Skip to content

Commit 9efd648

Browse files
Christian BuchnerChristian Buchner
Christian Buchner
authored and
Christian Buchner
committed
Merge branch 'master' of https://github.com/cbuchner1/ccminer 
2 parents 07cbafe + bdef3bb commit 9efd648

File tree

1 file changed

+82
-88
lines changed

1 file changed

+82
-88
lines changed

‎cuda_groestlcoin.cu‎

Lines changed: 82 additions & 88 deletions
Original file line numberDiff line numberDiff line change
@@ -7,6 +7,7 @@
77
#include <stdio.h>
88
#include <memory.h>
99

10+
// IMPORTANT: leave this enabled!
1011
#define USE_SHARED 1
1112

1213
// aus cpu-miner.c
@@ -15,35 +16,41 @@ extern int device_map[8];
1516
// aus heavy.cu
1617
extern cudaError_t MyStreamSynchronize(cudaStream_t stream, int situation, int thr_id);
1718

19+
// aus driver.c
20+
extern "C" void set_device(int device);
21+
1822
// Folgende Definitionen später durch header ersetzen
1923
typedef unsigned char uint8_t;
2024
typedef unsigned int uint32_t;
2125
typedef unsigned long long uint64_t;
2226

27+
// diese Struktur wird in der Init Funktion angefordert
28+
static cudaDeviceProp props;
29+
2330
// globaler Speicher für alle HeftyHashes aller Threads
2431
__constant__ uint32_t pTarget[8]; // Single GPU
2532
extern uint32_t *d_resultNonce[8];
2633

2734
__constant__ uint32_t groestlcoin_gpu_msg[32];
2835

36+
#define SPH_C32(x) ((uint32_t)(x ## U))
2937
#define SPH_T32(x) ((x) & SPH_C32(0xFFFFFFFF))
3038

3139
#define PC32up(j, r) ((uint32_t)((j) + (r)))
3240
#define PC32dn(j, r) 0
3341
#define QC32up(j, r) 0xFFFFFFFF
3442
#define QC32dn(j, r) (((uint32_t)(r) << 24) ^ SPH_T32(~((uint32_t)(j) << 24)))
3543

36-
#define B32_0(x) ((x) & 0xFF)
37-
#define B32_1(x) (((x) >> 8) & 0xFF)
38-
#define B32_2(x) (((x) >> 16) & 0xFF)
39-
#define B32_3(x) ((x) >> 24)
40-
41-
#define SPH_C32(x) ((uint32_t)(x ## U))
42-
#define C32e(x) ((SPH_C32(x) >> 24) \
43-
| ((SPH_C32(x) >> 8) & SPH_C32(0x0000FF00)) \
44-
| ((SPH_C32(x) << 8) & SPH_C32(0x00FF0000)) \
45-
| ((SPH_C32(x) << 24) & SPH_C32(0xFF000000)))
44+
#define B32_0(x) __byte_perm(x, 0, 0x4440)
45+
//((x) & 0xFF)
46+
#define B32_1(x) __byte_perm(x, 0, 0x4441)
47+
//(((x) >> 8) & 0xFF)
48+
#define B32_2(x) __byte_perm(x, 0, 0x4442)
49+
//(((x) >> 16) & 0xFF)
50+
#define B32_3(x) __byte_perm(x, 0, 0x4443)
51+
//((x) >> 24)
4652

53+
#if 0
4754
#if USE_SHARED
4855
#define T0up(x) (*((uint32_t*)mixtabs + ( (x))))
4956
#define T0dn(x) (*((uint32_t*)mixtabs + (256+(x))))
@@ -63,6 +70,18 @@ __constant__ uint32_t groestlcoin_gpu_msg[32];
6370
#define T3up(x) tex1Dfetch(t3up1, x)
6471
#define T3dn(x) tex1Dfetch(t3dn1, x)
6572
#endif
73+
#endif
74+
75+
// a healthy mix between shared and textured access provides the highest speed!
76+
#define T0up(x) (*((uint32_t*)mixtabs + ( (x))))
77+
#define T0dn(x) tex1Dfetch(t0dn1, x)
78+
#define T1up(x) tex1Dfetch(t1up1, x)
79+
#define T1dn(x) (*((uint32_t*)mixtabs + (768+(x))))
80+
#define T2up(x) tex1Dfetch(t2up1, x)
81+
#define T2dn(x) (*((uint32_t*)mixtabs + (1280+(x))))
82+
#define T3up(x) (*((uint32_t*)mixtabs + (1536+(x))))
83+
#define T3dn(x) tex1Dfetch(t3dn1, x)
84+
6685
texture<unsigned int, 1, cudaReadModeElementType> t0up1;
6786
texture<unsigned int, 1, cudaReadModeElementType> t0dn1;
6887
texture<unsigned int, 1, cudaReadModeElementType> t1up1;
@@ -81,21 +100,6 @@ extern uint32_t T2dn_cpu[];
81100
extern uint32_t T3up_cpu[];
82101
extern uint32_t T3dn_cpu[];
83102

