67-a87a-7c52f15ade04" +version = "0.1.3" + +[[Future]] +deps = ["Random"] +uuid = "9fa8497b-333b-5362-9e8d-4d0656e87820" + +[[GPUArrays]] +deps = ["AbstractFFTs", "Adapt", "LinearAlgebra", "Printf", "Random", "Serialization", "Statistics"] +git-tree-sha1 = "df5b8569904c5c10e84c640984cfff054b18c086" +uuid = "0c68f7d7-f131-5f86-a1c3-88cf8149b2d7" +version = "6.4.1" + +[[GPUCompiler]] +deps = ["DataStructures", "ExprTools", "InteractiveUtils", "LLVM", "Libdl", "Logging", "Scratch", "Serialization", "TimerOutputs", "UUIDs"] +git-tree-sha1 = "42d635f6d87af125b86288df3819f805fb4d851a" +uuid = "61eb1bfa-7361-4325-ad38-22787b887f55" +version = "0.11.5" + +[[InteractiveUtils]] +deps = ["Markdown"] +uuid = "b77e0a4c-d291-57a0-90e8-8db25a27a240" + +[[JLLWrappers]] +deps = ["Preferences"] +git-tree-sha1 = "642a199af8b68253517b80bd3bfd17eb4e84df6e" +uuid = "692b3bcd-3c85-4b1f-b108-f13ce0eb3210" +version = "1.3.0" + +[[LLVM]] +deps = ["CEnum", "Libdl", "Printf", "Unicode"] +git-tree-sha1 = "b499c68a45249b0385585c62f4a9b62b5db8e691" +uuid = "929cbde3-209d-540e-8aea-75f648917ca0" +version = "3.7.1" + +[[LazyArtifacts]] +deps = ["Artifacts", "Pkg"] +uuid = "4af54fe1-eca0-43a8-85a7-787d91b784e3" + +[[LibCURL]] +deps = ["LibCURL_jll", "MozillaCACerts_jll"] +uuid = "b27032c2-a3e7-50c8-80cd-2d36dbcbfd21" + +[[LibCURL_jll]] +deps = ["Artifacts", "LibSSH2_jll", "Libdl", "MbedTLS_jll", "Zlib_jll", "nghttp2_jll"] +uuid = "deac9b47-8bc7-5906-a0fe-35ac56dc84c0" + +[[LibGit2]] +deps = ["Base64", "NetworkOptions", "Printf", "SHA"] +uuid = "76f85450-5226-5b5a-8eaa-529ad045b433" + +[[LibSSH2_jll]] +deps = ["Artifacts", "Libdl", "MbedTLS_jll"] +uuid = "29816b5a-b9ab-546f-933c-edad1886dfa8" + +[[Libdl]] +uuid = "8f399da3-3557-5675-b5ff-fb832c97cbdb" + +[[LinearAlgebra]] +deps = ["Libdl"] +uuid = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e" + +[[LogExpFunctions]] +deps = ["DocStringExtensions", "LinearAlgebra"] +git-tree-sha1 = "1ba664552f1ef15325e68dc4c05c3ef8c2d5d885" +uuid = "2ab3a3ac-af41-5b50-aa03-7779005ae688" +version = "0.2.4" + +[[Logging]] +uuid = "56ddb016-857b-54e1-b83d-db4d58db5568" + +[[MacroTools]] +deps = ["Markdown", "Random"] +git-tree-sha1 = "6a8a2a625ab0dea913aba95c11370589e0239ff0" +uuid = "1914dd2f-81c6-5fcd-8719-6d5c9610ff09" +version = "0.5.6" + +[[Markdown]] +deps = ["Base64"] +uuid = "d6f4376e-aef5-505a-96c1-9c027394607a" + +[[MbedTLS_jll]] +deps = ["Artifacts", "Libdl"] +uuid = "c8ffd9c3-330d-5841-b78e-0817d7145fa1" + +[[Memoize]] +deps = ["MacroTools"] +git-tree-sha1 = "2b1dfcba103de714d31c033b5dacc2e4a12c7caa" +uuid = "c03570c3-d221-55d1-a50c-7939bbd78826" +version = "0.4.4" + +[[Mmap]] +uuid = "a63ad114-7e13-5084-954f-fe012c677804" + +[[MozillaCACerts_jll]] +uuid = "14a3606d-f60d-562e-9121-12d972cd8159" + +[[NUMA_jll]] +deps = ["Libdl", "Pkg"] +git-tree-sha1 = "778f9bd14400cff2c32ed357e12766ac0e3d766e" +uuid = "7f51dc2b-bb24-59f8-b771-bb1490e4195d" +version = "2.0.13+1" + +[[NetworkOptions]] +uuid = "ca575930-c2e3-43a9-ace4-1e988b2c1908" + +[[OpenSpecFun_jll]] +deps = ["Artifacts", "CompilerSupportLibraries_jll", "JLLWrappers", "Libdl", "Pkg"] +git-tree-sha1 = "13652491f6856acfd2db29360e1bbcd4565d04f1" +uuid = "efe28fd5-8261-553b-a9e1-b2916fc3738e" +version = "0.5.5+0" + +[[OrderedCollections]] +git-tree-sha1 = "85f8e6578bf1f9ee0d11e7bb1b1456435479d47c" +uuid = "bac558e1-5e72-5ebc-8fee-abe8a469f55d" +version = "1.4.1" + +[[Parameters]] +deps = ["OrderedCollections", "UnPack"] +git-tree-sha1 = "2276ac65f1e236e0a6ea70baff3f62ad4c625345" +uuid = "d96e819e-fc66-5662-9728-84c9c7592b0a" +version = "0.12.2" + +[[Pkg]] +deps = ["Artifacts", "Dates", "Downloads", "LibGit2", "Libdl", "Logging", "Markdown", "Printf", "REPL", "Random", "SHA", "Serialization", "TOML", "Tar", "UUIDs", "p7zip_jll"] +uuid = "44cfe95a-1eb2-52ea-b672-e2afdf69b78f" + +[[Preferences]] +deps = ["TOML"] +git-tree-sha1 = "00cfd92944ca9c760982747e9a1d0d5d86ab1e5a" +uuid = "21216c6a-2e73-6563-6e65-726566657250" +version = "1.2.2" + +[[Printf]] +deps = ["Unicode"] +uuid = "de0858da-6303-5e67-8744-51eddeeeb8d7" + +[[REPL]] +deps = ["InteractiveUtils", "Markdown", "Sockets", "Unicode"] +uuid = "3fa0cd96-eef1-5676-8a61-b3b8758bbffb" + +[[Random]] +deps = ["Serialization"] +uuid = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c" + +[[Random123]] +deps = ["Libdl", "Random", "RandomNumbers"] +git-tree-sha1 = "7c6710c8198fd4444b5eb6a3840b7d47bd3593c5" +uuid = "74087812-796a-5b5d-8853-05524746bad3" +version = "1.3.1" + +[[RandomNumbers]] +deps = ["Random", "Requires"] +git-tree-sha1 = "441e6fc35597524ada7f85e13df1f4e10137d16f" +uuid = "e6cf234a-135c-5ec9-84dd-332b85af5143" +version = "1.4.0" + +[[Reexport]] +git-tree-sha1 = "5f6c21241f0f655da3952fd60aa18477cf96c220" +uuid = "189a3867-3050-52da-a836-e630ba90ab69" +version = "1.1.0" + +[[Requires]] +deps = ["UUIDs"] +git-tree-sha1 = "4036a3bd08ac7e968e27c203d45f5fff15020621" +uuid = "ae029012-a4dd-5104-9daa-d747884805df" +version = "1.1.3" + +[[SHA]] +uuid = "ea8e919c-243c-51af-8825-aaa63cd721ce" + +[[Scratch]] +deps = ["Dates"] +git-tree-sha1 = "0b4b7f1393cff97c33891da2a0bf69c6ed241fda" +uuid = "6c6a2e73-6563-6170-7368-637461726353" +version = "1.1.0" + +[[Serialization]] +uuid = "9e88b42a-f829-5b0c-bbe9-9e923198166b" + +[[Setfield]] +deps = ["ConstructionBase", "Future", "MacroTools", "Requires"] +git-tree-sha1 = "d5640fc570fb1b6c54512f0bd3853866bd298b3e" +uuid = "efcf1570-3423-57d1-acb7-fd33fddbac46" +version = "0.7.0" + +[[SharedArrays]] +deps = ["Distributed", "Mmap", "Random", "Serialization"] +uuid = "1a1011a3-84de-559e-8e89-a11a2f7dc383" + +[[Sockets]] +uuid = "6462fe0b-24de-5631-8697-dd941f90decc" + +[[SparseArrays]] +deps = ["LinearAlgebra", "Random"] +uuid = "2f01184e-e22b-5df5-ae63-d93ebab69eaf" + +[[SpecialFunctions]] +deps = ["ChainRulesCore", "LogExpFunctions", "OpenSpecFun_jll"] +git-tree-sha1 = "a50550fa3164a8c46747e62063b4d774ac1bcf49" +uuid = "276daf66-3868-5448-9aa4-cd146d93841b" +version = "1.5.1" + +[[Statistics]] +deps = ["LinearAlgebra", "SparseArrays"] +uuid = "10745b16-79ce-11e8-11f9-7d13ad32a3b2" + +[[TOML]] +deps = ["Dates"] +uuid = "fa267f1f-6049-4f14-aa54-33bafae1ed76" + +[[Tar]] +deps = ["ArgTools", "SHA"] +uuid = "a4e569a6-e804-4fa4-b0f3-eef7a1d5b13e" + +[[Test]] +deps = ["InteractiveUtils", "Logging", "Random", "Serialization"] +uuid = "8dfed614-e22c-5e08-85e1-65c5234f0b40" + +[[TextWrap]] +git-tree-sha1 = "9250ef9b01b66667380cf3275b3f7488d0e25faf" +uuid = "b718987f-49a8-5099-9789-dcd902bef87d" +version = "1.0.1" + +[[TimerOutputs]] +deps = ["ExprTools", "Printf"] +git-tree-sha1 = "bf8aacc899a1bd16522d0350e1e2310510d77236" +uuid = "a759f4b9-e2f1-59dc-863e-4aeb61b1ea8f" +version = "0.5.9" + +[[UUIDs]] +deps = ["Random", "SHA"] +uuid = "cf7118a7-6976-5b1a-9a39-7adc72f591a4" + +[[UnPack]] +git-tree-sha1 = "387c1f73762231e86e0c9c5443ce3b4a0a9a0c2b" +uuid = "3a884ed6-31ef-47d7-9d2a-63182c4928ed" +version = "1.0.2" + +[[Unicode]] +uuid = "4ec0a83e-493e-50e2-b9ac-8f72acf5a8f5" + +[[XZ_jll]] +deps = ["Artifacts", "JLLWrappers", "Libdl", "Pkg"] +git-tree-sha1 = "9f76853ea2ba894054e24640abfb73d73e5a4cb5" +uuid = "ffd25f8a-64ca-5728-b0f7-c24cf3aae800" +version = "5.2.5+0" + +[[Zlib_jll]] +deps = ["Libdl"] +uuid = "83775a58-1f1d-513f-b197-d71354ab007a" + +[[argp_standalone_jll]] +deps = ["Artifacts", "JLLWrappers", "Libdl", "Pkg"] +git-tree-sha1 = "c4fa3457046fc93249b63e8319e743b6c8590609" +uuid = "c53206cc-00f7-50bf-ad1e-3ae1f6e49bc3" +version = "1.3.0+0" + +[[fts_jll]] +deps = ["Artifacts", "JLLWrappers", "Libdl", "Pkg"] +git-tree-sha1 = "78732b942383d2cb521df8a1a0814911144e663d" +uuid = "d65627f6-89bd-53e8-8ab5-8b75ff535eee" +version = "1.2.7+1" + +[[hsa_rocr_jll]] +deps = ["Artifacts", "Elfutils_jll", "JLLWrappers", "Libdl", "NUMA_jll", "Pkg", "Zlib_jll", "hsakmt_roct_jll"] +git-tree-sha1 = "42189f176d6ae4f37c0c0e652fec339bb0bfab5d" +uuid = "dd59ff1a-a01a-568d-8b29-0669330f116a" +version = "3.7.0+1" + +[[hsakmt_roct_jll]] +deps = ["Artifacts", "JLLWrappers", "Libdl", "NUMA_jll", "Pkg"] +git-tree-sha1 = "8a9ee6c091e952e4ea6585d15131d43f789ae041" +uuid = "1cecccd7-a9b6-5045-9cdc-a44c19b16d76" +version = "3.8.0+0" + +[[nghttp2_jll]] +deps = ["Artifacts", "Libdl"] +uuid = "8e850ede-7688-5339-a07c-302acd2aaf8d" + +[[obstack_jll]] +deps = ["Artifacts", "JLLWrappers", "Libdl", "Pkg"] +git-tree-sha1 = "1c4a6b66e934fc6db4649cb2910c72f53bbfea7e" +uuid = "c88a4935-d25e-5644-aacc-5db6f1b8ef79" +version = "1.2.2+0" + +[[p7zip_jll]] +deps = ["Artifacts", "Libdl"] +uuid = "3f19e933-33d8-53b3-aaab-bd5110c3b7a0" diff --git