Refactoring + SIMD/AVX support

8 years ago · 836361a66c
parent c60d4cbcaf
commit 836361a66c
1 changed files with 140 additions and 101 deletions
--- a/dotnet/WhiteRabbit.UnmanagedBridge/md5.cpp
+++ b/dotnet/WhiteRabbit.UnmanagedBridge/md5.cpp
@ -1,102 +1,139 @@
 #include "stdafx.h"
-
 #include "md5.h"

 #include "intrin.h"
-#include "immintrin.h"

 #pragma unmanaged

-#define OP_XOR(a, b) (a ^ b)
-#define OP_AND(a, b) (a & b)
-#define OP_OR(a, b) (a | b)
-#define OP_NEG(a) (~a)
-#define OP_ADD(a, b) (a + b)
-#define OP_ROT(a, r) (_rotl(a, r))
-#define OP_BLEND(a, b, x) (OP_OR(OP_AND(x, b), OP_AND(OP_NEG(x), a)))
+#if AVX2
+typedef __m256i MD5Vector;
+
+#define OP_XOR(a, b) _mm256_xor_si256(a, b)
+#define OP_AND(a, b) _mm256_and_si256(a, b)
+#define OP_ANDNOT(a, b) _mm256_andnot_si256(a, b)
+#define OP_OR(a, b) _mm256_or_si256(a, b)
+#define OP_ADD(a, b) _mm256_add_epi32(a, b)
+#define OP_ROT(a, r) OP_OR(_mm256_slli_epi32(a, r), _mm256_srli_epi32(a, 32 - (r)))
+#define OP_BLEND(a, b, x) OP_OR(OP_AND(x, b), OP_ANDNOT(x, a))
+
+#define CREATE_VECTOR(a) _mm256_set1_epi32(a)
+#define CREATE_VECTOR_FROM_INPUT(input, offset) _mm256_set1_epi32(input[offset])
+
+#define WRITE_TO_OUTPUT(a, output) \
+    output[0] = a.m256i_u32[0];
+#elif SIMD
+typedef __m128i MD5Vector;
+
+#define OP_XOR(a, b) _mm_xor_si128(a, b)
+#define OP_AND(a, b) _mm_and_si128(a, b)
+#define OP_ANDNOT(a, b) _mm_andnot_si128(a, b)
+#define OP_OR(a, b) _mm_or_si128(a, b)
+#define OP_ADD(a, b) _mm_add_epi32(a, b)
+#define OP_ROT(a, r) OP_OR(_mm_slli_epi32(a, r), _mm_srli_epi32(a, 32 - (r)))
+#define OP_BLEND(a, b, x) OP_OR(OP_AND(x, b), OP_ANDNOT(x, a))

+#define CREATE_VECTOR(a) _mm_set1_epi32(a)
+#define CREATE_VECTOR_FROM_INPUT(input, offset) _mm_set1_epi32(input[offset])
+
+#define WRITE_TO_OUTPUT(a, output) \
+    output[0] = a.m128i_u32[0];
+#else
 typedef unsigned int MD5Vector;

-#define CREATE_VECTOR(a) (a)
+#define OP_XOR(a, b) (a) ^ (b)
+#define OP_AND(a, b) (a) & (b)
+#define OP_ANDNOT(a, b) ~(a) & (b)
+#define OP_OR(a, b) (a) | (b)
+#define OP_ADD(a, b) (a) + (b)
+#define OP_ROT(a, r) _rotl(a, r)
+#define OP_BLEND(a, b, x) ((x) & (b)) | (~(x) & (a))
+
+#define CREATE_VECTOR(a) a
 #define CREATE_VECTOR_FROM_INPUT(input, offset) (input[offset])

+#define WRITE_TO_OUTPUT(a, output) \
+    output[0] = a;
+#endif
+
+#define OP_NEG(a) OP_ANDNOT(a, CREATE_VECTOR(0xffffffff))
+
 typedef struct {
-    MD5Vector K[64];
-    MD5Vector Init[4];
+    unsigned int K[64];
+    unsigned int Init[4];
 } MD5Parameters;

