[GitHub] [tvm] ekalda commented on a diff in pull request #11250: [microTVM][ARM] Add Relay tests for conv2d registered schedules

[GitBox <gi...@apache.org>]

ekalda commented on code in PR #11250:
URL: https://github.com/apache/tvm/pull/11250#discussion_r873863778


##########
python/tvm/autotvm/task/dispatcher.py:
##########
@@ -178,6 +178,50 @@ def update(self, target, workload, cfg):
         self._config = cfg
 
 
+class ApplyFixedConfig(DispatchContext):
+    """Apply a config of a deterministic schedule.
+
+    Parameters
+    ----------
+    tasks : list[tvm.autotvm.task.task.Task]
+        List of autoTVM tasks.
+    schedule_name : str
+        Name of schedule to use.
+    """
+
+    def __init__(self, tasks, schedule_name: str):
+        super(ApplyFixedConfig, self).__init__()
+        self._schedule_name = schedule_name
+        self._tasks = tasks
+        self.workload = None
+
+    def _query_inside(self, target, workload):
+        """Override query"""
+        self.workload = workload
+
+        # Creat a config from correct task

Review Comment:
   nit:
   ```suggestion
           # Create a config from correct task
   ```



##########
python/tvm/micro/testing/aot_test_utils.py:
##########
@@ -708,31 +708,52 @@ def compile_models(
 
     compiled_mods = list()
     for model in models:
-        with tvm.transform.PassContext(opt_level=3, config=config):
-            # TODO(Mousius) - Remove once executor/runtime are fully removed from Target
-            if use_runtime_executor:
-                executor_factory = tvm.relay.build(
-                    model.module,
-                    target,
-                    executor=executor,
-                    runtime=runtime,
-                    workspace_memory_pools=workspace_memory_pools,
-                    params=model.params,
-                    mod_name=model.name,
-                )
-                compiled_mods.append(
-                    AOTCompiledTestModel(model=model, executor_factory=executor_factory)
-                )
-            else:
-                executor_factory = tvm.relay.build(
-                    model.module,
-                    tvm.target.Target(target, host=target),
-                    params=model.params,
-                    mod_name=model.name,
-                )
-                compiled_mods.append(
-                    AOTCompiledTestModel(model=model, executor_factory=executor_factory)
-                )
+        if schedule_name:
+            # Testing with deterministic schedule
+            task_list = autotvm.task.extract_from_program(
+                model.module, target=target, params=model.params
+            )
+            with tvm.autotvm.apply_fixed_config(task_list, schedule_name):
+                with tvm.transform.PassContext(opt_level=3, config=config):
+                    if use_runtime_executor:
+                        executor_factory = tvm.relay.build(
+                            model.module,
+                            target,
+                            executor=executor,
+                            runtime=runtime,
+                            workspace_memory_pools=workspace_memory_pools,
+                            params=model.params,
+                            mod_name=model.name,
+                        )
+                        compiled_mods.append(
+                            AOTCompiledTestModel(model=model, executor_factory=executor_factory)
+                        )

Review Comment:
   Missing else block?



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org