Merge pull request fireice-uk#84 from fireice-uk/topic-autoconfig

CPU autoconfig

Author: fireice-uk

autoAdjust.hpp

@@ -0,0 +1,150 @@
+#pragma once
+#include "jconf.h"
+#include "console.h"
+
+#ifdef _WIN32
+#include <windows.h>
+#else
+#include <unistd.h>
+#endif // _WIN32
+
+// Mask bits between h and l and return the value
+// This enables us to put in values exactly like in the manual
+// For example EBX[31:22] is get_masked(cpu_info[1], 31, 22)
+inline int32_t get_masked(int32_t val, int32_t h, int32_t l)
+{
+	val &= (0x7FFFFFFF >> (31-(h-l))) << l;
+	return val >> l;
+}
+
+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");
+
+		if(!detectL3Size())
+			return;
+
+		if(L3KB_size < 1024 || L3KB_size > 102400)
+		{
+			printer::inst()->print_msg(L0, "Autoconf failed: L3 size sanity check failed - %u KB.", L3KB_size);
+			return;
+		}
+
+		printer::inst()->print_msg(L0, "Autoconf L3 size detected at %u KB.", L3KB_size);
+
+		detectCPUConf();
+
+		printer::inst()->print_msg(L0, "Autoconf core count detected as %u on %s.", corecnt,
+			linux_layout ? "Linux" : "Windows");
+
+		printer::inst()->print_str("\n**************** Copy&Paste ****************\n\n");
+		printer::inst()->print_str("\"cpu_threads_conf\" :\n[\n");
+
+		uint32_t aff_id = 0;
+		char strbuf[256];
+		for(uint32_t i=0; i < corecnt; i++)
+		{
+			bool double_mode;
+
+			if(L3KB_size <= 0)
+				break;
+
+			double_mode = L3KB_size / 2048 > (int32_t)(corecnt-i);
+
+			snprintf(strbuf, sizeof(strbuf), "   { \"low_power_mode\" : %s, \"no_prefetch\" : true, \"affine_to_cpu\" : %u },\n",
+				double_mode ? "true" : "false", aff_id);
+			printer::inst()->print_str(strbuf);
+
+			if(!linux_layout || old_amd)
+			{
+				aff_id += 2;
+
+				if(aff_id >= corecnt)
+					aff_id = 1;
+			}
+			else
+				aff_id++;
+
+			if(double_mode)
+				L3KB_size -= 4096;
+			else
+				L3KB_size -= 2048;
+		}
+
+		printer::inst()->print_str("]\n\n**************** Copy&Paste ****************\n");
+	}
+
+private:
+	bool detectL3Size()
+	{
+		int32_t cpu_info[4];
+		char cpustr[13] = {0};
+
+		jconf::cpuid(0, 0, cpu_info);
+		memcpy(cpustr, &cpu_info[1], 4);
+		memcpy(cpustr+4, &cpu_info[3], 4);
+		memcpy(cpustr+8, &cpu_info[2], 4);
+
+		if(strcmp(cpustr, "GenuineIntel") == 0)
+		{
+			jconf::cpuid(4, 3, cpu_info);
+
+			if(get_masked(cpu_info[0], 7, 5) != 3)
+			{
+				printer::inst()->print_msg(L0, "Autoconf failed: Couln't find L3 cache page.");
+				return false;
+			}
+
+			L3KB_size = ((get_masked(cpu_info[1], 31, 22) + 1) * (get_masked(cpu_info[1], 21, 12) + 1) *
+				(get_masked(cpu_info[1], 11, 0) + 1) * (cpu_info[2] + 1)) / 1024;
+
+			return true;
+		}
+		else if(strcmp(cpustr, "AuthenticAMD") == 0)
+		{
+			jconf::cpuid(0x80000006, 0, cpu_info);
+
+			L3KB_size = get_masked(cpu_info[3], 31, 18) * 512;
+
+			jconf::cpuid(1, 0, cpu_info);
+			if(get_masked(cpu_info[0], 11, 8) < 0x17) //0x17h is Zen
+				old_amd = true;
+
+			return true;
+		}
+		else
+		{
+			printer::inst()->print_msg(L0, "Autoconf failed: Unknown CPU type: %s.", cpustr);
+			return false;
+		}
+	}
+
+	void detectCPUConf()
+	{
+#ifdef _WIN32
+		SYSTEM_INFO info;
+		GetSystemInfo(&info);
+		corecnt = info.dwNumberOfProcessors;
+		linux_layout = false;
+#else
+		corecnt = sysconf(_SC_NPROCESSORS_ONLN);
+		linux_layout = true;
+#endif // _WIN32
+	}
+
+	int32_t L3KB_size = 0;
+	uint32_t corecnt;
+	bool old_amd = false;
+	bool linux_layout;
+};

cli-miner.cpp

@@ -26,6 +26,7 @@
 #include "jconf.h"
 #include "console.h"
 #include "donate-level.h"
+#include "autoAdjust.hpp"
 
