From 69f2911e04ffb1b19eef1fafb8c040af271f656e Mon Sep 17 00:00:00 2001 From: Tor Aamodt Date: Thu, 15 Jul 2010 18:09:46 -0800 Subject: creating branch for adding support for CUDA 3.x and Fermi [git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 6829] --- benchmarks/CUDA/STO/md5_kernel.cu | 995 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 995 insertions(+) create mode 100644 benchmarks/CUDA/STO/md5_kernel.cu (limited to 'benchmarks/CUDA/STO/md5_kernel.cu') diff --git a/benchmarks/CUDA/STO/md5_kernel.cu b/benchmarks/CUDA/STO/md5_kernel.cu new file mode 100644 index 0000000..419daa5 --- /dev/null +++ b/benchmarks/CUDA/STO/md5_kernel.cu @@ -0,0 +1,995 @@ +/*========================================================================== + MD5 KERNEL + +* Copyright (c) 2008, NetSysLab at the University of British Columbia +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* * Redistributions of source code must retain the above copyright +* notice, this list of conditions and the following disclaimer. +* * Redistributions in binary form must reproduce the above copyright +* notice, this list of conditions and the following disclaimer in the +* documentation and/or other materials provided with the distribution. +* * Neither the name of the University nor the +* names of its contributors may be used to endorse or promote products +* derived from this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY NetSysLab ``AS IS'' AND ANY +* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED +* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL NetSysLab BE LIABLE FOR ANY +* DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES +* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; +* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND +* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +DESCRIPTION + CPU version of the storeGPU library. + + +==========================================================================*/ + +/*========================================================================== + + INCLUDES + +==========================================================================*/ +#include +#include +#include "cust.h" + +/*========================================================================== + + DATA DECLARATIONS + +==========================================================================*/ + +/*-------------------------------------------------------------------------- + TYPE DEFINITIONS +--------------------------------------------------------------------------*/ +typedef struct { + unsigned long total[2]; /*!< number of bytes processed */ + unsigned long state[4]; /*!< intermediate digest state */ + unsigned char buffer[64]; /*!< data block being processed */ +} md5_context; + +/*-------------------------------------------------------------------------- + FUNCTION PROTOTYPES +--------------------------------------------------------------------------*/ + +/*-------------------------------------------------------------------------- + CONSTANTS +--------------------------------------------------------------------------*/ + +/*-------------------------------------------------------------------------- + GLOBAL VARIABLES +--------------------------------------------------------------------------*/ + +__device__ +const unsigned char md5_padding[64] = +{ + 0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 +}; + +/*-------------------------------------------------------------------------- + MACROS +--------------------------------------------------------------------------*/ +// 32-bit integer manipulation macros (little endian) +#ifndef GET_UINT32_LE +#define GET_UINT32_LE(n,b,i) \ +{ \ + (n) = ( (unsigned long) (b)[(i) ] ) \ + | ( (unsigned long) (b)[(i) + 1] << 8 ) \ + | ( (unsigned long) (b)[(i) + 2] << 16 ) \ + | ( (unsigned