GOSTSec
/
sgminer
mirror of https://github.com/GOSTSec/sgminer
1
0
Fork 0
Code Issues Releases Wiki Activity
Browse Source

Revert "Move the non cl_ variables into the cgpu info struct to allow creating a new cl state on reinit, preserving known GPU variables." 

This reverts commit 28880d0dc7.
nfactor-troky
Con Kolivas 13 years ago
parent
7c50bb02ab
commit
b1289a0159
4 changed files with 29 additions and 48 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 15
      main.c
  2. 5
      miner.h
  3. 51
      ocl.c
  4. 6
      ocl.h

15
main.c
Unescape View File

@ -2966,7 +2966,6 @@ static void *gpuminer_thread(void *userdata)
uint32_t *res, *blank_res; uint32_t *res, *blank_res;
double gpu_ms_average = 7; double gpu_ms_average = 7;
int gpu = dev_from_id(thr_id); int gpu = dev_from_id(thr_id);
struct cgpu_info *cgpu = mythr->cgpu;
size_t globalThreads[1]; size_t globalThreads[1];
size_t localThreads[1]; size_t localThreads[1];
@ -2978,7 +2977,7 @@ static void *gpuminer_thread(void *userdata)
struct work *work = make_work(); struct work *work = make_work();
unsigned int threads; unsigned int threads;
unsigned const int vectors = cgpu->vwidth; unsigned const int vectors = clState->preferred_vwidth;
unsigned int hashes; unsigned int hashes;
unsigned int hashes_done = 0; unsigned int hashes_done = 0;
@ -3015,7 +3014,7 @@ static void *gpuminer_thread(void *userdata)
} }
gettimeofday(&tv_start, NULL); gettimeofday(&tv_start, NULL);
localThreads[0] = cgpu->work_size; localThreads[0] = clState->work_size;
set_threads_hashes(vectors, &threads, &hashes, &globalThreads[0], set_threads_hashes(vectors, &threads, &hashes, &globalThreads[0],
localThreads[0]); localThreads[0]);
@ -3029,7 +3028,7 @@ static void *gpuminer_thread(void *userdata)
if (unlikely(status != CL_SUCCESS)) if (unlikely(status != CL_SUCCESS))
{ applog(LOG_ERR, "Error: clEnqueueWriteBuffer failed."); goto out; } { applog(LOG_ERR, "Error: clEnqueueWriteBuffer failed."); goto out; }
cgpu->status = LIFE_WELL; mythr->cgpu->status = LIFE_WELL;
if (opt_debug) if (opt_debug)
applog(LOG_DEBUG, "Popping ping in gpuminer thread"); applog(LOG_DEBUG, "Popping ping in gpuminer thread");
@ -3156,7 +3155,7 @@ static void *gpuminer_thread(void *userdata)
} }
if (unlikely(!gpu_devices[gpu])) { if (unlikely(!gpu_devices[gpu])) {
applog(LOG_WARNING, "Thread %d being disabled", thr_id); applog(LOG_WARNING, "Thread %d being disabled", thr_id);
mythr->rolling = cgpu->rolling = 0; mythr->rolling = mythr->cgpu->rolling = 0;
if (opt_debug) if (opt_debug)
applog(LOG_DEBUG, "Popping wakeup ping in gpuminer thread"); applog(LOG_DEBUG, "Popping wakeup ping in gpuminer thread");
@ -4047,15 +4046,13 @@ int main (int argc, char *argv[])
/* start GPU mining threads */ /* start GPU mining threads */
for (j = 0; j < nDevs * opt_g_threads; j++) { for (j = 0; j < nDevs * opt_g_threads; j++) {
int gpu = j % nDevs; int gpu = j % nDevs;
struct cgpu_info *cgpu;
gpus[gpu].is_gpu = 1; gpus[gpu].is_gpu = 1;
gpus[gpu].cpu_gpu = gpu; gpus[gpu].cpu_gpu = gpu;
thr = &thr_info[i]; thr = &thr_info[i];
thr->id = i; thr->id = i;
cgpu = &gpus[gpu]; thr->cgpu = &gpus[gpu];
thr->cgpu = cgpu;
thr->q = tq_new(); thr->q = tq_new();
if (!thr->q) if (!thr->q)
@ -4071,7 +4068,7 @@ int main (int argc, char *argv[])
} }
applog(LOG_INFO, "Init GPU thread %i", i); applog(LOG_INFO, "Init GPU thread %i", i);
clStates[i] = initCl(cgpu, name, sizeof(name)); clStates[i] = initCl(gpu, name, sizeof(name));
if (!clStates[i]) { if (!clStates[i]) {
applog(LOG_ERR, "Failed to init GPU thread %d", i); applog(LOG_ERR, "Failed to init GPU thread %d", i);
gpu_devices[i] = false; gpu_devices[i] = false;

5
miner.h
Unescape View File

@ -152,11 +152,6 @@ struct cgpu_info {
double efficiency; double efficiency;
double utility; double utility;
enum alive status; enum alive status;
int hasBitAlign;
unsigned int vwidth;
size_t max_work_size;
size_t work_size;
}; };
struct thr_info { struct thr_info {

