Migrate to KA 0.9

* Ported to KA 0.9 API * Project.toml for tests
exanauts · Apr 18, 2023 · 3ebc6bc · 3ebc6bc
1 parent c327078
commit 3ebc6bc
Show file tree

Hide file tree

Showing 10 changed files with 102 additions and 87 deletions.
diff --git a/Project.toml b/Project.toml
@@ -1,7 +1,7 @@
 name = "ExaTron"
 uuid = "28b18bf8-76f9-41ea-81fa-0f922810b349"
 authors = ["Youngdae Kim <[email protected]>", "François Pacaud <[email protected]>", "Kibaek Kim <[email protected]>", "Michel Schanen <[email protected]>"]
-version = "2.1.3"
+version = "3.0.0"
 
 [deps]
 DelimitedFiles = "8bb1440f-4735-579b-a4ab-409b98df4dab"
@@ -13,22 +13,7 @@ Requires = "ae029012-a4dd-5104-9daa-d747884805df"
 TOML = "fa267f1f-6049-4f14-aa54-33bafae1ed76"
 
 [compat]
-KernelAbstractions = "0.8"
+KernelAbstractions = "0.9"
 Requires = "1"
 TOML = "1"
-julia = "1.7"
-
-[extras]
-AMDGPU = "21141c5a-9bdb-4563-92ae-f87d6854732e"
-CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
-CUDAKernels = "72cfdca4-0801-4ab0-bf6a-d52aa10adc57"
-KernelAbstractions = "63c18a36-062a-441e-b654-da1e3ab1ce7c"
-PowerModels = "c36e90e8-916a-50a6-bd94-075b64ef4655"
-ROCKernels = "7eb9e9f0-4bd3-4c4c-8bef-26bd9629d9b9"
-Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
-SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
-StatsBase = "2913bbd2-ae8a-5f71-8c99-4fb6c76f3a91"
-Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
-
-[targets]
-test = ["AMDGPU", "CUDA", "CUDAKernels", "KernelAbstractions", "PowerModels", "Random", "ROCKernels", "SparseArrays", "StatsBase", "Test"]
+julia = "1.8"
diff --git a/examples/Project.toml b/examples/Project.toml
@@ -1,12 +1,10 @@
 [deps]
 AMDGPU = "21141c5a-9bdb-4563-92ae-f87d6854732e"
 CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
-CUDAKernels = "72cfdca4-0801-4ab0-bf6a-d52aa10adc57"
 ExaTronKernels = "dec458bd-9b7b-4867-ad20-20d9a112d608"
 KernelAbstractions = "63c18a36-062a-441e-b654-da1e3ab1ce7c"
 LazyArtifacts = "4af54fe1-eca0-43a8-85a7-787d91b784e3"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 PowerModels = "c36e90e8-916a-50a6-bd94-075b64ef4655"
 Printf = "de0858da-6303-5e67-8744-51eddeeeb8d7"
-ROCKernels = "7eb9e9f0-4bd3-4c4c-8bef-26bd9629d9b9"
 Requires = "ae029012-a4dd-5104-9daa-d747884805df"
diff --git a/examples/admm/acopf_admm_gpu.jl b/examples/admm/acopf_admm_gpu.jl
@@ -31,7 +31,7 @@ function get_generator_data(data::OPFData, ::KA.CPU)
     return pgmin,pgmax,qgmin,qgmax,c2,c1,c0
 end
 
-function get_generator_data(data::OPFData, ::CUDADevice)
+function get_generator_data(data::OPFData, ::CUDABackend)
     ngen = length(data.generators)
 
     pgmin = CuArray{Float64}(undef, ngen)
@@ -109,7 +109,7 @@ function get_bus_data(data::OPFData, ::KA.CPU)
     return FrStart,FrIdx,ToStart,ToIdx,GenStart,GenIdx,Pd,Qd
 end
 