a/JuliaStream.jl/Project.toml b/JuliaStream.jl/Project.toml new file mode 100644 index 0000000..0afa7d0 --- /dev/null +++ b/JuliaStream.jl/Project.toml @@ -0,0 +1,14 @@ +name = "JuliaStream" +uuid = "1bdcc9b7-f5ed-4705-bc7b-be1b748ec681" +authors = ["Wei-Chen Lin "] +version = "3.4.0" + +[deps] +AMDGPU = "21141c5a-9bdb-4563-92ae-f87d6854732e" +ArgParse = "c7e460c6-2fb9-53a9-8c5b-16f535851c63" +CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba" +Distributed = "8ba89e20-285c-5b6f-9357-94700520ee1b" +Parameters = "d96e819e-fc66-5662-9728-84c9c7592b0a" + +[compat] +julia = "1.6" diff --git a/JuliaStream.jl/README.md b/JuliaStream.jl/README.md new file mode 100644 index 0000000..7d94a34 --- /dev/null +++ b/JuliaStream.jl/README.md @@ -0,0 +1,30 @@ +JuliaStream.jl +============== + +This is an implementation of BabelStream in Julia which contains the following variants: + + * `PlainStream.jl` - Single threaded `for` + * `ThreadedStream.jl` - Threaded implementation with `Threads.@threads` macros + * `DistributedStream.jl` - Process based parallelism with `@distributed` macros + * `CUDAStream.jl` - Direct port of BabelStream's native CUDA implementation using [CUDA.jl](https://github.com/JuliaGPU/CUDA.jl) + * `AMDGPUStream.jl` - Direct port of BabelStream's native HIP implementation using [AMDGPU.jl](https://github.com/JuliaGPU/AMDGPU.jl) + +### Build & Run + +Prerequisites + + * Julia 1.6+ + +With Julia on path, run the benchmark with: + +```shell +> cd JuliaStream.jl +> julia --project -e 'import Pkg; Pkg.instantiate()' # only required on first run +> julia --project src/Stream.jl +``` + +**Important:** + * Julia is 1-indexed, so N > 1 in `--device N` + * Thread count for `ThreadedStream` must be set via the `JULIA_NUM_THREADS` environment variable (e.g `export JULIA_NUM_THREADS=$(nproc)`) otherwise it defaults to 1 + * You must *prepend* the number of processes needed for `DistributedStream`, e.g `julia -p$(nproc) --project src/DistributedStream.jl` + * Certain implementations such as CUDA and AMDGPU will do hardware detection at runtime and may download and/or compile further software packages for the platform. diff --git a/JuliaStream.jl/src/AMDGPUStream.jl b/JuliaStream.jl/src/AMDGPUStream.jl new file mode 100644 index 0000000..80f69b4 --- /dev/null +++ b/JuliaStream.jl/src/AMDGPUStream.jl @@ -0,0 +1,178 @@ +# AMDGPU.jl doesn't support CPU agents, so this isn't a feature-complete ROCmStream, only AMD GPUs +include("Stream.jl") +using AMDGPU + +const ROCData = StreamData{T,ROCArray{T}} where {T} +const TBSize = 1024::Int +const DotBlocks = 256::Int + +# AMDGPU.agents()'s internal iteration order isn't stable +function gpu_agents_in_repr_order() + # XXX if we select anything other than :gpu, we get + # HSA_STATUS_ERROR_INVALID_AGENT on the first kernel submission + sort(AMDGPU.get_agents(:gpu), by = repr) +end + +function devices() + try + map(repr, gpu_agents_in_repr_order()) + catch + # probably unsupported + [] + end +end + +function gridsize(data::ROCData{T})::Int where {T} + return data.size +end + +function make_stream( + arraysize::Int, + scalar::T, + device::Int, + silent::Bool, +)::ROCData{T} where {T} + + if arraysize % TBSize != 0 + error("arraysize ($(arraysize)) must be divisible by $(TBSize)!") + end + + # XXX AMDGPU doesn't expose an API for setting the default like CUDA.device!() + # but AMDGPU.get_default_agent returns DEFAULT_AGENT so we can do it by hand + AMDGPU.DEFAULT_AGENT[] = gpu_agents_in_repr_order()[device] + + data = ROCData{T}( + ROCArray{T}(undef, arraysize), + ROCArray{T}(undef, arraysize), + ROCArray{T}(undef, arraysize), + scalar, + arraysize, + ) + selected = AMDGPU.get_default_agent() + if !silent + println("Using GPU HSA device: $(AMDGPU.get_name(selected)) ($(repr(selected)))") + println("Kernel parameters : <<<$(gridsize(data)),$(TBSize)>>>") + end + return data +end + +function hard_wait(kernel) + # soft wait causes HSA_REFCOUNT overflow issues + AMDGPU.wait(kernel, soft = false) +end + +function init_arrays!(data::ROCData{T}, init::Tuple{T,T,T}) where {T} + AMDGPU.fill!(data.a, init[1]) + AMDGPU.fill!(data.b, init[2]) + AMDGPU.fill!(data.c, init[3]) +end + +function copy!