long) (b)[(i) + 3] << 24 ); \ +} +#endif + +#ifndef PUT_UINT32_LE +#define PUT_UINT32_LE(n,b,i) \ +{ \ + (b)[(i) ] = (unsigned char) ( (n) ); \ + (b)[(i) + 1] = (unsigned char) ( (n) >> 8 ); \ + (b)[(i) + 2] = (unsigned char) ( (n) >> 16 ); \ + (b)[(i) + 3] = (unsigned char) ( (n) >> 24 ); \ +} +#endif + +#ifdef FEATURE_SHARED_MEMORY +// current thread stride. +#define SHARED_MEMORY_INDEX(index) (32 * (index) + (threadIdx.x & 0x1F)) +#endif /* FEATURE_SHARED_MEMORY */ + + + +/*========================================================================== + + FUNCTIONS + +==========================================================================*/ + +/*-------------------------------------------------------------------------- + LOCAL FUNCTIONS +--------------------------------------------------------------------------*/ + + +#ifndef FEATURE_SHARED_MEMORY +/*=========================================================================== + +FUNCTION + +DESCRIPTION + MD5 context setup + +DEPENDENCIES + + +RETURN VALUE + + +===========================================================================*/ +__device__ +static void md5_starts( md5_context *ctx ) { + ctx->total[0] = 0; + ctx->total[1] = 0; + + ctx->state[0] = 0x67452301; + ctx->state[1] = 0xEFCDAB89; + ctx->state[2] = 0x98BADCFE; + ctx->state[3] = 0x10325476; +} + +/*=========================================================================== + +FUNCTION MD5_PROCESS + +DESCRIPTION + + +DEPENDENCIES + + +RETURN VALUE + + +===========================================================================*/ +__device__ +static void md5_process( md5_context *ctx, unsigned char data[64] ) { + + unsigned long A, B, C, D; + unsigned long *X = (unsigned long *)data; + + + GET_UINT32_LE( X[ 0], data, 0 ); + GET_UINT32_LE( X[ 1], data, 4 ); + GET_UINT32_LE( X[ 2], data, 8 ); + GET_UINT32_LE( X[ 3], data, 12 ); + GET_UINT32_LE( X[ 4], data, 16 ); + GET_UINT32_LE( X[ 5], data, 20 ); + GET_UINT32_LE( X[ 6], data, 24 ); + GET_UINT32_LE( X[ 7], data, 28 ); + GET_UINT32_LE( X[ 8], data, 32 ); + GET_UINT32_LE( X[ 9], data, 36 ); + GET_UINT32_LE( X[10], data, 40 ); + GET_UINT32_LE( X[11], data, 44 ); + GET_UINT32_LE( X[12], data, 48 ); + GET_UINT32_LE( X[13], data, 52 ); + GET_UINT32_LE( X[14], data, 56 ); + GET_UINT32_LE( X[15], data, 60 ); + +#undef S +#define S(x,n) ((x << n) | ((x & 0xFFFFFFFF) >> (32 - n))) + +#undef P +#define P(a,b,c,d,k,s,t) { \ + a += F(b,c,d) + X[k] + t; a = S(a,s) + b; \ + } \ + + A = ctx->state[0]; + B = ctx->state[1]; + C = ctx->state[2]; + D = ctx->state[3]; + +#define F(x,y,z) (z ^ (x & (y ^ z))) + + P( A, B, C, D, 0, 7, 0xD76AA478 ); + P( D, A, B, C, 1, 12, 0xE8C7B756 ); + P( C, D, A, B, 2, 17, 0x242070DB ); + P( B, C, D, A, 3, 22, 0xC1BDCEEE ); + P( A, B, C, D, 4, 7, 0xF57C0FAF ); + P( D, A, B, C, 5, 12, 0x4787C62A ); + P( C, D, A, B, 6, 17, 0xA8304613 ); + P( B, C, D, A, 7, 22, 0xFD469501 ); + P( A, B, C, D, 8, 7, 0x698098D8 ); + P( D, A, B, C, 9, 12, 0x8B44F7AF ); + P( C, D, A, B, 10, 17, 0xFFFF5BB1 ); + P( B, C, D, A, 11, 22, 0x895CD7BE ); + P( A, B, C, D, 12, 7, 0x6B901122 ); + P( D, A, B, C, 13, 12, 0xFD987193 ); + P( C, D, A, B, 14, 17, 0xA679438E ); + P( B, C, D, A, 15, 22, 0x49B40821 ); + +#undef F + +#define F(x,y,z) (y ^ (z & (x ^ y))) + + P( A, B, C, D, 1, 5, 0xF61E2562 ); + P( D, A, B, C, 6, 9, 0xC040B340 ); + P( C, D, A, B, 11, 14, 0x265E5A51 ); + P( B, C, D, A, 0, 20, 0xE9B6C7AA ); + P( A, B, C, D, 5, 5, 0xD62F105D ); + P( D, A, B, C, 10, 9, 0x02441453 ); + P( C, D, A, B, 15, 14, 0xD8A1E681 ); + P( B, C, D, A, 4, 20, 0xE7D3FBC8 ); + P( A, B, C, D, 9, 5, 0x21E1CDE6 ); + P( D, A, B, C, 14, 9, 0xC33707D6 ); + P( C, D, A, B, 3, 14, 0xF4D50D87 ); + P( B, C, D, A, 8, 20, 0x455A14ED ); + P( A, B, C, D, 13, 5, 0xA9E3E905 ); + P( D, A, B, C, 2, 9, 0xFCEFA3F8 ); + P( C, D, A, B, 7, 14, 0x676F02D9 ); + P( B, C, D, A, 12, 20, 0x8D2A4C8A ); + +#undef F + +#define F(x,y,z) (x ^ y ^ z) + + P( A, B, C, D, 5, 4, 0xFFFA3942 ); + P( D, A, B, C, 8, 11, 0x8771F681 ); + P( C, D, A, B, 11, 16, 0x6D9D6122 ); + P( B, C, D, A, 14, 23, 0xFDE5380C ); + P( A, B, C, D, 1, 4, 0xA4BEEA44 ); + P( D, A, B, C, 4, 11, 0x4BDECFA9 ); + P( C, D, A, B, 7, 16, 0xF6BB4B60 ); + P( B, C, D, A, 10, 23, 0xBEBFBC70 ); + P( A, B, C, D, 13, 4, 0x289B7EC6 ); + P( D, A, B, C, 0, 11, 0xEAA127FA ); + P( C, D, A, B, 3, 16, 0xD4EF3085 ); + P( B, C, D, A, 6, 23, 0x04881D05 ); + P( A, B, C, D, 9, 4, 0xD9D4D039 ); + P( D, A, B, C, 12, 11, 0xE6DB99E5 ); + P( C, D, A, B, 15, 16, 0x1FA27CF8 ); + P( B, C, D, A, 2, 23, 0xC4AC5665 ); + +#undef F + +#define F(x,y,z) (y ^ (x | ~z)) + + P( A, B, C, D, 0, 6, 0xF4292244 ); + P( D, A, B, C, 7, 10, 0x432AFF97 ); + P( C, D, A, B, 14, 15, 0xAB9423A7 ); + P( B, C, D, A, 5, 21, 0xFC93A039 ); + P( A, B, C, D, 12, 6, 0x655B59C3 ); + P( D, A, B, C, 3, 10, 0x8F0CCC92 ); + P( C, D, A, B, 10, 15, 0xFFEFF47D ); + P( B, C, D, A, 1, 21, 0x85845DD1 ); + P( A, B, C, D, 8, 6, 0x6FA87E4F ); + P( D, A, B, C, 15, 10, 0xFE2CE6E0 ); + P( C, D, A, B, 6, 15, 0xA3014314 ); + P( B, C, D, A, 13, 21, 0x4E0811A1 ); + P( A, B, C, D, 4, 6, 0xF7537E82 ); + P( D, A, B, C, 11, 10, 0xBD3AF235 ); + P( C, D, A, B, 2, 15, 0x2AD7D2BB ); + P( B, C, D, A, 9, 21, 0xEB86D391 ); + +#undef F + + ctx->state[0] += A; + ctx->state[1] += B; + ctx->state[2] += C; + ctx->state[3] += D; +} + +/*=========================================================================== + +FUNCTION MD5_UPDATE + +DESCRIPTION + MD5 process buffer + +DEPENDENCIES + + +RETURN VALUE + + +===========================================================================*/ +__device__ +static void md5_update( md5_context *ctx, unsigned char *input, int ilen ) { + int fill; + unsigned long left; + + if( ilen <= 0 ) + return; + + left = ctx->total[0] & 0x3F; + fill = 64 - left; + + ctx->total[0] += ilen; + ctx->total[0] &= 0xFFFFFFFF; + + if( ctx->total[0] < (unsigned long) ilen ) + ctx->total[1]++; + + if( left && ilen >= fill ) { + + // + /*memcpy( (void *) (ctx->buffer + left), + (void *) input, fill );*/ + for (int i = 0; i < fill; i++) { + ctx->buffer[i+left] = input[i]; + } + // + + md5_process( ctx, ctx->buffer ); + input += fill; + ilen -= fill; + left = 0; + } + + while( ilen >= 64 ) { + md5_process( ctx, input ); + input += 64; + ilen -= 64; + } + + if( ilen > 0 ) { + + // + /* memcpy( (void *) (ctx->buffer + left), + (void *) input, ilen );*/ + for (int i = 0; i < ilen; i++) { + ctx->buffer[i+left] = input[i]; + } + // + + } +} + +/*=========================================================================== + +FUNCTION MD5_FINISH + +DESCRIPTION + MD5 final digest + +DEPENDENCIES + None. + +RETURN VALUE + + +===========================================================================*/ +__device__ +void md5_finish( md5_context *ctx, unsigned char *output ) { + + unsigned long last, padn; + unsigned long high, low; + unsigned char msglen[8]; + + high = ( ctx->total[0] >> 29 ) | ( ctx->total[1] << 3 ); + low = ( ctx->total[0] << 3 ); + + PUT_UINT32_LE( low, msglen, 0 ); + PUT_UINT32_LE( high, msglen, 4 ); + + last = ctx->total[0] & 0x3F; + padn = ( last < 56 ) ? ( 56 - last ) : ( 120 - last ); + + md5_update( ctx, (unsigned char *) md5_padding, padn ); + md5_update( ctx, msglen, 8 ); + + + PUT_UINT32_LE( ctx->state[0], output, 0 ); +#ifndef FEATURE_REDUCED_HASH_SIZE + PUT_UINT32_LE( ctx->state[1], output, 4 ); + PUT_UINT32_LE( ctx->state[2], output, 8 ); + PUT_UINT32_LE( ctx->state[3], output, 12 ); +#endif +} + +/*=========================================================================== + +FUNCTION MD5_INTERNAL + +DESCRIPTION + Does the real md5 algorithm + +DEPENDENCIES + None + +RETURN VALUE + output is the hash result + +===========================================================================*/ +__device__ +static void md5_internal( unsigned char *input, int ilen, + unsigned char *output ) { + md5_context ctx; + + md5_starts( &ctx ); + md5_update( &ctx, input, ilen ); + md5_finish( &ctx, output ); + +} +#endif /* #ifndef FEATURE_SHARED_MEMORY */ + +#ifdef FEATURE_SHARED_MEMORY +/*=========================================================================== + +FUNCTION MD5_INTERNAL + +DESCRIPTION + Does the real md5 algorithm. + +DEPENDENCIES + None + +RETURN VALUE + output is the hash result + +===========================================================================*/ + +__device__ +static void md5_internal( unsigned int *input, unsigned int *sharedMemory, + int chunkSize, unsigned char *output ) { + + /* Number of passes (512 bit blocks) we have to do */ + int numberOfPasses = chunkSize / 64 + 1; + /* Used during the hashing process */ + unsigned long A, B, C, D; + /* Needed to do the little endian stuff */ + unsigned char *data = (unsigned char *)sharedMemory; + + /* Will hold the hash value through the + intermediate stages of MD5 algorithm */ + unsigned int state0 = 0x67452301; + unsigned int state1 = 0xEFCDAB89; + unsigned int state2 = 0x98BADCFE; + unsigned int state3 = 0x10325476; + + + /* Used to cache the shared memory index calculations, but testing showed + that it has no performance effect. */ + int x0 = SHARED_MEMORY_INDEX(0); + int x1 = SHARED_MEMORY_INDEX(1); + int x2 = SHARED_MEMORY_INDEX(2); + int x3 = SHARED_MEMORY_INDEX(3); + int x4 = SHARED_MEMORY_INDEX(4); + int x5 = SHARED_MEMORY_INDEX(5); + int x6 = SHARED_MEMORY_INDEX(6); + int x7 = SHARED_MEMORY_INDEX(7); + int x8 = SHARED_MEMORY_INDEX(8); + int x9 = SHARED_MEMORY_INDEX(9); + int x10 = SHARED_MEMORY_INDEX(10); + int x11 = SHARED_MEMORY_INDEX(11); + int x12 = SHARED_MEMORY_INDEX(12); + int x13 = SHARED_MEMORY_INDEX(13); + int x14 = SHARED_MEMORY_INDEX(14); + int x15 = SHARED_MEMORY_INDEX(15); + +#undef GET_CACHED_INDEX +#define GET_CACHED_INDEX(index) (x##index) + + + for( int index = 0 ; index < (numberOfPasses) ; index++ ) { + + /* Move data to the thread's shared memory space */ + sharedMemory[GET_CACHED_INDEX(0)] = input[0 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(1)] = input[1 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(2)] = input[2 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(3)] = input[3 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(4)] = input[4 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(5)] = input[5 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(6)] = input[6 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(7)] = input[7 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(8)] = input[8 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(9)] = input[9 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(10)] = input[10 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(11)] = input[11 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(12)] = input[12 + 16 * index]; + + /* Testing the code with and without this if statement shows that + it has no effect on performance. */ + if(index == numberOfPasses -1 ) { + /* The last pass will contain the size of the chunk size (according to + official MD5 algorithm). */ + sharedMemory[GET_CACHED_INDEX(13)] = 0x00000080; + sharedMemory[GET_CACHED_INDEX(14)] = chunkSize << 3; + sharedMemory[GET_CACHED_INDEX(15)] = chunkSize >> 29; + } else { + sharedMemory[GET_CACHED_INDEX(13)] = input[13 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(14)] = input[14 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(15)] = input[15 + 16 * index]; + } + + /* Get the little endian stuff done. */ + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(0)], + data, GET_CACHED_INDEX(0) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(1)], + data, GET_CACHED_INDEX(1) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(2)], + data, GET_CACHED_INDEX(2) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(3)], + data, GET_CACHED_INDEX(3) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(4)], + data, GET_CACHED_INDEX(4) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(5)], + data, GET_CACHED_INDEX(5) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(6)], + data, GET_CACHED_INDEX(6) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(7)], + data, GET_CACHED_INDEX(7) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(8)], + data, GET_CACHED_INDEX(8) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(9)], + data, GET_CACHED_INDEX(9) * 4 ); + GET_UINT32_LE( sharedMemory[GET_CACHED_INDEX(10)], + data, GET_CACHED_INDEX(10) * 4 ); + GET_UINT32_LE( sharedMemory[GET_CACHED_INDEX(11)], + data, GET_CACHED_INDEX(11) * 4 ); + GET_UINT32_LE( sharedMemory[GET_CACHED_INDEX(12)], + data, GET_CACHED_INDEX(12) * 4 ); + GET_UINT32_LE( sharedMemory[GET_CACHED_INDEX(13)], + data, GET_CACHED_INDEX(13) * 4 ); + GET_UINT32_LE( sharedMemory[GET_CACHED_INDEX(14)], + data, GET_CACHED_INDEX(14) * 4 ); + GET_UINT32_LE( sharedMemory[GET_CACHED_INDEX(15)], + data, GET_CACHED_INDEX(15) * 4 ); + + + /* Start the MD5 permutations */ +#undef S +#define S(x,n) ((x << n) | ((x & 0xFFFFFFFF) >> (32 - n))) +#undef P +#define P(a,b,c,d,k,s,t) { \ + a += F(b,c,d) + sharedMemory[GET_CACHED_INDEX(k)] + t; a = S(a,s) + b; \ + } \ + + A = state0; + B = state1; + C = state2; + D = state3; + +#undef F + +#define F(x,y,z) (z ^ (x & (y ^ z))) + + P( A, B, C, D, 0, 7, 0xD76AA478 ); + P( D, A, B, C, 1, 12, 0xE8C7B756 ); + P( C, D, A, B, 2, 17, 0x242070DB ); + P( B, C, D, A, 3, 22, 0xC1BDCEEE ); + P( A, B, C, D, 4, 7, 0xF57C0FAF ); + P( D, A, B, C, 5, 12, 0x4787C62A ); + P( C, D, A, B, 6, 17, 0xA8304613 ); + P( B, C, D, A, 7, 22, 0xFD469501 ); + P( A, B, C, D, 8, 7, 0x698098D8 ); + P( D, A, B, C, 9, 12, 0x8B44F7AF ); + P( C, D, A, B, 10, 17, 0xFFFF5BB1 ); + P( B, C, D, A, 11, 22, 0x895CD7BE ); + P( A, B, C, D, 12, 7, 0x6B901122 ); + P( D, A, B, C, 13, 12, 0xFD987193 ); + P( C, D, A, B, 14, 17, 0xA679438E ); + P( B, C, D, A, 15, 22, 0x49B40821 ); + +#undef F + +#define F(x,y,z) (y ^ (z & (x ^ y))) + + P( A, B, C, D, 1, 5, 0xF61E2562 ); + P( D, A, B, C, 6, 9, 0xC040B340 ); + P( C, D, A, B, 11, 14, 0x265E5A51 ); + P( B, C, D, A, 0, 20, 0xE9B6C7AA ); + P( A, B, C, D, 5, 5, 0xD62F105D ); + P( D, A, B, C, 10, 9, 0x02441453 ); + P( C, D, A, B, 15, 14, 0xD8A1E681 ); + P( B, C, D, A, 4, 20, 0xE7D3FBC8 ); + P( A, B, C, D, 9, 5, 0x21E1CDE6 ); + P( D, A, B, C, 14, 9, 0xC33707D6 ); + P( C, D, A, B, 3, 14, 0xF4D50D87 ); + P( B, C, D, A, 8, 20, 0x455A14ED ); + P( A, B, C, D, 13, 5, 0xA9E3E905 ); + P( D, A, B, C, 2, 9, 0xFCEFA3F8 ); + P( C, D, A, B, 7, 14, 0x676F02D9 ); + P( B, C, D, A, 12, 20, 0x8D2A4C8A ); + +#undef F + +#define F(x,y,z) (x ^ y ^ z) + + P( A, B, C, D, 5, 4, 0xFFFA3942 ); + P( D, A, B, C, 8, 11, 0x8771F681 ); + P( C, D, A, B, 11, 16, 0x6D9D6122 ); + P( B, C, D, A, 14, 23, 0xFDE5380C ); + P( A, B, C, D, 1, 4, 0xA4BEEA44 ); + P( D, A, B, C, 4, 11, 0x4BDECFA9 ); + P( C, D, A, B, 7, 16, 0xF6BB4B60 ); + P( B, C, D, A, 10, 23, 0xBEBFBC70 ); + P( A, B, C, D, 13, 4, 0x289B7EC6 ); + P( D, A, B, C, 0, 11, 0xEAA127FA ); + P( C, D, A, B, 3, 16, 0xD4EF3085 ); + P( B, C, D, A, 6, 23, 0x04881D05 ); + P( A, B, C, D, 9, 4, 0xD9D4D039 ); + P( D, A, B, C, 12, 11, 0xE6DB99E5 ); + P( C, D, A, B, 15, 16, 0x1FA27CF8 ); + P( B, C, D, A, 2, 23, 0xC4AC5665 ); + +#undef F + +#define F(x,y,z) (y ^ (x | ~z)) + + P( A, B, C, D, 0, 6, 0xF4292244 ); + P( D, A, B, C, 7, 10, 0x432AFF97 ); + P( C, D, A, B, 14, 15, 0xAB9423A7 ); + P( B, C, D, A, 5, 21, 0xFC93A039 ); + P( A, B, C, D, 12, 6, 0x655B59C3 ); + P( D, A, B, C, 3, 10, 0x8F0CCC92 ); + P( C, D, A, B, 10, 15, 0xFFEFF47D ); + P( B, C, D, A, 1, 21, 0x85845DD1 ); + P( A, B, C, D, 8, 6, 0x6FA87E4F ); + P( D, A, B, C, 15, 10, 0xFE2CE6E0 ); + P( C, D, A, B, 6, 15, 0xA3014314 ); + P( B, C, D, A, 13, 21, 0x4E0811A1 ); + P( A, B, C, D, 4, 6, 0xF7537E82 ); + P( D, A, B, C, 11, 10, 0xBD3AF235 ); + P( C, D, A, B, 2, 15, 0x2AD7D2BB ); + P( B, C, D, A, 9, 21, 0xEB86D391 ); + +#undef F + + state0 += A; + state1 += B; + state2 += C; + state3 += D; + } + + /* Got the hash, store it in the output buffer. */ + PUT_UINT32_LE( state0, output, 0 ); +#ifndef FEATURE_REDUCED_HASH_SIZE + PUT_UINT32_LE( state1, output, 4 ); + PUT_UINT32_LE( state2, output, 8 ); + PUT_UINT32_LE( state3, output, 12 ); +#endif + +} + +__device__ +static void md5_internal_overlap( unsigned int *input, unsigned int *sharedMemory, + int chunkSize, unsigned char *output ) { + + /* Number of passes (512 bit blocks) we have to do */ + int numberOfPasses = chunkSize / 64 + 1; + /* Used during the hashing process */ + unsigned long A, B, C, D; + /* Needed to do the little endian stuff */ + unsigned char *data = (unsigned char *)sharedMemory; + // number of padding bytes. + int numPadBytes = 0; + int numPadInt = 0; + //int numPadRemain = 0; + + /* Will hold the hash value through the + intermediate stages of MD5 algorithm */ + unsigned int state0 = 0x67452301; + unsigned int state1 = 0xEFCDAB89; + unsigned int state2 = 0x98BADCFE; + unsigned int state3 = 0x10325476; + + + /* Used to cache the shared memory index calculations, but testing showed + that it has no performance effect. */ + int x0 = SHARED_MEMORY_INDEX(0); + int x1 = SHARED_MEMORY_INDEX(1); + int x2 = SHARED_MEMORY_INDEX(2); + int x3 = SHARED_MEMORY_INDEX(3); + int x4 = SHARED_MEMORY_INDEX(4); + int x5 = SHARED_MEMORY_INDEX(5); + int x6 = SHARED_MEMORY_INDEX(6); + int x7 = SHARED_MEMORY_INDEX(7); + int x8 = SHARED_MEMORY_INDEX(8); + int x9 = SHARED_MEMORY_INDEX(9); + int x10 = SHARED_MEMORY_INDEX(10); + int x11 = SHARED_MEMORY_INDEX(11); + int x12 = SHARED_MEMORY_INDEX(12); + int x13 = SHARED_MEMORY_INDEX(13); + int x14 = SHARED_MEMORY_INDEX(14); + int x15 = SHARED_MEMORY_INDEX(15); + +#undef GET_CACHED_INDEX +#define GET_CACHED_INDEX(index) (x##index) + + + for( int index = 0 ; index < (numberOfPasses) ; index++ ) { + + if(index == numberOfPasses - 1 ) { + + numPadBytes = (64-12) - (chunkSize - (numberOfPasses-1)*64); + numPadInt = numPadBytes/sizeof(int); + /*numPadRemain = numPadBytes-numPadInt*sizeof(int); + printf("\nLast loop chunkSize = %d, numberOfPasses= %d and \nnumPadBytes = %d, numPadInt =%d, numPadRemain = %d\n", + chunkSize,numberOfPasses,numPadBytes,numPadInt,numPadRemain);*/ + + int i=0; + for(i = 0 ; i < numPadInt ; i++){ + sharedMemory[SHARED_MEMORY_INDEX(13-i)] = 0; + } + + int j=0; + for(j=0;j<(16-3-numPadInt);j++){ + //printf("j= %d\n",j); + sharedMemory[SHARED_MEMORY_INDEX(j)] = input[j + 16 * index]; + } + + + /* The last pass will contain the size of the chunk size (according to + official MD5 algorithm). */ + sharedMemory[SHARED_MEMORY_INDEX(13-i)] = 0x00000080; + //printf("the last one at %d\n",13-i); + + sharedMemory[GET_CACHED_INDEX(14)] = chunkSize << 3; + sharedMemory[GET_CACHED_INDEX(15)] = chunkSize >> 29; + } else { + /* Move data to the thread's shared memory space */ + //printf("Not last loop\n"); + sharedMemory[GET_CACHED_INDEX(0)] = input[0 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(1)] = input[1 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(2)] = input[2 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(3)] = input[3 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(4)] = input[4 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(5)] = input[5 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(6)] = input[6 