static int64_t opencl_scanhash(struct thr_info *thr, struct work *work,
int64_t __maybe_unused max_nonce)
{
const int thr_id = thr->id;
struct opencl_thread_data *thrdata = thr->cgpu_data;
struct cgpu_info *gpu = thr->cgpu;
_clState *clState = clStates[thr_id];
const cl_kernel *kernel = &clState->kernel;
const int dynamic_us = opt_dynamic_interval * 1000;
cl_int status;
size_t globalThreads[1];
size_t localThreads[1] = { clState->wsize };
int64_t hashes;
/* Windows' timer resolution is only 15ms so oversample 5x */
if (gpu->dynamic && (++gpu->intervals * dynamic_us) > 70000) {
struct timeval tv_gpuend;
double gpu_us;
gettimeofday(&tv_gpuend, NULL);
gpu_us = us_tdiff(&tv_gpuend, &gpu->tv_gpustart) / gpu->intervals;
if (gpu_us > dynamic_us) {
if (gpu->intensity > MIN_INTENSITY)
--gpu->intensity;
} else if (gpu_us < dynamic_us / 2) {
if (gpu->intensity < MAX_INTENSITY)
++gpu->intensity;
}
memcpy(&(gpu->tv_gpustart), &tv_gpuend, sizeof(struct timeval));
gpu->intervals = 0;
}
set_threads_hashes(clState->vwidth, &hashes, globalThreads, localThreads[0], &gpu->intensity);
if (hashes > gpu->max_hashes)
gpu->max_hashes = hashes;
status = thrdata->queue_kernel_parameters(clState, &work->blk, globalThreads[0]);
if (unlikely(status != CL_SUCCESS)) {
applog(LOG_ERR, "Error: clSetKernelArg of all params failed.");
return -1;
}
if (clState->goffset) {
size_t global_work_offset[1];
global_work_offset[0] = work->blk.nonce;
status = clEnqueueNDRangeKernel(clState->commandQueue, *kernel, 1, global_work_offset,
globalThreads, localThreads, 0, NULL, NULL);
} else
status = clEnqueueNDRangeKernel(clState->commandQueue, *kernel, 1, NULL,
globalThreads, localThreads, 0, NULL, NULL);
if (unlikely(status != CL_SUCCESS)) {
applog(LOG_ERR, "Error %d: Enqueueing kernel onto command queue. (clEnqueueNDRangeKernel)", status);
return -1;
}
status = clEnqueueReadBuffer(clState->commandQueue, clState->outputBuffer, CL_FALSE, 0,
BUFFERSIZE, thrdata->res, 0, NULL, NULL);
if (unlikely(status != CL_SUCCESS)) {
applog(LOG_ERR, "Error: clEnqueueReadBuffer failed error %d. (clEnqueueReadBuffer)", status);
return -1;
}
/* The amount of work scanned can fluctuate when intensity changes
* and since we do this one cycle behind, we increment the work more
* than enough to prevent repeating work */
work->blk.nonce += gpu->max_hashes;
/* This finish flushes the readbuffer set with CL_FALSE in clEnqueueReadBuffer */
clFinish(clState->commandQueue);
/* FOUND entry is used as a counter to say how many nonces exist */
if (thrdata->res[FOUND]) {
/* Clear the buffer again */
status = clEnqueueWriteBuffer(clState->commandQueue, clState->outputBuffer, CL_FALSE, 0,
BUFFERSIZE, blank_res, 0, NULL, NULL);
if (unlikely(status != CL_SUCCESS)) {
applog(LOG_ERR, "Error: clEnqueueWriteBuffer failed.");
return -1;
}
applog(LOG_DEBUG, "GPU %d found something?", gpu->device_id);
postcalc_hash_async(thr, work, thrdata->res);
memset(thrdata->res, 0, BUFFERSIZE);
/* This finish flushes the writebuffer set with CL_FALSE in clEnqueueWriteBuffer */
clFinish(clState->commandQueue);
}
return hashes;
}
static uint64_t opencl_scanhash(struct thr_info *thr, struct work *work,
uint64_t __maybe_unused max_nonce)
{
const int thr_id = thr->id;
struct opencl_thread_data *thrdata = thr->cgpu_data;
struct cgpu_info *gpu = thr->cgpu;
_clState *clState = clStates[thr_id];
const cl_kernel *kernel = &clState->kernel;
const int dynamic_us = opt_dynamic_interval * 1000;
cl_int status;
size_t globalThreads[1];
size_t localThreads[1] = { clState->wsize };
unsigned int threads;
unsigned int hashes;
/* This finish flushes the readbuffer set with CL_FALSE later */
clFinish(clState->commandQueue);
gettimeofday(&gpu->tv_gpuend, NULL);
if (gpu->dynamic) {
struct timeval diff;
suseconds_t gpu_us;
timersub(&gpu->tv_gpuend, &gpu->tv_gpustart, &diff);
gpu_us = diff.tv_sec * 1000000 + diff.tv_usec;
if (likely(gpu_us > 0)) {
gpu->gpu_us_average = (gpu->gpu_us_average + gpu_us * 0.63) / 1.63;
/* Try to not let the GPU be out for longer than
* opt_dynamic_interval in ms, but increase
* intensity when the system is idle in dynamic mode */
if (gpu->gpu_us_average > dynamic_us) {
if (gpu->intensity > MIN_INTENSITY)
--gpu->intensity;
} else if (gpu->gpu_us_average < dynamic_us / 2) {
if (gpu->intensity < MAX_INTENSITY)
++gpu->intensity;
}
}
}
set_threads_hashes(clState->vwidth, &threads, &hashes, globalThreads,
localThreads[0], gpu->intensity);
if (hashes > gpu->max_hashes)
gpu->max_hashes = hashes;
status = thrdata->queue_kernel_parameters(clState, &work->blk, globalThreads[0]);
if (unlikely(status != CL_SUCCESS)) {
applog(LOG_ERR, "Error: clSetKernelArg of all params failed.");
return 0;
}
/* MAXBUFFERS entry is used as a flag to say nonces exist */
if (thrdata->res[FOUND]) {
/* Clear the buffer again */
status = clEnqueueWriteBuffer(clState->commandQueue, clState->outputBuffer, CL_FALSE, 0,
BUFFERSIZE, blank_res, 0, NULL, NULL);
if (unlikely(status != CL_SUCCESS)) {
applog(LOG_ERR, "Error: clEnqueueWriteBuffer failed.");
return 0;
}
if (unlikely(thrdata->last_work)) {
applog(LOG_DEBUG, "GPU %d found something in last work?", gpu->device_id);
postcalc_hash_async(thr, thrdata->last_work, thrdata->res);
thrdata->last_work = NULL;
} else {
applog(LOG_DEBUG, "GPU %d found something?", gpu->device_id);
postcalc_hash_async(thr, work, thrdata->res);
}
memset(thrdata->res, 0, BUFFERSIZE);
clFinish(clState->commandQueue);
}
gettimeofday(&gpu->tv_gpustart, NULL);
if (clState->goffset) {
size_t global_work_offset[1];
global_work_offset[0] = work->blk.nonce;
status = clEnqueueNDRangeKernel(clState->commandQueue, *kernel, 1, global_work_offset,
globalThreads, localThreads, 0, NULL, NULL);
} else
status = clEnqueueNDRangeKernel(clState->commandQueue, *kernel, 1, NULL,
globalThreads, localThreads, 0, NULL, NULL);
if (unlikely(status != CL_SUCCESS)) {
applog(LOG_ERR, "Error: Enqueueing kernel onto command queue. (clEnqueueNDRangeKernel)");
return 0;
}
status = clEnqueueReadBuffer(clState->commandQueue, clState->outputBuffer, CL_FALSE, 0,
BUFFERSIZE, thrdata->res, 0, NULL, NULL);
if (unlikely(status != CL_SUCCESS)) {
applog(LOG_ERR, "Error: clEnqueueReadBuffer failed. (clEnqueueReadBuffer)");
return 0;
}
/* The amount of work scanned can fluctuate when intensity changes
* and since we do this one cycle behind, we increment the work more
* than enough to prevent repeating work */
work->blk.nonce += gpu->max_hashes;
return hashes;
}
