Merge pull request Syllo#413 from polluks/master

Update nvtop manpage with additional options

Author: Syllo

manpage/nvtop.in

@@ -8,8 +8,10 @@ nvtop \- interactive GPU process viewer
 
 .SH SYNOPSIS
 .B nvtop
-\fR[\fB\-hv\fR]
+\fR[\fB\-hvCfpPris\fR]
 \fR[\fB\-d\fR \fIdelay\fR]
+\fR[\fB\-c\fR \fIconfig-file\fR]
+\fR[\fB\-E\fR \fIseconds\fR]
 
 .SH DESCRIPTION
 nvtop is a ncurses\-based GPU status viewer for AMD, Intel and NVIDIA GPUs.

src/CMakeLists.txt

@@ -101,6 +101,19 @@ if(INTEL_SUPPORT)
   target_sources(nvtop PRIVATE extract_gpuinfo_intel.c)
   target_sources(nvtop PRIVATE extract_gpuinfo_intel_i915.c)
   target_sources(nvtop PRIVATE extract_gpuinfo_intel_xe.c)
+  
+  find_library(LIBXPUM
+    NAMES libxpum.so
+    PATHS /usr/lib /usr/lib/x86_64-linux-gnu /usr/lib64 /usr/local/lib
+    HINTS ${CMAKE_INSTALL_PREFIX}/lib ${CMAKE_INSTALL_PREFIX}/lib64
+  )
+  if(LIBXPUM)
+    message(STATUS "Found libxpum.so; Intel XPU Manager support enabled")
+    target_sources(nvtop PRIVATE extract_gpuinfo_intel_xpumanager.c)
+    target_link_libraries(nvtop PRIVATE "${LIBXPUM}")
+  else()
+    message(STATUS "libxpum.so not found; XPU Manager metrics will not be available")
+  endif()
 endif()
 
 if(V3D_SUPPORT)

src/extract_gpuinfo_intel.c

@@ -66,15 +66,20 @@ static struct gpu_info_intel *gpu_infos;
 
 __attribute__((constructor)) static void init_extract_gpuinfo_intel(void) { register_gpu_vendor(&gpu_vendor_intel); }
 
-bool gpuinfo_intel_init(void) { return true; }
+bool gpuinfo_intel_init(void) {
+    gpuinfo_intel_xpumanager_init();
+    return true;
+}
+
 void gpuinfo_intel_shutdown(void) {
-  for (unsigned i = 0; i < intel_gpu_count; ++i) {
-    struct gpu_info_intel *current = &gpu_infos[i];
-    if (current->card_fd)
-      close(current->card_fd);
-    nvtop_device_unref(current->card_device);
-    nvtop_device_unref(current->driver_device);
-  }
+    for (unsigned i = 0; i < intel_gpu_count; ++i) {
+        struct gpu_info_intel *current = &gpu_infos[i];
+        if (current->card_fd)
+          close(current->card_fd);
+        nvtop_device_unref(current->card_device);
+        nvtop_device_unref(current->driver_device);
+    }
+    gpuinfo_intel_xpumanager_shutdown();
 }
 
 const char *gpuinfo_intel_last_error_string(void) { return "Err"; }
@@ -109,6 +114,7 @@ static void add_intel_cards(struct nvtop_device *dev, struct list_head *devices,
   thisGPU->card_device = nvtop_device_ref(dev);
   thisGPU->driver_device = nvtop_device_ref(parent);
   thisGPU->hwmon_device = nvtop_device_get_hwmon(thisGPU->driver_device);
+  thisGPU->xpum_device_id = -1;
 
   const char *devname;
   if (nvtop_device_get_devname(thisGPU->card_device, &devname) >= 0)
@@ -118,6 +124,12 @@ static void add_intel_cards(struct nvtop_device *dev, struct list_head *devices,
   int retval = nvtop_device_get_property_value(thisGPU->driver_device, "PCI_SLOT_NAME", &pdev_val);
   assert(retval >= 0 && pdev_val != NULL && "Could not retrieve device PCI slot name");
   strncpy(thisGPU->base.pdev, pdev_val, PDEV_LEN);
+  
+  if (gpuinfo_intel_xpumanager_is_available()) {
+    gpuinfo_intel_xpumanager_get_device_id(pdev_val, &thisGPU->xpum_device_id);
+    gpuinfo_intel_xpumanager_populate_static_info(&thisGPU->base, thisGPU->xpum_device_id);
+  }
+  
   list_add_tail(&thisGPU->base.list, devices);
   // Register a fdinfo callback for this GPU
   processinfo_register_fdinfo_callback(parse_drm_fdinfo_intel, &thisGPU->base);
@@ -224,6 +236,13 @@ void gpuinfo_intel_refresh_dynamic_info(struct gpu_info *_gpu_info) {
 
   RESET_ALL(dynamic_info->valid);
 
+  if (gpu_info->xpum_device_id >= 0 && gpuinfo_intel_xpumanager_is_available()) {
+      gpuinfo_intel_xpumanager_refresh_dynamic_info(_gpu_info, gpu_info->xpum_device_id);
+      if (GPUINFO_DYNAMIC_FIELD_VALID(dynamic_info, gpu_util_rate)) {
+          return;
+      }
+  }
+
   // We are creating new devices because the device_get_sysattr_value caches its querries
   const char *syspath;
   nvtop_device *card_dev_noncached = NULL;

src/extract_gpuinfo_intel.h

@@ -1,5 +1,15 @@
-#include <stdint.h>
-#include <uthash.h>
+#include <stdio.h>
+#include <limits.h>
+#include <stdbool.h>
+#include <string.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <stdlib.h>
+#include <time.h>
+#include "list.h"
+#include "uthash.h"
+#include <nvtop/time.h>
+#include <nvtop/extract_gpuinfo_common.h>
 
 #define HASH_FIND_CLIENT(head, key_ptr, out_ptr) HASH_FIND(hh, head, key_ptr, sizeof(struct unique_cache_id), out_ptr)
 #define HASH_ADD_CLIENT(head, in_ptr) HASH_ADD(hh, head, client_id, sizeof(struct unique_cache_id), in_ptr)
@@ -59,10 +69,12 @@ struct gpu_info_intel {
 
   struct nvtop_device *driver_device;
   struct nvtop_device *hwmon_device;
-  struct intel_process_info_cache *last_update_process_cache, *current_update_process_cache; // Cached processes info
+  struct intel_process_info_cache *last_update_process_cache, *current_update_process_cache;
 
   struct nvtop_device *bridge_device;
 
+  int xpum_device_id;
+
   struct {
     unsigned energy_uj;
     struct timespec time;
@@ -74,3 +86,10 @@ extern void gpuinfo_intel_xe_refresh_dynamic_info(struct gpu_info *_gpu_info);
 
 extern bool parse_drm_fdinfo_intel_i915(struct gpu_info *info, FILE *fdinfo_file, struct gpu_process *process_info);
 extern bool parse_drm_fdinfo_intel_xe(struct gpu_info *info, FILE *fdinfo_file, struct gpu_process *process_info);
+
+extern bool gpuinfo_intel_xpumanager_init(void);
+extern void gpuinfo_intel_xpumanager_shutdown(void);
+extern bool gpuinfo_intel_xpumanager_get_device_id(const char *pci_bdf, int *device_id);
+extern void gpuinfo_intel_xpumanager_populate_static_info(struct gpu_info *gpu_info, int xpum_device_id);
+extern void gpuinfo_intel_xpumanager_refresh_dynamic_info(struct gpu_info *gpu_info, int xpum_device_id);
+extern bool gpuinfo_intel_xpumanager_is_available(void);