Add nvml for GPU monitoring (squashed)

  Based on mwhite73 <marvin.white@gmail.com> implementation

  Linked to the api system

  Also fix Makefile to support standard c++ files
  This prevent nvcc use without device code

Signed-off-by: Tanguy Pruvot <tanguy.pruvot@gmail.com>

Makefile.am

@@ -1,8 +1,9 @@
+# allow to use Host cuda functions in C/C++
+DEF_INCLUDES = @CUDA_INCLUDES@
 
+JANSSON_INCLUDES=
 if WANT_JANSSON
 JANSSON_INCLUDES= -I$(top_srcdir)/compat/jansson
-else
-JANSSON_INCLUDES=
 endif
 
 EXTRA_DIST		= autogen.sh README.txt LICENSE.txt \
@@ -17,7 +18,7 @@ ccminer_SOURCES		= elist.h miner.h compat.h \
 			  compat/inttypes.h compat/stdbool.h compat/unistd.h \
 			  compat/sys/time.h compat/getopt/getopt.h \
 			  cpu-miner.c util.c crc32.c hefty1.c scrypt.c \
-			  api.c hashlog.cpp stats.cpp cuda.cu \
+			  api.cpp hashlog.cpp stats.cpp cuda.cpp \
 			  heavy/heavy.cu \
 			  heavy/cuda_blake512.cu heavy/cuda_blake512.h \
 			  heavy/cuda_combine.cu heavy/cuda_combine.h \
@@ -49,19 +50,25 @@ ccminer_SOURCES		= elist.h miner.h compat.h \
 			  x17/x17.cu x17/cuda_x17_haval512.cu x17/cuda_x17_sha512.cu \
 			  x11/s3.cu
 
+if HAVE_NVML
+ccminer_SOURCES += nvml.cpp
+nvml_defs = -DUSE_WRAPNVML
+nvml_libs = -ldl
+endif
+
 if HAVE_WINDOWS
 ccminer_SOURCES += compat/winansi.c
 endif
 
-ccminer_LDFLAGS		= $(PTHREAD_FLAGS) @CUDA_LDFLAGS@
-ccminer_LDADD		= @LIBCURL@ @JANSSON_LIBS@ @PTHREAD_LIBS@ @WS2_LIBS@ @CUDA_LIBS@ @OPENMP_CFLAGS@ @LIBS@
-ccminer_CPPFLAGS	= @LIBCURL_CPPFLAGS@ @OPENMP_CFLAGS@ $(CPPFLAGS) $(PTHREAD_FLAGS) -fno-strict-aliasing $(JANSSON_INCLUDES) -DSCRYPT_KECCAK512 -DSCRYPT_CHACHA -DSCRYPT_CHOOSE_COMPILETIME
+ccminer_LDFLAGS  = $(PTHREAD_FLAGS) @CUDA_LDFLAGS@
+ccminer_LDADD    = @LIBCURL@ @JANSSON_LIBS@ @PTHREAD_LIBS@ @WS2_LIBS@ @CUDA_LIBS@ @OPENMP_CFLAGS@ @LIBS@ $(nvml_libs)
+ccminer_CPPFLAGS = @LIBCURL_CPPFLAGS@ @OPENMP_CFLAGS@ $(CPPFLAGS) $(PTHREAD_FLAGS) -fno-strict-aliasing $(JANSSON_INCLUDES) $(DEF_INCLUDES) $(nvml_defs) -DSCRYPT_KECCAK512 -DSCRYPT_CHACHA -DSCRYPT_CHOOSE_COMPILETIME
 
 nvcc_ARCH  = -gencode=arch=compute_50,code=\"sm_50,compute_50\"
 #nvcc_ARCH += -gencode=arch=compute_35,code=\"sm_35,compute_35\"
 #nvcc_ARCH += -gencode=arch=compute_30,code=\"sm_30,compute_30\"
 
-nvcc_FLAGS = $(nvcc_ARCH) -I . @CUDA_CFLAGS@
+nvcc_FLAGS = $(nvcc_ARCH) @CUDA_INCLUDES@ -I. @CUDA_CFLAGS@
 nvcc_FLAGS += $(JANSSON_INCLUDES) --ptxas-options="-v"
 
