ccminer/quark/cuda_bmw512.cu

#if 1

#include <cuda.h>
#include "cuda_runtime.h"
#include "device_launch_parameters.h"

#include <stdio.h>
#include <memory.h>

// Folgende Definitionen sp<EFBFBD>ter durch header ersetzen
typedef unsigned char uint8_t;
typedef unsigned int uint32_t;

// Endian Drehung f<EFBFBD>r 32 Bit Typen
/*
static __device__ uint32_t cuda_swab32(uint32_t x)
{
    return (((x << 24) & 0xff000000u) | ((x << 8) & 0x00ff0000u)
          | ((x >> 8) & 0x0000ff00u) | ((x >> 24) & 0x000000ffu));
}
*/
static __device__ uint32_t cuda_swab32(uint32_t x)
{
	return __byte_perm(x, 0, 0x0123);
}
// Endian Drehung f<EFBFBD>r 64 Bit Typen
static __device__ unsigned long long cuda_swab64(unsigned long long x) {
    uint32_t h = (x >> 32);
    uint32_t l = (x & 0xFFFFFFFFULL);
    return (((unsigned long long)cuda_swab32(l)) << 32) | ((unsigned long long)cuda_swab32(h));
}

// das Hi Word aus einem 64 Bit Typen extrahieren
static __device__ uint32_t HIWORD(const unsigned long long &x) {
#if __CUDA_ARCH__ >= 130
	return (uint32_t)__double2hiint(__longlong_as_double(x));
#else
	return (uint32_t)(x >> 32);
#endif
}

// das Hi Word in einem 64 Bit Typen ersetzen
static __device__ unsigned long long REPLACE_HIWORD(const unsigned long long &x, const uint32_t &y) {
	return (x & 0xFFFFFFFFULL) | (((unsigned long long)y) << 32ULL);
}

// das Lo Word aus einem 64 Bit Typen extrahieren
static __device__ uint32_t LOWORD(const unsigned long long &x) {
#if __CUDA_ARCH__ >= 130
	return (uint32_t)__double2loint(__longlong_as_double(x));
#else
	return (uint32_t)(x & 0xFFFFFFFFULL);
#endif
}

static __device__ unsigned long long MAKE_ULONGLONG(uint32_t LO, uint32_t HI)
{
#if __CUDA_ARCH__ >= 130
    return __double_as_longlong(__hiloint2double(HI, LO));
#else
	return (unsigned long long)LO | (((unsigned long long)HI) << 32ULL);
#endif
}

// das Lo Word in einem 64 Bit Typen ersetzen
static __device__ unsigned long long REPLACE_LOWORD(const unsigned long long &x, const uint32_t &y) {
	return (x & 0xFFFFFFFF00000000ULL) | ((unsigned long long)y);
}

// der Versuch, einen Wrapper f<EFBFBD>r einen aus 32 Bit Registern zusammengesetzten uin64_t Typen zu entferfen...
#if 1
typedef unsigned long long uint64_t;
#else
typedef class uint64
{
public:
	__device__ uint64()
	{
	}
	__device__ uint64(unsigned long long init)
	{
		val = make_uint2( LOWORD(init), HIWORD(init) );
	}
	__device__ uint64(uint32_t lo, uint32_t hi)
	{
		val = make_uint2( lo, hi );
	}
	__device__ const uint64 operator^(uint64 const& rhs) const
	{
		return uint64(val.x ^ rhs.val.x, val.y ^ rhs.val.y);
	}
	__device__ const uint64 operator|(uint64 const& rhs) const
	{
		return uint64(val.x | rhs.val.x, val.y | rhs.val.y);
	}
	__device__ const uint64 operator+(unsigned long long const& rhs) const
	{
		return *this+uint64(rhs);
	}
	__device__ const uint64 operator+(uint64 const& rhs) const
	{
		uint64 res;
		asm ("add.cc.u32      %0, %2, %4;\n\t"
			 "addc.cc.u32     %1, %3, %5;\n\t"
			 : "=r"(res.val.x), "=r"(res.val.y)
			 : "r"(    val.x), "r"(    val.y),
			   "r"(rhs.val.x), "r"(rhs.val.y));
		return res;
	}
	__device__ const uint64 operator-(uint64 const& rhs) const
	{
		uint64 res;
		asm ("sub.cc.u32      %0, %2, %4;\n\t"
			 "subc.cc.u32     %1, %3, %5;\n\t"
			 : "=r"(res.val.x), "=r"(res.val.y)
			 : "r"(    val.x), "r"(    val.y),
			   "r"(rhs.val.x), "r"(rhs.val.y));
		return res;
	}
	__device__ const uint64 operator<<(int n) const
	{
		return uint64(unsigned long long(*this)<<n);
	}
	__device__ const uint64 operator>>(int n) const
	{
		return uint64(unsigned long long(*this)>>n);
	}
	__device__ operator unsigned long long() const
	{
		return MAKE_ULONGLONG(val.x, val.y);
	}
	uint2 val;
} uint64_t;
#endif

