Add support for new performance levels

Also added tests for new performance levels and clean up some
formatting/style issues.


[ROCm/rocm_smi_lib commit: 767fa53d8c]

projects/rocm-smi-lib/include/rocm_smi/rocm_smi.h

@@ -42,8 +42,8 @@
  * DEALINGS WITH THE SOFTWARE.
  *
  */
-#ifndef ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_H_
-#define ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_H_
+#ifndef INCLUDE_ROCM_SMI_ROCM_SMI_H_
+#define INCLUDE_ROCM_SMI_ROCM_SMI_H_
 
 #ifdef __cplusplus
 extern "C" {
@@ -94,14 +94,21 @@ typedef enum {
   RSMI_DEV_PERF_LEVEL_AUTO = 0,       //!< Performance level is "auto"
   RSMI_DEV_PERF_LEVEL_FIRST = RSMI_DEV_PERF_LEVEL_AUTO,
 
-  RSMI_DEV_PERF_LEVEL_LOW,            //!< Keep PowerPlay levels "low",
-                                      //!< regardless of workload
-  RSMI_DEV_PERF_LEVEL_HIGH,           //!< Keep PowerPlay levels "high",
-                                      //!< regardless of workload
-  RSMI_DEV_PERF_LEVEL_MANUAL,         //!< Only use values defined by manually
-                                      //!< setting the RSMI_CLK_TYPE_SYS speed
+  RSMI_DEV_PERF_LEVEL_LOW,              //!< Keep PowerPlay levels "low",
+                                        //!< regardless of workload
+  RSMI_DEV_PERF_LEVEL_HIGH,             //!< Keep PowerPlay levels "high",
+                                        //!< regardless of workload
+  RSMI_DEV_PERF_LEVEL_MANUAL,           //!< Only use values defined by manually
+                                        //!< setting the RSMI_CLK_TYPE_SYS speed
+  RSMI_DEV_PERF_LEVEL_STABLE_STD,       //!< Stable power state with profiling
+                                        //!< clocks
+  RSMI_DEV_PERF_LEVEL_STABLE_PEAK,      //!< Stable power state with peak clocks
+  RSMI_DEV_PERF_LEVEL_STABLE_MIN_MCLK,  //!< Stable power state with minimum
+                                        //!< memory clock
+  RSMI_DEV_PERF_LEVEL_STABLE_MIN_SCLK,  //!< Stable power state with minimum
+                                        //!< system clock
 
-  RSMI_DEV_PERF_LEVEL_LAST = RSMI_DEV_PERF_LEVEL_MANUAL,
+  RSMI_DEV_PERF_LEVEL_LAST = RSMI_DEV_PERF_LEVEL_STABLE_MIN_SCLK,
 
   RSMI_DEV_PERF_LEVEL_UNKNOWN = 0x100   //!< Unknown performance level
 } rsmi_dev_perf_level;
@@ -762,4 +769,4 @@ rsmi_status_string(rsmi_status_t status, const char **status_string);
 #ifdef __cplusplus
 }
 #endif  // __cplusplus
-#endif  // ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_H_
+#endif  // INCLUDE_ROCM_SMI_ROCM_SMI_H_

projects/rocm-smi-lib/include/rocm_smi/rocm_smi_common.h

@@ -42,8 +42,8 @@
  * DEALINGS WITH THE SOFTWARE.
  *
  */
-#ifndef ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_COMMON_H_
-#define ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_COMMON_H_
+#ifndef INCLUDE_ROCM_SMI_ROCM_SMI_COMMON_H_
+#define INCLUDE_ROCM_SMI_ROCM_SMI_COMMON_H_
 
 #define DBG_FILE_ERROR(STR) \
   if (env_->debug_output_bitfield & RSMI_DEBUG_SYSFS_FILE_PATHS) { \
@@ -60,4 +60,4 @@ struct RocmSMI_env_vars {
     uint32_t debug_output_bitfield;
 };
 
