Skip to content

Commit

Permalink
nvml: last fixes for windows
Browse files Browse the repository at this point in the history 
note about x86 releases, x86 nvml.dll is not installed on Windows x64!
  • Loading branch information
@tpruvot
tpruvot committed Nov 13, 2014
1 parent 8d6b809 commit babaebc
Show file tree
Hide file tree
Showing 4 changed files with 59 additions and 87 deletions.
16 changes: 5 additions & 11 deletions api.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -114,8 +114,6 @@ extern uint32_t rejected_count;
static void gpustatus(int thr_id)
{
char buf[MYBUFSIZ];
float gt;
int gp, gf;

if (thr_id >= 0 && thr_id < gpu_threads) {
struct cgpu_info *cgpu = &thr_info[thr_id].gpu;
Expand All @@ -125,16 +123,12 @@ static void gpustatus(int thr_id)
#ifdef USE_WRAPNVML
// todo
if (1 || cgpu->has_monitoring) {
gf = gpu_fanpercent(cgpu);
gt = gpu_temp(cgpu);
gp = gpu_power(cgpu);
// gpu_clock(cgpu);
cgpu->gpu_temp = gpu_temp(cgpu);
cgpu->gpu_fan = gpu_fanpercent(cgpu);
cgpu->gpu_power = gpu_power(cgpu);
//cgpu->gpu_clock = gpu_clock(cgpu);
}
else
#endif
{
gt = 0.0; gf = gp = 0;
}

// todo: can be 0 if set by algo (auto)
if (opt_intensity == 0 && opt_work_size) {
Expand All @@ -155,7 +149,7 @@ static void gpustatus(int thr_id)

sprintf(buf, "GPU=%d;TEMP=%.1f;FAN=%d;POWER=%d;KHS=%.2f;"
"HWF=%d;I=%d|",
thr_id, gt, gf, gp, cgpu->khashes,
thr_id, cgpu->gpu_temp, cgpu->gpu_fan, cgpu->gpu_power, cgpu->khashes,
cgpu->hw_errors, cgpu->intensity);

strcat(buffer, buf);
Expand Down
22 changes: 6 additions & 16 deletions cpu-miner.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -56,21 +56,15 @@ BOOL WINAPI ConsoleHandler(DWORD);
#define HEAVYCOIN_BLKHDR_SZ 84
#define MNR_BLKHDR_SZ 80

// from cuda.cu
#ifdef __cplusplus
extern "C"
{
#endif
// from cuda.cpp
int cuda_num_devices();
void cuda_devicenames();
void cuda_devicereset();
int cuda_finddevice(char *name);
#ifdef __cplusplus
}
#endif

#ifdef USE_WRAPNVML
#include "nvml.h"
wrap_nvml_handle *hnvml = NULL;
#endif

#ifdef __linux /* Linux specific policy and affinity management */
Expand Down Expand Up @@ -247,10 +241,6 @@ uint32_t opt_work_size = 0; /* default */
char *opt_api_allow = "127.0.0.1"; /* 0.0.0.0 for all ips */
int opt_api_listen = 4068; /* 0 to disable */

#ifdef USE_WRAPNVML
wrap_nvml_handle *nvmlh = NULL;
#endif

#ifdef HAVE_GETOPT_LONG
#include <getopt.h>
#else
Expand Down Expand Up @@ -429,8 +419,8 @@ void proper_exit(int reason)
timeEndPeriod(1); // else never executed
#endif
#ifdef USE_WRAPNVML
if (nvmlh)
wrap_nvml_destroy(nvmlh);
if (hnvml)
wrap_nvml_destroy(hnvml);
#endif
exit(reason);
}
Expand Down Expand Up @@ -2140,8 +2130,8 @@ int main(int argc, char *argv[])
}

#ifdef USE_WRAPNVML
nvmlh = wrap_nvml_create();
if (nvmlh) {
hnvml = wrap_nvml_create();
if (hnvml) {
// todo: link threads info gpu
applog(LOG_INFO, "NVML GPU monitoring enabled.");
} else {
Expand Down
8 changes: 4 additions & 4 deletions cuda.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,7 @@ extern int device_map[8];
extern int device_sm[8];

// CUDA Devices on the System
extern "C" int cuda_num_devices()
int cuda_num_devices()
{
int version;
cudaError_t err = cudaDriverGetVersion(&version);
Expand All @@ -54,7 +54,7 @@ extern "C" int cuda_num_devices()
return GPU_N;
}

extern "C" void cuda_devicenames()
void cuda_devicenames()
{
cudaError_t err;
int GPU_N;
Expand All @@ -76,7 +76,7 @@ extern "C" void cuda_devicenames()
}