 # we're now targeting all major compute architectures within one binary.

README.txt

@@ -1,5 +1,5 @@
 
-ccMiner release 1.4.8-tpruvot (12 Nov 2014) - "API Stats"
+ccMiner release 1.4.9-tpruvot (Nov 2014) - "GPU Monitoring"
 ---------------------------------------------------------------
 
 ***************************************************************
@@ -155,9 +155,13 @@ features.
 
 >>> RELEASE HISTORY <<<
 
+  Nov. 13th 2014  v1.4.9
+                  Add nvml unit to monitor nvidia cards (api)
+                  API: small changes, bump v1.1
+
   Nov. 12th 2014  v1.4.8
                   Add a basic API and sample php json wrapper
-		  Add statsavg (def 20) and api-bind parameters
+                  Add statsavg (def 20) and api-bind parameters
                   Fix displayed hashrate for multi gpus systems
 
   Nov. 11th 2014  v1.4.7

api.c → api.cpp

@@ -8,7 +8,7 @@
  * Software Foundation; either version 2 of the License, or (at your option)
  * any later version.  See COPYING for more details.
  */
-#define APIVERSION "1.0"
+#define APIVERSION "1.1"
 
 #ifdef _MSC_VER
 # define  _WINSOCK_DEPRECATED_NO_WARNINGS
@@ -35,6 +35,10 @@
 #include "compat.h"
 #include "miner.h"
 
+#ifdef USE_WRAPNVML
+#include "nvml.h"
+#endif
+
 #ifndef _MSC_VER
 # include <errno.h>
 # include <sys/socket.h>
@@ -105,25 +109,26 @@ extern uint32_t rejected_count;
 
 #define gpu_threads opt_n_threads
 
-extern void get_currentalgo(char* buf, int sz);
-
 /***************************************************************/
 
 static void gpustatus(int thr_id)
 {
 	char buf[MYBUFSIZ];
 	float gt;
-	int gf, gp;
+	int gp, gf;
 
 	if (thr_id >= 0 && thr_id < gpu_threads) {
 		struct cgpu_info *cgpu = &thr_info[thr_id].gpu;
 
-#ifdef HAVE_HWMONITORING
+		cgpu->thr_id = thr_id;
+
+#ifdef USE_WRAPNVML
 		// todo
-		if (gpu->has_monitoring) {
-			gt = gpu_temp(gpu);
-			gf = gpu_fanspeed(gpu);
-			gp = gpu_fanpercent(gpu);
+		if (1 || cgpu->has_monitoring) {
+			gf = gpu_fanpercent(cgpu);
+			gt = gpu_temp(cgpu);
+			gp = gpu_power(cgpu);
+			// gpu_clock(cgpu);
 		}
 		else
 #endif
@@ -148,7 +153,7 @@ static void gpustatus(int thr_id)
 
 		cgpu->khashes = stats_get_speed(thr_id) / 1000.0;
 
-		sprintf(buf, "GPU=%d;TEMP=%.1f;FAN=%d;FANP=%d;KHS=%.2f;"
+		sprintf(buf, "GPU=%d;TEMP=%.1f;FAN=%d;POWER=%d;KHS=%.2f;"
 			"HWF=%d;I=%d|",
 			thr_id, gt, gf, gp, cgpu->khashes,
 			cgpu->hw_errors, cgpu->intensity);
@@ -162,14 +167,14 @@ static void gpustatus(int thr_id)
 static char *getsummary(char *params)
 {
 	char algo[64] = "";
-	time_t uptime = (time(NULL) - startup);
-	double accps = (60.0 * accepted_count) / (uptime ? (uint32_t) uptime : 1.0);
+	double uptime = difftime(time(NULL), startup);
+	double accps = (60.0 * accepted_count) / (uptime ? uptime : 1.0);
 
 	get_currentalgo(algo, sizeof(algo));
 
