Merge pull request #1610 from fireice-uk/dev2.4.4

release 2.4.4

1 files changed, 10 insertions, 30 deletions

diff --git a/xmrstak/backend/nvidia/minethd.cpp b/xmrstak/backend/nvidia/minethd.cpp
index 92f5f78..16171e1 100644
--- a/xmrstak/backend/nvidia/minethd.cpp
+++ b/xmrstak/backend/nvidia/minethd.cpp
@@ -181,7 +181,7 @@ std::vector<iBackend*>* minethd::thread_starter(uint32_t threadOffset, miner_wor
 		}
 		else
 			printer::inst()->print_msg(L1, "Starting NVIDIA GPU thread %d, no affinity.", i);
-		
+
 		minethd* thd = new minethd(pWork, i + threadOffset, cfg);
 		pvThreads->push_back(thd);
 
@@ -195,27 +195,6 @@ std::vector<iBackend*>* minethd::thread_starter(uint32_t threadOffset, miner_wor
 	return pvThreads;
 }
 
-void minethd::switch_work(miner_work& pWork)
-{
-	// iConsumeCnt is a basic lock-like polling mechanism just in case we happen to push work
-	// faster than threads can consume them. This should never happen in real life.
-	// Pool cant physically send jobs faster than every 250ms or so due to net latency.
-
-	while (globalStates::inst().iConsumeCnt.load(std::memory_order_seq_cst) < globalStates::inst().iThreadCount)
-		std::this_thread::sleep_for(std::chrono::milliseconds(100));
-
-	globalStates::inst().oGlobalWork = pWork;
-	globalStates::inst().iConsumeCnt.store(0, std::memory_order_seq_cst);
-	globalStates::inst().iGlobalJobNo++;
-}
-
-void minethd::consume_work()
-{
-	memcpy(&oWork, &globalStates::inst().oGlobalWork, sizeof(miner_work));
-	iJobNo++;
-	globalStates::inst().iConsumeCnt++;
-}
-
 void minethd::work_main()
 {
 	if(affinity >= 0) //-1 means no affinity
@@ -223,7 +202,7 @@ void minethd::work_main()
 
 	if(cuda_get_deviceinfo(&ctx) != 0 || cryptonight_extra_cpu_init(&ctx) != 1)
 	{
-		printer::inst()->print_msg(L0, "Setup failed for GPU %d. Exitting.\n", (int)iThreadNo);
+		printer::inst()->print_msg(L0, "Setup failed for GPU %d. Exiting.\n", (int)iThreadNo);
 		std::exit(0);
 	}
 
@@ -237,14 +216,12 @@ void minethd::work_main()
 	uint64_t iCount = 0;
 	cryptonight_ctx* cpu_ctx;
 	cpu_ctx = cpu::minethd::minethd_alloc_ctx();
-	
+
 	// start with root algorithm and switch later if fork version is reached
 	auto miner_algo = ::jconf::inst()->GetCurrentCoinSelection().GetDescription(1).GetMiningAlgoRoot();
 	cn_hash_fun hash_fun = cpu::minethd::func_selector(::jconf::inst()->HaveHardwareAes(), true /*bNoPrefetch*/, miner_algo);
-	
-	uint32_t iNonce;
 
-	globalStates::inst().iConsumeCnt++;
+	uint32_t iNonce;
 
 	uint8_t version = 0;
 	size_t lastPoolId = 0;
@@ -261,7 +238,7 @@ void minethd::work_main()
 			while (globalStates::inst().iGlobalJobNo.load(std::memory_order_relaxed) == iJobNo)
 				std::this_thread::sleep_for(std::chrono::milliseconds(100));
 
-			consume_work();
+			globalStates::inst().consume_work(oWork, iJobNo);
 			continue;
 		}
 		uint8_t new_version = oWork.getVersion();
@@ -298,8 +275,11 @@ void minethd::work_main()
 			if((round_ctr++ & 0xF) == 0)
 			{
 				globalStates::inst().calc_start_nonce(iNonce, oWork.bNiceHash, h_per_round * 16);
+				// check if the job is still valid, there is a small posibility that the job is switched
+				if(globalStates::inst().iGlobalJobNo.load(std::memory_order_relaxed) != iJobNo)
+					break;
 			}
-			
+
 			uint32_t foundNonce[10];
 			uint32_t foundCount;
 
@@ -337,7 +317,7 @@ void minethd::work_main()
 			std::this_thread::yield();
 		}
 
-		consume_work();
+		globalStates::inst().consume_work(oWork, iJobNo);
 	}
 }