51
ocl.c
Unescape View File

@ -267,16 +267,8 @@ void patch_opcodes(char *w, unsigned remaining)
_clState *initCQ(_clState *clState, unsigned int gpu) _clState *initCQ(_clState *clState, unsigned int gpu)
{ {
cl_int status = 0; cl_int status = 0;
cl_context_properties cps[3] = { CL_CONTEXT_PLATFORM, (cl_context_properties)platform, 0 };
clState->context = clCreateContextFromType(cps, CL_DEVICE_TYPE_GPU, NULL, NULL, &status);
if (status != CL_SUCCESS)
{
applog(LOG_ERR, "Error: Creating Context. (clCreateContextFromType)");
return NULL;
}
/* create a cl program executable for the device specified */ /* create a cl program executable for all the devices specified */
status = clBuildProgram(clState->program, 1, &devices[gpu], NULL, NULL, NULL); status = clBuildProgram(clState->program, 1, &devices[gpu], NULL, NULL, NULL);
if (status != CL_SUCCESS) if (status != CL_SUCCESS)
{ {
@ -320,9 +312,8 @@ _clState *initCQ(_clState *clState, unsigned int gpu)
return clState; return clState;
} }
_clState *initCl(struct cgpu_info *cgpu, char *name, size_t nameSize) _clState *initCl(unsigned int gpu, char *name, size_t nameSize)
{ {
unsigned int gpu = cgpu->cpu_gpu;
int patchbfi = 0; int patchbfi = 0;
cl_int status = 0; cl_int status = 0;
size_t nDevices; size_t nDevices;
@ -367,7 +358,7 @@ _clState *initCl(struct cgpu_info *cgpu, char *name, size_t nameSize)
} }
find = strstr(extensions, camo); find = strstr(extensions, camo);
if (find) if (find)
cgpu->hasBitAlign = patchbfi = 1; clState->hasBitAlign = patchbfi = 1;
status = clGetDeviceInfo(devices[gpu], CL_DEVICE_PREFERRED_VECTOR_WIDTH_INT, sizeof(cl_uint), (void *)&clState->preferred_vwidth, NULL); status = clGetDeviceInfo(devices[gpu], CL_DEVICE_PREFERRED_VECTOR_WIDTH_INT, sizeof(cl_uint), (void *)&clState->preferred_vwidth, NULL);
if (status != CL_SUCCESS) { if (status != CL_SUCCESS) {
@ -377,27 +368,26 @@ _clState *initCl(struct cgpu_info *cgpu, char *name, size_t nameSize)
if (opt_debug) if (opt_debug)
applog(LOG_DEBUG, "Preferred vector width reported %d", clState->preferred_vwidth); applog(LOG_DEBUG, "Preferred vector width reported %d", clState->preferred_vwidth);
status = clGetDeviceInfo(devices[gpu], CL_DEVICE_MAX_WORK_GROUP_SIZE, sizeof(size_t), (void *)&cgpu->max_work_size, NULL); status = clGetDeviceInfo(devices[gpu], CL_DEVICE_MAX_WORK_GROUP_SIZE, sizeof(size_t), (void *)&clState->max_work_size, NULL);
if (status != CL_SUCCESS) { if (status != CL_SUCCESS) {
applog(LOG_ERR, "Error: Failed to clGetDeviceInfo when trying to get CL_DEVICE_MAX_WORK_GROUP_SIZE"); applog(LOG_ERR, "Error: Failed to clGetDeviceInfo when trying to get CL_DEVICE_MAX_WORK_GROUP_SIZE");
return NULL; return NULL;
} }
if (opt_debug) if (opt_debug)
applog(LOG_DEBUG, "Max work group size reported %d", cgpu->max_work_size); applog(LOG_DEBUG, "Max work group size reported %d", clState->max_work_size);
/* For some reason 2 vectors is still better even if the card says /* For some reason 2 vectors is still better even if the card says
* otherwise, and many cards lie about their max so use 256 as max * otherwise, and many cards lie about their max so use 256 as max
* unless explicitly set on the command line */ * unless explicitly set on the command line */
cgpu->vwidth = clState->preferred_vwidth;
if (clState->preferred_vwidth > 1) if (clState->preferred_vwidth > 1)
cgpu->vwidth = 2; clState->preferred_vwidth = 2;
if (opt_vectors) if (opt_vectors)
cgpu->vwidth = opt_vectors; clState->preferred_vwidth = opt_vectors;
if (opt_worksize && opt_worksize <= cgpu->max_work_size) if (opt_worksize && opt_worksize <= clState->max_work_size)
cgpu->work_size = opt_worksize; clState->work_size = opt_worksize;
else else
cgpu->work_size = (cgpu->max_work_size <= 256 ? cgpu->max_work_size : 256) / clState->work_size = (clState->max_work_size <= 256 ? clState->max_work_size : 256) /
cgpu->vwidth; clState->preferred_vwidth;
/* Create binary filename based on parameters passed to opencl /* Create binary filename based on parameters passed to opencl
* compiler to ensure we only load a binary that matches what would * compiler to ensure we only load a binary that matches what would
@ -409,7 +399,7 @@ _clState *initCl(struct cgpu_info *cgpu, char *name, size_t nameSize)
char filename[16]; char filename[16];
if (chosen_kernel == KL_NONE) { if (chosen_kernel == KL_NONE) {
if (cgpu->hasBitAlign) if (clState->hasBitAlign)
chosen_kernel = KL_PHATK; chosen_kernel = KL_PHATK;
else else
chosen_kernel = KL_POCLBM; chosen_kernel = KL_POCLBM;
@ -452,14 +442,14 @@ _clState *initCl(struct cgpu_info *cgpu, char *name, size_t nameSize)
} }
strcat(binaryfilename, name); strcat(binaryfilename, name);
if (cgpu->hasBitAlign) if (clState->hasBitAlign)
strcat(binaryfilename, "bitalign"); strcat(binaryfilename, "bitalign");
strcat(binaryfilename, "v"); strcat(binaryfilename, "v");
sprintf(numbuf, "%d", cgpu->vwidth); sprintf(numbuf, "%d", clState->preferred_vwidth);
strcat(binaryfilename, numbuf); strcat(binaryfilename, numbuf);
strcat(binaryfilename, "w"); strcat(binaryfilename, "w");
sprintf(numbuf, "%d", (int)cgpu->work_size); sprintf(numbuf, "%d", (int)clState->work_size);
strcat(binaryfilename, numbuf); strcat(binaryfilename, numbuf);
strcat(binaryfilename, "long"); strcat(binaryfilename, "long");
sprintf(numbuf, "%d", (int)sizeof(long)); sprintf(numbuf, "%d", (int)sizeof(long));
@ -515,7 +505,7 @@ build:
memcpy(source, rawsource, pl); memcpy(source, rawsource, pl);
/* Patch the source file with the preferred_vwidth */ /* Patch the source file with the preferred_vwidth */
if (cgpu->vwidth > 1) { if (clState->preferred_vwidth > 1) {
char *find = strstr(source, "VECTORSX"); char *find = strstr(source, "VECTORSX");
if (unlikely(!find)) { if (unlikely(!find)) {
@ -523,7 +513,7 @@ build:
return NULL; return NULL;
} }
find += 7; // "VECTORS" find += 7; // "VECTORS"
if (cgpu->vwidth == 2) if (clState->preferred_vwidth == 2)
strncpy(find, "2", 1); strncpy(find, "2", 1);
else else
strncpy(find, "4", 1); strncpy(find, "4", 1);
@ -532,7 +522,7 @@ build:
} }
/* Patch the source file defining BITALIGN */ /* Patch the source file defining BITALIGN */
if (cgpu->hasBitAlign) { if (clState->hasBitAlign) {
char *find = strstr(source, "BITALIGNX"); char *find = strstr(source, "BITALIGNX");
if (unlikely(!find)) { if (unlikely(!find)) {
@ -690,11 +680,8 @@ built:
free(binaries); free(binaries);
free(binary_sizes); free(binary_sizes);
/* We throw everything out now and create the real context we're using in initCQ */
clReleaseContext(clState->context);
applog(LOG_INFO, "Initialising kernel %s with%s BFI_INT patching, %d vectors and worksize %d", applog(LOG_INFO, "Initialising kernel %s with%s BFI_INT patching, %d vectors and worksize %d",
filename, patchbfi ? "" : "out", cgpu->vwidth, cgpu->work_size); filename, patchbfi ? "" : "out", clState->preferred_vwidth, clState->work_size);
return initCQ(clState, gpu); return initCQ(clState, gpu);
} }