 	*buffer = '\0';
 	sprintf(buffer, "NAME=%s;VER=%s;API=%s;"
-		"ALGO=%s;KHS=%.2f;ACC=%d;REJ=%d;ACCMN=%.3f;UPTIME=%d|",
+		"ALGO=%s;KHS=%.2f;ACC=%d;REJ=%d;ACCMN=%.3f;UPTIME=%.1f|",
 		PACKAGE_NAME, PACKAGE_VERSION, APIVERSION,
 		algo, (double)global_hashrate / 1000.0,
 		accepted_count, rejected_count,
@@ -186,7 +191,7 @@ static char *getstats(char *params)
 }
 
 struct CMDS {
-	char *name;
+	const char *name;
 	char *(*func)(char *);
 } cmds[] = {
 	{ "summary", getsummary },
@@ -195,15 +200,18 @@ struct CMDS {
 
 #define CMDMAX 2
 
-static void send_result(SOCKETTYPE c, char *result)
+static int send_result(SOCKETTYPE c, char *result)
 {
 	int n;
 
-	if (result == NULL)
-		result = "";
+	if (!result) {
+		n = send(c, "", 1, 0);
+	} else {
+		// ignore failure - it's closed immediately anyway
+		n = send(c, result, strlen(result) + 1, 0);
+	}
 
-	// ignore failure - it's closed immediately anyway
-	n = send(c, result, strlen(result) + 1, 0);
+	return n;
 }
 
 /*
@@ -400,7 +408,8 @@ static void api()
 			if ((time(NULL) - bindstart) > 61)
 				break;
 			else {
-				applog(LOG_ERR, "API bind to port %d failed - trying again in 15sec", port);
+				if (!opt_quiet || opt_debug)
+					applog(LOG_WARNING, "API bind to port %d failed - trying again in 15sec", port);
 				sleep(15);
 			}
 		}
@@ -409,7 +418,7 @@ static void api()
 	}
 
 	if (bound == 0) {
-		applog(LOG_ERR, "API bind to port %d failed (%s)%s", port, binderror, UNAVAILABLE);
+		applog(LOG_WARNING, "API bind to port %d failed (%s)%s", port, binderror, UNAVAILABLE);
 		free(apisock);
 		return;
 	}

ccminer.vcxproj

@@ -87,7 +87,7 @@
       <Optimization>Disabled</Optimization>
       <RuntimeLibrary>MultiThreadedDebugDLL</RuntimeLibrary>
       <FunctionLevelLinking>true</FunctionLevelLinking>
-      <PreprocessorDefinitions>WIN32;_DEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;CURL_STATICLIB;SCRYPT_KECCAK512;SCRYPT_CHACHA;SCRYPT_CHOOSE_COMPILETIME;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <PreprocessorDefinitions>WIN32;_DEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;CURL_STATICLIB;USE_WRAPNVML;SCRYPT_KECCAK512;SCRYPT_CHACHA;SCRYPT_CHOOSE_COMPILETIME;%(PreprocessorDefinitions)</PreprocessorDefinitions>
       <AdditionalIncludeDirectories>.;compat;compat\curl-for-windows\curl\include;compat\jansson;compat\getopt;compat\pthreads;compat\curl-for-windows\openssl\openssl\include;compat\curl-for-windows\zlib;%(AdditionalIncludeDirectories);$(CudaToolkitIncludeDir)</AdditionalIncludeDirectories>
       <EnableParallelCodeGeneration>true</EnableParallelCodeGeneration>
     </ClCompile>
@@ -114,7 +114,7 @@
       <Optimization>Disabled</Optimization>
       <RuntimeLibrary>MultiThreadedDebugDLL</RuntimeLibrary>
       <FunctionLevelLinking>true</FunctionLevelLinking>
-      <PreprocessorDefinitions>WIN32;_DEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;CURL_STATICLIB;SCRYPT_KECCAK512;SCRYPT_CHACHA;SCRYPT_CHOOSE_COMPILETIME;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <PreprocessorDefinitions>WIN32;_DEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;CURL_STATICLIB;USE_WRAPNVML;SCRYPT_KECCAK512;SCRYPT_CHACHA;SCRYPT_CHOOSE_COMPILETIME;%(PreprocessorDefinitions)</PreprocessorDefinitions>
       <AdditionalIncludeDirectories>.;compat;compat\curl-for-windows\curl\include;compat\jansson;compat\getopt;compat\pthreads;compat\curl-for-windows\openssl\openssl\include;compat\curl-for-windows\zlib;%(AdditionalIncludeDirectories);$(CudaToolkitIncludeDir)</AdditionalIncludeDirectories>
       <StructMemberAlignment>8Bytes</StructMemberAlignment>
       <EnableParallelCodeGeneration>true</EnableParallelCodeGeneration>
@@ -150,7 +150,7 @@
       <CompileAsManaged>false</CompileAsManaged>
       <FunctionLevelLinking>true</FunctionLevelLinking>
       <IntrinsicFunctions>true</IntrinsicFunctions>
-      <PreprocessorDefinitions>WIN32;NDEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;CURL_STATICLIB;SCRYPT_KECCAK512;SCRYPT_CHACHA;SCRYPT_CHOOSE_COMPILETIME;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <PreprocessorDefinitions>WIN32;NDEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;CURL_STATICLIB;USE_WRAPNVML;SCRYPT_KECCAK512;SCRYPT_CHACHA;SCRYPT_CHOOSE_COMPILETIME;%(PreprocessorDefinitions)</PreprocessorDefinitions>
       <AdditionalIncludeDirectories>.;compat;compat\curl-for-windows\curl\include;compat\jansson;compat\getopt;compat\pthreads;compat\curl-for-windows\openssl\openssl\include;compat\curl-for-windows\zlib;%(AdditionalIncludeDirectories);$(CudaToolkitIncludeDir)</AdditionalIncludeDirectories>
       <EnableParallelCodeGeneration>true</EnableParallelCodeGeneration>
       <ExceptionHandling>SyncCThrow</ExceptionHandling>
@@ -193,7 +193,7 @@
       <CompileAsManaged>false</CompileAsManaged>
       <FunctionLevelLinking>true</FunctionLevelLinking>
       <IntrinsicFunctions>true</IntrinsicFunctions>
-      <PreprocessorDefinitions>WIN32;NDEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;CURL_STATICLIB;SCRYPT_KECCAK512;SCRYPT_CHACHA;SCRYPT_CHOOSE_COMPILETIME;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <PreprocessorDefinitions>WIN32;NDEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;CURL_STATICLIB;USE_WRAPNVML;SCRYPT_KECCAK512;SCRYPT_CHACHA;SCRYPT_CHOOSE_COMPILETIME;%(PreprocessorDefinitions)</PreprocessorDefinitions>
       <AdditionalIncludeDirectories>.;compat;compat\curl-for-windows\curl\include;compat\jansson;compat\getopt;compat\pthreads;compat\curl-for-windows\openssl\openssl\include;compat\curl-for-windows\zlib;%(AdditionalIncludeDirectories);$(CudaToolkitIncludeDir)</AdditionalIncludeDirectories>
     </ClCompile>
     <Link>
@@ -240,9 +240,8 @@
     <ClCompile Include="groestlcoin.cpp" />
     <ClCompile Include="hashlog.cpp" />
     <ClCompile Include="stats.cpp" />
-    <ClCompile Include="api.c">
-      <AdditionalOptions>/Tp %(AdditionalOptions)</AdditionalOptions>
-    </ClCompile>
+    <ClCompile Include="nvml.cpp" />
+    <ClCompile Include="api.cpp" />
     <ClCompile Include="hefty1.c" />
     <ClCompile Include="myriadgroestl.cpp" />
     <ClCompile Include="scrypt.c">
@@ -321,7 +320,7 @@
     <ClInclude Include="uint256.h" />
   </ItemGroup>
   <ItemGroup>
-    <CudaCompile Include="cuda.cu" />
+    <CudaCompile Include="cuda.cpp" />
     <CudaCompile Include="bitslice_transformations_quad.cu">
       <ExcludedFromBuild>true</ExcludedFromBuild>
     </CudaCompile>

