Merge branch 'master' into refactor_interface

Author: isazi

README.rst

@@ -34,7 +34,7 @@ Or both:
 
 More information about how to install Kernel Tuner and its
 dependencies can be found in the `installation guide 
-<http://kerneltuner.github.io/kernel_tuner/stable/install.html>`__
+<http://kerneltuner.github.io/kernel_tuner/stable/install.html>`__.
 
 Example usage
 -------------
@@ -97,16 +97,17 @@ Search strategies for tuning
 Kernel Tuner supports many optimization algorithms to accelerate the auto-tuning process. Currently 
 implemented search algorithms are: Brute Force (default), Nelder-Mead, Powell, CG, BFGS, L-BFGS-B, TNC, 
 COBYLA, SLSQP, Random Search, Basinhopping, Differential Evolution, a Genetic Algorithm, Particle Swarm 
-Optimization, the Firefly Algorithm, and Simulated Annealing.
+Optimization, the Firefly Algorithm, Simulated Annealing, Dual Annealing, Iterative Local Search,
+Multi-start Local Search, and Bayesian Optimization.
 
 .. image:: doc/gemm-amd-summary.png
     :width: 100%
     :align: center
 
 Using a search strategy is easy, you only need to specify to ``tune_kernel`` which strategy and method 
 you would like to use, for example ``strategy="genetic_algorithm"`` or ``strategy="basinhopping"``. 
-For a full overview of the supported search strategies and methods please see the `user 
-api documentation <http://kerneltuner.github.io/kernel_tuner/stable/user-api.html>`__.
+For a full overview of the supported search strategies and methods please see the 
+Kernel Tuner documentation on `Optimization Strategies <https://kerneltuner.github.io/kernel_tuner/stable/optimization.html>`__.
 
 Tuning host and kernel code
 ---------------------------

kernel_tuner/file_utils.py

@@ -155,7 +155,14 @@ def get_device_query(target):
         nvidia_smi_out = subprocess.run(["nvidia-smi", "--query", "-x"],
                                         capture_output=True)
         nvidia_smi = xmltodict.parse(nvidia_smi_out.stdout)
-        del nvidia_smi["nvidia_smi_log"]["gpu"]["processes"]
+        gpu_info = nvidia_smi["nvidia_smi_log"]["gpu"]
+        del_key = "processes"
+        # on multi-GPU systems gpu_info is a list
+        if isinstance(gpu_info, list):
+            for gpu in gpu_info:
+                del gpu[del_key]
+        elif isinstance(gpu_info, dict) and del_key in gpu_info:
+                del gpu_info[del_key]
         return nvidia_smi
     elif target == "amd":
         rocm_smi_out = subprocess.run(["rocm-smi", "--showallinfo", "--json"],

kernel_tuner/runners/sequential.py

@@ -95,7 +95,7 @@ def run(self, parameter_space, tuning_options):
                 params.update(result)
 
                 # only compute metrics on configs that have not errored
-                if isinstance(result[tuning_options.objective], ErrorConfig):
+                if tuning_options.objective in result and isinstance(result[tuning_options.objective], ErrorConfig):
                     logging.debug('kernel configuration was skipped silently due to compile or runtime failure')
                 elif tuning_options.metrics:
                     params = process_metrics(params, tuning_options.metrics)

kernel_tuner/util.py

@@ -758,6 +758,7 @@ def process_cache(cache, kernel_options, tuning_options, runner):
         c["problem_size"] = kernel_options.problem_size if not callable(kernel_options.problem_size) else "callable"
         c["tune_params_keys"] = list(tuning_options.tune_params.keys())
         c["tune_params"] = tuning_options.tune_params
+        c["objective"] = tuning_options.objective
         c["cache"] = {}
 
         contents = json.dumps(c, cls=NpEncoder, indent="")[:-3]    # except the last "}\n}"

test/test_runners.py

@@ -261,7 +261,8 @@ def test_runner(env):
     platform = 0
     iterations = 7
     verbose = False
-    objective = "time"
+    objective = "GFLOP/s"
+    metrics = OrderedDict({objective: lambda p: 1})
     opts = locals()
     kernel_options = Options([(k, opts.get(k, None))
                               for k in _kernel_options.keys()])

test/test_util_functions.py

@@ -550,7 +550,7 @@ def assert_open_cachefile_is_correctly_parsed(cache):
     delete_temp_file(cache)
 
     kernel_options = Options(kernel_name="test_kernel", problem_size=(1, 2))
-    tuning_options = Options(cache=cache, tune_params=Options(x=[1, 2, 3, 4]), simulation_mode=False)
+    tuning_options = Options(cache=cache, tune_params=Options(x=[1, 2, 3, 4]), simulation_mode=False, objective="time")
     runner = Options(dev=Options(name="test_device"), simulation_mode=False)
 
     try: