diff --git a/src/LatticeGPU.jl b/src/LatticeGPU.jl
index 6096e30..5ce9a18 100644
--- a/src/LatticeGPU.jl
+++ b/src/LatticeGPU.jl
@@ -41,5 +41,6 @@ using .YM
 export YMworkspace, GaugeParm, force0_wilson!, field, field_pln, randomize!, zero!, norm2
 export gauge_action, hamiltonian, plaquette, HMC!, OMF4!
 export wfl_euler, wfl_rk3, zfl_euler, zfl_rk3, Eoft_clover, Eoft_plaq, Qtop
+export sfcoupling
 
 end # module
diff --git a/src/YM/YM.jl b/src/YM/YM.jl
index 9699d74..8d4900a 100644
--- a/src/YM/YM.jl
+++ b/src/YM/YM.jl
@@ -125,4 +125,7 @@ export gauge_action, hamiltonian, plaquette, HMC!, OMF4!
 include("YMflow.jl")
 export wfl_euler, wfl_rk3, zfl_euler, zfl_rk3, Eoft_clover, Eoft_plaq, Qtop
 
+include("YMsf.jl")
+export sfcoupling
+
 end
diff --git a/src/YM/YMsf.jl b/src/YM/YMsf.jl
new file mode 100644
index 0000000..b5aba82
--- /dev/null
+++ b/src/YM/YMsf.jl
@@ -0,0 +1,68 @@
+###
+### "THE BEER-WARE LICENSE":
+### Alberto Ramos wrote this file. As long as you retain this 
+### notice you can do whatever you want with this stuff. If we meet some 
+### day, and you think this stuff is worth it, you can buy me a beer in 
+### return. <alberto.ramos@cern.ch>
+###
+### file:    YMsf.jl
+### created: Tue Oct 26 14:50:55 2021
+###                               
+
+const SR3   = 1.73205080756887729352744634151
+const SR3x2 = 3.46410161513775458705489268302
+
+function sfcoupling(U, lp::SpaceParm, gp::GaugeParm, ymws::YMworkspace)
+
+    if lp.iL[end] < 4
+        error("Array too small to store partial sums")
+    end
+    
+    tmp = zeros(T,lp.iL[end])
+    @timeit "SF coupling measurement" begin
+        CUDA.@sync begin
+            CUDA.@cuda threads=lp.bsz blocks=lp.rsz krnl_sfcoupling!(ymws.rm, U, gp.Ubnd, lp)
+        end
+    end
+
+    tp = ntuple(i->i, N-1)
+    V3 = prod(lp.iL[1:end-1])
+    tmp .= reshape(Array(CUDA.reduce(+, ymws.rm;dims=tp)),lp.iL[end])
+
+    dsdeta = (gp.beta/(2*gp.ng*V3))*(tmp[1] + tmp[end])
+    ddnu   = (gp.beta/(2*gp.ng*V3))*(tmp[2] + tmp[end-1])
+    
+    return dsdeta, ddnu
+end
+    
+function krnl_sfcoupling!(rm, U::AbstractArray{T}, Ubnd::T, lp::SpaceParm{N,M,B,D}) where {T,N,M,B,D}
+
+    b, r = CUDA.threadIdx().x, CUDA.blockIdx().x
+    it   = point_time((b, r), lp)
+    I    = point_coord((b,r), lp)
+    
+    rm[I] = zero(eltype(rm))
+    if (it == 1)
+        but, rut = up((b,r), N, lp)
+        IU = point_coord((but,rut), lp)
+        for id in 1:N-1
+            bu, ru = up((b,r), id, lp)
+
+            X = projalg(U[b,id,r]*U[bu,N,ru]/(U[b,N,r]*U[but,id,rut]))
+            rm[I]  += 3*X.t7 + SR3   * X.t8
+            rm[IU] += 2*X.t7 - SR3x2 * X.t8
+        end
+    elseif (it == lp.iL[end])
+        bdt, rdt = up((b,r), N, lp)
+        ID = point_coord((bdt,rdt), lp)
+        for id in 1:N-1
+            bu, ru = up((b,r), id, lp)
+
+            X = projalg(Ubnd/(U[b,id,r]*U[bu,N,ru])*U[b,N,r])
+            rm[I]  -= 3*X.t7 + SR3   * X.t8
+            rm[ID] += 2*X.t7 - SR3x2 * X.t8
+        end
+    end
+
+    return nothing
+end