More nvapi preparations and display skip reason in benchmark

include/shared.h

@@ -115,8 +115,8 @@ static inline int  CPU_ISSET (int num, cpu_set_t *cs) { return (cs->count & (1 <
  */
 
 #include "ext_ADL.h"
-#include "ext_nvml.h"
 #include "ext_nvapi.h"
+#include "ext_nvml.h"
 
 /**
  * shared stuff
@@ -1430,12 +1430,14 @@ void fsync (int fd);
 
 #ifdef HAVE_HWMON
 
-int hm_get_adapter_index_nvml (HM_ADAPTER_NVML nvGPUHandle[DEVICES_MAX]);
-
 int get_adapters_num_adl (void *adl, int *iNumberAdapters);
 
 int hm_get_adapter_index_adl (hm_attrs_t *hm_device, u32 *valid_adl_device_list, int num_adl_adapters, LPAdapterInfo lpAdapterInfo);
 
+int hm_get_adapter_index_nvapi (HM_ADAPTER_NVAPI nvapiGPUHandle[DEVICES_MAX]);
+
+int hm_get_adapter_index_nvml (HM_ADAPTER_NVML nvmlGPUHandle[DEVICES_MAX]);
+
 LPAdapterInfo hm_get_adapter_info_adl (void *adl, int iNumberAdapters);
 
 u32 *hm_get_list_valid_adl_adapters (int iNumberAdapters, int *num_adl_adapters, LPAdapterInfo lpAdapterInfo);

src/Makefile

@@ -152,11 +152,12 @@ NATIVE_OBJS              := obj/ext_OpenCL.NATIVE.o obj/shared.NATIVE.o obj/rp_k
 
 ifeq ($(UNAME),Linux)
 NATIVE_OBJS              += obj/ext_ADL.NATIVE.o
+NATIVE_OBJS              += obj/ext_nvapi.NATIVE.o
 NATIVE_OBJS              += obj/ext_nvml.NATIVE.o
 endif
 
-LINUX_32_OBJS            := obj/ext_OpenCL.LINUX.32.o obj/shared.LINUX.32.o obj/rp_kernel_on_cpu.LINUX.32.o obj/ext_ADL.LINUX.32.o obj/ext_nvml.LINUX.32.o
+LINUX_32_OBJS            := obj/ext_OpenCL.LINUX.32.o obj/shared.LINUX.32.o obj/rp_kernel_on_cpu.LINUX.32.o obj/ext_ADL.LINUX.32.o obj/ext_nvml.LINUX.32.o obj/ext_nvapi.LINUX.32.o
-LINUX_64_OBJS            := obj/ext_OpenCL.LINUX.64.o obj/shared.LINUX.64.o obj/rp_kernel_on_cpu.LINUX.64.o obj/ext_ADL.LINUX.64.o obj/ext_nvml.LINUX.64.o
+LINUX_64_OBJS            := obj/ext_OpenCL.LINUX.64.o obj/shared.LINUX.64.o obj/rp_kernel_on_cpu.LINUX.64.o obj/ext_ADL.LINUX.64.o obj/ext_nvml.LINUX.64.o obj/ext_nvapi.LINUX.32.o
 
 WIN_32_OBJS              := obj/ext_OpenCL.WIN.32.o   obj/shared.WIN.32.o   obj/rp_kernel_on_cpu.WIN.32.o   obj/ext_ADL.WIN.32.o   obj/ext_nvml.WIN.32.o   obj/ext_nvapi.WIN.32.o
 WIN_64_OBJS              := obj/ext_OpenCL.WIN.64.o   obj/shared.WIN.64.o   obj/rp_kernel_on_cpu.WIN.64.o   obj/ext_ADL.WIN.64.o   obj/ext_nvml.WIN.64.o   obj/ext_nvapi.WIN.64.o

src/ext_nvapi.c

@@ -13,11 +13,15 @@ int nvapi_init (NVAPI_PTR *nvapi)
 
   memset (nvapi, 0, sizeof (NVAPI_PTR));
 
