[v0.11] Adding uarch backend for intel iGPUs

CMakeLists.txt

@@ -54,7 +54,7 @@ set(CMAKE_CXX_FLAGS "${SANITY_FLAGS} -Wall -Wextra -pedantic -fstack-protector-a
 if(ENABLE_INTEL_BACKEND)
   target_compile_definitions(gpufetch PUBLIC BACKEND_INTEL)
 
-  add_library(intel_backend STATIC ${INTEL_DIR}/intel.cpp)
+  add_library(intel_backend STATIC ${INTEL_DIR}/intel.cpp ${INTEL_DIR}/pci.cpp ${INTEL_DIR}/uarch.cpp)
 
   if(NOT ${PCIUTILS_FOUND})
     add_dependencies(intel_backend pciutils)

src/common/master.cpp

@@ -20,6 +20,8 @@ struct gpu_list* get_gpu_list() {
   list->gpus = (struct gpu_info**) malloc(sizeof(struct info*) * MAX_GPUS);
 
 #ifdef BACKEND_CUDA
+  bool valid = true;
+
   while(valid) {
     list->gpus[idx] = get_gpu_info_cuda(idx);
     if(list->gpus[idx] != NULL) idx++;

src/common/pci.cpp

@@ -2,33 +2,39 @@
 #include "pci.hpp"
 #include <cstddef>
 
-/*
- * doc: https://wiki.osdev.org/PCI#Class_Codes
- *      https://pci-ids.ucw.cz/read/PC
- */
-#define VENDOR_ID_NVIDIA 0x10de
 #define CLASS_VGA_CONTROLLER 0x0300
 
-uint16_t pciutils_get_pci_vendor_id(struct pci_dev *devices) {
+uint16_t pciutils_get_pci_vendor_id(struct pci_dev *devices, int id) {
   for(struct pci_dev *dev=devices; dev != NULL; dev=dev->next) {
-    if(dev->vendor_id == VENDOR_ID_NVIDIA && dev->device_class == CLASS_VGA_CONTROLLER) {
+    if(dev->vendor_id == id && dev->device_class == CLASS_VGA_CONTROLLER) {
       return dev->vendor_id;
     }
   }
-  printErr("Unable to find a CUDA device using pciutils");
+
+  printErr("Unable to find a valid device for id %d using pciutils", id);
   return 0;
 }
 
-uint16_t pciutils_get_pci_device_id(struct pci_dev *devices) {
+uint16_t pciutils_get_pci_device_id(struct pci_dev *devices, int id) {
   for(struct pci_dev *dev=devices; dev != NULL; dev=dev->next) {
-   if(dev->vendor_id == VENDOR_ID_NVIDIA && dev->device_class == CLASS_VGA_CONTROLLER) {
+   if(dev->vendor_id == id && dev->device_class == CLASS_VGA_CONTROLLER) {
       return dev->device_id;
     }
   }
-  printErr("Unable to find a CUDA device using pciutils");
+
+  printErr("Unable to find a valid device for id %d using pciutils", id);
   return 0;
 }
 
+struct pci* get_pci_from_pciutils(struct pci_dev *devices, int id) {
+  struct pci* pci = (struct pci*) emalloc(sizeof(struct pci));
+
+  pci->vendor_id = pciutils_get_pci_vendor_id(devices, id);
+  pci->device_id = pciutils_get_pci_device_id(devices, id);
+
+  return pci;
+}
+
 struct pci_dev *get_pci_devices_from_pciutils() {
   struct pci_access *pacc;
   struct pci_dev *dev;

src/common/pci.hpp

@@ -6,8 +6,12 @@ extern "C" {
   #include <pci/pci.h>
 }
 
-uint16_t pciutils_get_pci_vendor_id(struct pci_dev *devices);
-uint16_t pciutils_get_pci_device_id(struct pci_dev *devices);
+struct pci {
+  uint16_t vendor_id;
+  uint16_t device_id;
+};
+
+struct pci* get_pci_from_pciutils(struct pci_dev *devices, int id);
 struct pci_dev *get_pci_devices_from_pciutils();
 
