[test] Update NVML with src from cudatoolkit

cscs-checks/cuda/nvml_check.py

@@ -4,22 +4,37 @@
 import reframe.utility.sanity as sn
 
 
-@rfm.required_version('>=2.14')
 @rfm.simple_test
 class NvmlCheck(rfm.RegressionTest):
+    ''' This test checks gpu modes with nvml:
+    * COMPUTE MODE:
+    result = nvmlDeviceGetComputeMode(device, &compute_mode);
+
+    * GPU OPERATION MODE (not supported since K20s, keeping as reminder):
+    result = nvmlDeviceGetGpuOperationMode(device, &gom_mode_current,
+                                                   &gom_mode_pending);
+    NVML_GOM_ALL_ON Everything is enabled and running at full speed.
+    NVML_GOM_COMPUTE Designed for running only compute tasks.
+                     Graphics operations < are not allowed.
+    NVML_GOM_LOW_DP Designed for running graphics applications that do not
+                    require < high bandwidth double precision.
+    '''
     def __init__(self):
         super().__init__()
         self.descr = 'check GPU compute mode'
         self.valid_systems = ['daint:gpu', 'dom:gpu']
         self.valid_prog_environs = ['PrgEnv-gnu']
-        self.sourcesdir = os.path.join(self.current_system.resourcesdir,
-                                       'CUDA', 'nvml')
         self.modules = ['craype-accel-nvidia60']
-        self.sourcepath = 'nvml.c'
         self.build_system = 'SingleSource'
+        self.sourcepath = 'example.c'
+        self.prebuild_cmd = [
+            'cp $CUDATOOLKIT_HOME/nvml/example/example.c .',
+            'patch -i ./nvml_example.patch'
+        ]
         self.build_system.ldflags = ['-lnvidia-ml']
         self.sanity_patterns = sn.assert_found(
-            r"compute\s+mode\s+'Exclusive Process'", self.stdout)
+            r"\s+Changing device.s compute mode from 'Exclusive Process' to ",
+            self.stdout)
 
         self.maintainers = ['AJ', 'VK']
-        self.tags = {'production'}
+        self.tags = {'production', 'craype'}

cscs-checks/cuda/src/nvml_example.patch

@@ -0,0 +1,10 @@
+--- example.c.ori	2019-10-11 11:57:51.000000000 +0200
++++ example.c	2019-10-11 11:58:24.000000000 +0200
+@@ -63,6 +63,7 @@
+     unsigned int device_count, i;
+
+     // First initialize NVML library
++    printf("NVML_API_VERSION=%d\n", NVML_API_VERSION);
+     result = nvmlInit();
+     if (NVML_SUCCESS != result)
+     {