Merge branch 'master' of github.com:RRZE-HPC/likwid

Makefile

@@ -617,7 +617,7 @@ install: install_daemon install_freq install_appdaemon install_container_helper
 	@mkdir -p $(PREFIX)/include
 	@chmod 755 $(PREFIX)/include
 	@install -m 644 $(SRC_DIR)/includes/likwid.h  $(PREFIX)/include/
-	@sed -i -e "s#<VERSION>#$(VERSION)#g" -e "s#<DATE>#$(DATE)#g" -e "s#<GITCOMMIT>#$(GITCOMMIT)#g" -e "s#<MINOR>#$(MINOR)#g" $(PREFIX)/include/likwid.h
+	@sed -i -e "s#<VERSION>#$(VERSION)#g" -e "s#<DATE>#$(DATE)#g" -e "s#<GITCOMMIT>#$(GITCOMMIT)#g" -e "s#<MINOR>#$(MINOR)#g" -e "s#VERSION.RELEASE.MINORVERSION#$(VERSION).$(RELEASE).$(MINOR)#g" -e "s#LIKWID_COMMIT GITCOMMIT#LIKWID_COMMIT \"$(GITCOMMIT)\"#g" $(PREFIX)/include/likwid.h
 	@install -m 644 $(SRC_DIR)/includes/likwid-marker.h  $(PREFIX)/include/
 	$(FORTRAN_INSTALL)
 	@echo "===> INSTALL groups to $(PREFIX)/share/likwid/perfgroups"

src/applications/likwid-sysfeatures.lua

@@ -61,6 +61,8 @@ local function usage()
     print_stdout("-s, --set <feature(s)>\t Set feature(s) to the given value")
     print_stdout("                      \t format: <category>.<name>=<value> or just <name>=<value> if unique")
     print_stdout("                      \t can be a comma-separated list of features")
+    print_stdout("    --saveall <file>  \t Save all available rw-features to file")
+    print_stdout("    --loadall <file>  \t Load all features from file")
     print_stdout("-O\t\t\t Output results in CSV")
     print_stdout("-V, --verbose <level>\t Set verbosity\n")
 end
@@ -121,7 +123,7 @@ local verbose = 0
 local output_csv = false
 
 -- parse the command line
-for opt,arg in likwid.getopt(arg, {"h","v","l","p","d:","g:","s:","a", "O","help","version","list", "print", "set:", "get:","all", "cpus:", "V:", "verbose:"}) do
+for opt,arg in likwid.getopt(arg, {"h","v","l","p","d:","g:","s:","a", "O","help","version","list", "print", "set:", "get:","all", "cpus:", "V:", "verbose:", "saveall:", "loadall:"}) do
     if (type(arg) == "string") then
         local s,e = arg:find("-");
         if s == 1 then
@@ -152,6 +154,10 @@ for opt,arg in likwid.getopt(arg, {"h","v","l","p","d:","g:","s:","a", "O","help
         getList = likwid.stringsplit(arg, ",")
     elseif opt == "s" or opt == "set" then
         setList = likwid.stringsplit(arg, ",")
+    elseif opt == "saveall" then
+        saveFeatures = arg
+    elseif opt == "loadall" then
+        loadFeatures = arg
     elseif opt == "?" then
         print_stderr("Invalid commandline option -"..arg)
         os.exit(1)
@@ -163,12 +169,12 @@ end
 
 
 -- validate command line input
-if (not printDevices) and (not listFeatures) and (not allFeatures) and (#getList == 0) and (#setList == 0) then
+if (not printDevices) and (not listFeatures) and (not allFeatures) and (not saveFeatures) and (not loadFeatures) and (#getList == 0) and (#setList == 0) then
     print_stderr("No operations specified, exiting...")
     os.exit(1)
 end
-if (printDevices or listFeatures or allFeatures) and (#getList > 0 or #setList > 0) then
-    print_stderr("Cannot list features and get/set at the same time")
+if (printDevices or listFeatures or allFeatures or saveFeatures or loadFeatures) and (#getList > 0 or #setList > 0) then
+    print_stderr("Cannot list features and get/set/load/save at the same time")
     os.exit(1)
 end
 if #devList == 0 then
@@ -184,12 +190,6 @@ if #devList == 0 then
     end
 end
 
---[[local access_mode = likwid.getAccessClientMode()
-if access_mode < 0 or access_mode > 1 then
-    print_stderr("Manipulation of HW features only for access mode 'direct' or 'accessdaemon'")
-    os.exit(1)
-end]]
-
 -- set verbosity
 if verbose > 0 and verbose <= 3 then
     likwid.setVerbosity(verbose)
@@ -206,20 +206,21 @@ end
 local ft_list = likwid.sysFeatures_list()
 