84-
#if __CUDA_ARCH__ < 350
85-
// Kepler (Compute 3.0)
86-
#define S(x, n) (((x) >> (n)) | ((x) << (32 - (n))))
87-
#else
88-
// Kepler (Compute 3.5)
89-
#define S(x, n) __funnelshift_r( x, x, n );
90-
#endif
91-
#define R(x, n) ((x) >> (n))
92-
#define Ch(x, y, z) ((x & (y ^ z)) ^ z)
93-
#define Maj(x, y, z) ((x & (y | z)) | (y & z))
94-
#define S0(x) (S(x, 2) ^ S(x, 13) ^ S(x, 22))
95-
#define S1(x) (S(x, 6) ^ S(x, 11) ^ S(x, 25))
96-
#define s0(x) (S(x, 7) ^ S(x, 18) ^ R(x, 3))
97-
#define s1(x) (S(x, 17) ^ S(x, 19) ^ R(x, 10))
98-
99103
#define SWAB32(x) ( ((x & 0x000000FF) << 24) | ((x & 0x0000FF00) << 8) | ((x & 0x00FF0000) >> 8) | ((x & 0xFF000000) >> 24) )
100104

101105

@@ -152,32 +156,25 @@ __device__ __forceinline__ void groestlcoin_perm_P(uint32_t *a, char *mixtabs)
152156
for(int k=0;k<16;k++) a[(k*2)+0] ^= PC32up(k * 0x10, 13); break;
153157
}
154158

155-
// RBTT
159+
// RBTT
156160
#pragma unroll 16
157-
for(int k=0;k<32;k+=2)
158-
{
159-
t[k + 0] = T0up( B32_0(a[k & 0x1f]) ) ^
160-
T1up( B32_1(a[(k + 2) & 0x1f]) ) ^
161-
T2up( B32_2(a[(k + 4) & 0x1f]) ) ^
162-
T3up( B32_3(a[(k + 6) & 0x1f]) ) ^
163-
T0dn( B32_0(a[(k + 9) & 0x1f]) ) ^
164-
T1dn( B32_1(a[(k + 11) & 0x1f]) ) ^
165-
T2dn( B32_2(a[(k + 13) & 0x1f]) ) ^
166-
T3dn( B32_3(a[(k + 23) & 0x1f]) );
167-
168-
t[k + 1] = T0dn( B32_0(a[k & 0x1f]) ) ^
169-
T1dn( B32_1(a[(k + 2) & 0x1f]) ) ^
170-
T2dn( B32_2(a[(k + 4) & 0x1f]) ) ^
171-
T3dn( B32_3(a[(k + 6) & 0x1f]) ) ^
172-
T0up( B32_0(a[(k + 9) & 0x1f]) ) ^
173-
T1up( B32_1(a[(k + 11) & 0x1f]) ) ^
174-
T2up( B32_2(a[(k + 13) & 0x1f]) ) ^
175-
T3up( B32_3(a[(k + 23) & 0x1f]) );
176-
}
161+
for(int k=0;k<32;k+=2)
162+
{
163+
uint32_t t0_0 = B32_0(a[(k ) & 0x1f]), t9_0 = B32_0(a[(k + 9) & 0x1f]);
164+
uint32_t t2_1 = B32_1(a[(k + 2) & 0x1f]), t11_1 = B32_1(a[(k + 11) & 0x1f]);
165+
uint32_t t4_2 = B32_2(a[(k + 4) & 0x1f]), t13_2 = B32_2(a[(k + 13) & 0x1f]);
166+
uint32_t t6_3 = B32_3(a[(k + 6) & 0x1f]), t23_3 = B32_3(a[(k + 23) & 0x1f]);
167+
168+
t[k + 0] = T0up( t0_0 ) ^ T1up( t2_1 ) ^ T2up( t4_2 ) ^ T3up( t6_3 ) ^
169+
T0dn( t9_0 ) ^ T1dn( t11_1 ) ^ T2dn( t13_2 ) ^ T3dn( t23_3 );
170+
171+
t[k + 1] = T0dn( t0_0 ) ^ T1dn( t2_1 ) ^ T2dn( t4_2 ) ^ T3dn( t6_3 ) ^
172+
T0up( t9_0 ) ^ T1up( t11_1 ) ^ T2up( t13_2 ) ^ T3up( t23_3 );
173+
}
177174
#pragma unroll 32
178-
for(int k=0;k<32;k++)
179-
a[k] = t[k];
180-
}
175+
for(int k=0;k<32;k++)
176+
a[k] = t[k];
177+
}
181178
}
182179

