Merge pull request #108 from psychocrypt/topic-hwloc

hwloc autoadjust

4 files changed, 243 insertions, 4 deletions

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 2373d85..876d68d 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -78,6 +78,29 @@ else()
 endif()
 
 ################################################################################
+# Find hwloc
+################################################################################
+
+option(HWLOC_ENABLE "Enable or disable the requirement of hwloc" ON)
+if(HWLOC_ENABLE)
+    find_path(HWLOC_INCLUDE_DIR hwloc.h
+        $ENV{HWLOC_ROOT}/inlcude
+        /opt/local/include
+        /usr/local/include
+        /usr/include
+    )
+    find_library(HWLOC NAMES hwloc)
+    if("${HWLOC}" STREQUAL "MHTD-NOTFOUND" OR ${HWLOC_INCLUDE_DIR} STREQUAL "HWLOC_INCLUDE_DIR-NOTFOUND")
+        message(FATAL_ERROR "hwloc NOT found: use `-DHWLOC_ENABLE=OFF` to build without hwloc support")
+    else()
+        set(LIBS ${LIBS} ${HWLOC})
+        include_directories(AFTER ${HWLOC_INCLUDE_DIR})
+    endif()
+else()
+    add_definitions("-DCONF_NO_HWLOC")
+endif()
+
+################################################################################
 # Compile & Link
 ################################################################################
 
diff --git a/README.md b/README.md
index 4c202f6..977e779 100644
--- a/README.md
+++ b/README.md
@@ -46,7 +46,7 @@ xXl2Nm/u3cPP/eQVrZz5H8eACwIv+LL1EV+9uLanWUa+IO5hHr3KElvKNKD6vN0=
 
 ### GNU Compiler
 ```
-    sudo apt-get install libmicrohttpd-dev libssl-dev cmake build-essential
+    sudo apt-get install libmicrohttpd-dev libssl-dev cmake build-essential libhwloc-dev
     cmake .
     make install
 ```
@@ -171,9 +171,12 @@ and install.
 - `MICROHTTPD_ENABLE` allow to disable/enable the dependency *microhttpd*
   - by default enabled
   - there is no *http* interface available if option is disabled: `cmake . -DMICROHTTPD_ENABLE=OFF`
-- `OpenSSL_ENABLE`allow to disable/enable the dependency *OpenSSL*
+- `OpenSSL_ENABLE` allow to disable/enable the dependency *OpenSSL*
   - by default enabled
-  - it is not possible to connect to a *https* secured pool if optin is disabled: `cmake . -DOpenSSL_ENABLE=OFF`
+  - it is not possible to connect to a *https* secured pool if option is disabled: `cmake . -DOpenSSL_ENABLE=OFF`
+- `HWLOC_ENABLE` allow to disable/enable the dependency *hwloc*
+  - by default enabled
+  - the config suggestion is not optimal if option is disabled: `cmake . -DHWLOC_ENABLE=OFF`
 