 -- print available devices
+device_types = {}
+device_types[likwid.hwthread] = "HWThread (T)"
+device_types[likwid.core] = "Core (C)"
+device_types[likwid.numa] = "NUMA (M)"
+device_types[likwid.die] = "Die (D)"
+device_types[likwid.socket] = "Socket (S)"
+device_types[likwid.node] = "Node (N)"
+if likwid.nvSupported() then
+    device_types[likwid.nvidia_gpu] = "Nvidia GPU (GN)"
+end
+if likwid.rocmSupported() then
+    device_types[likwid.amd_gpu] = "AMD GPU (GA)"
+end
+
 if printDevices then
-    device_types = {}
-    device_types[likwid.hwthread] = "HWThread (T)"
-    device_types[likwid.core] = "Core (C)"
-    device_types[likwid.numa] = "NUMA (M)"
-    device_types[likwid.die] = "Die (D)"
-    device_types[likwid.socket] = "Socket (S)"
-    device_types[likwid.node] = "Node (N)"
-    if likwid.nvSupported() then
-        device_types[likwid.nvidia_gpu] = "Nvidia GPU (GN)"
-    end
-    if likwid.rocmSupported() then
-        device_types[likwid.amd_gpu] = "AMD GPU (GA)"
-    end
     for devtype, name in pairs(device_types) do
         print(string.format("%s:", name))
         devices = likwid.getAvailableDevices(devtype)
@@ -390,5 +391,88 @@ if #setList > 0 and #devList > 0 then
     os.exit(0)
 end
 
+-- save all read/write features to file
+if saveFeatures then
+    local file = io.open(saveFeatures, "w")
+    -- iterate over all device types
+    for devtype, _ in pairs(device_types) do
+        -- iterate over all features
+        for _,f in pairs(ft_list) do
+            local full_name = string.format("%s.%s", f.Category, f.Name)
+            -- only allow matching device types and if feature is readable and writable
+            if f.TypeID ~= devtype or f.ReadOnly or f.WriteOnly then
+                goto next_feat
+            end
+
+            -- actually read the features
+            for _, dev in pairs(likwid.getAllDevices(devtype)) do
+                local lw_dev = likwid.createDevice(devtype, dev)
+                local v, err = likwid.sysFeatures_get(full_name, lw_dev)
+                if err then
+                    print_stderr(string.format("Failed to get feature '%s' on device %s:%d: %s", full_name, lw_dev:typeName(), lw_dev:id(), err))
+                    goto next_feat
+                end
+                file:write(string.format("%s.%s@%s=%s\n", f.Category, f.Name, lw_dev:id(), v))
+            end
+
+            ::next_feat::
+        end
+    end
+
+    file:close()
+
+    likwid.finalizeSysFeatures()
+    os.exit(0)
+end
+
+-- load all features from file
+if loadFeatures then
+    for line in io.lines(loadFeatures) do
+        -- split string like the following: cpu_freq.governor@5=schedutil
+        local part1 = likwid.stringsplit(line, "=")
+        if #part1 ~= 2 then
+            print_stderr("Invalid line: " .. line)
+            os.exit(1)
+        end
+        local part2 = likwid.stringsplit(part1[1], "@")
+        if #part2 ~= 2 then
+            print_stderr("Invalid line: " .. line)
+            os.exit(1)
+        end
+        part3 = likwid.stringsplit(part2[1], ".")
+        if #part3 ~= 2 then
+            print_stderr("Invalid line: " .. line)
+            os.exit(1)
+        end
+        local feat_cat = part3[1]
+        local feat_name = part3[2]
+        local dev_id = part2[2]
+        local value = part1[2]
+
+        local full_name = string.format("%s.%s", feat_cat, feat_name)
+
+        -- get device type of this particular feature
+        local devtype = nil
+        for _, f in pairs(ft_list) do
+            if f.Name == feat_name and f.Category == feat_cat then
+                devtype = f.TypeID
+                break
+            end
+        end
+        if not devtype then
+            print_stderr(string.format("Unknown feature: '%s'", full_name))
+            os.exit(1)
+        end
+        local lw_dev = likwid.createDevice(devtype, dev_id)
+        local success, err = likwid.sysFeatures_set(full_name, lw_dev, value)
+        if not success then
+            print_stderr(string.format("Failed to set feature '%s' on device %s:%d to %s: %s", full_name, lw_dev:typeName(), lw_dev:id(), value, err))
+        end
+    end
+
+    likwid.finalizeSysFeatures()
+    os.exit(0)
+end
+
 likwid.finalizeSysFeatures()
 os.exit(0)

src/applications/likwid.lua

@@ -189,6 +189,7 @@ likwid.finalizeSysFeatures = likwid_finalizeSysFeatures
 likwid.createDevice = likwid_createDevice
 likwid.createDevicesFromString = likwid_createDevicesFromString
 likwid.getAvailableDevices = likwid_getAvailableDevices
+likwid.getAllDevices = likwid_getAllDevices
 
 likwid.getCudaTopology = likwid_getCudaTopology
 likwid.putCudaTopology = likwid_putCudaTopology

src/includes/sysFeatures_amd_thermal.h

@@ -0,0 +1,8 @@
+#ifndef SYSFEATURES_AMD_THERMAL_H
+#define SYSFEATURES_AMD_THERMAL_H
+
+#include <sysFeatures_types.h>
+
+extern const _SysFeatureList likwid_sysft_amd_k10_cpu_thermal_feature_list;
+
+#endif //SYSFEATURES_AMD_THERMAL_H

src/luawid.c

@@ -3828,14 +3828,24 @@ static int lua_likwid_destroyDevice(lua_State *L)
     return 0;
 }
 
