Added timmings

Manifest.toml

@@ -310,9 +310,9 @@ uuid = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [[TimerOutputs]]
 deps = ["ExprTools", "Printf"]
-git-tree-sha1 = "209a8326c4f955e2442c07b56029e88bb48299c7"
+git-tree-sha1 = "7cb456f358e8f9d102a8b25e8dfedf58fa5689bc"
 uuid = "a759f4b9-e2f1-59dc-863e-4aeb61b1ea8f"
-version = "0.5.12"
+version = "0.5.13"
 
 [[UUIDs]]
 deps = ["Random", "SHA"]

Project.toml

@@ -7,3 +7,4 @@ version = "0.1.0"
 CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 StructArrays = "09ab397b-f2b6-538f-b94a-2f83cf4a842a"
+TimerOutputs = "a759f4b9-e2f1-59dc-863e-4aeb61b1ea8f"

src/YM/YM.jl

@@ -12,7 +12,7 @@
 
 module YM
 
-using CUDA, Random, StructArrays
+using CUDA, Random, StructArrays, TimerOutputs
 using ..Space
 using ..Groups
 using ..Fields
@@ -48,6 +48,7 @@ struct YMworkspace{T}
     rm # float   of volume
     function YMworkspace(::Type{G}, ::Type{T}, lp::SpaceParm) where {G <: Group, T <: AbstractFloat}
         
+        @timeit "Allocating YMWorkspace" begin
             if (G == SU2)
                 GRP = SU2
                 ALG = SU2alg
@@ -67,6 +68,7 @@ struct YMworkspace{T}
             end
             cs = scalar_field(Complex{T}, lp)
             rs = scalar_field(T, lp)
+        end
             
         return new{T}(GRP,ALG,T,f1, f2, mm, u1, cs, rs)
     end

src/YM/YMact.jl

@@ -305,10 +305,14 @@ the prefactor 1/g0^2, and assign it to the workspace force `ymws.frc1`
 function force_gauge(ymws::YMworkspace, U, c0, lp::SpaceParm)
 
     if abs(c0-1) < 1.0E-10
+        @timeit "Wilson gauge force" begin
             force_wilson_pln!(ymws.frc1, ymws.frc2, U, lp::SpaceParm)
+        end
     else
+        @timeit "Improved gauge force" begin
             force_wilson_pln!(ymws.frc1, ymws.frc2, U, lp::SpaceParm, c0)
         end
+    end
     return nothing
 end
 

src/YM/YMfields.jl

@@ -12,18 +12,22 @@
 function randomize!(f, lp::SpaceParm, ymws::YMworkspace) 
         
     if ymws.ALG == SU2alg
+        @timeit "Randomize SU(2) algebra field" begin
             m = CUDA.randn(ymws.PRC, lp.bsz,lp.ndim,3,lp.rsz)
             CUDA.@sync begin
                 CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_assign_SU2!(f,m,lp)
             end
+        end
         return nothing
     end
 
     if ymws.ALG == SU3alg
+        @timeit "Randomize SU(3) algebra field" begin
             m = CUDA.randn(ymws.PRC, lp.bsz,lp.ndim,8,lp.rsz)
             CUDA.@sync begin
                 CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_assign_SU3!(f,m,lp)
             end
+        end
         return nothing
     end
 

src/YM/YMflow.jl

@@ -63,6 +63,7 @@ end
 
 function flw_euler(U, ns, eps, c0, lp::SpaceParm, ymws::YMworkspace; add_zth=false)
     
+    @timeit "Integrating flow equations (Euler)" begin
         for i in 1:ns
             force_gauge(ymws, U, c0, lp)
             if add_zth
@@ -70,12 +71,14 @@ function flw_euler(U, ns, eps, c0, lp::SpaceParm, ymws::YMworkspace; add_zth=fal
             end
             U .= expm.(U, ymws.frc1, 2*eps)
         end
+    end
     
     return nothing
 end
 
 function flw_rk3(U, ns, eps, c0, lp::SpaceParm, ymws::YMworkspace; add_zth=false)
     