 #endif

src/common/uarch.hpp

@@ -0,0 +1,20 @@
+#ifndef __COMMON_UARCH__
+#define __COMMON_UARCH__
+
+typedef uint32_t GPUCHIP;
+typedef uint32_t MICROARCH;
+
+struct uarch {
+  int32_t cc_major;
+  int32_t cc_minor;
+  int32_t compute_capability;
+
+  MICROARCH uarch;
+  GPUCHIP chip;
+
+  int32_t process;
+  char* uarch_str;
+  char* chip_str;
+};
+
+#endif

src/cuda/chips.hpp

@@ -1,5 +1,5 @@
-#ifndef __GPUCHIPS__
-#define __GPUCHIPS__
+#ifndef __CUDA_GPUCHIPS__
+#define __CUDA_GPUCHIPS__
 
 typedef uint32_t GPUCHIP;
 

src/cuda/cuda.cpp

@@ -133,7 +133,7 @@ struct gpu_info* get_gpu_info_cuda(int gpu_idx) {
   strcpy(gpu->name, deviceProp.name);
 
   struct pci_dev *devices = get_pci_devices_from_pciutils();
-  gpu->pci = get_pci_from_pciutils(devices);
+  gpu->pci = get_pci_from_pciutils(devices, PCI_VENDOR_ID_NVIDIA);
   gpu->arch = get_uarch_from_cuda(gpu);
   gpu->cach = get_cache_info(deviceProp);
   gpu->mem = get_memory_info(gpu, deviceProp);

src/cuda/pci.cpp

@@ -10,20 +10,6 @@
    else if (pci->device_id == id) return chip;
 #define CHECK_PCI_END else { printBug("TODOO"); return CHIP_UNKNOWN; }
 
-struct pci {
-  uint16_t vendor_id;
-  uint16_t device_id;
-};
-
-struct pci* get_pci_from_pciutils(struct pci_dev *devices) {
-  struct pci* pci = (struct pci*) emalloc(sizeof(struct pci));
-
-  pci->vendor_id = pciutils_get_pci_vendor_id(devices);
-  pci->device_id = pciutils_get_pci_device_id(devices);
-
-  return pci;
-}
-
 /*
  * pci ids were retrieved using https://github.com/pciutils/pciids
  * and parsed using a custom script to take only the relevant

src/cuda/pci.hpp

@@ -6,6 +6,12 @@
 #include "../common/pci.hpp"
 #include "chips.hpp"
 
+/*
+ * doc: https://wiki.osdev.org/PCI#Class_Codes
+ *      https://pci-ids.ucw.cz/read/PC
+ */
+#define PCI_VENDOR_ID_NVIDIA 0x10de
+
 struct pci;
 
 struct pci* get_pci_from_pciutils(struct pci_dev *devices);

src/cuda/uarch.cpp

@@ -3,12 +3,11 @@
 #include <stdint.h>
 #include <cstddef>
 
+#include "../common/uarch.hpp"
 #include "../common/global.hpp"
 #include "../common/gpu.hpp"
 #include "chips.hpp"
 
-typedef uint32_t MICROARCH;
-
 // Any clock multiplier
 #define CM_ANY               -1
 
@@ -43,19 +42,6 @@ static const char *uarch_str[] = {
   /*[ARCH_AMPERE]     = */ "Ampere",
 };
 
-struct uarch {
-  int32_t cc_major;
-  int32_t cc_minor;
-  int32_t compute_capability;
-
-  MICROARCH uarch;
-  GPUCHIP chip;
-
-  int32_t process;
-  char* uarch_str;
-  char* chip_str;
-};
-
 #define CHECK_UARCH_START if (false) {}
 #define CHECK_UARCH(arch, chip_, str, uarch, process) \
    else if (arch->chip == chip_) fill_uarch(arch, str, uarch, process);

src/intel/chips.hpp

@@ -0,0 +1,13 @@
+#ifndef __INTEL_GPUCHIPS__
+#define __INTEL_GPUCHIPS__
+
+#include <stdint.h>
+
+typedef uint32_t GPUCHIP;
+
+enum {
+  CHIP_UNKNOWN_INTEL,
+  CHIP_UHDG_620,
+};
+
+#endif

src/intel/intel.cpp

@@ -3,6 +3,7 @@
 
 #include "intel.hpp"
 #include "uarch.hpp"
+#include "chips.hpp"
 #include "../common/pci.hpp"
 #include "../common/global.hpp"
 