(data::ROCData{T}) where {T} + function kernel(a, c) + i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x + @inbounds c[i] = a[i] + return + end + hard_wait(@roc groupsize = TBSize gridsize = gridsize(data) kernel(data.a, data.c)) +end + +function mul!(data::ROCData{T}) where {T} + function kernel(b, c, scalar) + i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x + @inbounds b[i] = scalar * c[i] + return + end + hard_wait( + @roc groupsize = TBSize gridsize = gridsize(data) kernel(data.b, data.c, data.scalar) + ) +end + +function add!(data::ROCData{T}) where {T} + function kernel(a, b, c) + i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x + @inbounds c[i] = a[i] + b[i] + return + end + hard_wait( + @roc groupsize = TBSize gridsize = gridsize(data) kernel(data.a, data.b, data.c) + ) +end + +function triad!(data::ROCData{T}) where {T} + function kernel(a, b, c, scalar) + i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x + @inbounds a[i] = b[i] + (scalar * c[i]) + return + end + hard_wait( + @roc groupsize = TBSize gridsize = gridsize(data) kernel( + data.a, + data.b, + data.c, + data.scalar, + ) + ) +end + +function nstream!(data::ROCData{T}) where {T} + function kernel(a, b, c, scalar) + i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x + @inbounds a[i] += b[i] + scalar * c[i] + return + end + hard_wait( + @roc groupsize = TBSize gridsize = gridsize(data) kernel( + data.a, + data.b, + data.c, + data.scalar, + ) + ) +end + +function dot(data::ROCData{T}) where {T} + function kernel(a, b, size, partial) + tb_sum = ROCDeviceArray((TBSize,), alloc_local(:reduce, T, TBSize)) + local_i = workitemIdx().x + @inbounds tb_sum[local_i] = 0.0 + + # do dot first + i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x + while i <= size + @inbounds tb_sum[local_i] += a[i] * b[i] + i += TBSize * DotBlocks # XXX don't use (workgroupDim().x * gridDimWG().x) here + end + + # then tree reduction + offset = workgroupDim().x ÷ 2 + while offset > 0 + sync_workgroup() + if (local_i - 1) < offset + @inbounds tb_sum[local_i] += tb_sum[local_i+offset] + end + offset ÷= 2 + end + + if (local_i == 1) + @inbounds partial[workgroupIdx().x] = tb_sum[local_i] + end + + return + end + partial_sum = ROCArray{T}(undef, DotBlocks) + hard_wait( + @roc groupsize = TBSize gridsize = TBSize * DotBlocks kernel( + data.a, + data.b, + data.size, + partial_sum, + ) + ) + return sum(partial_sum) +end + +function read_data(data::ROCData{T})::VectorData{T} where {T} + return VectorData{T}(data.a, data.b, data.c, data.scalar, data.size) +end + +main() \ No newline at end of file diff --git a/JuliaStream.jl/src/CUDAStream.jl b/JuliaStream.jl/src/CUDAStream.jl new file mode 100644 index 0000000..c4d0510 --- /dev/null +++ b/JuliaStream.jl/src/CUDAStream.jl @@ -0,0 +1,146 @@ +include("Stream.jl") +using CUDA + +const CuData = StreamData{T,CuArray{T}} where {T} +const TBSize = 1024::Int +const DotBlocks = 256::Int + +function devices() + return !CUDA.functional(false) ? [] : + map(d -> "$(CUDA.name(d)) ($(repr(d)))", CUDA.devices()) +end + +function blocks(data::CuData{T})::Int where {T} + return data.size ÷ TBSize +end + +function make_stream( + arraysize::Int, + scalar::T, + device::Int, + silent::Bool, +)::CuData{T} where {T} + + if arraysize % TBSize != 0 + error("arraysize ($(arraysize)) must be divisible by $(TBSize)!") + end + + # so CUDA's device is 0 indexed, so -1 from Julia + CUDA.device!(device - 1) + selected = CUDA.device() + # show_reason is set to true here so it dumps CUDA info + # for us regardless of whether it's functional + if !CUDA.functional(true) + error("Non-functional CUDA configuration") + end + data = CuData{T}( + CuArray{T}(undef, arraysize), + CuArray{T}(undef, arraysize), + CuArray{T}(undef, arraysize), + scalar, + arraysize, + ) + if !silent + println("Using CUDA device: $(CUDA.name(selected)) ($(repr(selected)))") + println("Kernel parameters: <<<$(blocks(data)),$(TBSize)>>>") + end + return data +end + +function init_arrays!(data::CuData{T}, init::Tuple{T,T,T}) where {T} + CUDA.fill!(data.a, init[1]) + CUDA.fill!(data.b, init[2]) + CUDA.fill!(data.c, init[3]) +end + +function copy!(data::CuData{T}) where {T} + function kernel(a, c) + i = (blockIdx().x - 1) * blockDim().x + threadIdx().x + @inbounds c[i] = a[i] + return + end + @cuda blocks = blocks(data) threads = TBSize kernel(data.a, data.c) + CUDA.synchronize() +end + +function mul!