 static const MD5Parameters Parameters = {
    {
-        CREATE_VECTOR(0xd76aa478),
-        CREATE_VECTOR(0xe8c7b756),
-        CREATE_VECTOR(0x242070db),
-        CREATE_VECTOR(0xc1bdceee),
-        CREATE_VECTOR(0xf57c0faf),
-        CREATE_VECTOR(0x4787c62a),
-        CREATE_VECTOR(0xa8304613),
-        CREATE_VECTOR(0xfd469501),
-        CREATE_VECTOR(0x698098d8),
-        CREATE_VECTOR(0x8b44f7af),
-        CREATE_VECTOR(0xffff5bb1),
-        CREATE_VECTOR(0x895cd7be),
-        CREATE_VECTOR(0x6b901122),
-        CREATE_VECTOR(0xfd987193),
-        CREATE_VECTOR(0xa679438e),
-        CREATE_VECTOR(0x49b40821),
-        CREATE_VECTOR(0xf61e2562),
-        CREATE_VECTOR(0xc040b340),
-        CREATE_VECTOR(0x265e5a51),
-        CREATE_VECTOR(0xe9b6c7aa),
-        CREATE_VECTOR(0xd62f105d),
-        CREATE_VECTOR(0x02441453),
-        CREATE_VECTOR(0xd8a1e681),
-        CREATE_VECTOR(0xe7d3fbc8),
-        CREATE_VECTOR(0x21e1cde6),
-        CREATE_VECTOR(0xc33707d6),
-        CREATE_VECTOR(0xf4d50d87),
-        CREATE_VECTOR(0x455a14ed),
-        CREATE_VECTOR(0xa9e3e905),
-        CREATE_VECTOR(0xfcefa3f8),
-        CREATE_VECTOR(0x676f02d9),
-        CREATE_VECTOR(0x8d2a4c8a),
-        CREATE_VECTOR(0xfffa3942),
-        CREATE_VECTOR(0x8771f681),
-        CREATE_VECTOR(0x6d9d6122),
-        CREATE_VECTOR(0xfde5380c),
-        CREATE_VECTOR(0xa4beea44),
-        CREATE_VECTOR(0x4bdecfa9),
-        CREATE_VECTOR(0xf6bb4b60),
-        CREATE_VECTOR(0xbebfbc70),
-        CREATE_VECTOR(0x289b7ec6),
-        CREATE_VECTOR(0xeaa127fa),
-        CREATE_VECTOR(0xd4ef3085),
-        CREATE_VECTOR(0x04881d05),
-        CREATE_VECTOR(0xd9d4d039),
-        CREATE_VECTOR(0xe6db99e5),
-        CREATE_VECTOR(0x1fa27cf8),
-        CREATE_VECTOR(0xc4ac5665),
-        CREATE_VECTOR(0xf4292244),
-        CREATE_VECTOR(0x432aff97),
-        CREATE_VECTOR(0xab9423a7),
-        CREATE_VECTOR(0xfc93a039),
-        CREATE_VECTOR(0x655b59c3),
-        CREATE_VECTOR(0x8f0ccc92),
-        CREATE_VECTOR(0xffeff47d),
-        CREATE_VECTOR(0x85845dd1),
-        CREATE_VECTOR(0x6fa87e4f),
-        CREATE_VECTOR(0xfe2ce6e0),
-        CREATE_VECTOR(0xa3014314),
-        CREATE_VECTOR(0x4e0811a1),
-        CREATE_VECTOR(0xf7537e82),
-        CREATE_VECTOR(0xbd3af235),
-        CREATE_VECTOR(0x2ad7d2bb),
-        CREATE_VECTOR(0xeb86d391),
+        0xd76aa478,
+        0xe8c7b756,
+        0x242070db,
+        0xc1bdceee,
+        0xf57c0faf,
+        0x4787c62a,
+        0xa8304613,
+        0xfd469501,
+        0x698098d8,
+        0x8b44f7af,
+        0xffff5bb1,
+        0x895cd7be,
+        0x6b901122,
+        0xfd987193,
+        0xa679438e,
+        0x49b40821,
+        0xf61e2562,
+        0xc040b340,
+        0x265e5a51,
+        0xe9b6c7aa,
+        0xd62f105d,
+        0x02441453,
+        0xd8a1e681,
+        0xe7d3fbc8,
+        0x21e1cde6,
+        0xc33707d6,
+        0xf4d50d87,
+        0x455a14ed,
+        0xa9e3e905,
+        0xfcefa3f8,
+        0x676f02d9,
+        0x8d2a4c8a,
+        0xfffa3942,
+        0x8771f681,
+        0x6d9d6122,
+        0xfde5380c,
+        0xa4beea44,
+        0x4bdecfa9,
+        0xf6bb4b60,
+        0xbebfbc70,
+        0x289b7ec6,
+        0xeaa127fa,
+        0xd4ef3085,
+        0x04881d05,
+        0xd9d4d039,
+        0xe6db99e5,
+        0x1fa27cf8,
+        0xc4ac5665,
+        0xf4292244,
+        0x432aff97,
+        0xab9423a7,
+        0xfc93a039,
+        0x655b59c3,
+        0x8f0ccc92,
+        0xffeff47d,
+        0x85845dd1,
+        0x6fa87e4f,
+        0xfe2ce6e0,
+        0xa3014314,
+        0x4e0811a1,
+        0xf7537e82,
+        0xbd3af235,
+        0x2ad7d2bb,
+        0xeb86d391,
    },
    {
-        CREATE_VECTOR(0x67452301),
-        CREATE_VECTOR(0xefcdab89),
-        CREATE_VECTOR(0x98badcfe),
-        CREATE_VECTOR(0x10325476),
+        0x67452301,
+        0xefcdab89,
+        0x98badcfe,
+        0x10325476,
    },
 };

