Browse Source

Do away with the flaky free_work api in the driver code which would often lose the work data in opencl and simply flush it before exiting the opencl scanhash.

nfactor-troky
Con Kolivas 12 years ago
parent
commit
217be6ed98
  1. 2
      cgminer.c
  2. 63
      driver-opencl.c
  3. 1
      miner.h

2
cgminer.c

@ -5535,8 +5535,6 @@ void *miner_thread(void *userdata) @@ -5535,8 +5535,6 @@ void *miner_thread(void *userdata)
while (1) {
mythr->work_restart = false;
if (api->free_work && likely(work->pool))
api->free_work(mythr, work);
get_work(work, mythr, thr_id);
cgpu->new_work = true;

63
driver-opencl.c

@ -1455,17 +1455,6 @@ static bool opencl_thread_init(struct thr_info *thr) @@ -1455,17 +1455,6 @@ static bool opencl_thread_init(struct thr_info *thr)
return true;
}
static void opencl_free_work(struct thr_info *thr, struct work *work)
{
const int thr_id = thr->id;
struct opencl_thread_data *thrdata = thr->cgpu_data;
_clState *clState = clStates[thr_id];
clFinish(clState->commandQueue);
if (thrdata->res[FOUND])
thrdata->last_work = copy_work(work);
}
static bool opencl_prepare_work(struct thr_info __maybe_unused *thr, struct work *work)
{
@ -1495,9 +1484,6 @@ static int64_t opencl_scanhash(struct thr_info *thr, struct work *work, @@ -1495,9 +1484,6 @@ static int64_t opencl_scanhash(struct thr_info *thr, struct work *work,
size_t localThreads[1] = { clState->wsize };
int64_t hashes;
/* This finish flushes the readbuffer set with CL_FALSE later */
clFinish(clState->commandQueue);
/* Windows' timer resolution is only 15ms so oversample 5x */
if (gpu->dynamic && (++gpu->intervals * dynamic_us) > 70000) {
struct timeval tv_gpuend;
@ -1520,28 +1506,6 @@ static int64_t opencl_scanhash(struct thr_info *thr, struct work *work, @@ -1520,28 +1506,6 @@ static int64_t opencl_scanhash(struct thr_info *thr, struct work *work,
if (hashes > gpu->max_hashes)
gpu->max_hashes = hashes;
/* 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;
}
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);
free_work(thrdata->last_work);
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);
}
status = thrdata->queue_kernel_parameters(clState, &work->blk, globalThreads[0]);
if (unlikely(status != CL_SUCCESS)) {
applog(LOG_ERR, "Error: clSetKernelArg of all params failed.");
@ -1574,6 +1538,32 @@ static int64_t opencl_scanhash(struct thr_info *thr, struct work *work, @@ -1574,6 +1538,32 @@ static int64_t opencl_scanhash(struct thr_info *thr, struct work *work,
* 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;
}
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);
free_work(thrdata->last_work);
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);
/* This finish flushes the writebuffer set with CL_FALSE in clEnqueueWriteBuffer */
clFinish(clState->commandQueue);
}
return hashes;
}
@ -1599,7 +1589,6 @@ struct device_api opencl_api = { @@ -1599,7 +1589,6 @@ struct device_api opencl_api = {
.get_statline = get_opencl_statline,
.thread_prepare = opencl_thread_prepare,
.thread_init = opencl_thread_init,
.free_work = opencl_free_work,
.prepare_work = opencl_prepare_work,
.scanhash = opencl_scanhash,
.thread_shutdown = opencl_thread_shutdown,

1
miner.h

@ -268,7 +268,6 @@ struct device_api { @@ -268,7 +268,6 @@ struct device_api {
bool (*thread_prepare)(struct thr_info*);
uint64_t (*can_limit_work)(struct thr_info*);
bool (*thread_init)(struct thr_info*);
void (*free_work)(struct thr_info*, struct work*);
bool (*prepare_work)(struct thr_info*, struct work*);
int64_t (*scanhash)(struct thr_info*, struct work*, int64_t);
void (*hw_error)(struct thr_info*);

Loading…
Cancel
Save