+    @timeit "Integrating flow equations (RK3)" begin
         for i in 1:ns
             e0 = eps/2
             force_gauge(ymws, U, c0, lp)
@@ -102,6 +105,7 @@ function flw_rk3(U, ns, eps, c0, lp::SpaceParm, ymws::YMworkspace; add_zth=false
             ymws.mom .= e1.*ymws.frc1 .- ymws.mom 
             U .= expm.(U, ymws.mom)
         end
+    end
                               
     return nothing
 end

src/YM/YMhmc.jl

@@ -18,14 +18,18 @@ Returns the value of the gauge plaquette action for the configuration U. The par
 function gauge_action(U, lp::SpaceParm, gp::GaugeParm{T}, ymws::YMworkspace{T}) where T <: AbstractFloat
 
     if abs(gp.c0-1) < 1.0E-10
+        @timeit "Wilson gauge action" begin
             CUDA.@sync begin
                 CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_plaq!(ymws.cm, U, lp)
             end
+        end
     else
+        @timeit "Improved gauge action" begin
             CUDA.@sync begin
                 CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_impr!(ymws.cm, U, gp.c0, (1-gp.c0)/8, lp)
             end
         end
+    end
     S = gp.beta*( prod(lp.iL)*lp.npls*(gp.c0 + (1-gp.c0)/8) -
                   CUDA.mapreduce(real, +, ymws.cm)/gp.ng )
 
@@ -34,22 +38,28 @@ end
 
 function plaquette(U, lp::SpaceParm, gp::GaugeParm, ymws::YMworkspace)
 
+    @timeit "Plaquette measurement" begin
         CUDA.@sync begin
             CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_plaq!(ymws.cm, U, lp)
         end
+    end
     
     return CUDA.mapreduce(real, +, ymws.cm)/(prod(lp.iL)*lp.npls)
 end
     
 function hamiltonian(mom, U, lp, gp, ymws)
+    @timeit "Computing Hamiltonian" begin
         K = CUDA.mapreduce(norm2, +, mom)/2 
         V = gauge_action(U, lp, gp, ymws)
+    end
         
     return K+V
 end
 
 function HMC!(U, eps, ns, lp::SpaceParm, gp::GaugeParm, ymws::YMworkspace{T}; noacc=false) where T
 
+    @timeit "HMC trayectory" begin
+        
         int = omf4(T, eps, ns)
         ymws.U1 .= U
         
@@ -74,11 +84,14 @@ function HMC!(U, eps, ns, lp::SpaceParm, gp::GaugeParm, ymws::YMworkspace{T}; no
             end
         end
         
+    end
     return dh, acc
 end
 
 function MD!(mom, U, int::IntrScheme{NI, T}, lp::SpaceParm, gp::GaugeParm{T}, ymws::YMworkspace{T}) where {NI, T <: AbstractFloat}
     
+    @timeit "MD evolution" begin
+
         ee = int.eps*gp.beta/gp.ng
         force_gauge(ymws, U, gp.c0, lp)
         mom .= mom .+ (int.r[1]*ee) .* ymws.frc1
@@ -101,6 +114,7 @@ function MD!(mom, U, int::IntrScheme{NI, T}, lp::SpaceParm, gp::GaugeParm{T}, ym
                 k += off
             end
         end
+    end
     
     return nothing
 end

src/main/times.jl

@@ -1,4 +1,4 @@
-using CUDA, Logging, StructArrays, Random
+using CUDA, Logging, StructArrays, Random, TimerOutputs
 
 CUDA.allowscalar(false)
 import Pkg
@@ -7,7 +7,7 @@ Pkg.activate("/lhome/ific/a/alramos/s.images/julia/workspace/LatticeGPU")
 using LatticeGPU
 
 # Set lattice/block size
-lp = SpaceParm{4}((32,32,32,32), (4,4,4,4))
+lp = SpaceParm{4}((16,16,16,16), (4,4,4,4))
 println("Space  Parameters: ", lp)
 
 # Seed RNG
@@ -93,5 +93,6 @@ println("Time for 100 steps of RK3 flow integrator: ")
 println("Action: ", gauge_action(U, lp, gp, ymws))
 println("## END improved action/flow measurements")
 
+print_timer(linechars = :ascii)
 println("END")