forked from cbuchner1/CudaMiner
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathsalsa_kernel.h
112 lines (89 loc) · 4.26 KB
/
salsa_kernel.h
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
#ifndef SALSA_KERNEL_H
#define SALSA_KERNEL_H
#include <stdbool.h>
#ifdef __cplusplus
extern "C" {
#endif
typedef unsigned int uint32_t; // define this as 32 bit type derived from int
// from cpu-miner.h
extern int device_map[8];
extern int device_interactive[8];
extern int device_batchsize[8];
extern int device_backoff[8];
extern int device_lookup_gap[8];
extern int device_texturecache[8];
extern int device_singlememory[8];
extern char *device_config[8];
extern char *device_name[8];
extern bool autotune;
// CUDA externals
extern int cuda_num_devices();
extern int cuda_finddevice(char *name);
extern void cuda_shutdown(int thr_id);
extern int cuda_throughput(int thr_id);
extern uint32_t *cuda_transferbuffer(int thr_id, int stream);
extern uint32_t *cuda_hashbuffer(int thr_id, int stream);
extern void cuda_scrypt_HtoD(int thr_id, uint32_t *X, int stream);
extern void cuda_scrypt_serialize(int thr_id, int stream);
extern void cuda_scrypt_core(int thr_id, int stream, unsigned int N);
extern void cuda_scrypt_done(int thr_id, int stream);
extern void cuda_scrypt_DtoH(int thr_id, uint32_t *X, int stream);
extern void cuda_scrypt_sync(int thr_id, int stream);
extern void cuda_scrypt_flush(int thr_id, int stream);
extern void cuda_prepare_keccak256(int thr_id, const uint32_t host_pdata[20], const uint32_t ptarget[8]);
extern bool cuda_do_keccak256(int thr_id, int stream, uint32_t *hash, uint32_t nonce, int throughput, bool do_d2h);
extern void computeGold(uint32_t *idata, uint32_t *reference, uint32_t *V);
extern void default_prepare_keccak256(int thr_id, const uint32_t host_pdata[20], const uint32_t ptarget[8]);
#ifdef __NVCC__
#include <cuda_runtime.h>
extern cudaError_t MyStreamSynchronize(cudaStream_t stream, int situation, int thr_id);
extern bool default_do_keccak256(dim3 grid, dim3 threads, int thr_id, int stream, uint32_t *hash, uint32_t nonce, int throughput, bool do_d2h);
#endif
#ifdef __cplusplus
}
// If we're in C++ mode, we're either compiling .cu files or scrypt.cpp
#ifdef __NVCC__
/**
* An pure virtual interface for a CUDA kernel implementation.
* TODO: encapsulate the kernel launch parameters in some kind of wrapper.
*/
class KernelInterface
{
public:
virtual void set_scratchbuf_constants(int MAXWARPS, uint32_t** h_V) = 0;
virtual bool run_kernel(dim3 grid, dim3 threads, int WARPS_PER_BLOCK, int thr_id, cudaStream_t stream, uint32_t* d_idata, uint32_t* d_odata, unsigned int N, unsigned int LOOKUP_GAP, bool interactive, bool benchmark, int texture_cache) = 0;
virtual bool bindtexture_1D(uint32_t *d_V, size_t size) { return true; }
virtual bool bindtexture_2D(uint32_t *d_V, int width, int height, size_t pitch) { return true; }
virtual bool unbindtexture_1D() { return true; }
virtual bool unbindtexture_2D() { return true; }
virtual char get_identifier() = 0;
virtual int get_major_version() { return 1; }
virtual int get_minor_version() { return 0; }
virtual int max_warps_per_block() = 0;
virtual int get_texel_width() = 0;
virtual bool no_textures() { return false; };
virtual bool single_memory() { return false; };
virtual int threads_per_wu() { return 1; }
virtual bool support_lookup_gap() { return false; }
virtual cudaSharedMemConfig shared_mem_config() { return cudaSharedMemBankSizeDefault; }
virtual cudaFuncCache cache_config() { return cudaFuncCachePreferNone; }
virtual void prepare_keccak256(int thr_id, const uint32_t host_pdata[20], const uint32_t ptarget[8]) {
default_prepare_keccak256(thr_id, host_pdata, ptarget);
}
virtual bool do_keccak256(dim3 grid, dim3 threads, int thr_id, int stream, uint32_t *hash, uint32_t nonce, int throughput, bool do_d2h = false) {
return default_do_keccak256(grid, threads, thr_id, stream, hash, nonce, throughput, do_d2h);
}
};
// Define work unit size
#define TOTAL_WARP_LIMIT 4096
#define WU_PER_WARP (32 / THREADS_PER_WU)
#define WU_PER_BLOCK (WU_PER_WARP*WARPS_PER_BLOCK)
#define WU_PER_LAUNCH (GRID_BLOCKS*WU_PER_BLOCK)
// make scratchpad size dependent on N and LOOKUP_GAP
#define SCRATCH (((N+LOOKUP_GAP-1)/LOOKUP_GAP)*32)
// Not performing error checking is actually bad, but...
#define checkCudaErrors(x) x
#define getLastCudaError(x)
#endif // #ifdef __NVCC__
#endif // #ifdef __cplusplus
#endif // #ifndef SALSA_KERNEL_H