small speed improvement

firoorg · May 20, 2019 · 99951cf · 99951cf
1 parent 34962b8
commit 99951cf
Show file tree

Hide file tree

Showing 25 changed files with 373 additions and 1,531 deletions.
diff --git a/JHA/cuda_jha_compactionTest.cu b/JHA/cuda_jha_compactionTest.cu
@@ -5,7 +5,7 @@
 #include <sm_30_intrinsics.h>
 
 #ifdef __INTELLISENSE__
-#define __shfl_up(a,b)
+#define __shfl_up_sync(c,a,b)
 #endif
 
 static uint32_t *d_tempBranch1Nonces[MAX_GPUS];
@@ -128,7 +128,7 @@ void jackpot_compactTest_gpu_SCAN(uint32_t *data, int width, uint32_t *partial_s
 
 	for (int i=1; i<=width; i*=2)
 	{
-		uint32_t n = __shfl_up((int)value, i, width);
+		uint32_t n = __shfl_up_sync(0xFFFFFFFF, (int)value, i, width);
 
 		if (lane_id >= i) value += n;
 	}
@@ -155,7 +155,7 @@ void jackpot_compactTest_gpu_SCAN(uint32_t *data, int width, uint32_t *partial_s
 
 		for (int i=1; i<=width; i*=2)
 		{
-			uint32_t n = __shfl_up((int)warp_sum, i, width);
+			uint32_t n = __shfl_up_sync(0xFFFFFFFF, (int)warp_sum, i, width);
 
 		if (lane_id >= i) warp_sum += n;
 		}

diff --git a/RUN-ZCOIN-MTP.cmd b/RUN-ZCOIN-MTP.cmd
@@ -1,9 +1,9 @@
 
-x64\Release\ccminer -a mtp -o  http://127.0.0.1:8382   -u djm34 -p password --coinbase-addr aDn7MMYjVQqenT11VFDYHfFdwXmSTRUTak  --no-getwork -i 20
+rem x64\Release\ccminer -a mtp -o  http://127.0.0.1:8382   -u djm34 -p password --coinbase-addr aDn7MMYjVQqenT11VFDYHfFdwXmSTRUTak  --no-getwork -i 20
 rem x64\Release\ccminer -a mtp -o  stratum+tcp://xzc.2miners.com:8080  -u  aDn7MMYjVQqenT11VFDYHfFdwXmSTRUTak -p 0 -i 21 -d 0,1 
-rem x64\Release\ccminer -a mtp -o  stratum+tcp://zcoin.mintpond.com:3000 -u  aDn7MMYjVQqenT11VFDYHfFdwXmSTRUTak.worker  -p 0,strict,verbose,d=1000  -i 20 
+x64\Release\ccminer -a mtp -o  stratum+tcp://zcoin.mintpond.com:3000 -u  aDn7MMYjVQqenT11VFDYHfFdwXmSTRUTak.worker  -p 0,strict,verbose,d=500  -i 21 
 
-rem x64\Release\ccminer -a mtp -o   stratum+tcp://pool.bibop.net:4000  -u  aDn7MMYjVQqenT11VFDYHfFdwXmSTRUTak -p 0,c=XZC -i 20 -d 1 
+rem x64\Release\ccminer -a mtp -o   stratum+tcp://pool.bibop.net:4001  -u  aDn7MMYjVQqenT11VFDYHfFdwXmSTRUTak -p 0,c=XZC,d=10 -i 20 -d 1 
 
 
 pause
diff --git a/argon2ref/core.h b/argon2ref/core.h
@@ -240,4 +240,7 @@ void fill_segment(const argon2_instance_t *instance,
 int fill_memory_blocks(argon2_instance_t *instance);
 
 int fill_memory_blocks_mtp(argon2_instance_t *instance);
+
+
+
 #endif
diff --git a/ccminer.cpp b/ccminer.cpp
@@ -1214,15 +1214,17 @@ static bool submit_upstream_work_mtp(CURL *curl, struct work *work, struct mtp *
 			params = json_dumps(val, 0);
 			json_decref(val);
 
