mirror of
https://github.com/GOSTSec/ccminer
synced 2025-08-26 13:51:51 +00:00
simd: then reindent the code
no changes, only error checks (cuda safe call)
This commit is contained in:
Tanguy Pruvot
parent
b465fe6825
commit
93f4409dde
@ -1,13 +1,14 @@
|
||||
// Parallelisierung:
|
||||
// Parallelization:
|
||||
//
|
||||
// FFT_8 wird 2 mal 8-fach parallel ausgeführt (in FFT_64)
|
||||
// und 1 mal 16-fach parallel (in FFT_128_full)
|
||||
// FFT_8 wird 2 times 8-fach parallel ausgeführt (in FFT_64)
|
||||
// and 1 time 16-fach parallel (in FFT_128_full)
|
||||
//
|
||||
// STEP8_IF und STEP8_MAJ beinhalten je zwei 8-fach parallele Operationen
|
||||
// STEP8_IF and STEP8_MAJ beinhalten je 2x 8-fach parallel Operations
|
||||
|
||||
#define TPB 64
|
||||
|
||||
#include "cuda_helper.h"
|
||||
#include <stdio.h>
|
||||
|
||||
// aus heavy.cu
|
||||
extern cudaError_t MyStreamSynchronize(cudaStream_t stream, int situation, int thr_id);
|
||||
@ -20,14 +21,14 @@ texture<uint4, 1, cudaReadModeElementType> texRef1D_128;
|
||||
|
||||
__constant__ uint32_t c_perm[8][8];
|
||||
const uint32_t h_perm[8][8] = {
|
||||
{ 2,3,6,7,0,1,4,5 },
|
||||
{ 6,7,2,3,4,5,0,1 },
|
||||
{ 7,6,5,4,3,2,1,0 },
|
||||
{ 1,0,3,2,5,4,7,6 },
|
||||
{ 0,1,4,5,6,7,2,3 },
|
||||
{ 6,7,2,3,0,1,4,5 },
|
||||
{ 6,7,0,1,4,5,2,3 },
|
||||
{ 4,5,2,3,6,7,0,1 }
|
||||
{ 2, 3, 6, 7, 0, 1, 4, 5 },
|
||||
{ 6, 7, 2, 3, 4, 5, 0, 1 },
|
||||
{ 7, 6, 5, 4, 3, 2, 1, 0 },
|
||||
{ 1, 0, 3, 2, 5, 4, 7, 6 },
|
||||
{ 0, 1, 4, 5, 6, 7, 2, 3 },
|
||||
{ 6, 7, 2, 3, 0, 1, 4, 5 },
|
||||
{ 6, 7, 0, 1, 4, 5, 2, 3 },
|
||||
{ 4, 5, 2, 3, 6, 7, 0, 1 }
|
||||
};
|
||||
|
||||
__constant__ uint32_t c_IV_512[32];
|
||||
@ -43,30 +44,30 @@ static const int h_FFT128_8_16_Twiddle[128] = {
|
||||
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
|
||||
1, 60, 2, 120, 4, -17, 8, -34, 16, -68, 32, 121, 64, -15, 128, -30,
|
||||
1, 46, 60, -67, 2, 92, 120, 123, 4, -73, -17, -11, 8, 111, -34, -22,
|
||||
1, -67, 120, -73, 8, -22, -68, -70, 64, 81, -30, -46, -2, -123, 17, -111,
|
||||
1, -118, 46, -31, 60, 116, -67, -61, 2, 21, 92, -62, 120, -25, 123, -122,
|
||||
1, 116, 92, -122, -17, 84, -22, 18, 32, 114, 117, -49, -30, 118, 67, 62,
|
||||
1, -67, 120, -73, 8, -22, -68, -70, 64, 81, -30, -46, -2,-123, 17,-111,
|
||||
1,-118, 46, -31, 60, 116, -67, -61, 2, 21, 92, -62, 120, -25, 123,-122,
|
||||
1, 116, 92,-122, -17, 84, -22, 18, 32, 114, 117, -49, -30, 118, 67, 62,
|
||||
1, -31, -67, 21, 120, -122, -73, -50, 8, 9, -22, -89, -68, 52, -70, 114,
|
||||
1, -61, 123, -50, -34, 18, -70, -99, 128, -98, 67, 25, 17, -9, 35, -79
