Detect dual GPU cards via the indirect information of - 1st card has a fan controller. 2nd card does not have a fan controller, cards share the same

device name, and are one bus ID apart. Use this information to create paired devices, displaying the fan speed on both devices and using the hottest
temperature from either GPU to control the overall fanspeed.

adl.c

@@ -94,7 +94,7 @@ static inline void unlock_adl(void)
 
 void init_adl(int nDevs)
 {
-	int i, j, devices = 0, last_adapter = -1, gpu = 0, dummy = 0;
+	int i, j, devices = 0, last_adapter = -1, gpu = 0, dummy = 0, prev_id = -1, prev_gpu = -1;
 
 #if defined (LINUX)
 	hDLL = dlopen( "libatiadlxx.so", RTLD_LAZY|RTLD_GLOBAL);
@@ -230,9 +230,11 @@ void init_adl(int nDevs)
 		/* From here on we know this device is a discrete device and
 		 * should support ADL */
 		ga = &gpus[gpu].adl;
+		ga->gpu = gpu;
 		ga->iAdapterIndex = iAdapterIndex;
 		ga->lpAdapterID = lpAdapterID;
 		ga->DefPerfLev = NULL;
+		ga->twin = NULL;
 
 		ga->lpThermalControllerInfo.iSize=sizeof(ADLThermalControllerInfo);
 		if (ADL_Overdrive5_ThermalDevices_Enum(iAdapterIndex, 0, &ga->lpThermalControllerInfo) != ADL_OK)
@@ -306,8 +308,21 @@ void init_adl(int nDevs)
 		ga->iMemoryClock = lpOdPerformanceLevels->aLevels[lev].iMemoryClock;
 		ga->iVddc = lpOdPerformanceLevels->aLevels[lev].iVddc;
 
-		if (ADL_Overdrive5_FanSpeedInfo_Get(iAdapterIndex, 0, &ga->lpFanSpeedInfo) != ADL_OK)
+		if (ADL_Overdrive5_FanSpeedInfo_Get(iAdapterIndex, 0, &ga->lpFanSpeedInfo) != ADL_OK) {
 			applog(LOG_INFO, "Failed to ADL_Overdrive5_FanSpeedInfo_Get");
+			/* This is our opportunity to detect the 2nd GPU in a
+			 * dual GPU device with a fan controller only on the
+			 * first */
+			if (prev_id >= 0 &&
+			    !strcmp(lpInfo[i].strAdapterName, lpInfo[prev_id].strAdapterName) &&
+			    lpInfo[i].iBusNumber == lpInfo[prev_id].iBusNumber + 1 &&
+			    gpus[prev_gpu].adl.has_fanspeed) {
+				applog(LOG_INFO, "2nd GPU of dual card device detected");
+				ga->twin = &gpus[prev_gpu].adl;
+				gpus[prev_gpu].adl.twin = ga;
+			}
+		} else
+			ga->has_fanspeed = true;
 
 		/* Save the fanspeed values as defaults in case we reset later */
 		ADL_Overdrive5_FanSpeed_Get(ga->iAdapterIndex, 0, &ga->DefFanSpeedValue);
@@ -343,6 +358,9 @@ void init_adl(int nDevs)
 			ga->managed = true;
 		}
 		ga->lasttemp = __gpu_temp(ga);
+
+		prev_id = i;
+		prev_gpu = gpu;
 	}
 }
 
@@ -465,6 +483,9 @@ int gpu_activity(int gpu)
 
 static inline int __gpu_fanspeed(struct gpu_adl *ga)
 {
+	if (!ga->has_fanspeed && ga->twin)
+		return __gpu_fanspeed(ga->twin);
+
 	if (!(ga->lpFanSpeedInfo.iFlags & ADL_DL_FANCTRL_SUPPORTS_RPM_READ))
 		return -1;
 	ga->lpFanSpeedValue.iSpeedType = ADL_DL_FANCTRL_SPEED_TYPE_RPM;
@@ -488,8 +509,11 @@ int gpu_fanspeed(int gpu)
 	return ret;
 }
 