-#endif  // ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_COMMON_H_
+#endif  // INCLUDE_ROCM_SMI_ROCM_SMI_COMMON_H_

projects/rocm-smi-lib/include/rocm_smi/rocm_smi_device.h

@@ -42,8 +42,8 @@
  * DEALINGS WITH THE SOFTWARE.
  *
  */
-#ifndef ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_DEVICE_H_
-#define ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_DEVICE_H_
+#ifndef INCLUDE_ROCM_SMI_ROCM_SMI_DEVICE_H_
+#define INCLUDE_ROCM_SMI_ROCM_SMI_DEVICE_H_
 #include <string>
 #include <memory>
 #include <utility>
@@ -53,6 +53,7 @@
 #include "rocm_smi/rocm_smi_monitor.h"
 #include "rocm_smi/rocm_smi_power_mon.h"
 #include "rocm_smi/rocm_smi_common.h"
+#include "rocm_smi/rocm_smi.h"
 
 namespace amd {
 namespace smi {
@@ -84,6 +85,7 @@ class Device {
     int writeDevInfo(DevInfoTypes type, std::string val);
     uint32_t index(void) const {return index_;}
     void set_index(uint32_t index) {index_ = index;}
+    static rsmi_dev_perf_level perfLvlStrToEnum(std::string s);
 
  private:
     std::shared_ptr<Monitor> monitor_;
@@ -100,4 +102,4 @@ class Device {
 }  // namespace smi
 }  // namespace amd
 
-#endif  // ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_DEVICE_H_
+#endif  // INCLUDE_ROCM_SMI_ROCM_SMI_DEVICE_H_

projects/rocm-smi-lib/include/rocm_smi/rocm_smi_main.h

@@ -42,8 +42,8 @@
  * DEALINGS WITH THE SOFTWARE.
  *
  */
-#ifndef ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_MAIN_H_
-#define ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_MAIN_H_
+#ifndef INCLUDE_ROCM_SMI_ROCM_SMI_MAIN_H_
+#define INCLUDE_ROCM_SMI_ROCM_SMI_MAIN_H_
 
 #include <vector>
 #include <memory>
@@ -95,4 +95,4 @@ class RocmSMI {
 }  // namespace smi
 }  // namespace amd
 
-#endif  // ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_MAIN_H_
+#endif  // INCLUDE_ROCM_SMI_ROCM_SMI_MAIN_H_

projects/rocm-smi-lib/include/rocm_smi/rocm_smi_monitor.h

@@ -42,8 +42,8 @@
  * DEALINGS WITH THE SOFTWARE.
  *
  */
-#ifndef ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_MONITOR_H_
-#define ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_MONITOR_H_
+#ifndef INCLUDE_ROCM_SMI_ROCM_SMI_MONITOR_H_
+#define INCLUDE_ROCM_SMI_ROCM_SMI_MONITOR_H_
 
 #include <string>
 #include <cstdint>
@@ -97,4 +97,4 @@ class Monitor {
 }  // namespace smi
 }  // namespace amd
 
-#endif  // ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_MONITOR_H_
+#endif  // INCLUDE_ROCM_SMI_ROCM_SMI_MONITOR_H_

projects/rocm-smi-lib/include/rocm_smi/rocm_smi_power_mon.h

@@ -42,8 +42,8 @@
  * DEALINGS WITH THE SOFTWARE.
  *
  */
-#ifndef ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_POWER_MON_H_
-#define ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_POWER_MON_H_
+#ifndef INCLUDE_ROCM_SMI_ROCM_SMI_POWER_MON_H_
+#define INCLUDE_ROCM_SMI_ROCM_SMI_POWER_MON_H_
 
 #include <string>
 #include <cstdint>
@@ -78,4 +78,4 @@ class PowerMon {
 }  // namespace smi
 }  // namespace amd
 