-function get_bus_data(data::OPFData, ::CUDADevice)
+function get_bus_data(data::OPFData, ::CUDABackend)
     nbus = length(data.buses)
 
     FrIdx = [l for b=1:nbus for l in data.FromLines[b]]
@@ -194,7 +194,7 @@ function get_branch_data(data::OPFData, device::KA.CPU)
             ybus.YttR, ybus.YttI, ybus.YtfR, ybus.YtfI, frBound, toBound
 end
 
-function get_branch_data(data::OPFData, device::CUDADevice)
+function get_branch_data(data::OPFData, device::CUDABackend)
     buses = data.buses
     lines = data.lines
     BusIdx = data.BusIdx
@@ -492,40 +492,44 @@ function admm_solve!(env::AdmmEnv, sol::SolutionOneLevel; iterlim=800, scale=1e-
                 break
             end
         else
-            wait(copy_data_kernel(env.device, 64, mod.nvar)(mod.nvar, sol.u_prev, sol.u_curr, dependencies=Event(env.device)))
-            wait(copy_data_kernel(env.device, 64, mod.nvar)(mod.nvar, sol.v_prev, sol.v_curr, dependencies=Event(env.device)))
-            wait(copy_data_kernel(env.device, 64, mod.nvar)(mod.nvar, sol.l_prev, sol.l_curr, dependencies=Event(env.device)))
+            copy_data_kernel(env.device, 64, mod.nvar)(mod.nvar, sol.u_prev, sol.u_curr)
+            copy_data_kernel(env.device, 64, mod.nvar)(mod.nvar, sol.v_prev, sol.v_curr)
+            copy_data_kernel(env.device, 64, mod.nvar)(mod.nvar, sol.l_prev, sol.l_curr)
+            KA.synchronize(env.device)
 
             generator_kernel(mod.gen_mod, data.baseMVA, sol.u_curr, sol.v_curr, sol.l_curr, sol.rho, env.device)
 
-            wait(polar_kernel(env.device, 32, mod.nline*32)(mod.n, mod.nline, mod.line_start, scale,
-                                                            sol.u_curr, sol.v_curr, sol.l_curr, sol.rho,
-                                                            shift_lines, env.membuf, mod.YffR, mod.YffI, mod.YftR, mod.YftI,
-                                                            mod.YttR, mod.YttI, mod.YtfR, mod.YtfI, mod.FrBound, mod.ToBound,
-                                                            dependencies=Event(env.device)
-                                                            )
+            polar_kernel(env.device, 32, mod.nline*32)(
+                mod.n, mod.nline, mod.line_start, scale,
+                sol.u_curr, sol.v_curr, sol.l_curr, sol.rho,
+                shift_lines, env.membuf, mod.YffR, mod.YffI, mod.YftR, mod.YftI,
+                mod.YttR, mod.YttI, mod.YtfR, mod.YtfI, mod.FrBound, mod.ToBound
             )
-            wait(bus_kernel(env.device, 32, mod.nbus)(data.baseMVA, mod.nbus, mod.gen_mod.gen_start, mod.line_start,
-                                                                           mod.FrStart, mod.FrIdx, mod.ToStart, mod.ToIdx, mod.GenStart,
-                                                                           mod.GenIdx, mod.Pd, mod.Qd, sol.u_curr, sol.v_curr, sol.l_curr,
-                                                                           sol.rho, mod.YshR, mod.YshI,
-                                                                           dependencies=Event(env.device)
-                                                                           )
+            KA.synchronize(env.device)
+
+            bus_kernel(env.device, 32, mod.nbus)(
+                data.baseMVA, mod.nbus, mod.gen_mod.gen_start, mod.line_start,
+                mod.FrStart, mod.FrIdx, mod.ToStart, mod.ToIdx, mod.GenStart,
+                mod.GenIdx, mod.Pd, mod.Qd, sol.u_curr, sol.v_curr, sol.l_curr,
+                sol.rho, mod.YshR, mod.YshI
             )
+            KA.synchronize(env.device)
 
