Browse Source

Fix windows linkage, C/C++ mismatch

master
Tanguy Pruvot 9 years ago
parent
commit
c7cfe0e2ca
  1. 3
      api.cpp
  2. 95
      cuda.cpp
  3. 5
      miner.h
  4. 3
      nvml.cpp

3
api.cpp

@ -101,8 +101,7 @@ extern int num_cpus; @@ -101,8 +101,7 @@ extern int num_cpus;
extern float cpu_temp(int);
extern uint32_t cpu_clock(int);
// cuda.cpp
int cuda_num_devices();
int cuda_gpu_clocks(struct cgpu_info *gpu);
extern int cuda_gpu_clocks(struct cgpu_info *gpu);
char driver_version[32] = { 0 };

95
cuda.cpp

@ -23,6 +23,11 @@ @@ -23,6 +23,11 @@
#include "cuda_runtime.h"
#ifdef __cplusplus
/* miner.h functions are declared in C type, not C++ */
extern "C" {
#endif
// CUDA Devices on the System
int cuda_num_devices()
{
@ -150,49 +155,6 @@ uint32_t device_intensity(int thr_id, const char *func, uint32_t defcount) @@ -150,49 +155,6 @@ uint32_t device_intensity(int thr_id, const char *func, uint32_t defcount)
return throughput;
}
// Zeitsynchronisations-Routine von cudaminer mit CPU sleep
// Note: if you disable all of these calls, CPU usage will hit 100%
typedef struct { double value[8]; } tsumarray;
cudaError_t MyStreamSynchronize(cudaStream_t stream, int situation, int thr_id)
{
cudaError_t result = cudaSuccess;
if (situation >= 0)
{
static std::map<int, tsumarray> tsum;
double a = 0.95, b = 0.05;
if (tsum.find(situation) == tsum.end()) { a = 0.5; b = 0.5; } // faster initial convergence
double tsync = 0.0;
double tsleep = 0.95 * tsum[situation].value[thr_id];
if (cudaStreamQuery(stream) == cudaErrorNotReady)
{
usleep((useconds_t)(1e6*tsleep));
struct timeval tv_start, tv_end;
gettimeofday(&tv_start, NULL);
result = cudaStreamSynchronize(stream);
gettimeofday(&tv_end, NULL);
tsync = 1e-6 * (tv_end.tv_usec-tv_start.tv_usec) + (tv_end.tv_sec-tv_start.tv_sec);
}
if (tsync >= 0) tsum[situation].value[thr_id] = a * tsum[situation].value[thr_id] + b * (tsleep+tsync);
}
else
result = cudaStreamSynchronize(stream);
return result;
}
int cuda_gpu_clocks(struct cgpu_info *gpu)
{
cudaDeviceProp props;
if (cudaGetDeviceProperties(&props, gpu->gpu_id) == cudaSuccess) {
gpu->gpu_clock = props.clockRate;
gpu->gpu_memclock = props.memoryClockRate;
gpu->gpu_mem = props.totalGlobalMem;
return 0;
}
return -1;
}
// if we use 2 threads on the same gpu, we need to reinit the threads
void cuda_reset_device(int thr_id, bool *init)
{
@ -228,6 +190,53 @@ int cuda_available_memory(int thr_id) @@ -228,6 +190,53 @@ int cuda_available_memory(int thr_id)
return (int) (mfree / (1024 * 1024));
}
#ifdef __cplusplus
} /* extern "C" */
#endif
int cuda_gpu_clocks(struct cgpu_info *gpu)
{
cudaDeviceProp props;
if (cudaGetDeviceProperties(&props, gpu->gpu_id) == cudaSuccess) {
gpu->gpu_clock = props.clockRate;
gpu->gpu_memclock = props.memoryClockRate;
gpu->gpu_mem = props.totalGlobalMem;
return 0;
}
return -1;
}
// Zeitsynchronisations-Routine von cudaminer mit CPU sleep
// Note: if you disable all of these calls, CPU usage will hit 100%
typedef struct { double value[8]; } tsumarray;
cudaError_t MyStreamSynchronize(cudaStream_t stream, int situation, int thr_id)
{
cudaError_t result = cudaSuccess;
if (situation >= 0)
{
static std::map<int, tsumarray> tsum;
double a = 0.95, b = 0.05;
if (tsum.find(situation) == tsum.end()) { a = 0.5; b = 0.5; } // faster initial convergence
double tsync = 0.0;
double tsleep = 0.95 * tsum[situation].value[thr_id];
if (cudaStreamQuery(stream) == cudaErrorNotReady)
{
usleep((useconds_t)(1e6*tsleep));
struct timeval tv_start, tv_end;
gettimeofday(&tv_start, NULL);
result = cudaStreamSynchronize(stream);
gettimeofday(&tv_end, NULL);
tsync = 1e-6 * (tv_end.tv_usec-tv_start.tv_usec) + (tv_end.tv_sec-tv_start.tv_sec);
}
if (tsync >= 0) tsum[situation].value[thr_id] = a * tsum[situation].value[thr_id] + b * (tsleep+tsync);
}
else
result = cudaStreamSynchronize(stream);
return result;
}
void cudaReportHardwareFailure(int thr_id, cudaError_t err, const char* func)
{
struct cgpu_info *gpu = &thr_info[thr_id].gpu;

5
miner.h

@ -39,9 +39,6 @@ extern "C" { @@ -39,9 +39,6 @@ extern "C" {
# include <malloc.h>
# define alloca _alloca
# elif !defined HAVE_ALLOCA
# ifdef __cplusplus
extern "C"
# endif
void *alloca (size_t);
# endif
#endif
@ -483,7 +480,7 @@ extern long device_sm[MAX_GPUS]; @@ -483,7 +480,7 @@ extern long device_sm[MAX_GPUS];
extern uint32_t gpus_intensity[MAX_GPUS];
extern int opt_cudaschedule;
// decl. from cuda.cpp
// cuda.cpp
int cuda_num_devices();
void cuda_devicenames();
void cuda_reset_device(int thr_id, bool *init);

3
nvml.cpp

@ -25,9 +25,6 @@ @@ -25,9 +25,6 @@
#include "nvml.h"
#include "cuda_runtime.h"
// cuda.cpp
int cuda_num_devices();
#ifdef USE_WRAPNVML
extern nvml_handle *hnvml;

Loading…
Cancel
Save