 ## PGP Key
 ```
diff --git a/autoAdjustHwloc.hpp b/autoAdjustHwloc.hpp
new file mode 100644
index 0000000..07c3b4a
--- /dev/null
+++ b/autoAdjustHwloc.hpp
@@ -0,0 +1,209 @@
+#pragma once
+
+#include "console.h"
+#include <hwloc.h>
+
+#ifdef _WIN32
+#include <windows.h>
+#else
+#include <unistd.h>
+#endif // _WIN32
+
+class autoAdjust
+{
+public:
+
+	autoAdjust()
+	{
+	}
+
+	void printConfig()
+	{
+		printer::inst()->print_str("The configuration for 'cpu_threads_conf' in your config file is 'null'.\n");
+		printer::inst()->print_str("The miner evaluates your system and prints a suggestion for the section `cpu_threads_conf` to the terminal.\n");
+		printer::inst()->print_str("The values are not optimal, please try to tweak the values based on notes in config.txt.\n");
+		printer::inst()->print_str("Please copy & paste the block within the asterisks to your config.\n\n");
+		printer::inst()->print_str("\n**************** Copy&Paste ****************\n\n");
+		printer::inst()->print_str("\"cpu_threads_conf\" :\n[\n");
+
+		int depth;
+		hwloc_topology_t topology;
+		hwloc_obj_t socket;
+
+
+		hwloc_topology_init(&topology);
+		hwloc_topology_load(topology);
+
+		depth = hwloc_get_type_depth(topology, HWLOC_OBJ_SOCKET);
+		if (depth == HWLOC_TYPE_DEPTH_UNKNOWN)
+		{
+			printf("*** The number of sockets is unknown\n");
+		}
+
+		for (int i = 0; i < hwloc_get_nbobjs_by_depth(topology, depth); i++)
+		{
+			socket = hwloc_get_obj_by_depth(topology, depth, i);
+
+			// search cacheprinter::inst()->print_str("\n**************** Copy&Paste ****************\n\n");
+			for (int j = 0; j < socket->arity; j++)
+			{
+				hwloc_obj_t nextLvl = socket->children[j];
+				findCache(topology, nextLvl);
+			}
+		}
+
+		/* Destroy topology object. */
+		hwloc_topology_destroy(topology);
+		
+		printer::inst()->print_str("],\n\n**************** Copy&Paste ****************\n");
+	}
+
+private:
+
+	/** count bits with ones */
+	template<typename T>
+	int compat_popcnt(T x)
+	{
+	  int count = 0;
+	  for( int i = 0; i < sizeof(T) - 1; ++i )
+		if( ( ( x >> i ) & 1 ) == 1 )
+		  ++count;
+	  return count;
+	}
+
+	int rightZeros(size_t v)
+	{
+		int c;
+		if (v)
+		{
+			v = (v ^ (v - 1)) >> 1;
+			for (c = 0; v; c++)
+			{
+				v >>= 1;
+			}
+		}
+		else
+		{
+			c = CHAR_BIT * sizeof (v);
+		}
+		return c;
+	}
+
+	inline void getConfig(hwloc_topology_t topology, hwloc_obj_t obj, size_t& numHashes, int& numCachesLeft)
+	{
+		if (obj->type == HWLOC_OBJ_CORE)
+		{
+			if (obj)
+			{
+				hwloc_cpuset_t cpuset;
+				/* Get a copy of its cpuset that we may modify. */
+				cpuset = hwloc_bitmap_dup(obj->cpuset);
+				size_t allcpu = hwloc_bitmap_to_ulong(cpuset);
+				/* Get only one logical processor (in case the core is
+				   SMT/hyperthreaded). */
+				hwloc_bitmap_singlify(cpuset);
+
+
+				size_t cpu = hwloc_bitmap_to_ulong(cpuset);
+				// move bit mask to right to allow to compare always the first bit
+				cpu >>= rightZeros(allcpu);
+
+
+				int nativeCores = compat_popcnt(cpu);
+				int numPus = obj->arity;
+				for (int i = 0; i < numPus && numHashes != 0; i++)
+				{
+					hwloc_obj_t pu = obj->children[i];
+					// only use native pu's
+					if (pu->type == HWLOC_OBJ_PU && cpu & 1)
+					{
+						// if no cache is available we give each native core a hash
+						int numUnit = numCachesLeft != 0 ? numCachesLeft : nativeCores;
+
+						// two hashes per native pu if number of hashes if larger than compute units
+						int power = numHashes > numUnit ? 2 : 1;
+						char strbuf[256];
+						//printf("------------------------------core %i -> %i mpu=%i %lu\n", pu->os_index, power, nativeCores, numHashes);
+						snprintf(strbuf, sizeof(strbuf), "   { \"low_power_mode\" : %s, \"no_prefetch\" : true, \"affine_to_cpu\" : %u },\n",
+							power == 2 ? "true" : "false", pu->os_index);
+						printer::inst()->print_str(strbuf);
+
+						// update number of free hashes
+						numHashes -= power;
+						cpu >>= 1;
+
+						// one cache is filled with hashes
+						if (numCachesLeft != 0) numCachesLeft--;
+					}
+				}
+			}
+		}
+		else
+		{
+			for (int i = 0; i < obj->arity; i++)
+				getConfig(topology, obj->children[i], numHashes, numCachesLeft);
+		}
+	}
+
+	inline void findCache(hwloc_topology_t topology, hwloc_obj_t obj)
+	{
+		if (obj->type == HWLOC_OBJ_CACHE)
+		{
+			size_t cacheSize = obj->attr->cache.size;
+			size_t numHashL3 = ( cacheSize + m_scratchPadMemSize/ 2llu ) / m_scratchPadMemSize;
+
+			/* check cache is exclusive or inclusive */
+			const char* value = hwloc_obj_get_info_by_name(obj, "Inclusive");
+
+
+			bool doL3 = true;
+			if (value == NULL || value[0] != 49 || cacheSize == 0)
+			{
+				size_t numHashes = 0;
+				int numL2 = obj->arity;
+				for (int k = 0; k < numL2; k++)
+				{
+					hwloc_obj_t l3Cache = obj->children[k];
+					size_t l2Cache = 0;
+
+					if (obj->type == HWLOC_OBJ_CACHE)
+						l2Cache = l3Cache->attr->cache.size;
+					else
+						break;
+
+					if (l2Cache < m_scratchPadMemSize)
+					{
+						// we need to start from L3
+						break;
+					}
+
+					// start from L2
+
+					/* if more hashes available than objects in the current depth of the topology
+					 * than divide with round down else round up
+					 */
+					int extraHash = numHashL3 > numL2 ? numHashL3 / numL2 : (numHashL3 + numL2 - 1) / numL2;
+					numHashL3 -= extraHash;
+					if (numHashL3 < 0)
+						numHashL3 = 0;
+					numHashes += extraHash;
+					//add L2 hashes
+					numHashes += ( l2Cache + m_scratchPadMemSize / 2llu ) / m_scratchPadMemSize;
+					int numCachesLeft = numL2;
+					getConfig(topology, l3Cache, numHashes, numCachesLeft);
+					doL3 = false;
+				}
+			}
+			if (doL3)
+			{
+				int numCachesLeft = obj->arity;
+				getConfig(topology, obj, numHashL3, numCachesLeft);
+			}
+		}
+		else
+			for (int j = 0; j < obj->arity; j++)
+				findCache(topology, obj->children[j]);
+	}
+
+	static constexpr size_t m_scratchPadMemSize = ( 2llu * 1024llu * 1024llu );
+};
diff --git a/cli-miner.cpp b/cli-miner.cpp
index 70a5228..2cbe6d9 100644
--- a/cli-miner.cpp
+++ b/cli-miner.cpp
@@ -26,7 +26,11 @@
 #include "jconf.h"
 #include "console.h"
 #include "donate-level.h"
-#include "autoAdjust.hpp"
+#ifndef CONF_NO_HWLOC
+#   include "autoAdjustHwloc.hpp"
+#else
+#   include "autoAdjust.hpp"
+#endif
 #include "version.h"
 
 #ifndef CONF_NO_HTTPD