json output for frequency data (#488)

- add docs - add variorum_get_frequency_json function interfaces - supports Intel CPUs, AMD CPUs, IBM Power9 CPUs, AMD GPUs, nVidia GPUs --------- Co-authored-by: Kyle Fan <[email protected]>
LLNL · Jan 15, 2024 · 576ab21 · 576ab21
1 parent 5118852
commit 576ab21
Show file tree

Hide file tree

Showing 42 changed files with 712 additions and 0 deletions.
diff --git a/src/docs/sphinx/api/json_support_functions.rst b/src/docs/sphinx/api/json_support_functions.rst
@@ -17,3 +17,5 @@ Defined in ``variorum/variorum.h``.
 .. doxygenfunction:: variorum_get_node_power_domain_info_json
 
 .. doxygenfunction:: variorum_get_thermals_json
+
+.. doxygenfunction:: variorum_get_node_frequency_json
diff --git a/src/examples/CMakeLists.txt b/src/examples/CMakeLists.txt
@@ -16,6 +16,7 @@ set(BASIC_EXAMPLES
     variorum-cap-socket-power-limit-example
     variorum-disable-turbo-example
     variorum-enable-turbo-example
+    variorum-get-node-frequency-json-example
     variorum-get-node-power-domain-info-json-example
     variorum-get-node-power-json-example
     variorum-get-node-thermal-json-example

diff --git a/src/examples/variorum-get-node-frequency-json-example.c b/src/examples/variorum-get-node-frequency-json-example.c
@@ -0,0 +1,45 @@
+// Copyright 2019-2023 Lawrence Livermore National Security, LLC and other
+// Variorum Project Developers. See the top-level LICENSE file for details.
+//
+// SPDX-License-Identifier: MIT
+
+#include <getopt.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <variorum.h>
+
+int main(int argc, char **argv)
+{
+    int ret;
+
+    const char *usage = "Usage: %s [-h] [-v]\n";
+    int opt;
+    while ((opt = getopt(argc, argv, "hv")) != -1)
+    {
+        switch (opt)
+        {
+            case 'h':
+                printf(usage, argv[0]);
+                return 0;
+            case 'v':
+                printf("%s\n", variorum_get_current_version());
+                return 0;
+            default:
+                fprintf(stderr, usage, argv[0]);
+                return -1;
+        }
+    }
+    char *s = NULL;
+    ret = variorum_get_node_frequency_json(&s);
+    if (ret != 0)
+    {
+        printf("Variorum get frequency json failure!\n");
+        free(s);
+        exit(-1);
+    }
+    puts(s);
+    free(s);
+
+    return ret;
+}
diff --git a/src/variorum/AMD/config_amd.c b/src/variorum/AMD/config_amd.c
@@ -77,6 +77,7 @@ int set_amd_func_ptrs(int idx)
             g_platform[idx].variorum_get_node_power_json = amd_cpu_epyc_get_node_power_json;
             g_platform[idx].variorum_get_node_power_domain_info_json =
                 amd_cpu_epyc_get_node_power_domain_info_json;
+            g_platform[idx].variorum_get_frequency_json = amd_cpu_epyc_get_json_boostlimit;
             break;
         default:
             fprintf(stdout, "ESMI not initialized, drivers not found. "

diff --git a/src/variorum/AMD/epyc.c b/src/variorum/AMD/epyc.c
@@ -499,6 +499,51 @@ int amd_cpu_epyc_print_boostlimit()
     return 0;
 }
 