-#endif  // ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_POWER_MON_H_
+#endif  // INCLUDE_ROCM_SMI_ROCM_SMI_POWER_MON_H_

projects/rocm-smi-lib/include/rocm_smi/rocm_smi_utils.h

@@ -42,8 +42,8 @@
  * DEALINGS WITH THE SOFTWARE.
  *
  */
-#ifndef ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_UTILS_H_
-#define ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_UTILS_H_
+#ifndef INCLUDE_ROCM_SMI_ROCM_SMI_UTILS_H_
+#define INCLUDE_ROCM_SMI_ROCM_SMI_UTILS_H_
 
 #include <string>
 #include <cstdint>
@@ -68,4 +68,4 @@ int WriteSysfsStr(std::string path, std::string val);
 }  // namespace smi
 }  // namespace amd
 
-#endif  // ROCM_SMI_LIB_INCLUDE_ROCM_SMI_ROCM_SMI_UTILS_H_
+#endif  // INCLUDE_ROCM_SMI_ROCM_SMI_UTILS_H_

projects/rocm-smi-lib/src/rocm_smi.cc

@@ -364,17 +364,8 @@ rsmi_dev_perf_level_get(uint32_t dv_ind, rsmi_dev_perf_level *perf) {
     return ret;
   }
 
-  if (val_str == "auto") {
-    *perf = RSMI_DEV_PERF_LEVEL_AUTO;
-  } else if (val_str == "low") {
-    *perf = RSMI_DEV_PERF_LEVEL_LOW;
-  } else if (val_str == "high") {
-    *perf = RSMI_DEV_PERF_LEVEL_HIGH;
-  } else if (val_str == "manual") {
-    *perf = RSMI_DEV_PERF_LEVEL_MANUAL;
-  } else {
-    *perf = RSMI_DEV_PERF_LEVEL_UNKNOWN;
-  }
+  *perf = amd::smi::Device::perfLvlStrToEnum(val_str);
+
   return ret;
   CATCH
 }

projects/rocm-smi-lib/src/rocm_smi_device.cc

@@ -69,6 +69,10 @@ static const char *kDevPerfLevelAutoStr = "auto";
 static const char *kDevPerfLevelLowStr = "low";
 static const char *kDevPerfLevelHighStr = "high";
 static const char *kDevPerfLevelManualStr = "manual";
+static const char *kDevPerfLevelStandardStr = "profile_standard";
+static const char *kDevPerfLevelMinMClkStr = "profile_min_mclk";
+static const char *kDevPerfLevelMinSClkStr = "profile_min_sclk";
+static const char *kDevPerfLevelPeakStr = "profile_peak";
 static const char *kDevPerfLevelUnknownStr = "unknown";
 
 static const std::map<DevInfoTypes, const char *> kDevAttribNameMap = {
@@ -85,6 +89,11 @@ static const std::map<rsmi_dev_perf_level, const char *> kDevPerfLvlMap = {
     {RSMI_DEV_PERF_LEVEL_LOW, kDevPerfLevelLowStr},
     {RSMI_DEV_PERF_LEVEL_HIGH, kDevPerfLevelHighStr},
     {RSMI_DEV_PERF_LEVEL_MANUAL, kDevPerfLevelManualStr},
+    {RSMI_DEV_PERF_LEVEL_STABLE_STD, kDevPerfLevelStandardStr},
+    {RSMI_DEV_PERF_LEVEL_STABLE_MIN_MCLK, kDevPerfLevelMinMClkStr},
+    {RSMI_DEV_PERF_LEVEL_STABLE_MIN_SCLK, kDevPerfLevelMinSClkStr},
+    {RSMI_DEV_PERF_LEVEL_STABLE_PEAK, kDevPerfLevelPeakStr},
+
     {RSMI_DEV_PERF_LEVEL_UNKNOWN, kDevPerfLevelUnknownStr},
 };
 
@@ -157,6 +166,18 @@ int Device::writeDevInfoStr(DevInfoTypes type, std::string valStr) {
   return 0;
 }
 