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(7)] = input[7 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(8)] = input[8 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(9)] = input[9 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(10)] = input[10 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(11)] = input[11 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(12)] = input[12 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(13)] = input[13 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(14)] = input[14 + 16 * index]; + sharedMemory[GET_CACHED_INDEX(15)] = input[15 + 16 * index]; + } + + /* Get the little endian stuff done. */ + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(0)], + data, GET_CACHED_INDEX(0) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(1)], + data, GET_CACHED_INDEX(1) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(2)], + data, GET_CACHED_INDEX(2) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(3)], + data, GET_CACHED_INDEX(3) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(4)], + data, GET_CACHED_INDEX(4) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(5)], + data, GET_CACHED_INDEX(5) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(6)], + data, GET_CACHED_INDEX(6) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(7)], + data, GET_CACHED_INDEX(7) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(8)], + data, GET_CACHED_INDEX(8) * 4 ); + GET_UINT32_LE( sharedMemory[ GET_CACHED_INDEX(9)], + data, GET_CACHED_INDEX(9) * 4 ); + GET_UINT32_LE( sharedMemory[GET_CACHED_INDEX(10)], + data, GET_CACHED_INDEX(10) * 4 ); + GET_UINT32_LE( sharedMemory[GET_CACHED_INDEX(11)], + data, GET_CACHED_INDEX(11) * 4 ); + GET_UINT32_LE( sharedMemory[GET_CACHED_INDEX(12)], + data, GET_CACHED_INDEX(12) * 4 ); + GET_UINT32_LE( sharedMemory[GET_CACHED_INDEX(13)], + data, GET_CACHED_INDEX(13) * 4 ); + GET_UINT32_LE( sharedMemory[GET_CACHED_INDEX(14)], + data, GET_CACHED_INDEX(14) * 4 ); + GET_UINT32_LE( sharedMemory[GET_CACHED_INDEX(15)], + data, GET_CACHED_INDEX(15) * 4 ); + + + /* Start the MD5 permutations */ +#undef S +#define S(x,n) ((x << n) | ((x & 0xFFFFFFFF) >> (32 - n))) +#undef P +#define P(a,b,c,d,k,s,t) { \ + a += F(b,c,d) + sharedMemory[GET_CACHED_INDEX(k)] + t; a = S(a,s) + b; \ + } \ + + A = state0; + B = state1; + C = state2; + D = state3; + +#undef F + +#define F(x,y,z) (z ^ (x & (y ^ z))) + + P( A, B, C, D, 0, 7, 0xD76AA478 ); + P( D, A, B, C, 1, 12, 0xE8C7B756 ); + P( C, D, A, B, 2, 17, 0x242070DB ); + P( B, C, D, A, 3, 22, 0xC1BDCEEE ); + P( A, B, C, D, 4, 7, 0xF57C0FAF ); + P( D, A, B, C, 5, 12, 0x4787C62A ); + P( C, D, A, B, 6, 17, 0xA8304613 ); + P( B, C, D, A, 7, 22, 0xFD469501 ); + P( A, B, C, D, 8, 7, 0x698098D8 ); + P( D, A, B, C, 9, 12, 0x8B44F7AF ); + P( C, D, A, B, 10, 17, 0xFFFF5BB1 ); + P( B, C, D, A, 11, 22, 0x895CD7BE ); + P( A, B, C, D, 12, 7, 0x6B901122 ); + P( D, A, B, C, 13, 12, 0xFD987193 ); + P( C, D, A, B, 14, 17, 0xA679438E ); + P( B, C, D, A, 15, 22, 0x49B40821 ); + +#undef F + +#define F(x,y,z) (y ^ (z & (x ^ y))) + + P( A, B, C, D, 1, 5, 0xF61E2562 ); + P( D, A, B, C, 6, 9, 0xC040B340 ); + P( C, D, A, B, 11, 14, 0x265E5A51 ); + P( B, C, D, A, 0, 20, 0xE9B6C7AA ); + P( A, B, C, D, 5, 5, 0xD62F105D ); + P( D, A, B, C, 10, 9, 0x02441453 ); + P( C, D, A, B, 15, 14, 0xD8A1E681 ); + P( B, C, D, A, 4, 20, 0xE7D3FBC8 ); + P( A, B, C, D, 9, 5, 0x21E1CDE6 ); + P( D, A, B, C, 14, 9, 0xC33707D6 ); + P( C, D, A, B, 3, 14, 0xF4D50D87 ); + P( B, C, D, A, 8, 20, 0x455A14ED ); + P( A, B, C, D, 13, 5, 0xA9E3E905 ); + P( D, A, B, C, 2, 9, 0xFCEFA3F8 ); + P( C, D, A, B, 7, 14, 0x676F02D9 ); + P( B, C, D, A, 12, 20, 0x8D2A4C8A ); + +#undef F + +#define F(x,y,z) (x ^ y ^ z) + + P( A, B, C, D, 5, 4, 0xFFFA3942 ); + P( D, A, B, C, 8, 11, 0x8771F681 ); + P( C, D, A, B, 11, 16, 0x6D9D6122 ); + P( B, C, D, A, 14, 23, 0xFDE5380C ); + P( A, B, C, D, 1, 4, 0xA4BEEA44 ); + P( D, A, B, C, 4, 11, 0x4BDECFA9 ); + P( C, D, A, B, 7, 16, 0xF6BB4B60 ); + P( B, C, D, A, 10, 23, 0xBEBFBC70 ); + P( A, B, C, D, 13, 4, 0x289B7EC6 ); + P( D, A, B, C, 0, 11, 0xEAA127FA ); + P( C, D, A, B, 3, 16, 0xD4EF3085 ); + P( B, C, D, A, 6, 23, 0x04881D05 ); + P( A, B, C, D, 9, 4, 0xD9D4D039 ); + P( D, A, B, C, 12, 11, 0xE6DB99E5 ); + P( C, D, A, B, 15, 16, 0x1FA27CF8 ); + P( B, C, D, A, 2, 23, 0xC4AC5665 ); + +#undef F + +#define F(x,y,z) (y ^ (x | ~z)) + + P( A, B, C, D, 0, 6, 0xF4292244 ); + P( D, A, B, C, 7, 10, 0x432AFF97 ); + P( C, D, A, B, 14, 15, 0xAB9423A7 ); + P( B, C, D, A, 5, 21, 0xFC93A039 ); + P( A, B, C, D, 12, 6, 0x655B59C3 ); + P( D, A, B, C, 3, 10, 0x8F0CCC92 ); + P( C, D, A, B, 10, 15, 0xFFEFF47D ); + P( B, C, D, A, 1, 21, 0x85845DD1 ); + P( A, B, C, D, 8, 6, 0x6FA87E4F ); + P( D, A, B, C, 15, 10, 0xFE2CE6E0 ); + P( C, D, A, B, 6, 15, 0xA3014314 ); + P( B, C, D, A, 13, 21, 0x4E0811A1 ); + P( A, B, C, D, 4, 6, 0xF7537E82 ); + P( D, A, B, C, 11, 10, 0xBD3AF235 ); + P( C, D, A, B, 2, 15, 0x2AD7D2BB ); + P( B, C, D, A, 9, 21, 0xEB86D391 ); + +#undef F + + state0 += A; + state1 += B; + state2 += C; + state3 += D; + } + + /* Got the hash, store it in the output buffer. */ + PUT_UINT32_LE( state0, output, 0 ); +#ifndef FEATURE_REDUCED_HASH_SIZE + PUT_UINT32_LE( state1, output, 4 ); + PUT_UINT32_LE( state2, output, 8 ); + PUT_UINT32_LE( state3, output, 12 ); +#endif + +} +#endif + +/*-------------------------------------------------------------------------- + GLOBAL FUNCTIONS +--------------------------------------------------------------------------*/ + +/*=========================================================================== + +FUNCTION MD5 + +DESCRIPTION + Main md5 hash function + +DEPENDENCIES + GPU must be initialized + +RETURN VALUE + output: the hash result + +===========================================================================*/ +__global__ +void md5( unsigned char *input, int chunkSize, int totalThreads, + int padSize, unsigned char *scratch) { + + int threadIndex = threadIdx.x + blockDim.x * blockIdx.x; + int chunkIndex = threadIndex * chunkSize; + int hashIndex = threadIndex * MD5_HASH_SIZE; + + if(threadIndex >= totalThreads) + return; + + if ((threadIndex == (totalThreads - 1)) && (padSize > 0)) { + for(int i = 0 ; i < padSize ; i++) + input[chunkIndex + chunkSize - padSize + i] = 0; + } + + +#ifdef FEATURE_SHARED_MEMORY + + __shared__ unsigned int sharedMemory[4 * 1024 - 32]; + + // 512 words are allocated for every warp of 32 threads + unsigned int *sharedMemoryIndex = sharedMemory + ((threadIdx.x >> 5) * 512); + unsigned int *inputIndex = (unsigned int *)(input + chunkIndex); + + md5_internal(inputIndex, sharedMemoryIndex, chunkSize, + scratch + hashIndex ); + +#else + md5_internal(input + chunkIndex, chunkSize, scratch + hashIndex ); +#endif /* FEATURE_SHARED_MEMORY */ + +} + + +__global__ +void md5_overlap( unsigned char *input, int chunkSize, int offset, + int totalThreads, int padSize, unsigned char *output ) { + + int threadIndex = threadIdx.x + blockDim.x * blockIdx.x; + int chunkIndex = threadIndex * offset; + int hashIndex = threadIndex * MD5_HASH_SIZE; + + + if(threadIndex >= totalThreads) + return; + + if ((threadIndex == (totalThreads - 1))) { + chunkSize-= padSize; + } + + +#ifdef FEATURE_SHARED_MEMORY + + __shared__ unsigned int sharedMemory[4 * 1024 - 32]; + + unsigned int *sharedMemoryIndex = sharedMemory + ((threadIdx.x >> 5) * 512); + unsigned int *inputIndex = (unsigned int *)(input + chunkIndex); + + md5_internal_overlap(inputIndex, sharedMemoryIndex, chunkSize, + output + hashIndex ); + +#else + md5_internal(input + chunkIndex, chunkSize, output + hashIndex ); +#endif /* FEATURE_SHARED_MEMORY */ + + +} + -- cgit v1.3