+  #ifdef _WIN
   #if __x86_64__
   nvapi->lib = hc_dlopen ("nvapi64.dll");
   #elif __x86__
   nvapi->lib = hc_dlopen ("nvapi.dll");
   #endif
+  #else
+  nvapi->lib = hc_dlopen ("nvapi.so", RTLD_NOW); // uhm yes, but .. yeah
+  #endif
 
   if (!nvapi->lib)
   {

src/hashcat.c

@@ -13319,6 +13319,7 @@ int main (int argc, char **argv)
      */
 
     int need_adl   = 0;
+    int need_nvapi = 0;
     int need_nvml  = 0;
 
     hc_device_param_t *devices_param = (hc_device_param_t *) mycalloc (DEVICES_MAX, sizeof (hc_device_param_t));
@@ -13577,7 +13578,7 @@ int main (int argc, char **argv)
 
         if (device_endian_little == CL_FALSE)
         {
-          if (data.quiet == 0) log_info ("Device #%u: WARNING: not little endian device", device_id + 1);
+          log_info ("Device #%u: WARNING: not little endian device", device_id + 1);
 
           device_param->skipped = 1;
         }
@@ -13590,7 +13591,7 @@ int main (int argc, char **argv)
 
         if (device_available == CL_FALSE)
         {
-          if (data.quiet == 0) log_info ("Device #%u: WARNING: device not available", device_id + 1);
+          log_info ("Device #%u: WARNING: device not available", device_id + 1);
 
           device_param->skipped = 1;
         }
@@ -13603,7 +13604,7 @@ int main (int argc, char **argv)
 
         if (device_compiler_available == CL_FALSE)
         {
-          if (data.quiet == 0) log_info ("Device #%u: WARNING: device no compiler available", device_id + 1);
+          log_info ("Device #%u: WARNING: device no compiler available", device_id + 1);
 
           device_param->skipped = 1;
         }
@@ -13616,7 +13617,7 @@ int main (int argc, char **argv)
 
         if ((device_execution_capabilities & CL_EXEC_KERNEL) == 0)
         {
-          if (data.quiet == 0) log_info ("Device #%u: WARNING: device does not support executing kernels", device_id + 1);
+          log_info ("Device #%u: WARNING: device does not support executing kernels", device_id + 1);
 
           device_param->skipped = 1;
         }
@@ -13633,14 +13634,14 @@ int main (int argc, char **argv)
 
         if (strstr (device_extensions, "base_atomics") == 0)
         {
-          if (data.quiet == 0) log_info ("Device #%u: WARNING: device does not support base atomics", device_id + 1);
+          log_info ("Device #%u: WARNING: device does not support base atomics", device_id + 1);
 
           device_param->skipped = 1;
         }
 
         if (strstr (device_extensions, "byte_addressable_store") == 0)
         {
-          if (data.quiet == 0) log_info ("Device #%u: WARNING: device does not support byte addressable store", device_id + 1);
+          log_info ("Device #%u: WARNING: device does not support byte addressable store", device_id + 1);
 
           device_param->skipped = 1;
         }
@@ -13655,7 +13656,7 @@ int main (int argc, char **argv)
 
         if (device_local_mem_size < 32768)
         {
-          if (data.quiet == 0) log_info ("Device #%u: WARNING: device local mem size is too small", device_id + 1);
+          log_info ("Device #%u: WARNING: device local mem size is too small", device_id + 1);
 
           device_param->skipped = 1;
         }
@@ -13670,9 +13671,10 @@ int main (int argc, char **argv)
         {
           if (device_param->device_vendor_id == VENDOR_ID_AMD_USE_INTEL)
           {
-            if (data.quiet == 0) log_info ("Device #%u: WARNING: not native intel opencl platform", device_id + 1);
+            log_info ("Device #%u: WARNING: not native intel opencl runtime, expect massive speed loss", device_id + 1);
+            log_info ("           You can use --force to override this but do not post error reports if you do so");
 
-            device_param->skipped = 1;
+            if (data.force == 0) device_param->skipped = 1;
           }
         }
 
@@ -13721,6 +13723,10 @@ int main (int argc, char **argv)
           if ((device_param->platform_vendor_id == VENDOR_ID_NV) && (device_param->device_vendor_id == VENDOR_ID_NV))
           {
             need_nvml = 1;
+
+            #ifdef _WIN
+            need_nvapi = 1;
+            #endif
           }
         }
 
