From 8ea7b87eed7eeefcb155b663517013308009cdf3 Mon Sep 17 00:00:00 2001
From: Jan Weidner <jw3126@gmail.com>
Date: Thu, 11 Jan 2018 15:50:58 +0100
Subject: [PATCH 1/2] unrolled transform!

---
 src/core.jl  | 53 +++++++++++++++++++++++++++++++++++++++++++++++++++-
 test/perf.jl | 16 ++++++++++++++++
 2 files changed, 68 insertions(+), 1 deletion(-)
 create mode 100644 test/perf.jl

diff --git a/src/core.jl b/src/core.jl
index f141497..35810d7 100644
--- a/src/core.jl
+++ b/src/core.jl
@@ -22,8 +22,9 @@ function conclude_round(a,b,c,d,f,g,pbuf,i)
     a,b,c,d
 end
 
+transform!(ctx::MD5_CTX) = transform_unrolled!(ctx)
 
-function transform!(context::MD5_CTX)
+function transform_baseline!(context::MD5_CTX)
    pbuf = buffer_pointer(context)
    a,b,c,d = context.state
      
@@ -50,6 +51,56 @@ function transform!(context::MD5_CTX)
     @inbounds context.state .+= [a,b,c,d]
 end
 
+@generated function transform_unrolled!(context::MD5_CTX)
+    ret = quote
+        pbuf = buffer_pointer(context)
+    end
+    ex  = quote
+        A = context.state[1]
+        B = context.state[2]
+        C = context.state[3]
+        D = context.state[4]
+    end
+    push!(ret.args, ex)
+    for i in 0:63
+        if 0 ≤ i ≤ 15
+            ex = :(F = (B & C) | ((~B) & D))
+            g = i
+        elseif 16 ≤ i ≤ 31
+            ex = :(F = (D & B) | ((~D) & C))
+            g = 5i + 1
+        elseif 32 ≤ i ≤ 47
+            ex = :(F = B ⊻ C ⊻ D)
+            g = 3i + 5
+        elseif 48 ≤ i ≤ 63
+            ex = :(F = C ⊻ (B | (~D)))
+            g = 7i
+        end
+        push!(ret.args, ex)
+        g = (g % 16) + 1
+        ex = quote
+            temp = D
+            D = C
+            C = B
+            inner = A + F + $(kk[i+1]) + unsafe_load(pbuf, $g)
+            rot_inner = lrot($(ss[i+1]), inner, 32)
+            B = B + rot_inner
+            A = temp
+        end
+        push!(ret.args, ex)
+    end
+
+    ex = quote
+        context.state[1] += A
+        context.state[2] += B
+        context.state[3] += C
+        context.state[4] += D
+    end
+    push!(ret.args, ex)
+    quote
+        @inbounds $ret
+    end
+end
 
 function digest!(context::T) where {T<:MD5_CTX}
     pad_remainder!(context)
diff --git a/test/perf.jl b/test/perf.jl
new file mode 100644
index 0000000..c12f398
--- /dev/null
+++ b/test/perf.jl
@@ -0,0 +1,16 @@
+import MD5
+using BenchmarkTools
+
+data = randstring(10^4)
+
+if Pkg.installed("Nettle") != nothing
+import Nettle
+@show Nettle.hexdigest("md5", data)
+display(@benchmark Nettle.hexdigest("md5", data))
+    println()
+end
+
+@show bytes2hex(MD5.md5(data))
+display(@benchmark MD5.md5(data))
+println()
+

From c310e5bd6b3c379c4546fd72a028e08e16a1f3fa Mon Sep 17 00:00:00 2001
From: Jan Weidner <jw3126@gmail.com>
Date: Thu, 11 Jan 2018 16:51:56 +0100
Subject: [PATCH 2/2] cleanup