(data::CuData{T}) where {T} + function kernel(b, c, scalar) + i = (blockIdx().x - 1) * blockDim().x + threadIdx().x + @inbounds b[i] = scalar * c[i] + return + end + @cuda blocks = blocks(data) threads = TBSize kernel(data.b, data.c, data.scalar) + CUDA.synchronize() +end + +function add!(data::CuData{T}) where {T} + function kernel(a, b, c) + i = (blockIdx().x - 1) * blockDim().x + threadIdx().x + @inbounds c[i] = a[i] + b[i] + return + end + @cuda blocks = blocks(data) threads = TBSize kernel(data.a, data.b, data.c) + CUDA.synchronize() +end + +function triad!(data::CuData{T}) where {T} + function kernel(a, b, c, scalar) + i = (blockIdx().x - 1) * blockDim().x + threadIdx().x + @inbounds a[i] = b[i] + (scalar * c[i]) + return + end + @cuda blocks = blocks(data) threads = TBSize kernel(data.a, data.b, data.c, data.scalar) + CUDA.synchronize() +end + +function nstream!(data::CuData{T}) where {T} + function kernel(a, b, c, scalar) + i = (blockIdx().x - 1) * blockDim().x + threadIdx().x + @inbounds a[i] += b[i] + scalar * c[i] + return + end + @cuda blocks = blocks(data) threads = TBSize kernel(data.a, data.b, data.c, data.scalar) + CUDA.synchronize() +end + +function dot(data::CuData{T}) where {T} + # direct port of the reduction in CUDAStream.cu + function kernel(a, b, size, partial) + tb_sum = @cuStaticSharedMem(T, TBSize) + local_i = threadIdx().x + @inbounds tb_sum[local_i] = 0.0 + + # do dot first + i = (blockIdx().x - 1) * blockDim().x + threadIdx().x + while i <= size + @inbounds tb_sum[local_i] += a[i] * b[i] + i += blockDim().x * gridDim().x + end + + # then tree reduction + offset = blockDim().x ÷ 2 + while offset > 0 + sync_threads() + if (local_i - 1) < offset + @inbounds tb_sum[local_i] += tb_sum[local_i+offset] + end + offset ÷= 2 + end + + if (local_i == 1) + @inbounds partial[blockIdx().x] = tb_sum[local_i] + end + + return + end + partial_sum = CuArray{T}(undef, DotBlocks) + @cuda blocks = DotBlocks threads = TBSize kernel(data.a, data.b, data.size, partial_sum) + CUDA.synchronize() + return sum(partial_sum) +end + +function read_data(data::CuData{T})::VectorData{T} where {T} + return VectorData{T}(data.a, data.b, data.c, data.scalar, data.size) +end + +main() \ No newline at end of file diff --git a/JuliaStream.jl/src/DistributedStream.jl b/JuliaStream.jl/src/DistributedStream.jl new file mode 100644 index 0000000..970c699 --- /dev/null +++ b/JuliaStream.jl/src/DistributedStream.jl @@ -0,0 +1,84 @@ +using Distributed + +include("Stream.jl") + +@everywhere include("StreamData.jl") +@everywhere using SharedArrays +@everywhere const SharedArrayData = StreamData{T,SharedArray{T}} where {T} + +function devices() + return ["CPU (localhost)"] +end + +function make_stream( + arraysize::Int, + scalar::T, + device::Int, + silent::Bool, +)::SharedArrayData{T} where {T} + if device != 1 + error("Only CPU device is supported") + end + + if !silent + println("Using max $(nworkers()) process(es) + 1 master") + end + return SharedArrayData{T}( + SharedArray{T}(arraysize), + SharedArray{T}(arraysize), + SharedArray{T}(arraysize), + scalar, + arraysize, + ) +end + +function init_arrays!(data::SharedArrayData{T}, init::Tuple{T,T,T}) where {T} + + @sync @distributed for i = 1:data.size + @inbounds data.a[i] = init[1] + @inbounds data.b[i] = init[2] + @inbounds data.c[i] = init[3] + end +end + +function copy!(data::SharedArrayData{T}) where {T} + @sync @distributed for i = 1:data.size + @inbounds data.c[i] = data.a[i] + end +end + +function mul!(data::SharedArrayData{T}) where {T} + @sync @distributed for i = 1:data.size + @inbounds data.b[i] = data.scalar * data.c[i] + end +end + +function add!(data::SharedArrayData{T}) where {T} + @sync @distributed for i = 1:data.size + @inbounds data.c[i] = data.a[i] + data.b[i] + end +end + +function triad!(data::SharedArrayData{T}) where {T} + @sync @distributed for i = 1:data.size + @inbounds data.a[i] = data.b[i] + (data.scalar * data.c[i]) + end +end + +function nstream!