+rsmi_dev_perf_level Device::perfLvlStrToEnum(std::string s) {
+  rsmi_dev_perf_level pl;
+
+  for (pl = RSMI_DEV_PERF_LEVEL_FIRST; pl <= RSMI_DEV_PERF_LEVEL_LAST; ) {
+    if (s == kDevPerfLvlMap.at(pl)) {
+      return pl;
+    }
+    pl = static_cast<rsmi_dev_perf_level>(static_cast<uint32_t>(pl) + 1);
+  }
+  return RSMI_DEV_PERF_LEVEL_UNKNOWN;
+}
+
 int Device::writeDevInfo(DevInfoTypes type, uint64_t val) {
   switch (type) {
     // The caller is responsible for making sure "val" is within a valid range
@@ -165,7 +186,7 @@ int Device::writeDevInfo(DevInfoTypes type, uint64_t val) {
       return writeDevInfoStr(type, std::to_string(val));
       break;
 
-    case kDevPerfLevel:  // string: "auto", "low", "high", "manual"
+    case kDevPerfLevel:  // string: "auto", "low", "high", "manual", ...
       return writeDevInfoStr(type,
                                  kDevPerfLvlMap.at((rsmi_dev_perf_level)val));
       break;

projects/rocm-smi-lib/tests/rocm_smi_test/CMakeLists.txt

@@ -198,7 +198,6 @@ aux_source_directory(${RSMITST_ROOT} rsmitstSources)
 # Header file include path
 
 include_directories(${RSMI_INC_DIR})
-include_directories(${RSMITST_ROOT})
 include_directories(${RSMITST_ROOT}/..)
 include_directories(${RSMITST_ROOT}/gtest/include)
 

projects/rocm-smi-lib/tests/rocm_smi_test/functional/rsmi_sanity.cc

@@ -48,9 +48,10 @@
 #include <string>
 #include <algorithm>
 #include <vector>
+#include <map>
 
 #include "rocm_smi/rocm_smi.h"
-#include "functional/rsmi_sanity.h"
+#include "rocm_smi_test/functional/rsmi_sanity.h"
 #include "gtest/gtest.h"
 
 static const uint32_t kNumBufferElements = 256;
@@ -86,7 +87,7 @@ static uint32_t gVerbosity = 3;
     } \
 }
 
-#define ALT_ASSERT_EQ(A,B) { \
+#define ALT_ASSERT_EQ(A, B) { \
     if ((A) != (B)) { \
       std::cout << "ASSERT failure: Expected " << #A << " == " << #B << \
         ", but got " << #A << " = " << (A) << ", and " << #B << " = " << \
@@ -334,6 +335,21 @@ static rsmi_status_t test_set_fan_speed(uint32_t dv_ind) {
   return ret;
 }
 
+static const std::map<rsmi_dev_perf_level, const char *> kDevPerfLvlNameMap = {
+    {RSMI_DEV_PERF_LEVEL_AUTO, "RSMI_DEV_PERF_LEVEL_AUTO"},
+    {RSMI_DEV_PERF_LEVEL_LOW, "RSMI_DEV_PERF_LEVEL_LOW"},
+    {RSMI_DEV_PERF_LEVEL_HIGH, "RSMI_DEV_PERF_LEVEL_HIGH"},
+    {RSMI_DEV_PERF_LEVEL_MANUAL, "RSMI_DEV_PERF_LEVEL_MANUAL"},
+    {RSMI_DEV_PERF_LEVEL_STABLE_STD, "RSMI_DEV_PERF_LEVEL_STABLE_STD"},
+    {RSMI_DEV_PERF_LEVEL_STABLE_MIN_MCLK,
+                                       "RSMI_DEV_PERF_LEVEL_STABLE_MIN_MCLK"},
+    {RSMI_DEV_PERF_LEVEL_STABLE_MIN_SCLK,
+                                       "RSMI_DEV_PERF_LEVEL_STABLE_MIN_SCLK"},
+    {RSMI_DEV_PERF_LEVEL_STABLE_PEAK, "RSMI_DEV_PERF_LEVEL_STABLE_PEAK"},
+
+    {RSMI_DEV_PERF_LEVEL_UNKNOWN, "RSMI_DEV_PERF_LEVEL_UNKNOWN"},
+};
+
 static rsmi_status_t test_set_perf_level(uint32_t dv_ind) {
   rsmi_status_t ret;
 
