[tvm] branch main updated: [Hexagon] Use VTCM while scheduling conv2d op (#13388)

[ju...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

junrushao pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git


The following commit(s) were added to refs/heads/main by this push:
     new bac450a645 [Hexagon] Use VTCM while scheduling conv2d op (#13388)
bac450a645 is described below

commit bac450a645c3f9e3a69a6f7af207cff462250bcf
Author: abhikran-quic <63...@users.noreply.github.com>
AuthorDate: Wed Nov 16 02:04:38 2022 +0530

    [Hexagon] Use VTCM while scheduling conv2d op (#13388)
    
    Use VTCM for E2E execution of conv2d to reduce tuning time.
---
 .../contrib/test_hexagon/metaschedule_e2e/test_resnet50_int8.py      | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/tests/python/contrib/test_hexagon/metaschedule_e2e/test_resnet50_int8.py b/tests/python/contrib/test_hexagon/metaschedule_e2e/test_resnet50_int8.py
index 9edf5877fd..91eb67bbf4 100644
--- a/tests/python/contrib/test_hexagon/metaschedule_e2e/test_resnet50_int8.py
+++ b/tests/python/contrib/test_hexagon/metaschedule_e2e/test_resnet50_int8.py
@@ -248,8 +248,9 @@ def _schedule_packed_8x8x32_conv2d():
 
         # Add cache for input and output activation layout transform,
         # note that weight is already in correct layout
-        input_cache = sch.cache_read(conv2d_block, 0, "global")  # pylint: disable=unused-variable
-        output_cache = sch.cache_write(outer_block, 0, "global")  # pylint: disable=unused-variable
+        # pylint: disable=unused-variable
+        input_cache = sch.cache_read(conv2d_block, 0, "global.vtcm")
+        output_cache = sch.cache_write(outer_block, 0, "global.vtcm")
         # Transform the layout of the input
         sch.transform_layout(
             conv2d_block, ("read", 0), index_map=index_map_nchw32c_nchw8h8w32c, pad_value=0