 #ifndef CONF_NO_HTTPD
 #	include "httpd.h"
@@ -99,6 +100,14 @@ int main(int argc, char *argv[])
 		return 0;
 	}
 
+	if(jconf::inst()->NeedsAutoconf())
+	{
+		autoAdjust adjust;
+		adjust.printConfig();
+		win_exit();
+		return 0;
+	}
+
 	if (!minethd::self_test())
 	{
 		win_exit();

config.txt

@@ -1,10 +1,3 @@
-/* 
- * Number of threads. You can configure them below. Cryptonight uses 2MB of memory, so the optimal setting 
- * here is the size of your L3 cache divided by 2. Intel mid-to-high end desktop processors have 2MB of L3
- * cache per physical core. Low end cpus can have 1.5 or 1 MB while Xeons can have 2, 2.5 or 3MB per core.
- */
-"cpu_thread_num" : 2,
-
 /*
  * Thread configuration for each thread. Make sure it matches the number above.
  * low_power_mode - This mode will double the cache usage, and double the single thread performance. It will 
@@ -19,11 +12,18 @@
  *                  even or odd numbered cpu numbers. For Linux it will be usually the lower CPU numbers, so for a 4 
  *                  physical core CPU you should select cpu numbers 0-3.
  *
+ * On the first run the miner will look at your system and suggest a basic configuration that will work,
+ * you can try to tweak it from there to get the best performance.
+ * 
+ * A filled out configuration should look like this:
+ * "cpu_threads_conf" :
+ * [ 
+ *      { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 0 },
+ *      { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 1 },
+ * ],
  */
-"cpu_threads_conf" : [ 
-	{ "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 0 },
-	{ "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 1 },
-],
+"cpu_threads_conf" : 
+null,
 
 /*
  * LARGE PAGE SUPPORT

jconf.cpp

@@ -45,7 +45,7 @@ using namespace rapidjson;
 /*
  * This enum needs to match index in oConfigValues, otherwise we will get a runtime error
  */
-enum configEnum { iCpuThreadNum, aCpuThreadsConf, sUseSlowMem, bNiceHashMode,
+enum configEnum { aCpuThreadsConf, sUseSlowMem, bNiceHashMode,
 	bTlsMode, bTlsSecureAlgo, sTlsFingerprint, sPoolAddr, sWalletAddr, sPoolPwd,
 	iCallTimeout, iNetRetry, iGiveUpLimit, iVerboseLevel, iAutohashTime,
 	sOutputFile, iHttpdPort, bPreferIpv4 };
@@ -56,10 +56,10 @@ struct configVal {
 	Type iType;
 };
 
-//Same order as in configEnum, as per comment above
+// Same order as in configEnum, as per comment above
+// kNullType means any type
 configVal oConfigValues[] = {
-	{ iCpuThreadNum, "cpu_thread_num", kNumberType },
-	{ aCpuThreadsConf, "cpu_threads_conf", kArrayType },
+	{ aCpuThreadsConf, "cpu_threads_conf", kNullType },
 	{ sUseSlowMem, "use_slow_memory", kStringType },
 	{ bNiceHashMode, "nicehash_nonce", kTrueType },
 	{ bTlsMode, "use_tls", kTrueType },
@@ -84,6 +84,8 @@ inline bool checkType(Type have, Type want)
 {
 	if(want == have)
 		return true;
+	else if(want == kNullType)
+		return true;
 	else if(want == kTrueType && have == kFalseType)
 		return true;
 	else if(want == kFalseType && have == kTrueType)
@@ -111,6 +113,9 @@ jconf::jconf()
 
 bool jconf::GetThreadConfig(size_t id, thd_cfg &cfg)
 {
+	if(!prv->configValues[aCpuThreadsConf]->IsArray())
+		return false;
+
 	if(id >= prv->configValues[aCpuThreadsConf]->Size())
 		return false;
 
@@ -206,7 +211,15 @@ bool jconf::PreferIpv4()
 
 size_t jconf::GetThreadCount()
 {
-	return prv->configValues[aCpuThreadsConf]->Size();
+	if(prv->configValues[aCpuThreadsConf]->IsArray())
+		return prv->configValues[aCpuThreadsConf]->Size();
+	else
+		return 0;
+}
+
+bool jconf::NeedsAutoconf()
+{
+	return !prv->configValues[aCpuThreadsConf]->IsArray();
 }
 
 uint64_t jconf::GetCallTimeout()
@@ -249,38 +262,34 @@ const char* jconf::GetOutputFile()
 	return prv->configValues[sOutputFile]->GetString();
 }
 