// aus heavy.cu
extern cudaError_t MyStreamSynchronize(cudaStream_t stream, int situation, int thr_id);

// die Message it Padding zur Berechnung auf der GPU
__constant__ uint64_t c_PaddedMessage80[16]; // padded message (80 bytes + padding)

#define SPH_C64(x)    ((uint64_t)(x ## ULL))

// aus heavy.cu
extern cudaError_t MyStreamSynchronize(cudaStream_t stream, int situation, int thr_id);

// diese 64 Bit Rotates werden unter Compute 3.5 (und besser) mit dem Funnel Shifter beschleunigt
#if __CUDA_ARCH__ >= 350
__forceinline__ __device__ uint64_t ROTL64(const uint64_t value, const int offset) {
    uint2 result;
    if(offset >= 32) {
        asm("shf.l.wrap.b32 %0, %1, %2, %3;" : "=r"(result.x) : "r"(__double2loint(__longlong_as_double(value))), "r"(__double2hiint(__longlong_as_double(value))), "r"(offset));
        asm("shf.l.wrap.b32 %0, %1, %2, %3;" : "=r"(result.y) : "r"(__double2hiint(__longlong_as_double(value))), "r"(__double2loint(__longlong_as_double(value))), "r"(offset));
    } else {
        asm("shf.l.wrap.b32 %0, %1, %2, %3;" : "=r"(result.x) : "r"(__double2hiint(__longlong_as_double(value))), "r"(__double2loint(__longlong_as_double(value))), "r"(offset));
        asm("shf.l.wrap.b32 %0, %1, %2, %3;" : "=r"(result.y) : "r"(__double2loint(__longlong_as_double(value))), "r"(__double2hiint(__longlong_as_double(value))), "r"(offset));
    }
    return  __double_as_longlong(__hiloint2double(result.y, result.x));
}
#else
#define ROTL64(x, n)        (((x) << (n)) | ((x) >> (64 - (n))))
#endif
#define SHL(x, n)            ((x) << (n))
#define SHR(x, n)            ((x) >> (n))

#define CONST_EXP2    q[i+0] + ROTL64(q[i+1], 5)  + q[i+2] + ROTL64(q[i+3], 11) + \
                    q[i+4] + ROTL64(q[i+5], 27) + q[i+6] + ROTL64(q[i+7], 32) + \
                    q[i+8] + ROTL64(q[i+9], 37) + q[i+10] + ROTL64(q[i+11], 43) + \
                    q[i+12] + ROTL64(q[i+13], 53) + (SHR(q[i+14],1) ^ q[i+14]) + (SHR(q[i+15],2) ^ q[i+15])

__device__ void Compression512(uint64_t *msg, uint64_t *hash)
{
    // Compression ref. implementation
    uint64_t tmp;
    uint64_t q[32];

