Commit 217be6ed98c907c3217c5eb587e15f53ff30e3d9
2012-11-25T00:37:46
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.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127
diff --git a/cgminer.c b/cgminer.c
index 5654154..9b71b36 100644
--- a/cgminer.c
+++ b/cgminer.c
@@ -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;
diff --git a/driver-opencl.c b/driver-opencl.c
index 9dc03a9..2dd0454 100644
--- a/driver-opencl.c
+++ b/driver-opencl.c
@@ -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,
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,
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,
* 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 = {
.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,
diff --git a/miner.h b/miner.h
index 0a97e4a..ba174a8 100644
--- a/miner.h
+++ b/miner.h
@@ -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*);