-			req = (char*)malloc(128 + 2 * 84 + strlen(work->txs) + strlen(params) + strlen(mtphashvalue_str) + strlen(mtpreserved_str)  + strlen(merkleroot_str)+ strlen(proofmtp_str) + strlen(blockmtp_str));
+			req = (char*)malloc(128 + 2 * 84 + strlen(work->txs) + strlen(params) + strlen(mtphashvalue_str) 
+									+ strlen(mtpreserved_str)  + strlen(merkleroot_str)+ strlen(proofmtp_str) + strlen(blockmtp_str));
 			sprintf(req,
 				"{\"method\": \"submitblock\", \"params\": [\"%s%s%s%s%s%s%s\", %s], \"id\":4}\r\n",
 				data_str, mtphashvalue_str, mtpreserved_str, merkleroot_str, blockmtp_str, proofmtp_str, work->txs, params);
 			free(params);
 
 		}
 		else {
-			req = (char*)malloc(128 + 2 * 84 + strlen(work->txs) + strlen(mtphashvalue_str) + strlen(mtpreserved_str)  + strlen(merkleroot_str) + strlen(proofmtp_str) + strlen(blockmtp_str) );
+			req = (char*)malloc(128 + 2 * 84 + strlen(work->txs) + strlen(mtphashvalue_str) + strlen(mtpreserved_str)  
+											 + strlen(merkleroot_str) + strlen(proofmtp_str) + strlen(blockmtp_str) );
 			sprintf(req,
 				"{\"method\": \"submitblock\", \"params\": [\"%s%s%s%s%s%s%s\"], \"id\":4}\r\n",
 				data_str, mtphashvalue_str, mtpreserved_str, merkleroot_str, blockmtp_str, proofmtp_str, work->txs);

diff --git a/ccminer.vcxproj b/ccminer.vcxproj
@@ -39,7 +39,7 @@
   </PropertyGroup>
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
   <ImportGroup Label="ExtensionSettings">
-    <Import Project="$(VCTargetsPath)\BuildCustomizations\CUDA 10.0.props" />
+    <Import Project="$(VCTargetsPath)\BuildCustomizations\CUDA 10.1.props" />
   </ImportGroup>
   <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
     <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
@@ -208,15 +208,15 @@
     </Link>
     <CudaCompile>
       <CInterleavedPTX>false</CInterleavedPTX>
-      <MaxRegCount>80</MaxRegCount>
-      <PtxAsOptionV>false</PtxAsOptionV>
+      <MaxRegCount>128</MaxRegCount>
+      <PtxAsOptionV>true</PtxAsOptionV>
       <Keep>true</Keep>
       <CodeGeneration>compute_52,sm_52;compute_61,sm_61;compute_75,sm_75</CodeGeneration>
       <Include>$(NVTOOLSEXT_PATH)\include;..\..\..\Common\C99</Include>
       <Optimization>O2</Optimization>
       <TargetMachinePlatform>64</TargetMachinePlatform>
-      <InterleaveSourceInPTX>false</InterleaveSourceInPTX>
-      <GenerateLineInfo>false</GenerateLineInfo>
+      <InterleaveSourceInPTX>true</InterleaveSourceInPTX>
+      <GenerateLineInfo>true</GenerateLineInfo>
     </CudaCompile>
     <CudaLink>
       <Optimization>O3</Optimization>
@@ -567,7 +567,7 @@
   </ItemGroup>
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
   <ImportGroup Label="ExtensionTargets">
-    <Import Project="$(VCTargetsPath)\BuildCustomizations\CUDA 10.0.targets" />
+    <Import Project="$(VCTargetsPath)\BuildCustomizations\CUDA 10.1.targets" />
   </ImportGroup>
   <!-- Copy the required dlls -->
   <Target Name="AfterBuild">