apache · tqchen · Sep 15, 2020 · Sep 15, 2020
diff --git a/include/tvm/auto_scheduler/compute_dag.h b/include/tvm/auto_scheduler/compute_dag.h
@@ -53,11 +53,11 @@ class AccessAnalyzerNode : public Object {
   using OperationMap = std::unordered_map<te::Operation, T, ObjectPtrHash, ObjectPtrEqual>;
 
   /*! \brief Map an operation to all operations it reads from.
-   * For each operation pair, use a two-dimentional array for multiple multi-dimentional accesses
+   * For each operation pair, use a two-dimensional array for multiple multi-dimensional accesses
    * The inner vector represents the indices of multi-dimensional access.*/
   OperationMap<OperationMap<std::vector<std::vector<PrimExpr>>>> read_from;
   /*! \brief Map an operation to all operations it is read by.
-   * For each operation pair, use a two-dimentional array for multiple multi-dimentional accesses
+   * For each operation pair, use a two-dimensional array for multiple multi-dimensional accesses
    * The inner vector represents the indices of multi-dimensional access.*/
   OperationMap<OperationMap<std::vector<std::vector<PrimExpr>>>> read_by;
   /*! \brief Store the number of common outer iterators for operation pairs that have
@@ -67,7 +67,7 @@ class AccessAnalyzerNode : public Object {
    *  (e.g., injective, broadcast and elementwise ops without reduction) */
   OperationMap<bool> is_simple_access;
   /*! \brief Store whether the operation is strictly inlineable
-   * (e.g., injective, broadcast and elementwise without reduction, branch or expenive operations)
+   * (e.g., injective, broadcast and elementwise without reduction, branch or expensive operations)
    */
   OperationMap<bool> is_strictly_inlineable;
   /*! \brief Store whether the operation needs multi-level tiling
@@ -98,8 +98,8 @@ class AccessAnalyzer : public ObjectRef {
   TVM_DLL bool IsSimpleAccess(const te::Operation& op) const;
 
   /*!
-   * \brief Return whether this operation is strictly inlinable
-   * (e.g., injective, broadcast and elementwise without reduction, branch or expenive operations)
+   * \brief Return whether this operation is strictly inlineable
+   * (e.g., injective, broadcast and elementwise without reduction, branch or expensive operations)
    * \param op The operation
    */
   TVM_DLL bool IsStrictlyInlineable(const te::Operation& op) const;
@@ -265,7 +265,7 @@ class ComputeDAG : public ObjectRef {
    * \brief Since some steps may change the ComputeDAG (e.g. CacheRead/CacheWrite), the initial
    * ComputeDAG may not be up-to-date. This function replays the given transform steps from the
    * initial state and returns an up-to-date ComputeDAG.
-   * \param steps The steps to be replaied. Usually we'll filter out the unused steps to speed up
+   * \param steps The steps to be replayed. Usually we'll filter out the unused steps to speed up
    * the replay process, since we only intend to get a ComputeDAG with the up-to-date op stage
    * structure.
    * \return The up-to-date ComputeDAG.

diff --git a/include/tvm/auto_scheduler/feature.h b/include/tvm/auto_scheduler/feature.h
@@ -100,9 +100,9 @@ void GetPerStoreFeaturesFromFile(const std::string& filename, int max_lines, int
 
 /*!
  * \brief Get per-store features from measurement input/result pairs
- * \param inputs The meaurement inputs
+ * \param inputs The measurement inputs
  * \param results The measurement results
- * \param skip_first_n_feature_extraction Skip feature extraction for the first n meaurement pairs
+ * \param skip_first_n_feature_extraction Skip feature extraction for the first n measurement pairs
  * \param max_n_bufs The maximum number of extracted buffers for one statement
  * \param features The returned feature vector. The innermost vector contains the
  * feature vectors for all BufferStoreNode statements

diff --git a/include/tvm/auto_scheduler/loop_state.h b/include/tvm/auto_scheduler/loop_state.h
@@ -458,7 +458,7 @@ namespace std {
 