-static inline int __gpu_fanpercent(struct gpu_adl *ga)
+static int __gpu_fanpercent(struct gpu_adl *ga)
 {
+	if (!ga->has_fanspeed && ga->twin)
+		return __gpu_fanpercent(ga->twin);
+
 	if (!(ga->lpFanSpeedInfo.iFlags & ADL_DL_FANCTRL_SUPPORTS_PERCENT_READ ))
 		return -1;
 	ga->lpFanSpeedValue.iSpeedType = ADL_DL_FANCTRL_SPEED_TYPE_PERCENT;
@@ -819,27 +843,13 @@ static int set_powertune(int gpu, int iPercentage)
 	return ret;
 }
 
-void gpu_autotune(int gpu, bool *enable)
+static void fan_autotune(int gpu, int temp, int fanpercent, bool *fan_optimal)
 {
-	int temp, fanpercent, engine, newpercent, newengine;
-	bool fan_optimal = true;
-	struct cgpu_info *cgpu;
-	struct gpu_adl *ga;
-
-	cgpu = &gpus[gpu];
-	ga = &cgpu->adl;
-
-	lock_adl();
-	ADL_Overdrive5_CurrentActivity_Get(ga->iAdapterIndex, &ga->lpActivity);
-	temp = __gpu_temp(ga);
-	newpercent = fanpercent = __gpu_fanpercent(ga);
-	unlock_adl();
-
-	newengine = engine = gpu_engineclock(gpu) * 100;
-
-	if (temp && fanpercent >= 0 && ga->autofan) {
+	struct cgpu_info *cgpu = &gpus[gpu];
+	struct gpu_adl *ga = &cgpu->adl;
 	int top = gpus[gpu].gpu_fan;
 	int bot = gpus[gpu].min_fan;
+	int newpercent = fanpercent;
 	int iMin = 0, iMax = 100;
 
 	get_fanrange(gpu, &iMin, &iMax);
@@ -872,6 +882,44 @@ void gpu_autotune(int gpu, bool *enable)
 	}
 }
 
+void gpu_autotune(int gpu, bool *enable)
+{
+	int temp, fanpercent, engine, newengine, twintemp;
+	bool fan_optimal = true;
+	struct cgpu_info *cgpu;
+	struct gpu_adl *ga;
+
+	cgpu = &gpus[gpu];
+	ga = &cgpu->adl;
+
+	lock_adl();
+	ADL_Overdrive5_CurrentActivity_Get(ga->iAdapterIndex, &ga->lpActivity);
+	temp = __gpu_temp(ga);
+	if (ga->twin)
+		twintemp = __gpu_temp(ga->twin);
+	fanpercent = __gpu_fanpercent(ga);
+	unlock_adl();
+
+	newengine = engine = gpu_engineclock(gpu) * 100;
+
+	if (temp && fanpercent >= 0 && ga->autofan) {
+		if (!ga->twin)
+			fan_autotune(gpu, temp, fanpercent, &fan_optimal);
+		else {
+			int hightemp, fan_gpu;
+
+			if (twintemp > temp)
+				hightemp = twintemp;
+			else
+				hightemp = temp;
+			if (ga->has_fanspeed)
+				fan_gpu = gpu;
+			else
+				fan_gpu = ga->twin->gpu;
+			fan_autotune(fan_gpu, hightemp, fanpercent, &fan_optimal);
+		}
+	}
+
 	if (engine && ga->autoengine) {
 		if (temp > ga->cutofftemp) {
 			applog(LOG_WARNING, "Hit thermal cutoff limit on GPU %d, disabling!", gpu);

miner.h

@@ -190,6 +190,7 @@ struct strategies {
 	const char *s;
 };
 
+struct cgpu_info;
 
 #ifdef HAVE_ADL
 struct gpu_adl {
@@ -221,10 +222,13 @@ struct gpu_adl {
 	int cutofftemp;
 	int minspeed;
 	int maxspeed;
+
+	int gpu;
+	bool has_fanspeed;
+	struct gpu_adl *twin;
 };
 #endif
 
-struct cgpu_info;
 struct thr_info;
 struct work;
 
@@ -276,6 +280,7 @@ struct cgpu_info {
 #ifdef HAVE_ADL
 	bool has_adl;
 	struct gpu_adl adl;
+
 	int gpu_engine;
 	int min_engine;
 	int gpu_fan;