// Can't be called directly in cpu-miner.c
extern "C" void cuda_devicereset()
void cuda_devicereset()
{
cudaDeviceReset();
}
Expand All @@ -103,7 +103,7 @@ static bool substringsearch(const char *haystack, const char *needle, int &match
}

// CUDA Gerät nach Namen finden (gibt Geräte-Index zurück oder -1)
extern "C" int cuda_finddevice(char *name)
int cuda_finddevice(char *name)
{
int num = cuda_num_devices();
int match = 0;
Expand Down
100 changes: 44 additions & 56 deletions nvml.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,6 @@

#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#ifndef _MSC_VER
#include <libgen.h>
#endif
Expand All @@ -34,7 +33,12 @@
#if defined(_MSC_VER) || defined(_WIN32) || defined(_WIN64)
#include <windows.h>
static void *wrap_dlopen(const char *filename) {
return (void *)LoadLibrary(filename);
HMODULE h = LoadLibrary(filename);
if (!h && opt_debug) {
applog(LOG_DEBUG, "dlopen(%d): failed to load %s",
GetLastError(), filename);
}
return (void*)h;
}
static void *wrap_dlsym(void *h, const char *sym) {
return (void *)GetProcAddress((HINSTANCE)h, sym);
Expand All @@ -46,9 +50,16 @@
#else
/* assume we can use dlopen itself... */
#include <dlfcn.h>
#include <errno.h>
static void *wrap_dlopen(const char *filename) {
return dlopen(filename, RTLD_NOW);
uintptr_t h = dlopen(filename, RTLD_NOW);
if (!h && opt_debug) {
applog(LOG_DEBUG, "dlopen(%d): failed to load %s",
errno, filename);
}
return (void*)h;
}

static void *wrap_dlsym(void *h, const char *sym) {
return dlsym(h, sym);
}
Expand All @@ -66,51 +77,27 @@ wrap_nvml_handle * wrap_nvml_create()
int i=0;
wrap_nvml_handle *nvmlh = NULL;

/*
* We use hard-coded library installation locations for the time being...
* No idea where or if libnvidia-ml.so is installed on MacOS X, a
* deep scouring of the filesystem on one of the Mac CUDA build boxes
* I used turned up nothing, so for now it's not going to work on OSX.
*/
#if defined(_WIN64)
/* 64-bit Windows */
#define libnvidia_ml "%PROGRAMFILES%/NVIDIA Corporation/NVSMI/nvml.dll"
#elif defined(_WIN32) || defined(_MSC_VER)
/* 32-bit Windows */
#define libnvidia_ml "%PROGRAMFILES%/NVIDIA Corporation/NVSMI/nvml.dll"
#elif defined(__linux) && (defined(__i386__) || defined(__ARM_ARCH_7A__))
/* 32-bit linux assumed */
#define libnvidia_ml "/usr/lib32/libnvidia-ml.so"
#elif defined(__linux)
/* 64-bit linux assumed */
#define libnvidia_ml "/usr/lib/libnvidia-ml.so"
#if defined(WIN32)
/* Windows (do not use slashes, else ExpandEnvironmentStrings will mix them) */
#define libnvidia_ml "%PROGRAMFILES%\\NVIDIA Corporation\\NVSMI\\nvml.dll"
#else
#error "Unrecognized platform: need NVML DLL path for this platform..."
/* linux assumed */
#define libnvidia_ml "libnvidia-ml.so"
#endif

#if WIN32
char tmp[512];
ExpandEnvironmentStringsA(libnvidia_ml, tmp, sizeof(tmp));
#if WIN32
ExpandEnvironmentStrings(libnvidia_ml, tmp, sizeof(tmp));
#else
char tmp[512] = libnvidia_ml;
strcpy(tmp, libnvidia_ml);
#endif

void *nvml_dll = wrap_dlopen(tmp);
if (nvml_dll == NULL) {
#ifdef WIN32
char lib[] = "nvml.dll";
#else
char lib[64] = { '\0' };
snprintf(lib, sizeof(lib), "%s", basename(tmp));
/* try dlopen without path, here /usr/lib/nvidia-340/libnvidia-ml.so */
nvml_dll = wrap_dlopen("nvml.dll");
if (nvml_dll == NULL)
#endif
nvml_dll = wrap_dlopen(lib);
if (opt_debug)
applog(LOG_DEBUG, "dlopen: %s=%p", lib, nvml_dll);
}
if (nvml_dll == NULL) {
if (opt_debug)
applog(LOG_DEBUG, "dlopen(%d): failed to load %s", errno, tmp);
return NULL;
}