-            wait(update_multiplier_kernel(env.device, 32, mod.nvar)(mod.nvar, sol.l_curr,
-                    sol.u_curr, sol.v_curr, sol.rho,
-                    dependencies=Event(env.device)
-                )
+            update_multiplier_kernel(env.device, 32, mod.nvar)(
+                mod.nvar, sol.l_curr,
+                sol.u_curr, sol.v_curr, sol.rho
             )
-            wait(primal_residual_kernel(env.device, 32, mod.nvar)(mod.nvar, sol.rp, sol.u_curr, sol.v_curr,
-                    dependencies=Event(env.device)
-                )
+            KA.synchronize(env.device)
+
+            primal_residual_kernel(env.device, 32, mod.nvar)(
+                mod.nvar, sol.rp, sol.u_curr, sol.v_curr
             )
-            wait(dual_residual_kernel(env.device, 32, mod.nvar)(mod.nvar, sol.rd, sol.v_prev, sol.v_curr, sol.rho,
-                    dependencies=Event(env.device)
-                )
+            KA.synchronize(env.device)
+
+            dual_residual_kernel(env.device, 32, mod.nvar)(
+                mod.nvar, sol.rd, sol.v_prev, sol.v_curr, sol.rho
             )
+            KA.synchronize(env.device)
 
             gpu_primres = norm(sol.rp)
             gpu_dualres = norm(sol.rd)

diff --git a/examples/admm/environment.jl b/examples/admm/environment.jl
@@ -278,7 +278,7 @@ AdmmEnv(opfdata::OPFData, device::KA.CPU, rho_pq, rho_va; options...) = AdmmEnv{
         opfdata, rho_pq, rho_va; device=device, options...
     )
 
-AdmmEnv(opfdata::OPFData, device::CUDADevice, rho_pq, rho_va; options...) = AdmmEnv{Float64, CuArray{Float64, 1}, CuArray{Int, 1}, CuArray{Float64, 2}}(
+AdmmEnv(opfdata::OPFData, device::CUDABackend, rho_pq, rho_va; options...) = AdmmEnv{Float64, CuArray{Float64, 1}, CuArray{Int, 1}, CuArray{Float64, 2}}(
         opfdata, rho_pq, rho_va; device=device, options...
     )
 

diff --git a/examples/admm/generator_kernel.jl b/examples/admm/generator_kernel.jl
@@ -52,11 +52,12 @@ function generator_kernel(
 )
     nblk = div(gen_mod.ngen, 32, RoundUp)
     nblk
-    wait(generator_kernel(device, 32, gen_mod.ngen)(baseMVA, gen_mod.ngen, gen_mod.gen_start,
-                u, v, l, rho, gen_mod.pgmin, gen_mod.pgmax, gen_mod.qgmin, gen_mod.qgmax, gen_mod.c2, gen_mod.c1,
-                dependencies=Event(device)
-                )
+    generator_kernel(device, 32, gen_mod.ngen)(
+        baseMVA, gen_mod.ngen, gen_mod.gen_start,
+        u, v, l, rho, gen_mod.pgmin, gen_mod.pgmax, gen_mod.qgmin, gen_mod.qgmax, gen_mod.c2, gen_mod.c1
     )
+    KA.synchronize(device)
+
     return 0.0
 end
 

diff --git a/examples/opf.jl b/examples/opf.jl
@@ -1,8 +1,6 @@
 using KernelAbstractions
 using AMDGPU
-using ROCKernels
 using CUDA
-using CUDAKernels
 
 # One day for Intel GPUs
 # using oneAPI
@@ -32,7 +30,7 @@ rho_va = parse(Float64, ARGS[2])
 max_iter = parse(Int, ARGS[3])
 # Indicate which GPU device to use
 device = CPU()
-# device = CUDADevice()
+# device = CUDABackend()
 # device = ROCDevice()
 # verbose = 0: No output
 # verbose = 1: Final result metrics