(data::SharedArrayData{T}) where {T} + @sync @distributed for i = 1:data.size + @inbounds data.a[i] += data.b[i] + data.scalar * data.c[i] + end +end + +function dot(data::SharedArrayData{T}) where {T} + return @distributed (+) for i = 1:data.size + @inbounds data.a[i] * data.b[i] + end +end + +function read_data(data::SharedArrayData{T})::VectorData{T} where {T} + return VectorData{T}(data.a, data.b, data.c, data.scalar, data.size) +end + +main() \ No newline at end of file diff --git a/JuliaStream.jl/src/JuliaStream.jl b/JuliaStream.jl/src/JuliaStream.jl new file mode 100644 index 0000000..e01d46d --- /dev/null +++ b/JuliaStream.jl/src/JuliaStream.jl @@ -0,0 +1,4 @@ +module JuliaStream +end + +println("Please run benchmarks directly via `julia --project src/Stream.jl`") \ No newline at end of file diff --git a/JuliaStream.jl/src/PlainStream.jl b/JuliaStream.jl/src/PlainStream.jl new file mode 100644 index 0000000..259a9b7 --- /dev/null +++ b/JuliaStream.jl/src/PlainStream.jl @@ -0,0 +1,64 @@ +include("Stream.jl") + +function devices() + return ["CPU"] +end + +function make_stream(arraysize::Int, scalar::T, device::Int, silent::Bool)::VectorData{T} where {T} + if device != 1 + error("Only CPU device is supported") + end + return VectorData{T}(1:arraysize, 1:arraysize, 1:arraysize, scalar, arraysize) +end + +function init_arrays!(data::VectorData{T}, init::Tuple{T,T,T}) where {T} + for i = 1:data.size + @inbounds data.a[i] = init[1] + @inbounds data.b[i] = init[2] + @inbounds data.c[i] = init[3] + end +end + +function copy!(data::VectorData{T}) where {T} + for i = 1:data.size + @inbounds data.c[i] = data.a[i] + end +end + +function mul!(data::VectorData{T}) where {T} + for i = 1:data.size + @inbounds data.b[i] = data.scalar * data.c[i] + end +end + +function add!(data::VectorData{T}) where {T} + for i = 1:data.size + @inbounds data.c[i] = data.a[i] + data.b[i] + end +end + +function triad!(data::VectorData{T}) where {T} + for i = 1:data.size + @inbounds data.a[i] = data.b[i] + (data.scalar * data.c[i]) + end +end + +function nstream!(data::VectorData{T}) where {T} + for i = 1:data.size + @inbounds data.a[i] += data.b[i] + data.scalar * data.c[i] + end +end + +function dot(data::VectorData{T}) where {T} + sum = zero(T) + for i = 1:data.size + @inbounds sum += data.a[i] * data.b[i] + end + return sum +end + +function read_data(data::VectorData{T})::VectorData{T} where {T} + return data +end + +main() \ No newline at end of file diff --git a/JuliaStream.jl/src/Stream.jl b/JuliaStream.jl/src/Stream.jl new file mode 100644 index 0000000..cce846b --- /dev/null +++ b/JuliaStream.jl/src/Stream.jl @@ -0,0 +1,292 @@ +using ArgParse +using Parameters +using Printf +using Base: Float64, Int + +include("StreamData.jl") + +const VectorData = StreamData{T,Vector{T}} where {T} + +struct Timings + copy::Vector{Float64} + mul::Vector{Float64} + add::Vector{Float64} + triad::Vector{Float64} + dot::Vector{Float64} + Timings(n) = new(zeros(n), zeros(n), zeros(n), zeros(n), zeros(n)) +end + +@enum Benchmark All Triad Nstream + +function run_all!(data::StreamData{T,C}, times::Int)::Tuple{Timings,T} where {T,C} + timings = Timings(times) + lastSum::T = 0 + for i = 1:times + @inbounds timings.copy[i] = @elapsed copy!(data) + @inbounds timings.mul[i] = @elapsed mul!(data) + @inbounds timings.add[i] = @elapsed add!(data) + @inbounds timings.triad[i] = @elapsed triad!(data) + @inbounds timings.dot[i] = @elapsed lastSum = dot(data) + end + return (timings, lastSum) +end + +function run_triad!(data::StreamData{T,C}, times::Int)::Float64 where {T,C} + return @elapsed for _ = 1:times + triad!(data) + end +end + +function run_nstream!(data::StreamData{T,C}, times::Int)::Vector{Float64} where {T,C} + timings::Vector{Float64} = zeros(times) + for i = 1:times + @inbounds timings[i] = @elapsed nstream!(data) + end + return timings +end + +function check_solutions( + data::StreamData{T,C}, + times::Int, + init::Tuple{T,T,T}, + benchmark::Benchmark, + dot::Union{T,Nothing}, +) where {T,C} + (gold_a, gold_b, gold_c) = init + for _ = 1:times + if benchmark == All + gold_c = gold_a + gold_b = data.scalar * gold_c + gold_c = gold_a + gold_b + gold_a = gold_b + data.scalar * gold_c + elseif benchmark == Triad + gold_a = gold_b + data.scalar * gold_c + elseif benchmark == Nstream + gold_a += gold_b + data.scalar * gold_c + else + error("Unknown benchmark", benchmark) + end + end + + tolerance = eps(T) * 100 + function validate_xs(name::String, xs::AbstractArray{T}, from::T) + error = (map(x -> abs(x - from), xs) |> sum) / length(xs) + failed = error > tolerance + if failed + println("Validation failed on $name. Average error $error") + end + !failed + end + a_valid = validate_xs("a", data.a, gold_a) + b_valid = validate_xs("b", data.b, gold_b) + c_valid = validate_xs("c", data.c, gold_c) + dot_valid = + dot !