Expand All @@ -120,9 +107,10 @@ wrap_nvml_handle * wrap_nvml_create()

nvmlh->nvmlInit = (wrap_nvmlReturn_t (*)(void))
wrap_dlsym(nvmlh->nvml_dll, "nvmlInit_v2");
if (!nvmlh->nvmlInit)
if (!nvmlh->nvmlInit) {
nvmlh->nvmlInit = (wrap_nvmlReturn_t (*)(void))
wrap_dlsym(nvmlh->nvml_dll, "nvmlInit");
}
nvmlh->nvmlDeviceGetCount = (wrap_nvmlReturn_t (*)(int *))
wrap_dlsym(nvmlh->nvml_dll, "nvmlDeviceGetCount_v2");
nvmlh->nvmlDeviceGetHandleByIndex = (wrap_nvmlReturn_t (*)(int, wrap_nvmlDevice_t *))
Expand Down Expand Up @@ -153,11 +141,10 @@ wrap_nvml_handle * wrap_nvml_create()
nvmlh->nvmlDeviceGetPciInfo == NULL ||
nvmlh->nvmlDeviceGetName == NULL ||
nvmlh->nvmlDeviceGetTemperature == NULL ||
nvmlh->nvmlDeviceGetFanSpeed == NULL ||
nvmlh->nvmlDeviceGetPowerUsage == NULL)
nvmlh->nvmlDeviceGetFanSpeed == NULL)
{
if (opt_debug)
applog(LOG_DEBUG, "Failed to obtain all required NVML function pointers");
applog(LOG_DEBUG, "Failed to obtain required NVML function pointers");
wrap_dlclose(nvmlh->nvml_dll);
free(nvmlh);
return NULL;
Expand Down Expand Up @@ -342,24 +329,26 @@ int wrap_nvml_destroy(wrap_nvml_handle *nvmlh)

/* api functions */

extern wrap_nvml_handle *nvmlh;
extern int device_map[8];
extern "C" {
wrap_nvml_handle *hnvml;
int device_map[8];
}

unsigned int gpu_fanpercent(struct cgpu_info *gpu)
{
unsigned int pct = 0;
if (nvmlh) {
wrap_nvml_get_fanpcnt(nvmlh, device_map[gpu->thr_id], &pct);
if (hnvml) {
wrap_nvml_get_fanpcnt(hnvml, device_map[gpu->thr_id], &pct);
}
return pct;
}

double gpu_temp(struct cgpu_info *gpu)
{
double tc = 0.0;
if (nvmlh) {
if (hnvml) {
unsigned int tmp = 0;
wrap_nvml_get_tempC(nvmlh, device_map[gpu->thr_id], &tmp);
wrap_nvml_get_tempC(hnvml, device_map[gpu->thr_id], &tmp);
tc = (double) tmp;
}
return tc;
Expand All @@ -368,27 +357,26 @@ double gpu_temp(struct cgpu_info *gpu)
unsigned int gpu_clock(struct cgpu_info *gpu)
{
unsigned int freq = 0;
if (nvmlh) {
wrap_nvml_get_clock(nvmlh, device_map[gpu->thr_id], NVML_CLOCK_SM, &freq);
if (hnvml) {
wrap_nvml_get_clock(hnvml, device_map[gpu->thr_id], NVML_CLOCK_SM, &freq);
}
return freq;
}

unsigned int gpu_power(struct cgpu_info *gpu)
{
unsigned int mw = 0;
if (nvmlh) {
wrap_nvml_get_power_usage(nvmlh, device_map[gpu->thr_id], &mw);
if (hnvml) {
wrap_nvml_get_power_usage(hnvml, device_map[gpu->thr_id], &mw);
}
return mw;
}

int gpu_pstate(struct cgpu_info *gpu)
{
int pstate = 0;
if (nvmlh) {
wrap_nvml_get_pstate(nvmlh, device_map[gpu->thr_id], &pstate);
//gpu->gpu_pstate = pstate;
if (hnvml) {
wrap_nvml_get_pstate(hnvml, device_map[gpu->thr_id], &pstate);
}
return pstate;
}
Expand Down Expand Up @@ -433,7 +421,7 @@ int gpu_pstate(struct cgpu_info *gpu)
* nvmlDeviceGetFanSpeed
nvmlDeviceGetGpuOperationMode
nvmlDeviceGetHandleByIndex
nvmlDeviceGetHandleByIndex_v2
* nvmlDeviceGetHandleByIndex_v2
nvmlDeviceGetHandleByPciBusId
nvmlDeviceGetHandleByPciBusId_v2
nvmlDeviceGetHandleBySerial
Expand All @@ -450,7 +438,7 @@ int gpu_pstate(struct cgpu_info *gpu)
nvmlDeviceGetMinorNumber
nvmlDeviceGetMultiGpuBoard
nvmlDeviceGetName
nvmlDeviceGetPciInfo
* nvmlDeviceGetPciInfo
nvmlDeviceGetPciInfo_v2
* nvmlDeviceGetPerformanceState
nvmlDeviceGetPersistenceMode
Expand Down

0 comments on commit babaebc

Please sign in to comment.