@@ -14042,6 +14048,28 @@ int main (int argc, char **argv)
         }
       }
 
+      if ((need_nvapi == 1) && (nvapi_init (nvapi) == 0))
+      {
+        data.hm_nvapi = nvapi;
+      }
+
+      if (data.hm_nvapi)
+      {
+        if (hm_NvAPI_Initialize (data.hm_nvapi) == NVAPI_OK)
+        {
+          HM_ADAPTER_NVAPI nvGPUHandle[DEVICES_MAX] = { 0 };
+
+          int tmp_in = hm_get_adapter_index_nvapi (nvGPUHandle);
+
+          int tmp_out = 0;
+
+          for (int i = 0; i < tmp_in; i++)
+          {
+            hm_adapters_nvapi[tmp_out++].adapter_index.nvapi = nvGPUHandle[i];
+          }
+        }
+      }
+
       if ((need_adl == 1) && (adl_init (adl) == 0))
       {
         data.hm_adl = adl;

src/shared.c

@@ -2672,31 +2672,6 @@ void fsync (int fd)
 
 #ifdef HAVE_HWMON
 
-int hm_get_adapter_index_nvml (HM_ADAPTER_NVML nvGPUHandle[DEVICES_MAX])
-{
-  int pGpuCount = 0;
-
-  for (uint i = 0; i < DEVICES_MAX; i++)
-  {
-    if (hm_NVML_nvmlDeviceGetHandleByIndex (data.hm_nvml, 1, i, &nvGPUHandle[i]) != NVML_SUCCESS) break;
-
-    // can be used to determine if the device by index matches the cuda device by index
-    // char name[100]; memset (name, 0, sizeof (name));
-    // hm_NVML_nvmlDeviceGetName (data.hm_nvml, nvGPUHandle[i], name, sizeof (name) - 1);
-
-    pGpuCount++;
-  }
-
-  if (pGpuCount == 0)
-  {
-    log_info ("WARN: No NVML adapters found");
-
-    return (0);
-  }
-
-  return (pGpuCount);
-}
-
 int get_adapters_num_adl (void *adl, int *iNumberAdapters)
 {
   if (hm_ADL_Adapter_NumberOfAdapters_Get ((ADL_PTR *) adl, iNumberAdapters) != ADL_OK) return -1;
@@ -2758,6 +2733,47 @@ LPAdapterInfo hm_get_adapter_info_adl (void *adl, int iNumberAdapters)
   return lpAdapterInfo;
 }
 
+int hm_get_adapter_index_nvapi (HM_ADAPTER_NVAPI nvapiGPUHandle[DEVICES_MAX])
+{
+  NvU32 pGpuCount;
+
+  if (hm_NvAPI_EnumPhysicalGPUs (data.hm_nvapi, nvapiGPUHandle, &pGpuCount) != NVAPI_OK) return (0);
+
+  if (pGpuCount == 0)
+  {
+    log_info ("WARN: No NvAPI adapters found");
+
+    return (0);
+  }
+
+  return (pGpuCount);
+}
+
+int hm_get_adapter_index_nvml (HM_ADAPTER_NVML nvmlGPUHandle[DEVICES_MAX])
+{
+  int pGpuCount = 0;
+
+  for (uint i = 0; i < DEVICES_MAX; i++)
+  {
+    if (hm_NVML_nvmlDeviceGetHandleByIndex (data.hm_nvml, 1, i, &nvmlGPUHandle[i]) != NVML_SUCCESS) break;
+
+    // can be used to determine if the device by index matches the cuda device by index
+    // char name[100]; memset (name, 0, sizeof (name));
+    // hm_NVML_nvmlDeviceGetName (data.hm_nvml, nvGPUHandle[i], name, sizeof (name) - 1);
+
+    pGpuCount++;
+  }
+
+  if (pGpuCount == 0)
+  {
+    log_info ("WARN: No NVML adapters found");
+
+    return (0);
+  }
+
+  return (pGpuCount);
+}
+
 /*
 //
 // does not help at all, since ADL does not assign different bus id, device id when we have multi GPU setups