 /*!
  * \brief The equal_to function for auto_scheduler::State.
- * This function checkes the equality by looking at the lowered string format of states.
+ * This function checks the equality by looking at the lowered string format of states.
  * If two states with different transform history have the same lowered string format,
  * they will be considered being equal.
  */

diff --git a/include/tvm/auto_scheduler/measure.h b/include/tvm/auto_scheduler/measure.h
@@ -71,7 +71,7 @@ enum class MeasureErrorNO : int {
   /*! \brief Timeout during run. */
   kRunTimeoutError = 7,
   /*! \brief Unknown error. */
-  kUnknonwError = 8,
+  kUnknownError = 8,
 };
 
 // Inputs and results of one measurement
@@ -104,7 +104,7 @@ class MeasureInput : public ObjectRef {
  public:
   /*!
    * \brief The constructor.
-   * \param task The SearchTeask of this measure.
+   * \param task The SearchTask of this measure.
    * \param state The State to be measured.
    */
   MeasureInput(SearchTask task, State state);
@@ -353,7 +353,7 @@ class LocalRunner : public ProgramRunner {
  public:
   /*!
    * \brief The constructor. See the corresponding class in python/tvm/auto_scheduler/measure.py
-   * for more detailed parameter explaination.
+   * for more detailed parameter explanation.
    * \param timeout The timeout limit (in second) for each run.
    * This is used in a wrapper of the multiprocessing.Process.join().
    * \param number The number of times to run the generated code for taking average.
@@ -401,7 +401,7 @@ class RPCRunner : public ProgramRunner {
  public:
   /*!
    * \brief The constructor. See the corresponding class in python/tvm/auto_scheduler/measure.py
-   * for more detailed parameter explaination.
+   * for more detailed parameter explanation.
    * \param key The key of the device registered in the RPC tracker.
    * \param host The host address of the RPC Tracker.
    * \param port The port of RPC Tracker.
@@ -445,7 +445,7 @@ class ProgramMeasurerNode : public Object {
   /*! \brief Verbosity level. 0 for silent, 1 to output information during program measuring. */
   int verbose;
   /*! \brief The number of max continuous error. */
-  int max_continous_error;
+  int max_continuous_error;
 
   /*! \brief Reset book keeping variables */
   void Reset();
@@ -472,7 +472,7 @@ class ProgramMeasurerNode : public Object {
                      Array<MeasureResult>* results);
 
   /*! \brief The default max continuous error setting. */
-  static const int DEFAULT_MAX_CONTINOUS_ERROR = 150;
+  static const int DEFAULT_MAX_CONTINUOUS_ERROR = 150;
 
   static constexpr const char* _type_key = "auto_scheduler.ProgramMeasurer";
   TVM_DECLARE_FINAL_OBJECT_INFO(ProgramMeasurerNode, Object);
@@ -491,11 +491,11 @@ class ProgramMeasurer : public ObjectRef {
    * \param callbacks MeasureCallback to be called after each measure batch.
    * \param verbose Verbosity level. 0 for silent, 1 to output information during program
    * measuring.
-   * \param max_continous_error The number of allowed maximum continuous error.
+   * \param max_continuous_error The number of allowed maximum continuous error.
    */
   ProgramMeasurer(ProgramBuilder builder, ProgramRunner runner,
                   Optional<Array<MeasureCallback>> callbacks, int verbose,
-                  int max_continous_error = -1);
+                  int max_continuous_error = -1);
 
   TVM_DEFINE_MUTABLE_OBJECT_REF_METHODS(ProgramMeasurer, ObjectRef, ProgramMeasurerNode);
 };

diff --git a/python/tvm/auto_scheduler/cost_model/cost_model.py b/python/tvm/auto_scheduler/cost_model/cost_model.py
@@ -55,7 +55,7 @@ def predict(self, search_task, states):
         ----------
         search_task : SearchTask
             The search task of states
-        statse : List[State]
+        states : List[State]
             The input states
 