+int amd_cpu_epyc_get_json_boostlimit(json_t *get_clock_obj_json)
+{
+    char *val = getenv("VARIORUM_LOG");
+    if (val != NULL && atoi(val) == 1)
+    {
+        printf("Running %s\n\n", __FUNCTION__);
+    }
+
+    int socket, core,  ret;
+    uint32_t boostlimit;
+
+    int num_sockets = g_platform[P_AMD_CPU_IDX].num_sockets;
+    int total_cores = g_platform[P_AMD_CPU_IDX].total_cores;
+    int cores_per_socket = total_cores / num_sockets;
+    int current_core = 0;
+
+    for (socket = 0; socket < num_sockets; ++socket)
+    {
+        char socket_name[16];
+        snprintf(socket_name, 16, "socket_%d", socket);
+        json_t *socket_obj = json_object_get(get_clock_obj_json, socket_name);
+        if (socket_obj == NULL)
+        {
+            socket_obj = json_object();
+            json_object_set_new(get_clock_obj_json, socket_name, socket_obj);
+        }
+
+        json_t *cpu_obj = json_object();
+        json_object_set_new(socket_obj, "CPU", cpu_obj);
+
+        json_t *core_obj = json_object();
+        json_object_set_new(cpu_obj, "core", core_obj);
+
+        for (core = 0; core < cores_per_socket; ++core)
+        {
+            ret = esmi_core_boostlimit_get(current_core, &boostlimit);
+            char core_avg_string[24];
+            snprintf(core_avg_string, 24, "core_%d_avg_freq_mhz", current_core);
+            json_object_set_new(core_obj, core_avg_string, json_real(boostlimit));
+            current_core++;
+        }
+    }
+    return 0;
+}
+
 int amd_cpu_epyc_set_each_core_boostlimit(int boostlimit)
 {
     char *val = getenv("VARIORUM_LOG");

diff --git a/src/variorum/AMD/epyc.h b/src/variorum/AMD/epyc.h
@@ -54,4 +54,8 @@ int amd_cpu_epyc_get_node_power_domain_info_json(
     char **get_domain_obj_str
 );
 
+int amd_cpu_epyc_get_json_boostlimit(
+    json_t *get_clock_obj_json
+);
+
 #endif
diff --git a/src/variorum/AMD_GPU/amd_gpu_power_features.c b/src/variorum/AMD_GPU/amd_gpu_power_features.c
@@ -579,6 +579,92 @@ void get_clocks_data(int chipid, int total_sockets, int verbose, FILE *output)
     }
 }
 
+void get_clocks_json(int chipid, int total_sockets, json_t *output)
+{
+    rsmi_status_t ret;
+    uint32_t num_devices;
+    int gpus_per_socket;
+    char socketID[16];
+
+    snprintf(socketID, 16, "socket_%d", chipid);
+
+    ret = rsmi_init(0);
+    if (ret != RSMI_STATUS_SUCCESS)
+    {
+        variorum_error_handler("Could not initialize RSMI",
+                               VARIORUM_ERROR_PLATFORM_ENV,
+                               getenv("HOSTNAME"), __FILE__, __FUNCTION__,
+                               __LINE__);
+    }
+
+    ret = rsmi_num_monitor_devices(&num_devices);
+    if (ret != RSMI_STATUS_SUCCESS)
+    {
+        variorum_error_handler("Could not get number of GPU devices",
+                               VARIORUM_ERROR_PLATFORM_ENV,
+                               getenv("HOSTNAME"), __FILE__, __FUNCTION__,
+                               __LINE__);
+    }
+
+    gpus_per_socket = num_devices / total_sockets;
+
+    json_t *socket_obj = json_object_get(output, socketID);
+    if (socket_obj == NULL)
+    {
+        socket_obj = json_object();
+        json_object_set_new(output, socketID, socket_obj);
+    }
+
+    json_t *gpu_obj = json_object();
+    json_object_set_new(socket_obj, "GPU", gpu_obj);
+
+    for (int i = chipid * gpus_per_socket;
+         i < (chipid + 1) * gpus_per_socket; i++)
+    {
+        rsmi_frequencies_t f_sys, f_mem;
+        uint32_t f_sys_val, f_mem_val;
+
+        ret = rsmi_dev_gpu_clk_freq_get(i, RSMI_CLK_TYPE_SYS, &f_sys);
+        if (ret != RSMI_STATUS_SUCCESS)
+        {
+            variorum_error_handler("RSMI API was not successful",
+                                   VARIORUM_ERROR_PLATFORM_ENV,
+                                   getenv("HOSTNAME"), __FILE__, __FUNCTION__,
+                                   __LINE__);
+        }
+
+        ret = rsmi_dev_gpu_clk_freq_get(i, RSMI_CLK_TYPE_MEM, &f_mem);
+        if (ret != RSMI_STATUS_SUCCESS)
+        {
+            variorum_error_handler("RSMI API was not successful",
+                                   VARIORUM_ERROR_PLATFORM_ENV,
+                                   getenv("HOSTNAME"), __FILE__, __FUNCTION__,
+                                   __LINE__);
+        }
+
+        f_sys_val = f_sys.frequency[f_sys.current] / (1000 * 1000); // Convert to MHz
+        f_mem_val = f_mem.frequency[f_mem.current] / (1000 * 1000); // Convert to MHz
+
+        char gpu_clock_string[32];
+        snprintf(gpu_clock_string, 32, "gpu_%d_freq_mhz", i);
+
+        char gpu_mem_clock_string[32];
+        snprintf(gpu_mem_clock_string, 32, "gpu_%d_mem_freq_mhz", i);
+
+        json_object_set_new(gpu_obj, gpu_clock_string, json_integer(f_sys_val));
+        json_object_set_new(gpu_obj, gpu_mem_clock_string, json_integer(f_mem_val));
+    }
+
+    ret = rsmi_shut_down();
+    if (ret != RSMI_STATUS_SUCCESS)
+    {
+        variorum_error_handler("Could not shutdown RSMI",
+                               VARIORUM_ERROR_PLATFORM_ENV,
+                               getenv("HOSTNAME"), __FILE__, __FUNCTION__,
+                               __LINE__);
+    }
+}
+
 void get_gpu_utilization_data(int chipid, int total_sockets, int verbose,
                               FILE *output)
 {

