From 012fb557f4cdac5ab2aed9b469b9c8bfd3441c45 Mon Sep 17 00:00:00 2001
From: americast <sayan.sinha@iitkgp.ac.in>
Date: Sat, 7 Jul 2018 09:49:47 +0530
Subject: [PATCH 1/2] relu_nnpack

---
 src/activation.jl | 21 ++++++++++++++++++++-
 1 file changed, 20 insertions(+), 1 deletion(-)

diff --git a/src/activation.jl b/src/activation.jl
index d5c5092ec..cb6ff19f3 100644
--- a/src/activation.jl
+++ b/src/activation.jl
@@ -45,8 +45,27 @@ const logsigmoid = logσ
 [Rectified Linear Unit](https://en.wikipedia.org/wiki/Rectifier_(neural_networks))
 activation function.
 """
-relu(x) = max(zero(x), x)
+relu_(x) = max(zero(x), x)
+
+function relu_(x, out)
+         ptp = ccall((:pthreadpool_create, :libnnpack), Ptr{Void}, (Csize_t,), 1)
+         input = Cfloat.(x)
+       
+         ccall((:nnp_initialize,"libnnpack"),Void,(),)
+         ccall((:nnp_relu_output,"libnnpack"),Void,(Csize_t, Csize_t, Ptr{Cfloat}, Ptr{Cfloat}, Cfloat, Ptr{Void}), Csize_t(size(input, 2)), Csize_t(size(input, 1)), input, out, Cfloat(0), ptp)
+       
+         return convert(typeof(x), out)
+end
 
+function relu(x)
+  @show length(size(x))
+  if (is_linux() || is_mac()) && length(size(x)) > 0
+    out = zeros(Cfloat, size(x))
+    return relu_(x, out)
+  else
+    return relu_(x)
+  end
+end
 
 """
     leakyrelu(x) = max(0.01x, x)

From 29f25ece627e76f3d50f5b7970941c5c9eedb360 Mon Sep 17 00:00:00 2001
From: americast <sayan.sinha@iitkgp.ac.in>
Date: Sun, 8 Jul 2018 12:05:12 +0530
Subject: [PATCH 2/2] Fixes

---
 src/activation.jl | 13 ++++---------
 1 file changed, 4 insertions(+), 9 deletions(-)

diff --git a/src/activation.jl b/src/activation.jl
index cb6ff19f3..d852b5db5 100644
--- a/src/activation.jl
+++ b/src/activation.jl
@@ -45,26 +45,21 @@ const logsigmoid = logσ
 [Rectified Linear Unit](https://en.wikipedia.org/wiki/Rectifier_(neural_networks))
 activation function.
 """
-relu_(x) = max(zero(x), x)
-
-function relu_(x, out)
+function relu_(x)
          ptp = ccall((:pthreadpool_create, :libnnpack), Ptr{Void}, (Csize_t,), 1)
          input = Cfloat.(x)
-       
+         out = zeros(Cfloat, size(x))
          ccall((:nnp_initialize,"libnnpack"),Void,(),)
          ccall((:nnp_relu_output,"libnnpack"),Void,(Csize_t, Csize_t, Ptr{Cfloat}, Ptr{Cfloat}, Cfloat, Ptr{Void}), Csize_t(size(input, 2)), Csize_t(size(input, 1)), input, out, Cfloat(0), ptp)
        
-         return convert(typeof(x), out)
+         return out
 end
 
 function relu(x)
-  @show length(size(x))
   if (is_linux() || is_mac()) && length(size(x)) > 0
-    out = zeros(Cfloat, size(x))
-    return relu_(x, out)
-  else
     return relu_(x)
   end
+  return max(zero(x), x)
 end
 
 """