== nothing ? + begin + gold_sum = gold_a * gold_b * data.size + error = abs((dot - gold_sum) / gold_sum) + failed = error > 1.0e-8 + if failed + println( + "Validation failed on sum. Error $error \nSum was $dot but should be $gold_sum", + ) + end + !failed + end : true + + a_valid && b_valid && c_valid && dot_valid +end + +@with_kw mutable struct Config + list::Bool = false + impl::String = "threaded" + device::Int = 1 + numtimes::Int = 100 + arraysize::Int = 33554432 + float::Bool = false + triad_only::Bool = false + nstream_only::Bool = false + csv::Bool = false + mibibytes::Bool = false +end + +function parse_options(given::Config) + s = ArgParseSettings() + @add_arg_table s begin + "--list" + help = "List available devices" + action = :store_true + "--device", "-d" + help = "Select device at DEVICE, NOTE: Julia is 1-indexed" + arg_type = Int + default = given.device + "--numtimes", "-n" + help = "Run the test NUMTIMES times (NUM >= 2)" + arg_type = Int + default = given.numtimes + "--arraysize", "-s" + help = "Use ARRAYSIZE elements in the array" + arg_type = Int + default = given.arraysize + "--float" + help = "Use floats (rather than doubles)" + action = :store_true + "--triad_only" + help = "Only run triad" + action = :store_true + "--nstream_only" + help = "Only run nstream" + action = :store_true + "--csv" + help = "Output as csv table" + action = :store_true + "--mibibytes" + help = "Use MiB=2^20 for bandwidth calculation (default MB=10^6)" + action = :store_true + end + args = parse_args(s) + # surely there's a better way than doing this: + for (arg, val) in args + setproperty!(given, Symbol(arg), val) + end +end + +const DefaultInit = (0.1, 0.2, 0.0) +const DefaultScalar = 0.4 +const Version = "3.4.0" + +function main() + + config::Config = Config() + parse_options(config) + + if config.list + ds = devices() + for (i, device) in enumerate(ds) + println("[$i] $(device)") + end + exit(0) + end + + ds = devices() + if config.device < 1 || config.device > length(ds) + error( + "Device $(config.device) out of range (1..$(length(ds))), NOTE: Julia is 1-indexed", + ) + end + + if config.float + type = Float32 + else + type = Float64 + end + + if config.nstream_only && !config.triad_only + benchmark = Nstream + elseif !config.nstream_only && config.triad_only + benchmark = Triad + elseif !config.nstream_only && !config.triad_only + benchmark = All + elseif config.nstream_only && config.triad_only + error("Both triad and nstream are enabled, pick one or omit both to run all benchmarks") + else + error("Invalid config: $(repr(config))") + end + + array_bytes = config.arraysize * sizeof(type) + total_bytes = array_bytes * 3 + (mega_scale, mega_suffix, giga_scale, giga_suffix) = + !config.mibibytes ? (1.0e-6, "MB", 1.0e-9, "GB") : (2^-20, "MiB", 2^-30, "GiB") + + if !config.csv + println("""BabelStream + Version: $Version + Implementation: Julia; $(config.impl)""") + println("Running kernels $(config.numtimes) times") + if benchmark == Triad + println("Number of elements: $(config.arraysize)") + end + println("Precision: $(config.float ? "float" : "double")") + r1 = n -> round(n; digits = 1) + println( + "Array size: $(r1(mega_scale * array_bytes)) $mega_suffix(=$(r1(giga_scale * array_bytes)) $giga_suffix)", + ) + println( + "Total size: $(r1(mega_scale * total_bytes)) $mega_suffix(=$(r1(giga_scale * total_bytes)) $giga_suffix)", + ) + end + + function mk_row(xs::Vector{Float64}, name::String, total_bytes::Int) + tail = Base.rest(xs) + min = Iterators.minimum(tail) + max = Iterators.maximum(tail) + avg = Iterators.sum(tail) / Iterators.length(tail) + mbps = mega_scale * total_bytes / min + if config.csv + return [ + ("function", name), + ("num_times", config.numtimes), + ("n_elements", config.arraysize), + ("sizeof", total_bytes), + ("max_m$( config.mibibytes ? "i" : "")bytes_per_sec", mbps), + ("min_runtime", min), + ("max_runtime", max), + ("avg_runtime", avg), + ] + else + return [ + ("Function", name), + ("M$(config.mibibytes ? "i" : "")Bytes/sec", round(mbps; digits = 3)), + ("Min (sec)", round(min; digits = 5)), + ("Max", round(max; digits = 5)), + ("Average", round(avg; digits = 5)), + ] + end + end + + function tabulate(rows::Vector{Tuple{String,Any}}...) + header = Base.first(rows) + padding = config.csv ? 