@@ -11,8 +12,11 @@ struct gpu_info* get_gpu_info_intel() {
   const char* name = "UHD Graphics XXX";
 
   gpu->vendor = GPU_VENDOR_INTEL;
-  gpu->name = (char *) emalloc(sizeof(char) * (strlen(name) + 1));
-  strcpy(gpu->name, name);
+
+  struct pci_dev *devices = get_pci_devices_from_pciutils();
+  gpu->pci = get_pci_from_pciutils(devices, PCI_VENDOR_ID_INTEL);
+  gpu->arch = get_uarch_from_pci(gpu->pci);
+  gpu->name = get_name_from_uarch(gpu->arch);
 
   return gpu;
 }

src/intel/pci.cpp

@@ -0,0 +1,17 @@
+#include <stdio.h>
+
+#include "pci.hpp"
+#include "chips.hpp"
+#include "../common/global.hpp"
+#include "../common/pci.hpp"
+
+#define CHECK_PCI_START if (false) {}
+#define CHECK_PCI(pci, id, chip) \
+   else if (pci->device_id == id) return chip;
+#define CHECK_PCI_END else { printBug("TODOO"); return CHIP_UNKNOWN_INTEL; }
+
+GPUCHIP get_chip_from_pci(struct pci* pci) {
+  CHECK_PCI_START
+  CHECK_PCI(pci, 0x5917, CHIP_UHDG_620)
+  CHECK_PCI_END
+}

src/intel/pci.hpp

@@ -0,0 +1,20 @@
+#ifndef __PCI_INTEL__
+#define __PCI_INTEL__
+
+#include <stdint.h>
+
+#include "../common/pci.hpp"
+#include "chips.hpp"
+
+/*
+ * doc: https://wiki.osdev.org/PCI#Class_Codes
+ *      https://pci-ids.ucw.cz/read/PC
+ */
+#define PCI_VENDOR_ID_INTEL 0x8086
+
+struct pci;
+
+struct pci* get_pci_from_pciutils(struct pci_dev *devices);
+GPUCHIP get_chip_from_pci(struct pci* pci);
+
+#endif

src/intel/uarch.cpp

@@ -0,0 +1,59 @@
+#include <stdint.h>
+#include <cstddef>
+#include <string.h>
+
+#include "../common/uarch.hpp"
+#include "../common/global.hpp"
+#include "../common/gpu.hpp"
+#include "chips.hpp"
+
+// Data not available
+#define NA                   -1
+
+// Unknown manufacturing process
+#define UNK                  -1
+
+// MICROARCH values
+enum {
+  UARCH_UNKNOWN,
+  UARCH_GEN9,
+  UARCH_GEN9_5,
+};
+
+static const char *uarch_str[] = {
+  /*[ARCH_UNKNOWN    = */ STRING_UNKNOWN,
+  /*[ARCH_GEN9]      = */ "Gen9",
+  /*[ARCH_GEN9.5]    = */ "Gen9.5",
+};
+
+#define CHECK_UARCH_START if (false) {}
+#define CHECK_UARCH(arch, chip_, str, uarch, process) \
+   else if (arch->chip == chip_) fill_uarch(arch, str, uarch, process);
+#define CHECK_UARCH_END else { printBug("map_chip_to_uarch: Unknown chip id: %d", arch->chip); fill_uarch(arch, STRING_UNKNOWN, UARCH_UNKNOWN, 0); }
+
+void fill_uarch(struct uarch* arch, char const *str, MICROARCH u, uint32_t process) {
+  arch->chip_str = (char *) emalloc(sizeof(char) * (strlen(str)+1));
+  strcpy(arch->chip_str, str);
+  arch->uarch = u;
+  arch->process = process;
+}
+
+void map_chip_to_uarch(struct uarch* arch) {
+  CHECK_UARCH_START
+  CHECK_UARCH(arch, CHIP_UHDG_620, "UHD Graphics 620", UARCH_GEN9_5, 14)
+  CHECK_UARCH_END
+}
+
+struct uarch* get_uarch_from_pci(struct pci* pci) {
+  struct uarch* arch = (struct uarch*) emalloc(sizeof(struct uarch));
+
+  arch->chip_str = NULL;
+  arch->chip = get_chip_from_pci(pci);
+  map_chip_to_uarch(arch);
+
+  return arch;
+}
+
+char* get_name_from_uarch(struct uarch* arch) {
+  return arch->chip_str;
+}

src/intel/uarch.hpp

@@ -3,4 +3,9 @@
 
 #include "../common/gpu.hpp"
 
+struct uarch;
+
+struct uarch* get_uarch_from_pci(struct pci* pci);
+char* get_name_from_uarch(struct uarch* arch);
+
 #endif