    tmp = (msg[ 5] ^ hash[ 5]) - (msg[ 7] ^ hash[ 7]) + (msg[10] ^ hash[10]) + (msg[13] ^ hash[13]) + (msg[14] ^ hash[14]);
    q[0] = (SHR(tmp, 1) ^ SHL(tmp, 3) ^ ROTL64(tmp,  4) ^ ROTL64(tmp, 37)) + hash[1];
    tmp = (msg[ 6] ^ hash[ 6]) - (msg[ 8] ^ hash[ 8]) + (msg[11] ^ hash[11]) + (msg[14] ^ hash[14]) - (msg[15] ^ hash[15]);
    q[1] = (SHR(tmp, 1) ^ SHL(tmp, 2) ^ ROTL64(tmp, 13) ^ ROTL64(tmp, 43)) + hash[2];
    tmp = (msg[ 0] ^ hash[ 0]) + (msg[ 7] ^ hash[ 7]) + (msg[ 9] ^ hash[ 9]) - (msg[12] ^ hash[12]) + (msg[15] ^ hash[15]);
    q[2] = (SHR(tmp, 2) ^ SHL(tmp, 1) ^ ROTL64(tmp, 19) ^ ROTL64(tmp, 53)) + hash[3];
    tmp = (msg[ 0] ^ hash[ 0]) - (msg[ 1] ^ hash[ 1]) + (msg[ 8] ^ hash[ 8]) - (msg[10] ^ hash[10]) + (msg[13] ^ hash[13]);
    q[3] = (SHR(tmp, 2) ^ SHL(tmp, 2) ^ ROTL64(tmp, 28) ^ ROTL64(tmp, 59)) + hash[4];
    tmp = (msg[ 1] ^ hash[ 1]) + (msg[ 2] ^ hash[ 2]) + (msg[ 9] ^ hash[ 9]) - (msg[11] ^ hash[11]) - (msg[14] ^ hash[14]);
    q[4] = (SHR(tmp, 1) ^ tmp) + hash[5];
    tmp = (msg[ 3] ^ hash[ 3]) - (msg[ 2] ^ hash[ 2]) + (msg[10] ^ hash[10]) - (msg[12] ^ hash[12]) + (msg[15] ^ hash[15]);
    q[5] = (SHR(tmp, 1) ^ SHL(tmp, 3) ^ ROTL64(tmp,  4) ^ ROTL64(tmp, 37)) + hash[6];
    tmp = (msg[ 4] ^ hash[ 4]) - (msg[ 0] ^ hash[ 0]) - (msg[ 3] ^ hash[ 3]) - (msg[11] ^ hash[11]) + (msg[13] ^ hash[13]);
    q[6] = (SHR(tmp, 1) ^ SHL(tmp, 2) ^ ROTL64(tmp, 13) ^ ROTL64(tmp, 43)) + hash[7];
    tmp = (msg[ 1] ^ hash[ 1]) - (msg[ 4] ^ hash[ 4]) - (msg[ 5] ^ hash[ 5]) - (msg[12] ^ hash[12]) - (msg[14] ^ hash[14]);
    q[7] = (SHR(tmp, 2) ^ SHL(tmp, 1) ^ ROTL64(tmp, 19) ^ ROTL64(tmp, 53)) + hash[8];
    tmp = (msg[ 2] ^ hash[ 2]) - (msg[ 5] ^ hash[ 5]) - (msg[ 6] ^ hash[ 6]) + (msg[13] ^ hash[13]) - (msg[15] ^ hash[15]);
    q[8] = (SHR(tmp, 2) ^ SHL(tmp, 2) ^ ROTL64(tmp, 28) ^ ROTL64(tmp, 59)) + hash[9];
    tmp = (msg[ 0] ^ hash[ 0]) - (msg[ 3] ^ hash[ 3]) + (msg[ 6] ^ hash[ 6]) - (msg[ 7] ^ hash[ 7]) + (msg[14] ^ hash[14]);
    q[9] = (SHR(tmp, 1) ^ tmp) + hash[10];
    tmp = (msg[ 8] ^ hash[ 8]) - (msg[ 1] ^ hash[ 1]) - (msg[ 4] ^ hash[ 4]) - (msg[ 7] ^ hash[ 7]) + (msg[15] ^ hash[15]);
    q[10] = (SHR(tmp, 1) ^ SHL(tmp, 3) ^ ROTL64(tmp,  4) ^ ROTL64(tmp, 37)) + hash[11];
    tmp = (msg[ 8] ^ hash[ 8]) - (msg[ 0] ^ hash[ 0]) - (msg[ 2] ^ hash[ 2]) - (msg[ 5] ^ hash[ 5]) + (msg[ 9] ^ hash[ 9]);
    q[11] = (SHR(tmp, 1) ^ SHL(tmp, 2) ^ ROTL64(tmp, 13) ^ ROTL64(tmp, 43)) + hash[12];
    tmp = (msg[ 1] ^ hash[ 1]) + (msg[ 3] ^ hash[ 3]) - (msg[ 6] ^ hash[ 6]) - (msg[ 9] ^ hash[ 9]) + (msg[10] ^ hash[10]);
    q[12] = (SHR(tmp, 2) ^ SHL(tmp, 1) ^ ROTL64(tmp, 19) ^ ROTL64(tmp, 53)) + hash[13];
    tmp = (msg[ 2] ^ hash[ 2]) + (msg[ 4] ^ hash[ 4]) + (msg[ 7] ^ hash[ 7]) + (msg[10] ^ hash[10]) + (msg[11] ^ hash[11]);
    q[13] = (SHR(tmp, 2) ^ SHL(tmp, 2) ^ ROTL64(tmp, 28) ^ ROTL64(tmp, 59)) + hash[14];
    tmp = (msg[ 3] ^ hash[ 3]) - (msg[ 5] ^ hash[ 5]) + (msg[ 8] ^ hash[ 8]) - (msg[11] ^ hash[11]) - (msg[12] ^ hash[12]);
    q[14] = (SHR(tmp, 1) ^ tmp) + hash[15];
    tmp = (msg[12] ^ hash[12]) - (msg[ 4] ^ hash[ 4]) - (msg[ 6] ^ hash[ 6]) - (msg[ 9] ^ hash[ 9]) + (msg[13] ^ hash[13]);
    q[15] = (SHR(tmp, 1) ^ SHL(tmp, 3) ^ ROTL64(tmp, 4) ^ ROTL64(tmp, 37)) + hash[0];

    // Expand 1
#pragma unroll 2
    for(int i=0;i<2;i++)
    {
        q[i+16] =
        (SHR(q[i], 1) ^ SHL(q[i], 2) ^ ROTL64(q[i], 13) ^ ROTL64(q[i], 43)) +
        (SHR(q[i+1], 2) ^ SHL(q[i+1], 1) ^ ROTL64(q[i+1], 19) ^ ROTL64(q[i+1], 53)) +
        (SHR(q[i+2], 2) ^ SHL(q[i+2], 2) ^ ROTL64(q[i+2], 28) ^ ROTL64(q[i+2], 59)) +
        (SHR(q[i+3], 1) ^ SHL(q[i+3], 3) ^ ROTL64(q[i+3],  4) ^ ROTL64(q[i+3], 37)) +
        (SHR(q[i+4], 1) ^ SHL(q[i+4], 2) ^ ROTL64(q[i+4], 13) ^ ROTL64(q[i+4], 43)) +
        (SHR(q[i+5], 2) ^ SHL(q[i+5], 1) ^ ROTL64(q[i+5], 19) ^ ROTL64(q[i+5], 53)) +
        (SHR(q[i+6], 2) ^ SHL(q[i+6], 2) ^ ROTL64(q[i+6], 28) ^ ROTL64(q[i+6], 59)) +
        (SHR(q[i+7], 1) ^ SHL(q[i+7], 3) ^ ROTL64(q[i+7],  4) ^ ROTL64(q[i+7], 37)) +
        (SHR(q[i+8], 1) ^ SHL(q[i+8], 2) ^ ROTL64(q[i+8], 13) ^ ROTL64(q[i+8], 43)) +
        (SHR(q[i+9], 2) ^ SHL(q[i+9], 1) ^ ROTL64(q[i+9], 19) ^ ROTL64(q[i+9], 53)) +
        (SHR(q[i+10], 2) ^ SHL(q[i+10], 2) ^ ROTL64(q[i+10], 28) ^ ROTL64(q[i+10], 59)) +
        (SHR(q[i+11], 1) ^ SHL(q[i+11], 3) ^ ROTL64(q[i+11],  4) ^ ROTL64(q[i+11], 37)) +
        (SHR(q[i+12], 1) ^ SHL(q[i+12], 2) ^ ROTL64(q[i+12], 13) ^ ROTL64(q[i+12], 43)) +
        (SHR(q[i+13], 2) ^ SHL(q[i+13], 1) ^ ROTL64(q[i+13], 19) ^ ROTL64(q[i+13], 53)) +
        (SHR(q[i+14], 2) ^ SHL(q[i+14], 2) ^ ROTL64(q[i+14], 28) ^ ROTL64(q[i+14], 59)) +
        (SHR(q[i+15], 1) ^ SHL(q[i+15], 3) ^ ROTL64(q[i+15],  4) ^ ROTL64(q[i+15], 37)) +
        ((    ((i+16)*(0x0555555555555555ull)) + ROTL64(msg[i], i+1) +
            ROTL64(msg[i+3], i+4) - ROTL64(msg[i+10], i+11) ) ^ hash[i+7]);
    }

#pragma unroll 4
    for(int i=2;i<6;i++) {
        q[i+16] = CONST_EXP2 + 
        ((    ((i+16)*(0x0555555555555555ull)) + ROTL64(msg[i], i+1) +
            ROTL64(msg[i+3], i+4) - ROTL64(msg[i+10], i+11) ) ^ hash[i+7]);
    }
#pragma unroll 3
    for(int i=6;i<9;i++) {
        q[i+16] = CONST_EXP2 + 
        ((    ((i+16)*(0x0555555555555555ull)) + ROTL64(msg[i], i+1) +
            ROTL64(msg[i+3], i+4) - ROTL64(msg[i-6], (i-6)+1) ) ^ hash[i+7]);
    }
#pragma unroll 4
    for(int i=9;i<13;i++) {
        q[i+16] = CONST_EXP2 + 
        ((    ((i+16)*(0x0555555555555555ull)) + ROTL64(msg[i], i+1) +
            ROTL64(msg[i+3], i+4) - ROTL64(msg[i-6], (i-6)+1) ) ^ hash[i-9]);
    }
#pragma unroll 3
    for(int i=13;i<16;i++) {
        q[i+16] = CONST_EXP2 + 
        ((    ((i+16)*(0x0555555555555555ull)) + ROTL64(msg[i], i+1) +
            ROTL64(msg[i-13], (i-13)+1) - ROTL64(msg[i-6], (i-6)+1) ) ^ hash[i-9]);
    }

    uint64_t XL64 = q[16]^q[17]^q[18]^q[19]^q[20]^q[21]^q[22]^q[23];
    uint64_t XH64 = XL64^q[24]^q[25]^q[26]^q[27]^q[28]^q[29]^q[30]^q[31];

    hash[0] =                       (SHL(XH64, 5) ^ SHR(q[16],5) ^ msg[ 0]) + (    XL64    ^ q[24] ^ q[ 0]);
    hash[1] =                       (SHR(XH64, 7) ^ SHL(q[17],8) ^ msg[ 1]) + (    XL64    ^ q[25] ^ q[ 1]);
    hash[2] =                       (SHR(XH64, 5) ^ SHL(q[18],5) ^ msg[ 2]) + (    XL64    ^ q[26] ^ q[ 2]);
    hash[3] =                       (SHR(XH64, 1) ^ SHL(q[19],5) ^ msg[ 3]) + (    XL64    ^ q[27] ^ q[ 3]);
    hash[4] =                       (SHR(XH64, 3) ^     q[20]    ^ msg[ 4]) + (    XL64    ^ q[28] ^ q[ 4]);
    hash[5] =                       (SHL(XH64, 6) ^ SHR(q[21],6) ^ msg[ 5]) + (    XL64    ^ q[29] ^ q[ 5]);
    hash[6] =                       (SHR(XH64, 4) ^ SHL(q[22],6) ^ msg[ 6]) + (    XL64    ^ q[30] ^ q[ 6]);
    hash[7] =                       (SHR(XH64,11) ^ SHL(q[23],2) ^ msg[ 7]) + (    XL64    ^ q[31] ^ q[ 7]);

    hash[ 8] = ROTL64(hash[4], 9) + (    XH64     ^     q[24]    ^ msg[ 8]) + (SHL(XL64,8) ^ q[23] ^ q[ 8]);
    hash[ 9] = ROTL64(hash[5],10) + (    XH64     ^     q[25]    ^ msg[ 9]) + (SHR(XL64,6) ^ q[16] ^ q[ 9]);
    hash[10] = ROTL64(hash[6],11) + (    XH64     ^     q[26]    ^ msg[10]) + (SHL(XL64,6) ^ q[17] ^ q[10]);
    hash[11] = ROTL64(hash[7],12) + (    XH64     ^     q[27]    ^ msg[11]) + (SHL(XL64,4) ^ q[18] ^ q[11]);
    hash[12] = ROTL64(hash[0],13) + (    XH64     ^     q[28]    ^ msg[12]) + (SHR(XL64,3) ^ q[19] ^ q[12]);
    hash[13] = ROTL64(hash[1],14) + (    XH64     ^     q[29]    ^ msg[13]) + (SHR(XL64,4) ^ q[20] ^ q[13]);
    hash[14] = ROTL64(hash[2],15) + (    XH64     ^     q[30]    ^ msg[14]) + (SHR(XL64,7) ^ q[21] ^ q[14]);
    hash[15] = ROTL64(hash[3],16) + (    XH64     ^     q[31]    ^ msg[15]) + (SHR(XL64,2) ^ q[22] ^ q[15]);
}
static __constant__ uint64_t d_constMem[16];
static uint64_t h_constMem[16] = {
	SPH_C64(0x8081828384858687),
    SPH_C64(0x88898A8B8C8D8E8F),
    SPH_C64(0x9091929394959697),
    SPH_C64(0x98999A9B9C9D9E9F),
    SPH_C64(0xA0A1A2A3A4A5A6A7),
    SPH_C64(0xA8A9AAABACADAEAF),
    SPH_C64(0xB0B1B2B3B4B5B6B7),
    SPH_C64(0xB8B9BABBBCBDBEBF),
    SPH_C64(0xC0C1C2C3C4C5C6C7),
    SPH_C64(0xC8C9CACBCCCDCECF),
    SPH_C64(0xD0D1D2D3D4D5D6D7),
    SPH_C64(0xD8D9DADBDCDDDEDF),
    SPH_C64(0xE0E1E2E3E4E5E6E7),
    SPH_C64(0xE8E9EAEBECEDEEEF),
    SPH_C64(0xF0F1F2F3F4F5F6F7),
    SPH_C64(0xF8F9FAFBFCFDFEFF)
};

__global__ void quark_bmw512_gpu_hash_64(int threads, uint32_t startNounce, uint64_t *g_hash, uint32_t *g_nonceVector)
{
    int thread = (blockDim.x * blockIdx.x + threadIdx.x);
    if (thread < threads)
    {
        uint32_t nounce = (g_nonceVector != NULL) ? g_nonceVector[thread] : (startNounce + thread);

        int hashPosition = nounce - startNounce;
        uint64_t *inpHash = &g_hash[8 * hashPosition];

        // Init
        uint64_t h[16];
		/*
        h[ 0] = SPH_C64(0x8081828384858687);
        h[ 1] = SPH_C64(0x88898A8B8C8D8E8F);
        h[ 2] = SPH_C64(0x9091929394959697);
        h[ 3] = SPH_C64(0x98999A9B9C9D9E9F);
        h[ 4] = SPH_C64(0xA0A1A2A3A4A5A6A7);
        h[ 5] = SPH_C64(0xA8A9AAABACADAEAF);
        h[ 6] = SPH_C64(0xB0B1B2B3B4B5B6B7);
        h[ 7] = SPH_C64(0xB8B9BABBBCBDBEBF);
        h[ 8] = SPH_C64(0xC0C1C2C3C4C5C6C7);
        h[ 9] = SPH_C64(0xC8C9CACBCCCDCECF);
        h[10] = SPH_C64(0xD0D1D2D3D4D5D6D7);
        h[11] = SPH_C64(0xD8D9DADBDCDDDEDF);
        h[12] = SPH_C64(0xE0E1E2E3E4E5E6E7);
        h[13] = SPH_C64(0xE8E9EAEBECEDEEEF);
        h[14] = SPH_C64(0xF0F1F2F3F4F5F6F7);
        h[15] = SPH_C64(0xF8F9FAFBFCFDFEFF);
		*/
#pragma unroll 16
		for(int i=0;i<16;i++)
			h[i] = d_constMem[i];
        // Nachricht kopieren (Achtung, die Nachricht hat 64 Byte,
        // BMW arbeitet mit 128 Byte!!!
        uint64_t message[16];
#pragma unroll 8
        for(int i=0;i<8;i++)
            message[i] = inpHash[i];
#pragma unroll 6
        for(int i=9;i<15;i++)
            message[i] = 0;

        // Padding einf<EFBFBD>gen (Byteorder?!?)
        message[8] = SPH_C64(0x80);
        // L<EFBFBD>nge (in Bits, d.h. 64 Byte * 8 = 512 Bits
        message[15] = SPH_C64(512);

        // Compression 1
        Compression512(message, h);

        // Final
#pragma unroll 16
        for(int i=0;i<16;i++)
            message[i] = 0xaaaaaaaaaaaaaaa0ull + (uint64_t)i;

        Compression512(h, message);

        // fertig
        uint64_t *outpHash = &g_hash[8 * hashPosition];

