This is an automated email from the ASF dual-hosted git repository.
syfeng pushed a change to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git
from b9c76f56dc [Runtime] Clean TVM stacktrace in error messages (#15161)
add c8f5595ca5 [MetaSchedule] Introduce MMA Tensor Core Multilevel Tiling
(#14673)
No new revisions were added by this update.
Summary of changes:
include/tvm/tir/schedule/schedule.h | 18 +
include/tvm/tir/transform.h | 12 +
python/tvm/tir/schedule/schedule.py | 40 +
python/tvm/tir/tensor_intrin/cuda.py | 318 +++++
python/tvm/tir/transform/transform.py | 22 +
src/driver/driver_api.cc | 6 +-
.../feature_extractor/per_store_feature.cc | 3 +-
src/meta_schedule/postproc/verify_gpu_code.cc | 3 +-
.../schedule_rule/multi_level_tiling.cc | 9 +-
.../schedule_rule/multi_level_tiling.h | 15 +
.../multi_level_tiling_tensor_core.cc | 325 ++++-
src/meta_schedule/schedule_rule/schedule_rule.cc | 15 +
src/runtime/thread_storage_scope.h | 21 +
src/tir/schedule/concrete_schedule.cc | 10 +
src/tir/schedule/concrete_schedule.h | 3 +
src/tir/schedule/primitive.h | 38 +
src/tir/schedule/primitive/compute_inline.cc | 19 +
src/tir/schedule/primitive/sampling.cc | 106 ++
src/tir/schedule/schedule.cc | 2 +
src/tir/schedule/traced_schedule.cc | 16 +
src/tir/schedule/traced_schedule.h | 3 +
src/tir/transforms/inject_permuted_layout.cc | 266 ++++
src/tir/transforms/inject_software_pipeline.cc | 7 +
src/tir/transforms/memhammer_intermediate_stage.cc | 3 +-
src/tir/transforms/memhammer_lower_auto_copy.cc | 6 +-
src/tir/transforms/memhammer_rewrite_rule.h | 16 +
src/tir/transforms/memhammer_tensorcore_rewrite.cc | 210 ++++
src/tir/transforms/transform_mma_buffer_layout.cc | 192 +++
...meta_schedule_mma_m16n8k8_auto_tensorization.py | 1315 ++++++++++++++++++++
...test_tir_transform_memhammer_lower_auto_copy.py | 127 ++
30 files changed, 3066 insertions(+), 80 deletions(-)
create mode 100644 src/tir/transforms/inject_permuted_layout.cc
create mode 100644 src/tir/transforms/transform_mma_buffer_layout.cc
create mode 100644
tests/python/unittest/test_meta_schedule_mma_m16n8k8_auto_tensorization.py