[AUTOTVM] Misc fix to document and style (#2035)

apps/benchmark/util.py

@@ -34,7 +34,7 @@ def get_network(name, batch_size, dtype='float32'):
     elif name == 'mobilenet_v2':
         net, params = nnvm.testing.mobilenet_v2.get_workload(batch_size=batch_size, dtype=dtype)
     elif name == 'inception_v3':
-        input_shape = (1, 3, 299, 299)
+        input_shape = (batch_size, 3, 299, 299)
         net, params = nnvm.testing.inception_v3.get_workload(batch_size=batch_size, dtype=dtype)
     elif "resnet" in name:
         n_layer = int(name.split('-')[1])

docs/install/from_source.rst

@@ -124,13 +124,15 @@ TVM package
 ~~~~~~~~~~~
 
 The python package is located at `tvm/python`
-There are several ways to install the package:
+There are two ways to install the package:
 
-1. Set the environment variable `PYTHONPATH` to tell python where to find
+Method 1
+   This method is **recommended for developers** who may change the codes.
+
+   Set the environment variable `PYTHONPATH` to tell python where to find
    the library. For example, assume we cloned `tvm` on the home directory
    `~`. then we can added the following line in `~/.bashrc`.
-   It is **recommended for developers** who may change the codes.
-   The changes will be immediately reflected once you pulled the code and rebuild the project (no need to call ``setup`` again)
+   The changes will be immediately reflected once you pull the code and rebuild the project (no need to call ``setup`` again)
 
    .. code:: bash
 
@@ -138,7 +140,8 @@ There are several ways to install the package:
        export PYTHONPATH=$TVM_HOME/python:$TVM_HOME/topi/python:$TVM_HOME/nnvm/python:${PYTHONPATH}
 
 
-2. Install tvm python bindings by `setup.py`:
+Method 2
+   Install tvm python bindings by `setup.py`:
 
    .. code:: bash
 

python/tvm/autotvm/measure/measure_methods.py

@@ -551,7 +551,9 @@ def check_remote(target, device_key, host=None, port=None, priority=100, timeout
     """
     def _check():
         remote = request_remote(device_key, host, port, priority)
-        remote.context(str(target))
+        ctx = remote.context(str(target))
+        while not ctx.exist:  # wait until we get an available device
+            pass
     t = threading.Thread(target=_check,)
     t.start()
     t.join(timeout)

python/tvm/autotvm/record.py

@@ -252,13 +252,13 @@ Usage:
 This record executable module has three modes.
 
 * Print log file in readable format
-e.g. python -m autotvm.record --mode read --i collect_conv.log --begin 0 --end 5 --ir --code
+e.g. python -m tvm.autotvm.record --mode read --i collect_conv.log --begin 0 --end 5 --ir --code
 
 * Extract history best from a large log file
-e.g. python -m autotvm.record --mode pick --i collect.log
+e.g. python -m tvm.autotvm.record --mode pick --i collect.log
 
 * Split a log file into separate files, each of which contains only a single wkl
-e.g. python -m autotvm.record --mode split --i collect.log
+e.g. python -m tvm.autotvm.record --mode split --i collect.log
 """
 if __name__ == '__main__':
     parser = argparse.ArgumentParser()

python/tvm/autotvm/task/dispatcher.py

@@ -292,17 +292,13 @@ class ApplyHistoryBest(DispatchContext):
                         best_by_targetkey[key] = (inp, res)
 
             # use model as key to build best map
-            for opt in inp.target.options:
-                if opt.startswith("-model"):
-                    model = opt[7:]
-                    key = (model, inp.task.workload)
-                    if key not in best_by_model:
-                        best_by_model[key] = (inp, res)
-                    else:
-                        _, other_res = best_by_model[key]
-                        if np.mean(other_res.costs) > np.mean(res.costs):
-                            best_by_model[key] = (inp, res)
-                    break
+            key = (inp.target.model, inp.task.workload)
+            if key not in best_by_model:
+                best_by_model[key] = (inp, res)
+            else:
+                _, other_res = best_by_model[key]
+                if np.mean(other_res.costs) > np.mean(res.costs):
+                    best_by_model[key] = (inp, res)
 
         logger.debug("Finish loading %d records", counter)
 
@@ -313,14 +309,11 @@ class ApplyHistoryBest(DispatchContext):
                                " above the dispatcher call. So does other target. ")
 
         # first try matching by model
-        for opt in target.options:
-            if opt.startswith("-model"):
-                model = opt[7:]
-                key = (model, workload)
-                if key in self._best_user_defined:
-                    return self._best_user_defined[key]
-                if key in self.best_by_model:
-                    return self.best_by_model[key][0].config
+        key = (target.model, workload)
+        if key in self._best_user_defined:
+            return self._best_user_defined[key]
+        if key in self.best_by_model:
+            return self.best_by_model[key][0].config
 
         # then try matching by target key
         for k in target.keys:
@@ -333,11 +326,9 @@ class ApplyHistoryBest(DispatchContext):
         return None
 
     def update(self, target, workload, cfg):
-        for opt in target.options:
-            if opt.startswith("-model"):
-                model = opt[7:]
-                key = (model, workload)
-                self._best_user_defined[key] = cfg
+        model = target.model
+        key = (model, workload)
+        self._best_user_defined[key] = cfg
 
         for k in target.keys:
             key = (k, workload)

python/tvm/relay/testing/init.py

@@ -1,7 +1,8 @@
 """Initializer of parameters."""
+import numpy as np
+
 import tvm
 from tvm import relay
-import numpy as np
 
 class Initializer(object):
     """The base class of an initializer."""
@@ -103,7 +104,7 @@ class Xavier(Initializer):
             raise ValueError("Incorrect factor type")
         # Hack for mobilenet, because there is less connectivity
         if "depthwise" in name:
-            factor = 3 * 3
+            factor = hw_scale
         scale = np.sqrt(self.magnitude / factor)
         if self.rnd_type == "uniform":
             arr[:] = np.random.uniform(-scale, scale, size=arr.shape)

python/tvm/target.py

@@ -419,7 +419,7 @@ def intel_graphics(model='unknown', options=None):
     return _api_internal._TargetCreate("opencl", *opts)
 
 
-def opengl(options=None):
+def opengl(model='unknown', options=None):
     """Returns a OpenGL target.
 
     Parameters
@@ -427,8 +427,8 @@ def opengl(options=None):
     options : str or list of str
         Additional options
     """
-    options = _merge_opts([], options)
-    return _api_internal._TargetCreate("opengl", *options)
+    opts = _merge_opts(["-model=%s" % model], options)
+    return _api_internal._TargetCreate("opengl", *opts)
 
 
 def arm_cpu(model='unknown', options=None):