|
||||
};
|
||||
|
||||
__constant__ int c_FFT256_2_128_Twiddle[128];
|
||||
static const int h_FFT256_2_128_Twiddle[128] = {
|
||||
1, 41, -118, 45, 46, 87, -31, 14,
|
||||
60, -110, 116, -127, -67, 80, -61, 69,
|
||||
1, 41,-118, 45, 46, 87, -31, 14,
|
||||
60,-110, 116,-127, -67, 80, -61, 69,
|
||||
2, 82, 21, 90, 92, -83, -62, 28,
|
||||
120, 37, -25, 3, 123, -97, -122, -119,
|
||||
4, -93, 42, -77, -73, 91, -124, 56,
|
||||
120, 37, -25, 3, 123, -97,-122,-119,
|
||||
4, -93, 42, -77, -73, 91,-124, 56,
|
||||
-17, 74, -50, 6, -11, 63, 13, 19,
|
||||
8, 71, 84, 103, 111, -75, 9, 112,
|
||||
-34, -109, -100, 12, -22, 126, 26, 38,
|
||||
16, -115, -89, -51, -35, 107, 18, -33,
|
||||
-34,-109,-100, 12, -22, 126, 26, 38,
|
||||
16,-115, -89, -51, -35, 107, 18, -33,
|
||||
-68, 39, 57, 24, -44, -5, 52, 76,
|
||||
32, 27, 79, -102, -70, -43, 36, -66,
|
||||
121, 78, 114, 48, -88, -10, 104, -105,
|
||||
32, 27, 79,-102, -70, -43, 36, -66,
|
||||
121, 78, 114, 48, -88, -10, 104,-105,
|
||||
64, 54, -99, 53, 117, -86, 72, 125,
|
||||
-15, -101, -29, 96, 81, -20, -49, 47,
|
||||
128, 108, 59, 106, -23, 85, -113, -7,
|
||||
-15,-101, -29, 96, 81, -20, -49, 47,
|
||||
128, 108, 59, 106, -23, 85,-113, -7,
|
||||
-30, 55, -58, -65, -95, -40, -98, 94
|
||||
};
|
||||
|
||||
@ -76,6 +77,7 @@ static const int h_FFT256_2_128_Twiddle[128] = {
|
||||
#define IF(x, y, z) ((((y) ^ (z)) & (x)) ^ (z))
|
||||
|
||||
#define MAJ(x, y, z) (((z) & (y)) | (((z) | (y)) & (x)))
|
||||
|
||||
#include "x11/simd_functions.cu"
|
||||
|
||||
/********************* Message expansion ************************/
|
||||
@ -84,24 +86,26 @@ static const int h_FFT256_2_128_Twiddle[128] = {
|
||||
* Reduce modulo 257; result is in [-127; 383]
|
||||
* REDUCE(x) := (x&255) - (x>>8)
|
||||
*/
|
||||
#define REDUCE(x) (((x)&255) - ((x)>>8))
|
||||
#define REDUCE(x) \
|
||||
(((x)&255) - ((x)>>8))
|
||||
|
||||
/*
|
||||
* Reduce from [-127; 383] to [-128; 128]
|
||||
* EXTRA_REDUCE_S(x) := x<=128 ? x : x-257
|
||||
*/
|
||||
#define EXTRA_REDUCE_S(x) \
|
||||
((x)<=128 ? (x) : (x)-257)
|
||||
((x)<=128 ? (x) : (x)-257)
|
||||
|
||||
/*
|
||||
* Reduce modulo 257; result is in [-128; 128]
|
||||
*/
|
||||
#define REDUCE_FULL_S(x) \
|
||||
EXTRA_REDUCE_S(REDUCE(x))
|
||||
EXTRA_REDUCE_S(REDUCE(x))
|
||||
|
||||
__device__ __forceinline__ void FFT_8(int *y, int stripe) {
|
||||
__device__ __forceinline__
|
||||
void FFT_8(int *y, int stripe) {
|
||||
|
||||
/*
|
||||
/*
|
||||
* FFT_8 using w=4 as 8th root of unity
|
||||
* Unrolled decimation in frequency (DIF) radix-2 NTT.