#pragma unroll 8
        for(int i=0;i<8;i++)
            outpHash[i] = message[i+8];
    }
}

__global__ void quark_bmw512_gpu_hash_80(int threads, uint32_t startNounce, uint64_t *g_hash)
{
    int thread = (blockDim.x * blockIdx.x + threadIdx.x);
    if (thread < threads)
    {
        uint32_t nounce = startNounce + thread;

        // Init
        uint64_t h[16];
#pragma unroll 16
		for(int i=0;i<16;i++)
			h[i] = d_constMem[i];

        // Nachricht kopieren (Achtung, die Nachricht hat 64 Byte,
        // BMW arbeitet mit 128 Byte!!!
        uint64_t message[16];
#pragma unroll 16
        for(int i=0;i<16;i++)
            message[i] = c_PaddedMessage80[i];

        // die Nounce durch die thread-spezifische ersetzen
        message[9] = REPLACE_HIWORD(message[9], cuda_swab32(nounce));

        // Compression 1
        Compression512(message, h);

        // Final
#pragma unroll 16
        for(int i=0;i<16;i++)
            message[i] = 0xaaaaaaaaaaaaaaa0ull + (uint64_t)i;

        Compression512(h, message);

        // fertig
        uint64_t *outpHash = &g_hash[8 * thread];

#pragma unroll 8
        for(int i=0;i<8;i++)
            outpHash[i] = message[i+8];
    }
}

// Setup-Funktionen
__host__ void quark_bmw512_cpu_init(int thr_id, int threads)
{
    // nix zu tun ;-)
	// jetzt schon :D
	cudaMemcpyToSymbol( d_constMem,
                        h_constMem,
                        sizeof(h_constMem),
                        0, cudaMemcpyHostToDevice);
}

// Bmw512 f<EFBFBD>r 80 Byte grosse Eingangsdaten
__host__ void quark_bmw512_cpu_setBlock_80(void *pdata)
{
	// Message mit Padding bereitstellen
	// lediglich die korrekte Nonce ist noch ab Byte 76 einzusetzen.
	unsigned char PaddedMessage[128];
	memcpy(PaddedMessage, pdata, 80);
	memset(PaddedMessage+80, 0, 48);
	uint64_t *message = (uint64_t*)PaddedMessage;
	// Padding einf<EFBFBD>gen (Byteorder?!?)
	message[10] = SPH_C64(0x80);
	// L<EFBFBD>nge (in Bits, d.h. 80 Byte * 8 = 640 Bits
	message[15] = SPH_C64(640);

	// die Message zur Berechnung auf der GPU
	cudaMemcpyToSymbol( c_PaddedMessage80, PaddedMessage, 16*sizeof(uint64_t), 0, cudaMemcpyHostToDevice);
}

__host__ void quark_bmw512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order)
{
    const int threadsperblock = 256;

    // berechne wie viele Thread Blocks wir brauchen
    dim3 grid((threads + threadsperblock-1)/threadsperblock);
    dim3 block(threadsperblock);

    // Gr<EFBFBD><EFBFBD>e des dynamischen Shared Memory Bereichs
    size_t shared_size = 0;

//    fprintf(stderr, "threads=%d, %d blocks, %d threads per block, %d bytes shared\n", threads, grid.x, block.x, shared_size);

    quark_bmw512_gpu_hash_64<<<grid, block, shared_size>>>(threads, startNounce, (uint64_t*)d_hash, d_nonceVector);
    MyStreamSynchronize(NULL, order, thr_id);
}

__host__ void quark_bmw512_cpu_hash_80(int thr_id, int threads, uint32_t startNounce, uint32_t *d_hash, int order)
{
    const int threadsperblock = 256;

    // berechne wie viele Thread Blocks wir brauchen
    dim3 grid((threads + threadsperblock-1)/threadsperblock);
    dim3 block(threadsperblock);

    // Gr<EFBFBD><EFBFBD>e des dynamischen Shared Memory Bereichs
    size_t shared_size = 0;

//    fprintf(stderr, "threads=%d, %d blocks, %d threads per block, %d bytes shared\n", threads, grid.x, block.x, shared_size);

    quark_bmw512_gpu_hash_80<<<grid, block, shared_size>>>(threads, startNounce, (uint64_t*)d_hash);
    MyStreamSynchronize(NULL, order, thr_id);
}

#endif