This is an automated email from the ASF dual-hosted git repository.
tqchen pushed a change to branch unity-staging
in repository https://gitbox.apache.org/repos/asf/tvm.git
discard 7e4d71cc1e [Unity] Statement rewriter for DataflowBlock (#14043)
discard 413329dfa2 [Unity] Relax dataflow pattern language (matching) (#14041)
omit b42f8a44cd [Unity] Update tests to adapt to latest TVMScript syntax
(#14039)
omit 50bcd1b0af [Unity] Disallow inline prim_func in relax IR (#14040)
omit e0775c8c95 [Unity][Pass] Block-level static memory planning (#14038)
omit cd55803ae1 [Unity] Initial PyTorch Frontend (#14037)
omit e06d67213e [Unity][Op] Add ShapeExpr Tests for Reshape Op (#14035)
omit d789a293f5 [Unity][Pass] Operator legalization (#14029)
omit 8ca086b227 [Unity][TVMScript] Move tir/relax import in script out of
__init__.py (#14033)
omit 407e7f43b6 [Unity][Pass] Wellformed Analysis (#14032)
omit 001a414484 [Unity][BlockBuilder] CallTE convert PrimValue args
(#14028)
omit 16e0c5f064 [Unity][Pass] Normalize Pass (#14031)
omit 53b3dc82a0 [Unity] Relay -> Relax translator (#14026)
omit dcb6ae9786 [Unity][Pass][TuningAPI] Introduce TuningAPI and
MetaSchedule pass (#14014)
omit a4dc336bd3 [Unity][Pass] BindParams pass, FoldConstant pass (#14016)
omit 617f21e260 [Unity][VM] Supporting "compiled" exec mode. (#14015)
omit 87a911f1c3 [Unity][Pass] LambdaLift pass (#14012)
omit dca96539f8 [Unity][Pass] Operator Fusion Passes (#14001)
omit 04d68873ed [Unity] NestedMsg Support utility (#13995)
omit 9c4e7e3944 [Unity] Relax op: manipulation (#13989)
omit 07dd06c3bf [Unity] Relax op: search (#13992)
omit 096f8464e9 [Unity] Relax op: linear algebra (#13988)
omit e980513ad7 [Unity] Relax op: creation (#13984)
omit 25b1fae67d [Unity] Relax op: neural networks (#13993)
omit e12a97e23c [Unity] Relax op: statistical (#13991)
omit 5e71c8060e [Unity] Relax op: arithmetic, comparison (#13983)
omit af46f0df85 [Unity] Relax op: image (#13994)
omit e561dd93db [Unity] Relax op: set (#13990)
omit baa08eec21 [Unity] Relax op: datatype (#13986)
omit d3870a7229 [Unity] Relax op: index (#13987)
omit bfa3b9c6d1 [Unity][TVMScript] Use explicit `R.shape` in TVMScript
(#13979)
omit a9a21ed988 [Unity] e2e Relax minimum build flow (#13961)
omit 3b79924383 [Unity] Relax VM shape lowering pass (#13956)
omit 48a4f2d4e1 [Unity] Relax VM codegen (#13954)
omit d15126636a [Unity] Relax TVMScript Printer (#13944)
omit 1f7aed4236 [Unity] Relax TVMScript Parser. (#13932)
omit e824efe4ac [Unity] Relax BlockBuilder and ExprMutator (#13926)
omit 4433f192bd [Unity] Basic StructInfo Analysis and Expr construction
(#13916)
omit 366c89d897 [Unity][CI] Unity specific jenkins setup (do not upstream
to main) (#13910)
omit d25251fb91 [Unity][IR] First-class StructInfo (#13907)
omit 1528a8db87 [Unity] Relax expressions and types (#13901)
omit adc73e35ca [Unity] Relax VM (#13878)
add 9f28b1da7a [WEB] Update web runtime to support latest emcc (#14046)
add c429a3b6a1 [WEB] WebGPU Codegen (#14048)