ccminer.vcxproj.filters

@@ -192,7 +192,10 @@
     <ClCompile Include="stats.cpp">
       <Filter>Source Files</Filter>
     </ClCompile>
-    <ClCompile Include="api.c">
+    <ClCompile Include="api.cpp">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+    <ClCompile Include="nvml.cpp">
       <Filter>Source Files</Filter>
     </ClCompile>
   </ItemGroup>
@@ -313,7 +316,7 @@
     </ClInclude>
   </ItemGroup>
   <ItemGroup>
-    <CudaCompile Include="cuda.cu">
+    <CudaCompile Include="cuda.cpp">
       <Filter>Source Files\CUDA</Filter>
     </CudaCompile>
     <CudaCompile Include="cuda_fugue256.cu">
@@ -482,4 +485,4 @@
       <Filter>Source Files\CUDA\x11</Filter>
     </CudaCompile>
   </ItemGroup>
-</Project>
+</Project>

configure.ac

@@ -1,4 +1,4 @@
-AC_INIT([ccminer], [1.4.8])
+AC_INIT([ccminer], [1.4.9])
 
 AC_PREREQ([2.59c])
 AC_CANONICAL_SYSTEM
@@ -142,19 +142,32 @@ dnl Setup CUDA paths
 AC_ARG_WITH([cuda],
    [  --with-cuda=PATH    prefix where cuda is installed [default=/usr/local/cuda]])
 
+AC_ARG_WITH([nvml],
+   [  --with-nvml=PATH    prefix where libnvml is installed [default=/usr/lib]])
+
+AM_CONDITIONAL([HAVE_NVML], [test -n "$with_nvml"])
+
 if test -n "$with_cuda"
 then
-   CUDA_CFLAGS="-I$with_cuda/include $CUDA_CFLAGS"
-   CUDA_LIBS="-lcudart"
-   CUDA_LDFLAGS="-L$with_cuda/lib$SUFFIX"
-   NVCC="$with_cuda/bin/nvcc"
+  CUDA_INCLUDES="-I$with_cuda/include"
+  CUDA_LIBS="-lcudart"
+  CUDA_LDFLAGS="-L$with_cuda/lib$SUFFIX"
+  NVCC="$with_cuda/bin/nvcc"
 else
-   CUDA_CFLAGS="-I/usr/local/cuda/include $CUDA_CFLAGS"
-   CUDA_LIBS="-lcudart -static-libstdc++"
-   CUDA_LDFLAGS="-L/usr/local/cuda/lib$SUFFIX"
-   NVCC="nvcc"
+  CUDA_INCLUDES="-I/usr/local/cuda/include"
+  CUDA_LIBS="-lcudart -static-libstdc++"
+  CUDA_LDFLAGS="-L/usr/local/cuda/lib$SUFFIX"
+  NVCC="nvcc"
 fi
+
+if test -n "$with_nvml" ; then
+  NVML_LIBPATH=$with_nvml
+  CUDA_LDFLAGS="$CUDA_LDFLAGS -ldl"
+fi
+AC_SUBST(NVML_LIBPATH)
+
 AC_SUBST(CUDA_CFLAGS)
+AC_SUBST(CUDA_INCLUDES)
 AC_SUBST(CUDA_LIBS)
 AC_SUBST(CUDA_LDFLAGS)
 AC_SUBST(NVCC)

configure.sh

@@ -7,5 +7,5 @@
 
 extracflags="-march=native -D_REENTRANT -falign-functions=16 -falign-jumps=16 -falign-labels=16"
 
-CUDA_CFLAGS="-O3 -Xcompiler -Wall" ./configure CXXFLAGS="-O3 $extracflags" --with-cuda=/usr/local/cuda
+CUDA_CFLAGS="-O3 -Xcompiler -Wall" ./configure CXXFLAGS="-O3 $extracflags" --with-cuda=/usr/local/cuda --with-nvml=libnvidia-ml.so