         Returns
@@ -126,7 +126,7 @@ def predict(self, task, states):
         ----------
         search_task : SearchTask
             The search task of states
-        statse : List[State]
+        states : List[State]
             The input states
 
         Returns
@@ -143,7 +143,7 @@ def predict_stages(self, task, states):
         ----------
         search_task : SearchTask
             The search task of states
-        statse : List[State]
+        states : List[State]
             The input states
 
         Returns
@@ -166,7 +166,7 @@ def predict_stages(self, task, states):
           ...
           int    n_stage_i;                 // the number of stages in states[i]
           float  stage_scores_1[n_stage_i]; // the scores for all stages in states[i]
-          ...  // untill i == N - 1
+          ...  // until i == N - 1
         }
         To implement this format, we also store int as float, so we can store all numbers
         into a single float array.

diff --git a/python/tvm/auto_scheduler/cost_model/xgb_model.py b/python/tvm/auto_scheduler/cost_model/xgb_model.py
@@ -44,7 +44,6 @@ def __init__(self):
     def get(self, key, matrix, default=None):
         """
         Get an attribute of a xgb.DMatrix
-
         Parameters
         ----------
         key: str
@@ -59,7 +58,6 @@ def get(self, key, matrix, default=None):
     def set(self, key, matrix, value):
         """
         Set an attribute for a xgb.DMatrix
-
         Parameters
         ----------
         key: str
@@ -77,15 +75,12 @@ def set(self, key, matrix, value):
 
 class XGBModel(PythonBasedModel):
     """Train a XGBoost model to predict the normalized throughputs of programs.
-
     Let the normalized throughput be the score of a program (higher is better). We predict
     the (approximiate) score of a program = the sum of the scores of all stages in this program.
     i.e. score(P) = score_s0 + score_s1 + ... + score_sn,
     where score_si is the score of Stage i in Program P.
-
     We extract feature for each stage and let the xgboost predict the score for each stage.
     We then sum up the predictions as the score of the whole program.
-
     We use RMSE as the loss function.  i.e. loss(P, y) = 1/2 * (score(P) - y)^2,
     where P is the program and y is the normalized throughput according to
     the ground truth (measurement).
@@ -123,7 +118,6 @@ def __init__(self, verbose_eval=25, num_warmup_sample=100, seed=None):
     def update(self, inputs, results):
         """Update the cost model according to new measurement results (training data).
         XGBoost does not support incremental training, so we re-train a new model every time.
-
         Parameters
         ----------
         inputs : List[MeasureInput]
@@ -175,14 +169,12 @@ def update(self, inputs, results):
 
     def predict(self, task, states):
         """Predict the scores of states
-
         Parameters
         ----------
         search_task : SearchTask
             The search task of states
         statse : List[State]
             The input states
-
         Returns
         -------
         scores: List[float]
@@ -205,24 +197,20 @@ def predict(self, task, states):
 
     def predict_stages(self, task, states):
         """Predict the scores of all stages in states. This is the breakdown version of `predict`.
-
         Parameters
         ----------
         search_task : SearchTask
             The search task of states
         statse : List[State]
             The input states
-
         Returns
         -------
         scores: List[float]
             The predicted scores for all stages in all states in the packed format
-
         Note
         ----
         For faster data copy between c++ and python, the python part returns scores in a
         single flatten array using a packed format. The c++ part then unpacks the flatten array.
-
         The packed format is:
         {
           float  scores[N];                 // scores[i] is the score for states[i].
@@ -269,7 +257,6 @@ def predict_stages(self, task, states):
     def update_from_file(self, file_name, n_lines=None):
         """Load measure records from a log file to update the cost model.
         This function can be used to pre-train the cost model with history log files.
-
         Parameters
         ----------
         file_name: str
@@ -283,7 +270,6 @@ def update_from_file(self, file_name, n_lines=None):
 