+void jconf::cpuid(uint32_t eax, int32_t ecx, int32_t val[4])
+{
+	memset(val, 0, sizeof(int32_t)*4);
+
+#ifdef _WIN32
+	__cpuidex(val, eax, ecx);
+#else
+	__cpuid_count(eax, ecx, val[0], val[1], val[2], val[3]);
+#endif
+}
+
 bool jconf::check_cpu_features()
 {
 	constexpr int AESNI_BIT = 1 << 25;
 	constexpr int SSE2_BIT = 1 << 26;
 	constexpr int BMI2_BIT = 1 << 8;
-
-	int cpu_info[4];
+	int32_t cpu_info[4];
 	bool bHaveSse2;
 
-#ifdef _WIN32
-	__cpuid(cpu_info, 1);
-#else
-	__cpuid(1, cpu_info[0], cpu_info[1], cpu_info[2], cpu_info[3]);
-#endif
+	cpuid(1, 0, cpu_info);
 
 	bHaveAes = (cpu_info[2] & AESNI_BIT) != 0;
 	bHaveSse2 = (cpu_info[3] & SSE2_BIT) != 0;
 
-#ifdef _WIN32
-	__cpuidex(cpu_info, 7, 0);
-#else
-	__cpuid_count(7, 0, cpu_info[0], cpu_info[1], cpu_info[2], cpu_info[3]);
-#endif
+	cpuid(7, 0, cpu_info);
 
 	bHaveBmi2 = (cpu_info[1] & BMI2_BIT) != 0;
 
-	if(!bHaveAes)
-		printer::inst()->print_msg(L0, "Your CPU doesn't support hardware AES. Don't expect high hashrates.");
-
-	if(bHaveBmi2)
-		printer::inst()->print_msg(L0, "CPU supports BMI2 instructions. Faster multiplication enabled.");
-
 	return bHaveSse2;
 }
 
@@ -384,23 +393,8 @@ bool jconf::parse_config(const char* sFilename)
 		}
 	}
 
-	size_t n_thd = prv->configValues[aCpuThreadsConf]->Size();
-	if(prv->configValues[iCpuThreadNum]->GetUint64() != n_thd)
-	{
-		printer::inst()->print_msg(L0,
-			"Invalid config file. Your CPU config array has %llu members, while you want to use %llu threads.",
-			int_port(n_thd), int_port(prv->configValues[iCpuThreadNum]->GetUint64()));
-		return false;
-	}
-
-	if(NiceHashMode() && n_thd >= 32)
-	{
-		printer::inst()->print_msg(L0, "You need to use less than 32 threads in NiceHash mode.");
-		return false;
-	}
-
 	thd_cfg c;
-	for(size_t i=0; i < n_thd; i++)
+	for(size_t i=0; i < GetThreadCount(); i++)
 	{
 		if(!GetThreadConfig(i, c))
 		{
@@ -409,6 +403,12 @@ bool jconf::parse_config(const char* sFilename)
 		}
 	}
 
+	if(NiceHashMode() && GetThreadCount() >= 32)
+	{
+		printer::inst()->print_msg(L0, "You need to use less than 32 threads in NiceHash mode.");
+		return false;
+	}
+
 	if(GetSlowMemSetting() == unknown_value)
 	{
 		printer::inst()->print_msg(L0,
@@ -457,5 +457,15 @@ bool jconf::parse_config(const char* sFilename)
 #endif // _WIN32
 
 	printer::inst()->set_verbose_level(prv->configValues[iVerboseLevel]->GetUint64());
+
+	if(!NeedsAutoconf())
+	{
+		if(!bHaveAes)
+			printer::inst()->print_msg(L0, "Your CPU doesn't support hardware AES. Don't expect high hashrates.");
+
+		if(bHaveBmi2)
+			printer::inst()->print_msg(L0, "CPU supports BMI2 instructions. Faster multiplication enabled.");
+	}
+
 	return true;
 }

jconf.h

@@ -29,6 +29,7 @@ class jconf
 
 	size_t GetThreadCount();
 	bool GetThreadConfig(size_t id, thd_cfg &cfg);
+	bool NeedsAutoconf();
 
 	slow_mem_cfg GetSlowMemSetting();
 
@@ -58,6 +59,8 @@ class jconf
 	inline bool HaveHardwareAes() { return bHaveAes; }
 	inline bool HaveMulx() { return bHaveBmi2; }
 
+	static void cpuid(uint32_t eax, int32_t ecx, int32_t val[4]);
+
 private:
 	jconf();
 	static jconf* oInst;

xmr-stak-cpu.cbp

@@ -68,6 +68,7 @@
 			<Add library="crypto" />
 			<Add library="ssl" />
 		</Linker>
+		<Unit filename="autoAdjust.hpp" />
 		<Unit filename="cli-miner.cpp" />
 		<Unit filename="console.cpp" />
 		<Unit filename="console.h" />