6
ocl.h
Unescape View File

@ -7,7 +7,6 @@
#else #else
#include <CL/cl.h> #include <CL/cl.h>
#endif #endif
#include "miner.h"
typedef struct { typedef struct {
cl_context context; cl_context context;
@ -15,13 +14,16 @@ typedef struct {
cl_command_queue commandQueue; cl_command_queue commandQueue;
cl_program program; cl_program program;
cl_mem outputBuffer; cl_mem outputBuffer;
int hasBitAlign;
cl_uint preferred_vwidth; cl_uint preferred_vwidth;
size_t max_work_size;
size_t work_size;
} _clState; } _clState;
extern char *file_contents(const char *filename, int *length); extern char *file_contents(const char *filename, int *length);
extern int clDevicesNum(); extern int clDevicesNum();
extern int preinit_devices(void); extern int preinit_devices(void);
extern _clState *initCQ(_clState *clState, unsigned int gpu); extern _clState *initCQ(_clState *clState, unsigned int gpu);
extern _clState *initCl(struct cgpu_info *cgpu, char *name, size_t nameSize); extern _clState *initCl(unsigned int gpu, char *name, size_t nameSize);
#endif /* HAVE_OPENCL */ #endif /* HAVE_OPENCL */
#endif /* __OCL_H__ */ #endif /* __OCL_H__ */

Write Preview
Loading…
Cancel
Save