     def save(self, file_name: str):
         """Save the model to a file
-
         Parameters
         ----------
         file_name: str
@@ -293,7 +279,6 @@ def save(self, file_name: str):
 
     def load(self, file_name: str):
         """Load the model from a file
-
         Parameters
         ----------
         file_name: str
@@ -307,12 +292,10 @@ def load(self, file_name: str):
 
 def feature_to_pack_sum_xgbmatrix(xs):
     """Convert an extracted multi-stage feature vector to a xgbmatrx in pack-sum format
-
     Parameters
     ----------
     xs: np.ndarray
         The feature vector
-
     Returns
     -------
     dmatrix: xgb.DMatrix
@@ -333,7 +316,6 @@ def feature_to_pack_sum_xgbmatrix(xs):
 
 def pack_sum_xgbmatrix(xs, ys, gids=None, weights=None):
     """Convert (feature, label) pairs into a xgb matrix with pack-sum format
-
     Parameters
     ----------
     xs: np.ndarray
@@ -344,7 +326,6 @@ def pack_sum_xgbmatrix(xs, ys, gids=None, weights=None):
         Group id (task id)
     weights: Optional[np.ndarray]
         The weight of samples
-
     Returns
     -------
     dmatrix: xgb.DMatrix
@@ -390,14 +371,12 @@ def pack_sum_xgbmatrix(xs, ys, gids=None, weights=None):
 
 def predict_throughput_pack_sum(raw_preds, pack_ids):
     """Predict the throughputs for predictions in pack-sum format
-
     Parameters
     ----------
     raw_preds: np.ndarray
         The raw predictions
     pack_ids: List[int]
         The pack id for predictions
-
     Returns
     -------
     throughputs: np.ndarray
@@ -410,14 +389,12 @@ def predict_throughput_pack_sum(raw_preds, pack_ids):
 def pack_sum_square_error(preds, dtrain):
     """Implement square error loss on pack-sum format as
      a custom objective function for xgboost.
-
     Parameters
     ----------
     preds: np.ndarray
         The predicitons
     dtrain: xgb.DMatrix
         The training set
-
     Returns
     -------
     gradient: np.ndarray
@@ -441,14 +418,12 @@ def pack_sum_square_error(preds, dtrain):
 
 def pack_sum_rmse(raw_preds, labels):
     """Evaluate RMSE (rooted mean square error) in the pack-sum format
-
     Parameters
     ----------
     raw_preds: np.ndarray
         The raw prediction
     labels: xgb.DMatrix
         The groud-truth label matrix
-
     Returns
     -------
     name: str
@@ -462,27 +437,23 @@ def pack_sum_rmse(raw_preds, labels):
 
 def pack_sum_average_peak_score(N):
     """Return the evaluation function for average-peak-score@N
-
     Parameters
     ----------
     N: int
         The "N" in "average-peak-score@N"
-
     Returns
     -------
     The evaluation function
     """
 
     def feval(preds, labels):
         """Evaluate average-peak-score@N in the pack-sum format
-
         Parameters
         ----------
         raw_preds: np.ndarray
             The raw prediction
         labels: xgb.DMatrix
             The groud-truth label matrix
-
         Returns
         -------
         name: str

diff --git a/python/tvm/auto_scheduler/feature.py b/python/tvm/auto_scheduler/feature.py
@@ -79,7 +79,7 @@ def unpack_feature(byte_arr: bytearray) -> Tuple[np.ndarray, np.ndarray, np.ndar
       float features_i[size[i]];  // The features for record i
       ... // until i == n - 1
 
-      float throuputs[sizes[n]];  // The normalized throughputs for n records
+      float throughputs[sizes[n]];  // The normalized throughputs for n records
       int   task_ids[size[n+1];   // The task ids for n records
 
     }