From 70d76737c3a81f3ca1930c7633b45d6c791716ee Mon Sep 17 00:00:00 2001
From: "Steven G. Johnson" <stevenj@mit.edu>
Date: Wed, 21 Dec 2016 17:11:43 -0500
Subject: [PATCH] update laplace benchmark to use fused dotops

---
 src/broadcast/BroadcastBenchmarks.jl | 20 +++-----------------
 src/problem/Laplacian.jl             | 12 ++++++++++--
 src/shootout/nbody_vec.jl            | 16 ++--------------
 src/utils/CompatUtils.jl             | 20 ++++++++++++++++++++
 4 files changed, 35 insertions(+), 33 deletions(-)
 create mode 100644 src/utils/CompatUtils.jl

diff --git a/src/broadcast/BroadcastBenchmarks.jl b/src/broadcast/BroadcastBenchmarks.jl
index 1cadf64c..778c74ba 100644
--- a/src/broadcast/BroadcastBenchmarks.jl
+++ b/src/broadcast/BroadcastBenchmarks.jl
@@ -1,28 +1,14 @@
 module BroadcastBenchmarks
 
-include(joinpath(dirname(@__FILE__), "..", "utils", "RandUtils.jl"))
+include(joinpath("..", "utils", "RandUtils.jl"))
+include(joinpath("..", "utils", "CompatUtils.jl"))
 
-using .RandUtils
+using .RandUtils, .CompatUtils
 using BenchmarkTools
 using Compat
 
 const SUITE = BenchmarkGroup()
 
-# work around lack of Compat support for .= (Compat.jl issue #285)
-if VERSION < v"0.5.0-dev+5575" #17510
-    macro dotcompat(ex)
-        if Meta.isexpr(ex, :comparison, 3) && ex.args[2] == :.=
-            :(copy!($(esc(ex.args[1])), $(esc(ex.args[3]))))
-        else
-            esc(ex)
-        end
-    end
-else
-    macro dotcompat(ex)
-        esc(ex)
-    end
-end
-
 ###########################################################################
 
 g = addgroup!(SUITE, "fusion", ["broadcast!", "array"])
diff --git a/src/problem/Laplacian.jl b/src/problem/Laplacian.jl
index f676c374..b5ef76aa 100644
--- a/src/problem/Laplacian.jl
+++ b/src/problem/Laplacian.jl
@@ -4,6 +4,9 @@ using Compat
 
 import Compat: UTF8String, view
 
+include(joinpath("..", "utils", "CompatUtils.jl"))
+using .CompatUtils
+
 ################################
 # Sparse Matrix * Dense Vector #
 ################################
@@ -76,7 +79,7 @@ function perf_laplace_iter_vec(N)
     Niter = 2^10
     dx2 = dy2 = 0.1*0.1
     for i = 1:Niter
-        u[2:N-1, 2:N-1] = ((u[1:N-2, 2:N-1] + u[3:N, 2:N-1])*dy2 + (u[2:N-1, 1:N-2] + u[2:N-1, 3:N])*dx2) * (1 / (2*(dx2+dy2)))
+        @dotcompat u[2:N-1, 2:N-1] .= ((u[1:N-2, 2:N-1] .+ u[3:N, 2:N-1]).*dy2 .+ (u[2:N-1, 1:N-2] .+ u[2:N-1, 3:N]).*dx2) .* (1 / (2*(dx2+dy2)))
     end
     return u
 end
@@ -86,8 +89,13 @@ function perf_laplace_iter_sub(N)
     u[1,:] = 1
     Niter = 2^10
     dx2 = dy2 = 0.1*0.1
+    u0 = view(u, 2:N-1, 2:N-1)
+    u1 = view(u, 1:N-2, 2:N-1)
+    u2 = view(u, 3:N,   2:N-1)
+    u3 = view(u, 2:N-1, 1:N-2)
+    u4 = view(u, 2:N-1, 3:N)
     for i = 1:Niter
-        u[2:N-1, 2:N-1] = ((view(u, 1:N-2, 2:N-1) + view(u,3:N, 2:N-1))*dy2 + (view(u,2:N-1, 1:N-2) + view(u, 2:N-1, 3:N))*dx2) * (1 / (2*(dx2+dy2)))
+        @dotcompat u0 .= ((u1 .+ u2).*dy2 .+ (u3 .+ u4).*dx2) .* (1 / (2*(dx2+dy2)))
     end
     return u
 end
diff --git a/src/shootout/nbody_vec.jl b/src/shootout/nbody_vec.jl
index a069eb4f..02b51d9d 100644
--- a/src/shootout/nbody_vec.jl
+++ b/src/shootout/nbody_vec.jl
@@ -6,20 +6,8 @@
 
 module NBodyVec
 
-# work around lack of Compat support for .= (Compat.jl issue #285)
-if VERSION < v"0.5.0-dev+5575" #17510
-    macro dotcompat(ex)
-        if Meta.isexpr(ex, :comparison, 3) && ex.args[2] == :.=
-            :(copy!($(esc(ex.args[1])), $(esc(ex.args[3]))))
-        else
-            esc(ex)
-        end
-    end
-else
-    macro dotcompat(ex)
-        esc(ex)
-    end
-end
+include(joinpath("..", "utils", "CompatUtils.jl"))
+using .CompatUtils
 
 # Constants
 const solar_mass = 4 * pi * pi
diff --git a/src/utils/CompatUtils.jl b/src/utils/CompatUtils.jl
new file mode 100644
index 00000000..3d79de04
--- /dev/null
+++ b/src/utils/CompatUtils.jl
@@ -0,0 +1,20 @@
+module CompatUtils
+
+export @dotcompat
+
+# work around lack of Compat support for .= (Compat.jl issue #285)
+if VERSION < v"0.5.0-dev+5575" #17510
+    macro dotcompat(ex)
+        if Meta.isexpr(ex, :comparison, 3) && ex.args[2] == :.=
+            :(copy!($(esc(ex.args[1])), $(esc(ex.args[3]))))
+        else
+            esc(ex)
+        end
+    end
+else
+    macro dotcompat(ex)
+        esc(ex)
+    end
+end
+
+end