-static int lua_likwid_getAvailableDevices(lua_State *L)
+static int lua_likwid_getDevices(lua_State *L, bool all)
 {
     const LikwidDeviceType type = luaL_checknumber(L, 1);
     char **id_list = NULL;
     size_t id_list_len = 0;
-    int err = likwid_device_get_available(type, &id_list, &id_list_len);
-    if (err < 0)
-        luaL_error(L, "likwid_device_get_available failed: %s", strerror(-err));
+    int err;
+    if (all)
+    {
+        err = likwid_device_get_all(type, &id_list, &id_list_len);
+        if (err < 0)
+            luaL_error(L, "likwid_device_get_all failed: %s", strerror(-err));
+    }
+    else
+    {
+        err = likwid_device_get_available(type, &id_list, &id_list_len);
+        if (err < 0)
+            luaL_error(L, "likwid_device_get_available failed: %s", strerror(-err));
+    }
 
     lua_newtable(L);
 
@@ -3850,6 +3860,16 @@ static int lua_likwid_getAvailableDevices(lua_State *L)
     return 1;
 }
 
+static int lua_likwid_getAvailableDevices(lua_State *L)
+{
+    return lua_likwid_getDevices(L, false);
+}
+
+static int lua_likwid_getAllDevices(lua_State *L)
+{
+    return lua_likwid_getDevices(L, true);
+}
+
 static int lua_likwiddevice_get_typeId(lua_State *L)
 {
     const LikwidDevice_t dev = *(LikwidDevice_t *)luaL_checkudata(L, 1, "LikwidDevice_t");
@@ -4249,6 +4269,7 @@ int __attribute__((visibility("default"))) luaopen_liblikwid(lua_State *L) {
     lua_register(L, "likwid_createDevicesFromString",lua_likwid_createDevicesFromString);
     lua_register(L, "likwid_createDevice",lua_likwid_createDevice);
     lua_register(L, "likwid_getAvailableDevices",lua_likwid_getAvailableDevices);
+    lua_register(L, "likwid_getAllDevices",lua_likwid_getAllDevices);
 #endif /* LIKWID_WITH_SYSFEATURES */
 #ifdef __MIC__
   setuid(0);

src/sysFeatures.c

@@ -186,10 +186,7 @@ int likwid_sysft_init(void)
 #ifdef LIKWID_WITH_NVMON
     err = likwid_sysft_init_nvml(&_feature_list);
     if (err < 0)
-    {
-        ERROR_PRINT(Failed to initialize SysFeatures nvml module);
-        return err;
-    }
+        DEBUG_PRINT(DEBUGLEV_INFO, Failed to initialize SysFeatures nvml module);
 #endif
 
     DEBUG_PRINT(DEBUGLEV_DEVELOP, Initialized %d features, _feature_list.num_features);

src/sysFeatures_amd.c

@@ -37,6 +37,7 @@
 #include <bitUtil.h>
 #include <sysFeatures_amd_hsmp.h>
 #include <sysFeatures_amd_rapl.h>
+#include <sysFeatures_amd_thermal.h>
 #include <sysFeatures_common.h>
 #include <topology.h>
 #include <registers.h>
@@ -282,13 +283,15 @@ static const _SysFeatureList* amd_k17_cpu_feature_inputs[] = {
     &amd_k17_cpu_speculation_feature_list,
     //&amd_k19_cpu_l1dflush_feature_list,
     &amd_k17_cpu_hwconfig_feature_list,
+    &likwid_sysft_amd_k10_cpu_thermal_feature_list,
     NULL,
 };
 
 static const _SysFeatureList* amd_k19_zen3_cpu_feature_inputs[] = {
     &amd_k19_cpu_prefetch_feature_list,
     &amd_k19_cpu_speculation_feature_list,
     &amd_k17_cpu_hwconfig_feature_list,
+    &likwid_sysft_amd_k10_cpu_thermal_feature_list,
     NULL,
 };
 
@@ -297,6 +300,7 @@ static const _SysFeatureList* amd_k19_zen4_cpu_feature_inputs[] = {
     &amd_k19_cpu_speculation_feature_list,
     &amd_k19_cpu_l1dflush_feature_list,
     &amd_k17_cpu_hwconfig_feature_list,
+    &likwid_sysft_amd_k10_cpu_thermal_feature_list,
     NULL,
 };