[GitHub] [tvm] insop commented on a change in pull request #7070: Add autoscheduler support to tvmc

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

insop commented on a change in pull request #7070:
URL: https://github.com/apache/tvm/pull/7070#discussion_r544037266



##########
File path: python/tvm/driver/tvmc/autotuner.py
##########
@@ -133,6 +127,85 @@ def add_tune_parser(subparsers):
         default=None,
         help="change the data layout of the whole graph",
     )
+    parser.add_argument(
+        "--enable-autoscheduler",
+        help="enable tuning the graph through the autoscheduler",
+        action="store_true",
+    )
+
+    auto_scheduler_group = parser.add_argument_group(
+        "Autoscheduler options",
+        "Autoscheduler options, used when --enabled-auto-scheduler is provided",
+    )
+
+    auto_scheduler_group.add_argument(
+        "--cache-line-bytes",
+        type=int,
+        default=64,
+        help="the size of cache line in bytes",
+    )
+    auto_scheduler_group.add_argument(
+        "--num-cores",
+        type=int,
+        default=4,
+        help="the number of device cores",
+    )
+    auto_scheduler_group.add_argument(
+        "--vector-unit-bytes",
+        type=int,
+        default=16,
+        help="the width of vector units in bytes",
+    )
+    auto_scheduler_group.add_argument(
+        "--max-shared-memory-per-block",
+        type=int,
+        default=0,
+        help="the max shared memory per block in bytes",
+    )
+    auto_scheduler_group.add_argument(
+        "--max-local-memory-per-block",
+        type=int,
+        default=0,
+        help="the max local memory per block in bytes",
+    )
+    auto_scheduler_group.add_argument(
+        "--max-threads-per-block",
+        type=int,
+        default=0,
+        help="the max number of threads per block",
+    )
+    auto_scheduler_group.add_argument(
+        "--max-vthread-extent",
+        type=int,
+        default=0,
+        help="the max vthread extent",
+    )
+    auto_scheduler_group.add_argument(
+        "--warp-size",
+        type=int,
+        default=0,
+        help="the thread numbers of a warp",
+    )
+    auto_scheduler_group.add_argument(
+        "--include-simple-tasks",
+        help="Whether to extract simple tasks that do not include complicated ops",
+        action="store_true",
+    )
+    auto_scheduler_group.add_argument(
+        "--log-estimated-latency",
+        help="Whether to log the estimated latency to the file after tuning a task",

Review comment:
       Nit, all other help text starts with low case, except the above two texts.




----------------------------------------------------------------
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.

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