183180
__device__ __forceinline__ void groestlcoin_perm_Q(uint32_t *a, char *mixtabs)
@@ -233,32 +230,25 @@ __device__ __forceinline__ void groestlcoin_perm_Q(uint32_t *a, char *mixtabs)
233230
for(int k=0;k<16;k++) { a[(k*2)+0] ^= QC32up(k * 0x10, 13); a[(k*2)+1] ^= QC32dn(k * 0x10, 13);} break;
234231
}
235232

236-
// RBTT
233+
// RBTT
237234
#pragma unroll 16
238-
for(int k=0;k<32;k+=2)
239-
{
240-
t[k + 0] = T0up( B32_0(a[(k + 2) & 0x1f]) ) ^
241-
T1up( B32_1(a[(k + 6) & 0x1f]) ) ^
242-
T2up( B32_2(a[(k + 10) & 0x1f]) ) ^
243-
T3up( B32_3(a[(k + 22) & 0x1f]) ) ^
244-
T0dn( B32_0(a[(k + 1) & 0x1f]) ) ^
245-
T1dn( B32_1(a[(k + 5) & 0x1f]) ) ^
246-
T2dn( B32_2(a[(k + 9) & 0x1f]) ) ^
247-
T3dn( B32_3(a[(k + 13) & 0x1f]) );
248-
249-
t[k + 1] = T0dn( B32_0(a[(k + 2) & 0x1f]) ) ^
250-
T1dn( B32_1(a[(k + 6) & 0x1f]) ) ^
251-
T2dn( B32_2(a[(k + 10) & 0x1f]) ) ^
252-
T3dn( B32_3(a[(k + 22) & 0x1f]) ) ^
253-
T0up( B32_0(a[(k + 1) & 0x1f]) ) ^
254-
T1up( B32_1(a[(k + 5) & 0x1f]) ) ^
255-
T2up( B32_2(a[(k + 9) & 0x1f]) ) ^
256-
T3up( B32_3(a[(k + 13) & 0x1f]) );
257-
}
235+
for(int k=0;k<32;k+=2)
236+
{
237+
uint32_t t2_0 = B32_0(a[(k + 2) & 0x1f]), t1_0 = B32_0(a[(k + 1) & 0x1f]);
238+
uint32_t t6_1 = B32_1(a[(k + 6) & 0x1f]), t5_1 = B32_1(a[(k + 5) & 0x1f]);
239+
uint32_t t10_2 = B32_2(a[(k + 10) & 0x1f]), t9_2 = B32_2(a[(k + 9) & 0x1f]);
240+
uint32_t t22_3 = B32_3(a[(k + 22) & 0x1f]), t13_3 = B32_3(a[(k + 13) & 0x1f]);
241+
242+
t[k + 0] = T0up( t2_0 ) ^ T1up( t6_1 ) ^ T2up( t10_2 ) ^ T3up( t22_3 ) ^
243+
T0dn( t1_0 ) ^ T1dn( t5_1 ) ^ T2dn( t9_2 ) ^ T3dn( t13_3 );
244+
245+
t[k + 1] = T0dn( t2_0 ) ^ T1dn( t6_1 ) ^ T2dn( t10_2 ) ^ T3dn( t22_3 ) ^
246+
T0up( t1_0 ) ^ T1up( t5_1 ) ^ T2up( t9_2 ) ^ T3up( t13_3 );
247+
}
258248
#pragma unroll 32
259-
for(int k=0;k<32;k++)
260-
a[k] = t[k];
261-
}
249+
for(int k=0;k<32;k++)
250+
a[k] = t[k];
251+
}
262252
}
263253
#if USE_SHARED
264254
__global__ void /* __launch_bounds__(256) */
@@ -271,14 +261,17 @@ __global__ void
271261
#if USE_SHARED
272262
extern __shared__ char mixtabs[];
273263