diff --git a/src/variorum/AMD_GPU/amd_gpu_power_features.h b/src/variorum/AMD_GPU/amd_gpu_power_features.h
@@ -59,4 +59,10 @@ void get_thermals_json(
     json_t *output
 );
 
+void get_clocks_json(
+    int chipid,
+    int total_sockets,
+    json_t *output
+);
+
 #endif
diff --git a/src/variorum/AMD_GPU/config_amd_gpu.c b/src/variorum/AMD_GPU/config_amd_gpu.c
@@ -34,6 +34,7 @@ int set_amd_gpu_func_ptrs(int idx)
         g_platform[idx].variorum_print_gpu_utilization =
             amd_gpu_instinct_get_gpu_utilization;
         g_platform[idx].variorum_get_thermals_json = amd_gpu_instinct_get_thermals_json;
+        g_platform[idx].variorum_get_frequency_json = amd_gpu_instinct_get_clocks_json;
         /* Initialize control interfaces */
         g_platform[idx].variorum_cap_each_gpu_power_limit =
             amd_gpu_instinct_cap_each_gpu_power_limit;

diff --git a/src/variorum/AMD_GPU/instinctGPU.c b/src/variorum/AMD_GPU/instinctGPU.c
@@ -120,6 +120,27 @@ int amd_gpu_instinct_get_clocks(int verbose)
     return 0;
 }
 
+int amd_gpu_instinct_get_clocks_json(json_t *get_clock_obj_json)
+{
+    char *val = getenv("VARIORUM_LOG");
+    if (val != NULL && atoi(val) == 1)
+    {
+        printf("Running %s\n", __FUNCTION__);
+    }
+
+    unsigned iter = 0;
+    unsigned nsockets;
+
+    variorum_get_topology(&nsockets, NULL, NULL, P_AMD_GPU_IDX);
+
+    for (iter = 0; iter < nsockets; iter++)
+    {
+        get_clocks_json(iter, nsockets, get_clock_obj_json);
+    }
+
+    return 0;
+}
+
 int amd_gpu_instinct_get_gpu_utilization(int verbose)
 {
     char *val = getenv("VARIORUM_LOG");

diff --git a/src/variorum/AMD_GPU/instinctGPU.h b/src/variorum/AMD_GPU/instinctGPU.h
@@ -37,4 +37,8 @@ int amd_gpu_instinct_get_thermals_json(
     json_t *get_thermal_obj
 );
 