@ -122,59 +159,59 @@ inline MD5Vector LeftRotate(MD5Vector x)
 }

 template<int r>
-inline MD5Vector Step1(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, MD5Vector k, MD5Vector w)
+inline MD5Vector Step1(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, unsigned int k, MD5Vector w)
 {
-    return OP_ADD(b, LeftRotate<r>(OP_ADD(Blend(d, c, b), OP_ADD(k, OP_ADD(a, w)))));
+    return OP_ADD(b, LeftRotate<r>(OP_ADD(Blend(d, c, b), OP_ADD(CREATE_VECTOR(k), OP_ADD(a, w)))));
 }

 template<int r>
-inline MD5Vector Step1(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, MD5Vector k)
+inline MD5Vector Step1(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, unsigned int k)
 {
-    return OP_ADD(b, LeftRotate<r>(OP_ADD(Blend(d, c, b), OP_ADD(k, a))));
+    return OP_ADD(b, LeftRotate<r>(OP_ADD(Blend(d, c, b), OP_ADD(CREATE_VECTOR(k), a))));
 }

 template<int r>
-inline MD5Vector Step2(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, MD5Vector k, MD5Vector w)
+inline MD5Vector Step2(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, unsigned int k, MD5Vector w)
 {
-    return OP_ADD(c, LeftRotate<r>(OP_ADD(Blend(d, c, b), OP_ADD(k, OP_ADD(a, w)))));
+    return OP_ADD(c, LeftRotate<r>(OP_ADD(Blend(d, c, b), OP_ADD(CREATE_VECTOR(k), OP_ADD(a, w)))));
 }

 template<int r>
-inline MD5Vector Step2(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, MD5Vector k)
+inline MD5Vector Step2(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, unsigned int k)
 {
-    return OP_ADD(c, LeftRotate<r>(OP_ADD(Blend(d, c, b), OP_ADD(k, a))));
+    return OP_ADD(c, LeftRotate<r>(OP_ADD(Blend(d, c, b), OP_ADD(CREATE_VECTOR(k), a))));
 }

 template<int r>
-inline MD5Vector Step3(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, MD5Vector k, MD5Vector w)
+inline MD5Vector Step3(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, unsigned int k, MD5Vector w)
 {
-    return OP_ADD(b, LeftRotate<r>(OP_ADD(Xor(b, c, d), OP_ADD(k, OP_ADD(a, w)))));
+    return OP_ADD(b, LeftRotate<r>(OP_ADD(Xor(b, c, d), OP_ADD(CREATE_VECTOR(k), OP_ADD(a, w)))));
 }

 template<int r>
-inline MD5Vector Step3(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, MD5Vector k)
+inline MD5Vector Step3(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, unsigned int k)
 {
-    return OP_ADD(b, LeftRotate<r>(OP_ADD(Xor(b, c, d), OP_ADD(k, a))));
+    return OP_ADD(b, LeftRotate<r>(OP_ADD(Xor(b, c, d), OP_ADD(CREATE_VECTOR(k), a))));
 }

 template<int r>
-inline MD5Vector Step4(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, MD5Vector k, MD5Vector w)
+inline MD5Vector Step4(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, unsigned int k, MD5Vector w)
 {
-    return OP_ADD(b, LeftRotate<r>(OP_ADD(I(c, b, d), OP_ADD(k, OP_ADD(a, w)))));
+    return OP_ADD(b, LeftRotate<r>(OP_ADD(I(c, b, d), OP_ADD(CREATE_VECTOR(k), OP_ADD(a, w)))));
 }

 template<int r>
-inline MD5Vector Step4(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, MD5Vector k)
+inline MD5Vector Step4(MD5Vector a, MD5Vector b, MD5Vector c, MD5Vector d, unsigned int k)
 {
-    return OP_ADD(b, LeftRotate<r>(OP_ADD(I(c, b, d), OP_ADD(k, a))));
+    return OP_ADD(b, LeftRotate<r>(OP_ADD(I(c, b, d), OP_ADD(CREATE_VECTOR(k), a))));
 }

 void md5(unsigned int * input, unsigned int * output)
 {
-    MD5Vector a = Parameters.Init[0];
-    MD5Vector b = Parameters.Init[1];
-    MD5Vector c = Parameters.Init[2];
-    MD5Vector d = Parameters.Init[3];
+    MD5Vector a = CREATE_VECTOR(Parameters.Init[0]);
+    MD5Vector b = CREATE_VECTOR(Parameters.Init[1]);
+    MD5Vector c = CREATE_VECTOR(Parameters.Init[2]);
+    MD5Vector d = CREATE_VECTOR(Parameters.Init[3]);

    MD5Vector inputVectors[8] = {
        CREATE_VECTOR_FROM_INPUT(input, 0),
@ -252,6 +289,8 @@ void md5(unsigned int * input, unsigned int * output)
    b = Step4<21>(b, c, d, a, Parameters.K[59]);
    a = Step4< 6>(a, b, c, d, Parameters.K[60], inputVectors[4]);

-    output[0] = OP_ADD(Parameters.Init[0], a);
+    a = OP_ADD(CREATE_VECTOR(Parameters.Init[0]), a);
+
+    WRITE_TO_OUTPUT(a, output);
 }
 #pragma managed