You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by sy...@apache.org on 2023/12/16 04:45:49 UTC
(tvm) branch test_all_cases_on_unity updated: fix cutlass gen
This is an automated email from the ASF dual-hosted git repository.
syfeng pushed a commit to branch test_all_cases_on_unity
in repository https://gitbox.apache.org/repos/asf/tvm.git
The following commit(s) were added to refs/heads/test_all_cases_on_unity by this push:
new 242587613a fix cutlass gen
242587613a is described below
commit 242587613a3a2ca2d97098fe5302310b8de3a4c5
Author: Siyuan Feng <hz...@sjtu.edu.cn>
AuthorDate: Sat Dec 16 12:45:31 2023 +0800
fix cutlass gen
---
gallery/how_to/work_with_relay/using_pipeline_executor.py | 3 +--
python/tvm/contrib/cutlass/gen_tensor_op.py | 2 +-
2 files changed, 2 insertions(+), 3 deletions(-)
diff --git a/gallery/how_to/work_with_relay/using_pipeline_executor.py b/gallery/how_to/work_with_relay/using_pipeline_executor.py
index 97496b86dc..8bb53cc743 100644
--- a/gallery/how_to/work_with_relay/using_pipeline_executor.py
+++ b/gallery/how_to/work_with_relay/using_pipeline_executor.py
@@ -109,8 +109,7 @@ subgraphs = graph_split(net["main"], split_config, params)
cutlass = tvm.target.Target(
{
"kind": "cutlass",
- "sm": 75,
- # "sm": int(tvm.target.Target("cuda").arch.split("_")[1]),
+ "sm": int(tvm.target.Target("cuda").arch.split("_")[1]),
"use_3xtf32": True,
"split_k_slices": [1],
"profile_all_alignments": False,
diff --git a/python/tvm/contrib/cutlass/gen_tensor_op.py b/python/tvm/contrib/cutlass/gen_tensor_op.py
index e8f9bc9261..c6aa4c3ab1 100644
--- a/python/tvm/contrib/cutlass/gen_tensor_op.py
+++ b/python/tvm/contrib/cutlass/gen_tensor_op.py
@@ -562,7 +562,7 @@ def instantiate_template(func_name, annotations, func_args):
elif "dense" in func_name or "matmul" in func_name:
batched = "batch" in annotations
- transposed = "transposed" in func_name
+ transposed = "transposed" in func_name or "dense" in func_name
lhs_arg_idx = _get_optional_int_annotation(annotations, "lhs_arg_idx", 0)
rhs_arg_idx = _get_optional_int_annotation(annotations, "rhs_arg_idx", 1)
bias_arg_idx = _get_optional_int_annotation(annotations, "bias_arg_idx", None)