+int amd_gpu_instinct_get_clocks_json(
+    json_t *get_clocks_obj_json
+);
+
 #endif
diff --git a/src/variorum/IBM/Power9.c b/src/variorum/IBM/Power9.c
@@ -571,3 +571,64 @@ int ibm_cpu_p9_get_node_power_domain_info_json(char **get_domain_obj_str)
 
     return 0;
 }
+
+int ibm_cpu_p9_get_node_frequency_json(json_t *get_frequency_obj_json)
+{
+    char *val = ("VARIORUM_LOG");
+    if (val != NULL && atoi(val) == 1)
+    {
+        printf("Running %s\n", __FUNCTION__);
+    }
+
+    void *buf;
+    int fd;
+    int rc;
+    int bytes;
+    unsigned iter = 0;
+    unsigned nsockets;
+
+#ifdef VARIORUM_WITH_IBM_CPU
+    variorum_get_topology(&nsockets, NULL, NULL, P_IBM_CPU_IDX);
+#endif
+
+    fd = open("/sys/firmware/opal/exports/occ_inband_sensors", O_RDONLY);
+    if (fd < 0)
+    {
+        printf("Failed to open occ_inband_sensors file\n");
+        return -1;
+    }
+
+    for (iter = 0; iter < nsockets; iter++)
+    {
+        lseek(fd, iter * OCC_SENSOR_DATA_BLOCK_SIZE, SEEK_SET);
+
+        buf = malloc(OCC_SENSOR_DATA_BLOCK_SIZE);
+        if (!buf)
+        {
+            printf("Failed to allocate\n");
+            return -1;
+        }
+
+        for (rc = bytes = 0; bytes < OCC_SENSOR_DATA_BLOCK_SIZE; bytes += rc)
+        {
+            rc = read(fd, buf + bytes, OCC_SENSOR_DATA_BLOCK_SIZE - bytes);
+
+            if (!rc || rc < 0)
+            {
+                break;
+            }
+        }
+
+        if (bytes != OCC_SENSOR_DATA_BLOCK_SIZE)
+        {
+            printf("Failed to read data\n");
+            free(buf);
+            return -1;
+        }
+        json_get_frequency_sensors(iter, get_frequency_obj_json, buf);
+        free(buf);
+    }
+
+    close(fd);
+    return 0;
+}
diff --git a/src/variorum/IBM/Power9.h b/src/variorum/IBM/Power9.h
@@ -44,4 +44,8 @@ int ibm_cpu_p9_get_node_thermal_json(
     json_t *get_thermal_obj
 );
 
+int ibm_cpu_p9_get_node_frequency_json(
+    json_t *get_frequency_obj_json
+);
+
 #endif
diff --git a/src/variorum/IBM/config_ibm.c b/src/variorum/IBM/config_ibm.c
@@ -37,6 +37,8 @@ int set_ibm_func_ptrs(int idx)
         g_platform[idx].variorum_get_node_power_domain_info_json =
             ibm_cpu_p9_get_node_power_domain_info_json;
         g_platform[idx].variorum_get_thermals_json = ibm_cpu_p9_get_node_thermal_json;
+        g_platform[idx].variorum_get_frequency_json =
+            ibm_cpu_p9_get_node_frequency_json;
     }
     else
     {
Original file line number	Diff line number	Diff line change
Expand Up		@@ -17,3 +17,5 @@ Defined in ``variorum/variorum.h``.
		.. doxygenfunction:: variorum_get_node_power_domain_info_json

		.. doxygenfunction:: variorum_get_thermals_json

		.. doxygenfunction:: variorum_get_node_frequency_json