Added L2 cache size property to OpenCL, HIP and mocked PyCUDA backends

Author: fjwillemsen

kernel_tuner/backends/hip.py

@@ -59,6 +59,7 @@ def __init__(self, device=0, iterations=7, compiler_options=None, observers=None
 
         self.name = self.hipProps._name.decode('utf-8')
         self.max_threads = self.hipProps.maxThreadsPerBlock
+        self.cache_size_L2 = self.hipProps.l2CacheSize
         self.device = device
         self.compiler_options = compiler_options or []
         self.iterations = iterations

kernel_tuner/backends/opencl.py

@@ -45,6 +45,10 @@ def __init__(
         self.max_threads = self.ctx.devices[0].get_info(
             cl.device_info.MAX_WORK_GROUP_SIZE
         )
+        # TODO the L2 cache size request fails
+        # self.cache_size_L2 = self.ctx.devices[0].get_info(
+        #     cl.device_affinity_domain.L2_CACHE
+        # )
         self.compiler_options = compiler_options or []
 
         # observer stuff

test/test_pycuda_mocked.py

@@ -13,7 +13,8 @@ def setup_mock(drv):
     context = Mock()
     devprops = {'MAX_THREADS_PER_BLOCK': 1024,
                 'COMPUTE_CAPABILITY_MAJOR': 5,
-                'COMPUTE_CAPABILITY_MINOR': 5}
+                'COMPUTE_CAPABILITY_MINOR': 5,
+                'L2_CACHE_SIZE': 4096}
     context.return_value.get_device.return_value.get_attributes.return_value = devprops
     context.return_value.get_device.return_value.compute_capability.return_value = "55"
     drv.Device.return_value.retain_primary_context.return_value = context()