@@ -345,30 +361,41 @@ static rsmi_status_t test_set_perf_level(uint32_t dv_ind) {
   CHK_ERR_RET(ret)
 
   IF_VERB(STANDARD) {
-    std::cout << "\t**Original Perf Level:" << (uint32_t)orig_pfl << std::endl;
+    std::cout << "\t**Original Perf Level:" <<
+                                 kDevPerfLvlNameMap.at(orig_pfl) << std::endl;
   }
 
-  pfl = (rsmi_dev_perf_level)((orig_pfl + 1) % (RSMI_DEV_PERF_LEVEL_LAST + 1));
+  uint32_t pfl_i = static_cast<uint32_t>(RSMI_DEV_PERF_LEVEL_FIRST);
+  for (; pfl_i <=  static_cast<uint32_t>(RSMI_DEV_PERF_LEVEL_LAST); pfl_i++) {
+    if (pfl_i == static_cast<uint32_t>(orig_pfl)) {
+      continue;
+    }
 
-  IF_VERB(STANDARD) {
-    std::cout << "Set Performance Level to " << (uint32_t)pfl << " ..." <<
-                                                                   std::endl;
-  }
-  ret = rsmi_dev_perf_level_set(dv_ind, pfl);
-  CHK_ERR_RET(ret)
-  ret = rsmi_dev_perf_level_get(dv_ind, &pfl);
-  CHK_ERR_RET(ret)
-  IF_VERB(STANDARD) {
-    std::cout << "\t**New Perf Level:" << (uint32_t)pfl << std::endl;
-    std::cout << "Reset Perf level to " << orig_pfl << " ..." << std::endl;
+    IF_VERB(STANDARD) {
+      std::cout << "Set Performance Level to " <<
+          kDevPerfLvlNameMap.at(static_cast<rsmi_dev_perf_level>(pfl_i)) <<
+                                                          " ..." << std::endl;
+    }
+    ret = rsmi_dev_perf_level_set(dv_ind,
+                                     static_cast<rsmi_dev_perf_level>(pfl_i));
+    CHK_ERR_RET(ret)
+    ret = rsmi_dev_perf_level_get(dv_ind, &pfl);
+    CHK_ERR_RET(ret)
+    IF_VERB(STANDARD) {
+      std::cout << "\t**New Perf Level:" << kDevPerfLvlNameMap.at(pfl) <<
+                                                                    std::endl;
+    }
   }
+  std::cout << "Reset Perf level to " << kDevPerfLvlNameMap.at(orig_pfl) <<
+                                                          " ..." << std::endl;
   ret = rsmi_dev_perf_level_set(dv_ind, orig_pfl);
   CHK_ERR_RET(ret)
   ret = rsmi_dev_perf_level_get(dv_ind, &pfl);
   CHK_ERR_RET(ret)
 