---
 LICENSE.md  |  2 +-
 src/core.jl | 73 ++++++++---------------------------------------------
 2 files changed, 11 insertions(+), 64 deletions(-)

diff --git a/LICENSE.md b/LICENSE.md
index 24d9bc6..c0e8d44 100644
--- a/LICENSE.md
+++ b/LICENSE.md
@@ -1,6 +1,6 @@
 The MD5.jl package is licensed under the MIT "Expat" License:
 
-> Copyright (c) 2018: Lyndon White.
+> Copyright (c) 2018: Lyndon White and Jan Weidner.
 > Copyright (c) 2014: Elliot Saba. (SHA.jl)
 > Derived from the RSA Data Security, Inc. MD5 Message-Digest Algorithm,  Copyright (C) 1991-2
 > 
diff --git a/src/core.jl b/src/core.jl
index 35810d7..c9af954 100644
--- a/src/core.jl
+++ b/src/core.jl
@@ -1,65 +1,12 @@
-# Nonlinear functions, in order to encourage inlining, these sadly are not an array of lambdas
-f_round0(b,c,d) = SHA.Round0(b,c,d) #UInt32((b & c) | (~b & d))
-f_round1(b,c,d) = UInt32((b & d) | (c & ~d))
-f_round2(b,c,d) = SHA.Round1And3(b,c,d) # return UInt32(b ⊻ c ⊻ d) #xors
-f_round3(b,c,d) = UInt32(c ⊻ (b | ~d))
-
-g_round0(i) = i
-g_round1(i) = (5i+1) % 16
-g_round2(i) = (3i + 5) % 16
-g_round3(i) = 7i % 16
-
-
-function conclude_round(a,b,c,d,f,g,pbuf,i)
-    @inbounds s = ss[i+1]
-    @inbounds k = kk[i+1]
-    @inbounds m = unsafe_load(pbuf, g+1)
-    f = f + a + k + m
-    a = d
-    d = c
-    c = b
-    b = b + lrot(s,f, 32)
-    a,b,c,d
-end
-
-transform!(ctx::MD5_CTX) = transform_unrolled!(ctx)
-
-function transform_baseline!(context::MD5_CTX)
-   pbuf = buffer_pointer(context)
-   a,b,c,d = context.state
-     
-   for i in 0:15
-        f = f_round0(b,c,d)
-        g = g_round0(i)
-        a,b,c,d = conclude_round(a, b, c, d, f, g, pbuf, i)
-   end
-   for i in 16:31
-        f = f_round1(b,c,d)
-        g = g_round1(i)
-        a,b,c,d = conclude_round(a, b, c, d, f, g, pbuf, i)
-   end
-   for i in 32:47
-        f = f_round2(b,c,d)
-        g = g_round2(i)
-        a,b,c,d = conclude_round(a, b, c, d, f, g, pbuf, i)
-   end
-   for i in 48:63
-        f = f_round3(b,c,d)
-        g = g_round3(i)
-        a,b,c,d = conclude_round(a, b, c, d, f, g, pbuf, i)
-    end
-    @inbounds context.state .+= [a,b,c,d]
-end
-
-@generated function transform_unrolled!(context::MD5_CTX)
+@generated function transform!(context::MD5_CTX)
     ret = quote
         pbuf = buffer_pointer(context)
     end
     ex  = quote
-        A = context.state[1]
-        B = context.state[2]
-        C = context.state[3]
-        D = context.state[4]
+        @inbounds A = context.state[1]
+        @inbounds B = context.state[2]
+        @inbounds C = context.state[3]
+        @inbounds D = context.state[4]
     end
     push!(ret.args, ex)
     for i in 0:63
@@ -91,14 +38,14 @@ end
     end
 
     ex = quote
-        context.state[1] += A
-        context.state[2] += B
-        context.state[3] += C
-        context.state[4] += D
+        @inbounds context.state[1] += A
+        @inbounds context.state[2] += B
+        @inbounds context.state[3] += C
+        @inbounds context.state[4] += D
     end
     push!(ret.args, ex)
     quote
-        @inbounds $ret
+        $ret
     end
 end