|
||||
* Output data is in revbin_permuted order.
|
||||
@ -110,20 +114,20 @@ __device__ __forceinline__ void FFT_8(int *y, int stripe) {
|
||||
#define X(i) y[stripe*i]
|
||||
|
||||
#define DO_REDUCE(i) \
|
||||
X(i) = REDUCE(X(i))
|
||||
X(i) = REDUCE(X(i))
|
||||
|
||||
#define DO_REDUCE_FULL_S(i) \
|
||||
do { \
|
||||
X(i) = REDUCE(X(i)); \
|
||||
X(i) = EXTRA_REDUCE_S(X(i)); \
|
||||
X(i) = REDUCE(X(i)); \
|
||||
X(i) = EXTRA_REDUCE_S(X(i)); \
|
||||
} while(0)
|
||||
|
||||
#define BUTTERFLY(i,j,n) \
|
||||
do { \
|
||||
int u= X(i); \
|
||||
int v= X(j); \
|
||||
X(i) = u+v; \
|
||||
X(j) = (u-v) << (2*n); \
|
||||
int u= X(i); \
|
||||
int v= X(j); \
|
||||
X(i) = u+v; \
|
||||
X(j) = (u-v) << (2*n); \
|
||||
} while(0)
|
||||
|
||||
BUTTERFLY(0, 4, 0);
|
||||
@ -163,17 +167,16 @@ X(j) = (u-v) << (2*n); \
|
||||
|
||||
__device__ __forceinline__ void FFT_16(int *y) {
|
||||
|
||||
/*
|
||||
/**
|
||||
* FFT_16 using w=2 as 16th root of unity
|
||||
* Unrolled decimation in frequency (DIF) radix-2 NTT.
|
||||
* Output data is in revbin_permuted order.
|
||||
*/
|
||||
|
||||
#define DO_REDUCE_FULL_S(i) \
|
||||
do { \
|
||||
y[i] = REDUCE(y[i]); \
|
||||
y[i] = EXTRA_REDUCE_S(y[i]); \
|
||||
} while(0)
|
||||
do { \
|
||||
y[i] = REDUCE(y[i]); \
|
||||
y[i] = EXTRA_REDUCE_S(y[i]); \
|
||||
} while(0)
|
||||
|
||||
int u,v;
|
||||
|
||||
@ -186,8 +189,8 @@ y[i] = EXTRA_REDUCE_S(y[i]); \
|
||||
// BUTTERFLY(6, 14, 6);
|
||||
// BUTTERFLY(7, 15, 7);
|
||||
{
|
||||
u= y[0]; // 0..7
|
||||
v= y[1]; // 8..15
|
||||
u = y[0]; // 0..7
|
||||
v = y[1]; // 8..15
|
||||
y[0] = u+v;
|
||||
y[1] = (u-v) << (threadIdx.x&7);
|
||||
}
|
||||
@ -204,8 +207,8 @@ y[i] = EXTRA_REDUCE_S(y[i]); \
|
||||
// BUTTERFLY( 2, 6, 4);
|
||||
// BUTTERFLY( 3, 7, 6);
|
||||
{
|
||||
u= __shfl((int)y[0], (threadIdx.x&3),8); // 0,1,2,3 0,1,2,3
|
||||
v= __shfl((int)y[0],4+(threadIdx.x&3),8); // 4,5,6,7 4,5,6,7
|
||||
u = __shfl((int)y[0], (threadIdx.x&3),8); // 0,1,2,3 0,1,2,3
|
||||
v = __shfl((int)y[0],4+(threadIdx.x&3),8); // 4,5,6,7 4,5,6,7
|
||||
y[0] = ((threadIdx.x&7) < 4) ? (u+v) : ((u-v) << (2*(threadIdx.x&3)));
|
||||
}
|
||||
|
||||
@ -214,8 +217,8 @@ y[i] = EXTRA_REDUCE_S(y[i]); \
|
||||
// BUTTERFLY(10, 14, 4);
|
||||
// BUTTERFLY(11, 15, 6);
|
||||
{
|
||||
u= __shfl((int)y[1], (threadIdx.x&3),8); // 8,9,10,11 8,9,10,11
|
||||
v= __shfl((int)y[1],4+(threadIdx.x&3),8); // 12,13,14,15 12,13,14,15
|
||||
u = __shfl((int)y[1], (threadIdx.x&3),8); // 8,9,10,11 8,9,10,11
|
||||
v = __shfl((int)y[1],4+(threadIdx.x&3),8); // 12,13,14,15 12,13,14,15
|
||||
y[1] = ((threadIdx.x&7) < 4) ? (u+v) : ((u-v) << (2*(threadIdx.x&3)));
|
||||
}
|
||||
|
||||
@ -233,8 +236,8 @@ y[i] = EXTRA_REDUCE_S(y[i]); \
|
||||
// BUTTERFLY( 4, 6, 0);
|
||||
// BUTTERFLY( 5, 7, 4);
|
||||
{
|
||||
u= __shfl((int)y[0], (threadIdx.x&5),8); // 0,1,0,1 4,5,4,5
|
||||
v= __shfl((int)y[0],2+(threadIdx.x&5),8); // 2,3,2,3 6,7,6,7
|
||||
u = __shfl((int)y[0], (threadIdx.x&5),8); // 0,1,0,1 4,5,4,5
|
||||
v = __shfl((int)y[0],2+(threadIdx.x&5),8); // 2,3,2,3 6,7,6,7
|
||||
y[0] = ((threadIdx.x&3) < 2) ? (u+v) : ((u-v) << (4*(threadIdx.x&1)));
|
||||
}
|
||||
|
||||
@ -243,8 +246,8 @@ y[i] = EXTRA_REDUCE_S(y[i]); \
|
||||
// BUTTERFLY(12, 14, 0);
|
||||
// BUTTERFLY(13, 15, 4);
|
||||
{
|
||||
u= __shfl((int)y[1], (threadIdx.x&5),8); // 8,9,8,9 12,13,12,13
|
||||
v= __shfl((int)y[1],2+(threadIdx.x&5),8); // 10,11,10,11 14,15,14,15
|
||||
u = __shfl((int)y[1], (threadIdx.x&5),8); // 8,9,8,9 12,13,12,13
|
||||
v = __shfl((int)y[1],2+(threadIdx.x&5),8); // 10,11,10,11 14,15,14,15
|
||||
y[1] = ((threadIdx.x&3) < 2) ? (u+v) : ((u-v) << (4*(threadIdx.x&1)));
|
||||
}
|
||||
|
||||
@ -253,8 +256,8 @@ y[i] = EXTRA_REDUCE_S(y[i]); \
|
||||
// BUTTERFLY( 4, 5, 0);
|
||||
// BUTTERFLY( 6, 7, 0);
|
||||
{
|
||||
u= __shfl((int)y[0], (threadIdx.x&6),8); // 0,0,2,2 4,4,6,6
|
||||
v= __shfl((int)y[0],1+(threadIdx.x&6),8); // 1,1,3,3 5,5,7,7
|
||||
u = __shfl((int)y[0], (threadIdx.x&6),8); // 0,0,2,2 4,4,6,6
|
||||
v = __shfl((int)y[0],1+(threadIdx.x&6),8); // 1,1,3,3 5,5,7,7
|
||||
y[0] = ((threadIdx.x&1) < 1) ? (u+v) : (u-v);
|
||||
}
|
||||
|
||||
@ -263,8 +266,8 @@ y[i] = EXTRA_REDUCE_S(y[i]); \
|
||||
// BUTTERFLY(12, 13, 0);
|
||||
// BUTTERFLY(14, 15, 0);
|
||||
{
|
||||
u= __shfl((int)y[1], (threadIdx.x&6),8); // 8,8,10,10 12,12,14,14
|
||||
v= __shfl((int)y[1],1+(threadIdx.x&6),8); // 9,9,11,11 13,13,15,15
|
||||
u = __shfl((int)y[1], (threadIdx.x&6),8); // 8,8,10,10 12,12,14,14
|
||||
v = __shfl((int)y[1],1+(threadIdx.x&6),8); // 9,9,11,11 13,13,15,15
|
||||
y[1] = ((threadIdx.x&1) < 1) ? (u+v) : (u-v);
|
||||
}
|
||||
|
||||
@ -274,7 +277,9 @@ y[i] = EXTRA_REDUCE_S(y[i]); \
|
||||
#undef DO_REDUCE_FULL_S
|
||||
}
|
||||
|
||||
__device__ __forceinline__ void FFT_128_full(int y[128]) {
|
||||
__device__ __forceinline__
|
||||
void FFT_128_full(int y[128])
|
||||
{
|
||||
int i;
|
||||
|
||||
FFT_8(y+0,2); // eight parallel FFT8's
|
||||
@ -289,11 +294,9 @@ __device__ __forceinline__ void FFT_128_full(int y[128]) {
|
||||
FFT_16(y+2*i); // eight sequential FFT16's, each one executed in parallel by 8 threads
|
||||
}
|
||||
|
||||
|
||||
__device__ __forceinline__ void FFT_256_halfzero(int y[256]) {
|
||||
|
||||
int i;
|
||||
|
||||
__device__ __forceinline__
|
||||
void FFT_256_halfzero(int y[256])
|
||||
{
|
||||
/*
|
||||
* FFT_256 using w=41 as 256th root of unity.
|
||||
* Decimation in frequency (DIF) NTT.
|
||||
@ -303,10 +306,10 @@ __device__ __forceinline__ void FFT_256_halfzero(int y[256]) {
|
||||
const int tmp = y[15];
|
||||
|
||||
#pragma unroll 8
|
||||
for (i=0; i<8; i++)
|
||||
for (int i=0; i<8; i++)
|
||||
y[16+i] = REDUCE(y[i] * c_FFT256_2_128_Twiddle[8*i+(threadIdx.x&7)]);
|
||||
#pragma unroll 8
|
||||
for (i=8; i<16; i++)
|
||||
for (int i=8; i<16; i++)
|
||||
y[16+i] = 0;
|
||||
|
||||
/* handle X^255 with an additional butterfly */
|
||||
@ -323,19 +326,18 @@ __device__ __forceinline__ void FFT_256_halfzero(int y[256]) {
|
||||
|
||||
/***************************************************/
|
||||
|
||||
__device__ __forceinline__ void Expansion(const uint32_t *data, uint4 *g_temp4)
|
||||
__device__ __forceinline__
|
||||
void Expansion(const uint32_t *data, uint4 *g_temp4)
|
||||
{
|
||||
int i;
|
||||
|
||||
/* Message Expansion using Number Theoretical Transform similar to FFT */
|
||||
int expanded[32];
|
||||
#pragma unroll 4
|
||||
for (i=0; i < 4; i++) {
|
||||
for (int i=0; i < 4; i++) {
|
||||
expanded[ i] = __byte_perm(__shfl((int)data[0], 2*i, 8), __shfl((int)data[0], (2*i)+1, 8), threadIdx.x&7)&0xff;
|
||||
expanded[4+i] = __byte_perm(__shfl((int)data[1], 2*i, 8), __shfl((int)data[1], (2*i)+1, 8), threadIdx.x&7)&0xff;
|
||||
}
|
||||
#pragma unroll 8
|
||||
for (i=8; i < 16; i++)
|
||||
for (int i=8; i < 16; i++)
|
||||
expanded[i] = 0;
|
||||
|
||||
FFT_256_halfzero(expanded);
|
||||
@ -447,7 +449,6 @@ __device__ __forceinline__ void Expansion(const uint32_t *data, uint4 *g_temp4)
|
||||
//{ 14, 78, 46, 110, 30, 94, 62, 126 }, { 15, 79, 47, 111, 31, 95, 63, 127 },
|
||||
//{ 2, 66, 34, 98, 18, 82, 50, 114 }, { 3, 67, 35, 99, 19, 83, 51, 115 },
|
||||
|
||||
|
||||
bool sel = ((threadIdx.x+2)&7) >= 4; // 2,3,4,5
|
||||
|
||||
P1 = sel?expanded[0]:expanded[1]; Q1 = __shfl(P1, threadIdx.x^1, 8);
|
||||
@ -474,7 +475,6 @@ __device__ __forceinline__ void Expansion(const uint32_t *data, uint4 *g_temp4)
|
||||
// 0 8 4 12 2 10 6 14 0 8 4 12 2 10 6 14 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1
|
||||
// 1 9 5 13 3 11 7 15 1 9 5 13 3 11 7 15 4 4 4 4 4 4 4 4 5 5 5 5 5 5 5 5
|
||||
|
||||
|
||||
P1 = sel?expanded[1]:expanded[0]; Q1 = __shfl(P1, threadIdx.x^1, 8);
|
||||
Q2 = sel?expanded[3]:expanded[2]; P2 = __shfl(Q2, threadIdx.x^1, 8);
|
||||
P = even? P1 : P2; Q = even? Q1 : Q2;
|
||||
@ -552,7 +552,7 @@ __device__ __forceinline__ void Expansion(const uint32_t *data, uint4 *g_temp4)
|
||||
}
|
||||
|
||||
/***************************************************/
|
||||
// Die Hash-Funktion
|
||||
|
||||
__global__ void __launch_bounds__(TPB,4)
|
||||
x11_simd512_gpu_expand_64(int threads, uint32_t startNounce, uint64_t *g_hash, uint32_t *g_nonceVector, uint4 *g_temp4)
|
||||
{
|
||||
@ -567,9 +567,10 @@ x11_simd512_gpu_expand_64(int threads, uint32_t startNounce, uint64_t *g_hash, u
|
||||
|
||||
// Hash einlesen und auf 8 Threads und 2 Register verteilen
|
||||
uint32_t Hash[2];
|
||||
#pragma unroll 2
|
||||
|
||||
#pragma unroll 2
|
||||
for (int i=0; i<2; i++)
|
||||
Hash[i] = inpHash[8*i+(threadIdx.x&7)];
|
||||
Hash[i] = inpHash[8*i + (threadIdx.x & 7)];
|
||||
|
||||
// Puffer für expandierte Nachricht
|
||||
uint4 *temp4 = &g_temp4[64 * hashPosition];
|
||||
@ -622,20 +623,13 @@ x11_simd512_gpu_final_64(int threads, uint32_t startNounce, uint64_t *g_hash, ui
|
||||
}
|
||||
}
|
||||
|
||||
// Setup-Funktionen
|
||||
__host__ void x11_simd512_cpu_init(int thr_id, int threads)
|
||||
__host__
|
||||
void x11_simd512_cpu_init(int thr_id, int threads)
|
||||
{
|
||||
cudaMalloc( &d_state[thr_id], 32*sizeof(int)*threads );
|
||||
cudaMalloc( &d_temp4[thr_id], 64*sizeof(uint4)*threads );
|
||||
CUDA_SAFE_CALL(cudaMalloc(&d_state[thr_id], 32*sizeof(int)*threads));
|
||||
CUDA_SAFE_CALL(cudaMalloc(&d_temp4[thr_id], 64*sizeof(uint4)*threads));
|
||||
|
||||
// Textur für 128 Bit Zugriffe
|
||||
cudaChannelFormatDesc channelDesc128 = cudaCreateChannelDesc<uint4>();
|
||||
texRef1D_128.normalized = 0;
|
||||
texRef1D_128.filterMode = cudaFilterModePoint;
|
||||
texRef1D_128.addressMode[0] = cudaAddressModeClamp;
|
||||
cudaBindTexture(NULL, &texRef1D_128, d_temp4[thr_id], &channelDesc128, 64*sizeof(uint4)*threads);
|
||||
|
||||
cudaMemcpyToSymbol( c_perm, h_perm, sizeof(h_perm), 0, cudaMemcpyHostToDevice);
|
||||
cudaMemcpyToSymbol(c_perm, h_perm, sizeof(h_perm), 0, cudaMemcpyHostToDevice);
|
||||
cudaMemcpyToSymbol(c_IV_512, h_IV_512, sizeof(h_IV_512), 0, cudaMemcpyHostToDevice);
|
||||
cudaMemcpyToSymbol(c_FFT128_8_16_Twiddle, h_FFT128_8_16_Twiddle, sizeof(h_FFT128_8_16_Twiddle), 0, cudaMemcpyHostToDevice);
|
||||
cudaMemcpyToSymbol(c_FFT256_2_128_Twiddle, h_FFT256_2_128_Twiddle, sizeof(h_FFT256_2_128_Twiddle), 0, cudaMemcpyHostToDevice);
|
||||
@ -644,28 +638,31 @@ __host__ void x11_simd512_cpu_init(int thr_id, int threads)
|
||||
cudaMemcpyToSymbol(d_cw1, h_cw1, sizeof(h_cw1), 0, cudaMemcpyHostToDevice);
|
||||
cudaMemcpyToSymbol(d_cw2, h_cw2, sizeof(h_cw2), 0, cudaMemcpyHostToDevice);
|
||||
cudaMemcpyToSymbol(d_cw3, h_cw3, sizeof(h_cw3), 0, cudaMemcpyHostToDevice);
|
||||
|
||||
// Texture for 128-Bit Zugriffe
|
||||
cudaChannelFormatDesc channelDesc128 = cudaCreateChannelDesc<uint4>();
|
||||
texRef1D_128.normalized = 0;
|
||||
texRef1D_128.filterMode = cudaFilterModePoint;
|
||||
texRef1D_128.addressMode[0] = cudaAddressModeClamp;
|
||||
CUDA_SAFE_CALL(cudaBindTexture(NULL, &texRef1D_128, d_temp4[thr_id], &channelDesc128, 64*sizeof(uint4)*threads));
|
||||
}
|
||||
|
||||
__host__ void x11_simd512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order)
|
||||
__host__
|
||||
void x11_simd512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order)
|
||||
{
|
||||
const int threadsperblock = TPB;
|
||||
|
||||
// Größe des dynamischen Shared Memory Bereichs
|
||||
size_t shared_size = 0;
|
||||
|
||||
// berechne wie viele Thread Blocks wir brauchen
|
||||
dim3 block(threadsperblock);
|
||||
|
||||
dim3 grid8(((threads + threadsperblock-1)/threadsperblock)*8);
|
||||
x11_simd512_gpu_expand_64<<<grid8, block, shared_size>>>(threads, startNounce, (uint64_t*)d_hash, d_nonceVector, d_temp4[thr_id]);
|
||||
|
||||
x11_simd512_gpu_expand_64 <<<grid8, block>>> (threads, startNounce, (uint64_t*)d_hash, d_nonceVector, d_temp4[thr_id]);
|
||||
|
||||
dim3 grid((threads + threadsperblock-1)/threadsperblock);
|
||||
|
||||
// künstlich die Occupancy limitieren, um das totale Erschöpfen des Texture Cache zu vermeiden
|
||||
x11_simd512_gpu_compress1_64<<<grid, block, shared_size>>>(threads, startNounce, (uint64_t*)d_hash, d_nonceVector, d_temp4[thr_id], d_state[thr_id]);
|
||||
x11_simd512_gpu_compress2_64<<<grid, block, shared_size>>>(threads, startNounce, (uint64_t*)d_hash, d_nonceVector, d_temp4[thr_id], d_state[thr_id]);
|
||||
x11_simd512_gpu_compress1_64 <<<grid, block>>> (threads, startNounce, (uint64_t*)d_hash, d_nonceVector, d_temp4[thr_id], d_state[thr_id]);
|
||||
x11_simd512_gpu_compress2_64 <<<grid, block>>> (threads, startNounce, (uint64_t*)d_hash, d_nonceVector, d_temp4[thr_id], d_state[thr_id]);
|
||||
|
||||
x11_simd512_gpu_final_64<<<grid, block, shared_size>>>(threads, startNounce, (uint64_t*)d_hash, d_nonceVector, d_temp4[thr_id], d_state[thr_id]);
|
||||
x11_simd512_gpu_final_64 <<<grid, block>>> (threads, startNounce, (uint64_t*)d_hash, d_nonceVector, d_temp4[thr_id], d_state[thr_id]);
|
||||
|
||||
MyStreamSynchronize(NULL, order, thr_id);
|
||||
}
|
||||
|
||||
@ -147,7 +147,6 @@ extern "C" int scanhash_x11(int thr_id, uint32_t *pdata,
|
||||
if (!init[thr_id])
|
||||
{
|
||||
CUDA_SAFE_CALL(cudaSetDevice(device_map[thr_id]));
|
||||
// Konstanten kopieren, Speicher belegen
|
||||
CUDA_SAFE_CALL(cudaMalloc(&d_hash[thr_id], 16 * sizeof(uint32_t) * throughput));
|
||||
|
||||
quark_blake512_cpu_init(thr_id, throughput);
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user