0 : 12 + sep = config.csv ? "," : "" + map(x -> rpad(x[1], padding), header) |> x -> join(x, sep) |> println + for row in rows + map(x -> rpad(x[2], padding), row) |> x -> join(x, sep) |> println + end + end + + init::Tuple{type,type,type} = DefaultInit + scalar::type = DefaultScalar + + data = make_stream(config.arraysize, scalar, config.device, config.csv) + + init_arrays!(data, init) + if benchmark == All + (timings, sum) = run_all!(data, config.numtimes) + valid = check_solutions(read_data(data), config.numtimes, init, benchmark, sum) + tabulate( + mk_row(timings.copy, "Copy", 2 * array_bytes), + mk_row(timings.mul, "Mul", 2 * array_bytes), + mk_row(timings.add, "Add", 3 * array_bytes), + mk_row(timings.triad, "Triad", 3 * array_bytes), + mk_row(timings.dot, "Dot", 2 * array_bytes), + ) + elseif benchmark == Nstream + timings = run_nstream!(data, config.numtimes) + valid = check_solutions(read_data(data), config.numtimes, init, benchmark, nothing) + tabulate(mk_row(timings, "Nstream", 4 * array_bytes)) + elseif benchmark == Triad + elapsed = run_triad!(data, config.numtimes) + valid = check_solutions(read_data(data), config.numtimes, init, benchmark, nothing) + total_bytes = 3 * array_bytes * config.numtimes + bandwidth = mega_scale * (total_bytes / elapsed) + println("Runtime (seconds): $(round(elapsed; digits=5))") + println("Bandwidth ($giga_suffix/s): $(round(bandwidth; digits=3)) ") + else + error("Bad benchmark $(benchmark)") + end + + if !valid + exit(1) + end + +end diff --git a/JuliaStream.jl/src/StreamData.jl b/JuliaStream.jl/src/StreamData.jl new file mode 100644 index 0000000..07498fe --- /dev/null +++ b/JuliaStream.jl/src/StreamData.jl @@ -0,0 +1,8 @@ + +struct StreamData{T,C<:AbstractArray{T}} + a::C + b::C + c::C + scalar::T + size::Int + end \ No newline at end of file diff --git a/JuliaStream.jl/src/ThreadedStream.jl b/JuliaStream.jl/src/ThreadedStream.jl new file mode 100644 index 0000000..fb995e6 --- /dev/null +++ b/JuliaStream.jl/src/ThreadedStream.jl @@ -0,0 +1,72 @@ +include("Stream.jl") + +function devices() + return ["CPU"] +end + +function make_stream( + arraysize::Int, + scalar::T, + device::Int, + silent::Bool, +)::VectorData{T} where {T} + if device != 1 + error("Only CPU device is supported") + end + if !silent + println("Using max $(Threads.nthreads()) threads") + end + return VectorData{T}(1:arraysize, 1:arraysize, 1:arraysize, scalar, arraysize) +end + +function init_arrays!(data::VectorData{T}, init::Tuple{T,T,T}) where {T} + Threads.@threads for i = 1:data.size + @inbounds data.a[i] = init[1] + @inbounds data.b[i] = init[2] + @inbounds data.c[i] = init[3] + end +end + +function copy!(data::VectorData{T}) where {T} + Threads.@threads for i = 1:data.size + @inbounds data.c[i] = data.a[i] + end +end + +function mul!(data::VectorData{T}) where {T} + Threads.@threads for i = 1:data.size + @inbounds data.b[i] = data.scalar * data.c[i] + end +end + +function add!(data::VectorData{T}) where {T} + Threads.@threads for i = 1:data.size + @inbounds data.c[i] = data.a[i] + data.b[i] + end +end + +function triad!(data::VectorData{T}) where {T} + Threads.@threads for i = 1:data.size + @inbounds data.a[i] = data.b[i] + (data.scalar * data.c[i]) + end +end + +function nstream!(data::VectorData{T}) where {T} + Threads.@threads for i = 1:data.size + @inbounds data.a[i] += data.b[i] + data.scalar * data.c[i] + end +end + +function dot(data::VectorData{T}) where {T} + partial = zeros(T, Threads.nthreads()) + Threads.@threads for i = 1:data.size + @inbounds partial[Threads.threadid()] += data.a[i] * data.b[i] + end + return sum(partial) +end + +function read_data(data::VectorData{T})::VectorData{T} where {T} + return data +end + +main() \ No newline at end of file