   IF_VERB(STANDARD) {
-    std::cout << "\t**New Perf Level:" << (uint32_t)pfl << std::endl;
+    std::cout << "\t**New Perf Level:" << kDevPerfLvlNameMap.at(pfl) <<
+                                                                    std::endl;
   }
   return ret;
 }
@@ -553,14 +580,20 @@ void TestSanity::Run(void) {
       print_temp_metric(RSMI_TEMP_CURRENT, "Current Temp.");
       print_temp_metric(RSMI_TEMP_MAX, "Temperature max value");
       print_temp_metric(RSMI_TEMP_MIN, "Temperature min value");
-      print_temp_metric(RSMI_TEMP_MAX_HYST, "Temperature hysteresis value for max limit");
-      print_temp_metric(RSMI_TEMP_MIN_HYST, "Temperature hysteresis value for min limit");
+      print_temp_metric(RSMI_TEMP_MAX_HYST,
+                                "Temperature hysteresis value for max limit");
+      print_temp_metric(RSMI_TEMP_MIN_HYST,
+                                "Temperature hysteresis value for min limit");
       print_temp_metric(RSMI_TEMP_CRITICAL, "Temperature critical max value");
-      print_temp_metric(RSMI_TEMP_CRITICAL_HYST, "Temperature hysteresis value for critical limit");
-      print_temp_metric(RSMI_TEMP_EMERGENCY, "Temperature emergency max value");
-      print_temp_metric(RSMI_TEMP_EMERGENCY_HYST, "Temperature hysteresis value for emergency limit");
+      print_temp_metric(RSMI_TEMP_CRITICAL_HYST,
+                           "Temperature hysteresis value for critical limit");
+      print_temp_metric(RSMI_TEMP_EMERGENCY,
+                                           "Temperature emergency max value");
+      print_temp_metric(RSMI_TEMP_EMERGENCY_HYST,
+                          "Temperature hysteresis value for emergency limit");
       print_temp_metric(RSMI_TEMP_CRIT_MIN, "Temperature critical min value");
-      print_temp_metric(RSMI_TEMP_CRIT_MIN_HYST, "Temperature hysteresis value for critical min value");
+      print_temp_metric(RSMI_TEMP_CRIT_MIN_HYST,
+                       "Temperature hysteresis value for critical min value");
       print_temp_metric(RSMI_TEMP_OFFSET, "Temperature offset");
       print_temp_metric(RSMI_TEMP_LOWEST, "Historical minimum temperature");
       print_temp_metric(RSMI_TEMP_HIGHEST, "Historical maximum temperature");

projects/rocm-smi-lib/tests/rocm_smi_test/test_base.h

@@ -42,8 +42,8 @@
  * DEALINGS WITH THE SOFTWARE.
  *
  */
-#ifndef ROCM_SMI_LIB_TESTS_ROCM_SMI_TEST_TEST_BASE_H_
-#define ROCM_SMI_LIB_TESTS_ROCM_SMI_TEST_TEST_BASE_H_
+#ifndef TESTS_ROCM_SMI_TEST_TEST_BASE_H_
+#define TESTS_ROCM_SMI_TEST_TEST_BASE_H_
 
 #include <string>
 
@@ -102,4 +102,4 @@ class TestBase {
   uint32_t verbosity_;   ///< How much additional output to produce
 };
 
-#endif  // ROCM_SMI_LIB_TESTS_ROCM_SMI_TEST_TEST_BASE_H_
+#endif  // TESTS_ROCM_SMI_TEST_TEST_BASE_H_

projects/rocm-smi-lib/tests/rocm_smi_test/test_common.h

@@ -43,8 +43,8 @@
  *
  */
 
-#ifndef ROCM_SMI_LIB_TESTS_ROCM_SMI_TEST_TEST_COMMON_H_
-#define ROCM_SMI_LIB_TESTS_ROCM_SMI_TEST_TEST_COMMON_H_
+#ifndef TESTS_ROCM_SMI_TEST_TEST_COMMON_H_
+#define TESTS_ROCM_SMI_TEST_TEST_COMMON_H_
 
 #include <memory>
 #include <vector>
@@ -63,4 +63,4 @@ uint32_t ProcessCmdline(RSMITstGlobals* test, int arg_cnt, char** arg_list);
 void DumpMonitorInfo(const TestBase *test);
 #endif
 
-#endif  // ROCM_SMI_LIB_TESTS_ROCM_SMI_TEST_TEST_COMMON_H_
+#endif  // TESTS_ROCM_SMI_TEST_TEST_COMMON_H_