This is an automated email from the ASF dual-hosted git repository.
github-bot pushed a change to branch nightly
in repository https://gitbox.apache.org/repos/asf/tvm.git
from 443fa20e20 [RUNTIME] Update Module and Registry to use String
Container (#14902)
add 9da026194f [OpenCL][Adreno] Fix conv2d when output channels < 4
(#14996)
add 8543cec133 [Hexagon] Remove "c" as aot_host_target
tvm/contrib/hexagon/pytest_pl… (#14997)
add b13be936a0 [DataType] Initial support of fp8 (e4m3/e5m2) (#14863)
add 1608ca82ce [CMAKE] Introduce dummy build as an option (#15000)
add ca30b13668 [DOCKER][ADRENO] Docker setup changes for multi user
environments (#15003)
add 7f02606af2 [OPENCL] Always use convert_T for type conversion (#14972)
add 00126b0484 [IR,TE,TIR] Use f-strings for string formatting, NFC
(#14990)
No new revisions were added by this update.
Summary of changes:
CMakeLists.txt | 27 ++-
cmake/modules/LibInfo.cmake | 1 +
docker/install/ubuntu_install_python_package.sh | 3 +-
include/tvm/runtime/data_type.h | 34 +++
include/tvm/tir/op.h | 3 +-
include/tvm/tir/transform.h | 14 ++
python/gen_requirements.py | 1 +
python/tvm/_ffi/runtime_ctypes.py | 22 ++
python/tvm/contrib/hexagon/pytest_plugin.py | 2 +-
python/tvm/contrib/nvcc.py | 17 ++
python/tvm/ir/container.py | 4 +-
python/tvm/ir/expr.py | 6 +-
python/tvm/ir/json_compact.py | 4 +-
python/tvm/relay/op/strategy/adreno.py | 47 +++-
python/tvm/runtime/ndarray.py | 19 ++
python/tvm/te/hybrid/module.py | 2 +-
python/tvm/te/hybrid/parser.py | 10 +-
python/tvm/te/hybrid/preprocessor.py | 4 +-
python/tvm/te/hybrid/utils.py | 6 +-
python/tvm/te/operation.py | 11 +-
python/tvm/te/schedule.py | 2 +-
python/tvm/te/tag.py | 2 +-
python/tvm/te/tensor.py | 4 +-
python/tvm/tir/buffer.py | 10 +-
python/tvm/tir/ir_builder.py | 2 +-
python/tvm/tir/schedule/schedule.py | 170 +++++--------
python/tvm/tir/schedule/state.py | 9 +-
python/tvm/tir/tensor_intrin/arm_cpu.py | 8 +-
python/tvm/tir/tensor_intrin/cuda.py | 17 +-
python/tvm/tir/transform/transform.py | 28 +++
src/driver/driver_api.cc | 2 +
src/ir/expr.cc | 14 +-
src/support/libinfo.cc | 1 +
src/support/scalars.h | 12 +
src/target/source/codegen_cuda.cc | 17 ++
src/target/source/codegen_cuda.h | 5 +-
src/target/source/codegen_opencl.cc | 10 +-
src/tir/op/op.cc | 7 +
src/tir/transforms/dtype_conversion.cc | 101 ++++++++
src/tir/transforms/dtype_conversion.h | 165 +++++++++++++
...6_legalize.cc => unsupported_dtype_legalize.cc} | 264 +++++++++++++--------
.../opencl_texture/test_conv2d_nchw_texture.py | 30 +++
.../opencl_texture/test_conv2d_nhwc_texture.py | 30 +++
.../test_depthwise_conv2d_nchw_texture.py | 31 +++
.../test_depthwise_conv2d_nhwc_texture.py | 31 +++
tests/python/unittest/test_datatype_nv_fp8.py | 104 ++++++++
.../python/unittest/test_target_codegen_opencl.py | 80 +++----
.../unittest/test_target_texture_codegen_opencl.py | 38 +--
.../unittest/test_te_schedule_bound_inference.py | 56 +++--
.../unittest/test_tir_transform_bf16_legalize.py | 5 +-
.../unittest/test_tir_transform_fp8_legalize.py | 224 +++++++++++++++++
tests/scripts/setup-adreno-env.sh | 13 +-
tests/scripts/task_python_adreno.sh | 13 +-
53 files changed, 1348 insertions(+), 394 deletions(-)
create mode 100644 src/tir/transforms/dtype_conversion.cc
create mode 100644 src/tir/transforms/dtype_conversion.h
rename src/tir/transforms/{bf16_legalize.cc => unsupported_dtype_legalize.cc}
(69%)
create mode 100644 tests/python/unittest/test_datatype_nv_fp8.py
create mode 100644 tests/python/unittest/test_tir_transform_fp8_legalize.py
