[v0.23] Sort PCI devices; this makes the devices list to match CUDA driver ordering, which fixes a bug when there was more than one NVIDIA GPU

src/common/main.cpp

@@ -8,7 +8,7 @@
 #include "../cuda/cuda.hpp"
 #include "../cuda/uarch.hpp"
 
-static const char* VERSION = "0.22";
+static const char* VERSION = "0.23";
 
 void print_help(char *argv[]) {
   const char **t = args_str;

src/common/master.cpp

@@ -1,6 +1,7 @@
 #include <cstdlib>
 #include <cstdio>
 
+#include "pci.hpp"
 #include "global.hpp"
 #include "colors.hpp"
 #include "master.hpp"
@@ -16,6 +17,7 @@ struct gpu_list {
 
 struct gpu_list* get_gpu_list() {
   int idx = 0;
+  struct pci_dev *devices = get_pci_devices_from_pciutils();
   struct gpu_list* list = (struct gpu_list*) malloc(sizeof(struct gpu_list));
   list->num_gpus = 0;
   list->gpus = (struct gpu_info**) malloc(sizeof(struct info*) * MAX_GPUS);
@@ -24,7 +26,7 @@ struct gpu_list* get_gpu_list() {
   bool valid = true;
 
   while(valid) {
-    list->gpus[idx] = get_gpu_info_cuda(idx);
+    list->gpus[idx] = get_gpu_info_cuda(devices, idx);
     if(list->gpus[idx] != NULL) idx++;
     else valid = false;
   }
@@ -33,7 +35,7 @@ struct gpu_list* get_gpu_list() {
 #endif
 
 #ifdef BACKEND_INTEL
-  list->gpus[idx] = get_gpu_info_intel();
+  list->gpus[idx] = get_gpu_info_intel(devices);
   if(list->gpus[idx] != NULL) list->num_gpus++;
 #endif
 

src/common/pci.cpp

@@ -1,3 +1,4 @@
+#include "sort.hpp"
 #include "global.hpp"
 #include "pci.hpp"
 #include "../cuda/pci.hpp"
@@ -10,6 +11,14 @@
 #define PCI_VENDOR_ID_AMD    0x1002
 #define CLASS_VGA_CONTROLLER 0x0300
 
+void debug_devices(struct pci_dev *devices) {
+  int idx = 0;
+  for(struct pci_dev *dev=devices; idx < 5 && dev != NULL; dev=dev->next) {
+    printf("%04x:%02x:%02x.%d\n", dev->domain, dev->bus, dev->dev, dev->func);
+    idx++;
+  }
+}
+
 bool pciutils_is_vendor_id_present(struct pci_dev *devices, int id) {
   for(struct pci_dev *dev=devices; dev != NULL; dev=dev->next) {
     if(dev->vendor_id == id && dev->device_class == CLASS_VGA_CONTROLLER) {
@@ -21,14 +30,19 @@ bool pciutils_is_vendor_id_present(struct pci_dev *devices, int id) {
   return false;
 }
 
-uint16_t pciutils_get_pci_device_id(struct pci_dev *devices, int id) {
+uint16_t pciutils_get_pci_device_id(struct pci_dev *devices, int id, int idx) {
+  int curr = 0;
+
   for(struct pci_dev *dev=devices; dev != NULL; dev=dev->next) {
-   if(dev->vendor_id == id && dev->device_class == CLASS_VGA_CONTROLLER) {
-      return dev->device_id;
+    if(dev->vendor_id == id && dev->device_class == CLASS_VGA_CONTROLLER) {
+      if(curr == idx) {
+        return dev->device_id;
+      }
+      curr++;
     }
   }
 
-  printErr("Unable to find a valid device for device id 0x%.4X using pciutils", id);
+  printErr("Unable to find a valid device for device id 0x%.4X with idx %d using pciutils", id, idx);
   return 0;
 }
 
@@ -48,13 +62,13 @@ void pciutils_set_pci_bus(struct pci* pci, struct pci_dev *devices, int id) {
   if(!found) printErr("Unable to find a valid device for id 0x%.4X using pciutils", id);
 }
 
-struct pci* get_pci_from_pciutils(struct pci_dev *devices, int id) {
+struct pci* get_pci_from_pciutils(struct pci_dev *devices, int id, int idx) {
   struct pci* pci = (struct pci*) emalloc(sizeof(struct pci));
 
   // TODO: Refactor this; instead of 2xGet + 1xSet, do it better
   if(pciutils_is_vendor_id_present(devices, id)) {
     pci->vendor_id = id;
-    pci->device_id = pciutils_get_pci_device_id(devices, id);
+    pci->device_id = pciutils_get_pci_device_id(devices, id, idx);
     pciutils_set_pci_bus(pci, devices, id);
     return pci;
   }
@@ -75,6 +89,8 @@ struct pci_dev *get_pci_devices_from_pciutils() {
     pci_fill_info(dev, PCI_FILL_IDENT | PCI_FILL_BASES | PCI_FILL_CLASS);
   }
 
+  sort_pci_devices(&pacc->devices);
+
   return pacc->devices;
 }
 

src/common/pci.hpp

@@ -15,7 +15,7 @@ struct pci {
   uint16_t func;
 };
 
-struct pci* get_pci_from_pciutils(struct pci_dev *devices, int id);
+struct pci* get_pci_from_pciutils(struct pci_dev *devices, int id, int idx);
 struct pci_dev *get_pci_devices_from_pciutils();
 void print_gpus_list_pci();
 

src/common/sort.cpp

@@ -0,0 +1,61 @@
+#include <cstdio>
+#include <cstdlib>
+#include "pci.hpp"
+#include "global.hpp"
+
+// Code inspired in lspci.c
+int compare_them(const void *A, const void *B) {
+  const struct pci_dev *a = *(struct pci_dev **) A;
+  const struct pci_dev *b = *(struct pci_dev **) B;
+
+  if (a->domain < b->domain)
+    return -1;
+  if (a->domain > b->domain)
+    return 1;
+  if (a->bus < b->bus)
+    return -1;
+  if (a->bus > b->bus)
+    return 1;
+  if (a->dev < b->dev)
+    return -1;
+  if (a->dev > b->dev)
+    return 1;
+  if (a->func < b->func)
+    return -1;
+  if (a->func > b->func)
+    return 1;
+
+  return 0;
+}
+
+void sort_pci_devices(struct pci_dev **devices) {
+  int i = 0;
+  struct pci_dev **arr;
+
+  int cnt = 0;
+  for(struct pci_dev *dev=*devices; dev != NULL; dev=dev->next) {
+    cnt++;
+  }
+
+  arr = (struct pci_dev **) emalloc(sizeof(struct pci_dev *) * cnt);
+  for(struct pci_dev *dev=*devices; dev != NULL; dev=dev->next) {
+    arr[i] = dev;
+    i++;
+  }
+
+  qsort(arr, cnt, sizeof(struct pci_dev *), compare_them);
+
+  struct pci_dev *ptr = *devices;
+  struct pci_dev *ptrb = *devices;
+  for(i = 0; i < cnt; i++) {
+    ptr = arr[i];
+    if(i > 0) {
+      ptrb->next = ptr;
+    }
+    ptrb = ptr;
+  }
+
+  ptr->next = NULL;
+  *devices = arr[0];
+  free(arr);
+}

src/common/sort.hpp

@@ -0,0 +1,7 @@
+#ifndef __SORT_PCI__
+#define __SORT_PCI__
+
+void sort_pci_devices(struct pci_dev **first_dev);
+
+#endif
+