274-
*((uint32_t*)mixtabs + ( threadIdx.x)) = tex1Dfetch(t0up1, threadIdx.x);
275-
*((uint32_t*)mixtabs + (256+threadIdx.x)) = tex1Dfetch(t0dn1, threadIdx.x);
276-
*((uint32_t*)mixtabs + (512+threadIdx.x)) = tex1Dfetch(t1up1, threadIdx.x);
277-
*((uint32_t*)mixtabs + (768+threadIdx.x)) = tex1Dfetch(t1dn1, threadIdx.x);
278-
*((uint32_t*)mixtabs + (1024+threadIdx.x)) = tex1Dfetch(t2up1, threadIdx.x);
279-
*((uint32_t*)mixtabs + (1280+threadIdx.x)) = tex1Dfetch(t2dn1, threadIdx.x);
280-
*((uint32_t*)mixtabs + (1536+threadIdx.x)) = tex1Dfetch(t3up1, threadIdx.x);
281-
*((uint32_t*)mixtabs + (1792+threadIdx.x)) = tex1Dfetch(t3dn1, threadIdx.x);
264+
if (threadIdx.x < 256)
265+
{
266+
*((uint32_t*)mixtabs + ( threadIdx.x)) = tex1Dfetch(t0up1, threadIdx.x);
267+
*((uint32_t*)mixtabs + (256+threadIdx.x)) = tex1Dfetch(t0dn1, threadIdx.x);
268+
*((uint32_t*)mixtabs + (512+threadIdx.x)) = tex1Dfetch(t1up1, threadIdx.x);
269+
*((uint32_t*)mixtabs + (768+threadIdx.x)) = tex1Dfetch(t1dn1, threadIdx.x);
270+
*((uint32_t*)mixtabs + (1024+threadIdx.x)) = tex1Dfetch(t2up1, threadIdx.x);
271+
*((uint32_t*)mixtabs + (1280+threadIdx.x)) = tex1Dfetch(t2dn1, threadIdx.x);
272+
*((uint32_t*)mixtabs + (1536+threadIdx.x)) = tex1Dfetch(t3up1, threadIdx.x);
273+
*((uint32_t*)mixtabs + (1792+threadIdx.x)) = tex1Dfetch(t3dn1, threadIdx.x);
274+
}
282275

283276
__syncthreads();
284277
#endif
@@ -407,8 +400,11 @@ __global__ void
407400
// Setup-Funktionen
408401
__host__ void groestlcoin_cpu_init(int thr_id, int threads)
409402
{
410-
cudaSetDevice(device_map[thr_id]);
411-
cudaDeviceSetCacheConfig( cudaFuncCachePreferShared );
403+
cudaSetDevice(device_map[thr_id]);
404+
405+
cudaGetDeviceProperties(&props, device_map[thr_id]);
406+
407+
cudaDeviceSetCacheConfig( cudaFuncCachePreferL1 );
412408
// Texturen mit obigem Makro initialisieren
413409
texDef(t0up1, d_T0up, T0up_cpu, sizeof(uint32_t)*256);
414410
texDef(t0dn1, d_T0dn, T0dn_cpu, sizeof(uint32_t)*256);
@@ -452,11 +448,9 @@ __host__ void groestlcoin_cpu_setBlock(int thr_id, void *data, void *pTargetIn)
452448

453449
__host__ void groestlcoin_cpu_hash(int thr_id, int threads, uint32_t startNounce, void *outputHashes, uint32_t *nounce)
454450
{
455-
#if USE_SHARED
456-
const int threadsperblock = 256; // Alignment mit mixtab Grösse. NICHT ÄNDERN
457-
#else
458-
const int threadsperblock = 512; // so einstellen wie gewünscht ;-)
459-
#endif
451+
// Compute 3.x und 5.x Geräte am besten mit 768 Threads ansteuern,
452+
// alle anderen mit 512 Threads.
453+
int threadsperblock = (props.major >= 3) ? 768 : 512;
460454

461455
// berechne wie viele Thread Blocks wir brauchen
462456
dim3 grid((threads + threadsperblock-1)/threadsperblock);

0 commit comments

Comments
 (0)