add 697c724e92 [Runtime] Add missing Type2Str for TVMByteArray (#14051)
add 5562d906f9 [BYOC] Update CUTLASS backend (SIMT support and codegen
clean up) (#14056)
new 6bb0c05943 [Unity] Relax VM (#13878)
new 666b839d42 [Unity] Relax expressions and types (#13901)
new 746e2b58a8 [Unity][IR] First-class StructInfo (#13907)
new 10622618ab [Unity][CI] Unity specific jenkins setup (do not upstream
to main) (#13910)
new 9f24dac1ac [Unity] Basic StructInfo Analysis and Expr construction
(#13916)
new 14c057ba9e [Unity] Relax BlockBuilder and ExprMutator (#13926)
new ed3653bfb2 [Unity] Relax TVMScript Parser. (#13932)
new 45a9bfd1df [Unity] Relax TVMScript Printer (#13944)
new 8fb1c9c577 [Unity] Relax VM codegen (#13954)
new ff1a02c66a [Unity] Relax VM shape lowering pass (#13956)
new c21a7ddfcc [Unity] e2e Relax minimum build flow (#13961)
new de46e9c368 [Unity][TVMScript] Use explicit `R.shape` in TVMScript
(#13979)
new 9f13b7bd6c [Unity] Relax op: index (#13987)
new 3e4af0dc9b [Unity] Relax op: datatype (#13986)
new a3d25f0a16 [Unity] Relax op: set (#13990)
new 0be100160f [Unity] Relax op: image (#13994)
new 05116eec7a [Unity] Relax op: arithmetic, comparison (#13983)
new 0063f899db [Unity] Relax op: statistical (#13991)
new 7dd27f0758 [Unity] Relax op: neural networks (#13993)
new ae81980963 [Unity] Relax op: creation (#13984)
new c3dfa324f9 [Unity] Relax op: linear algebra (#13988)
new aaac5c44bc [Unity] Relax op: search (#13992)
new 4e2cb4d468 [Unity] Relax op: manipulation (#13989)
new de32194595 [Unity] NestedMsg Support utility (#13995)
new 6fa4490d8c [Unity][Pass] Operator Fusion Passes (#14001)
new f0e22c9b1a [Unity][Pass] LambdaLift pass (#14012)
new 55810737be [Unity][VM] Supporting "compiled" exec mode. (#14015)
new 3928e9b63c [Unity][Pass] BindParams pass, FoldConstant pass (#14016)
new 8dd819fcf2 [Unity][Pass][TuningAPI] Introduce TuningAPI and
MetaSchedule pass (#14014)
new 18ebd15791 [Unity] Relay -> Relax translator (#14026)
new cf51272588 [Unity][Pass] Normalize Pass (#14031)
new eec6b0f2f8 [Unity][BlockBuilder] CallTE convert PrimValue args
(#14028)
new 29fe7a6c28 [Unity][Pass] Wellformed Analysis (#14032)
new 2b4e4d930a [Unity][TVMScript] Move tir/relax import in script out of
__init__.py (#14033)
new 77604b464b [Unity][Pass] Operator legalization (#14029)
new a952bdb26b [Unity][Op] Add ShapeExpr Tests for Reshape Op (#14035)
new 405ef9dbf6 [Unity] Initial PyTorch Frontend (#14037)
new a3ecb2f726 [Unity][Pass] Block-level static memory planning (#14038)
new 59527567cc [Unity] Disallow inline prim_func in relax IR (#14040)
new 3a909e2b95 [Unity] Update tests to adapt to latest TVMScript syntax
(#14039)
new 82f7a5e049 [Unity] Relax dataflow pattern language (matching) (#14041)
new b3cf43af34 [Unity] Statement rewriter for DataflowBlock (#14043)
new 73a45e3022 [Unity][Pass] FuseOps FuseTIR fixes (#14044)
new e9a6e49f3a [Unity][TVMScript] Overload `__neg__` for relax expr
(#14045)
new 41684ee567 [Unity][VM] Add per-op profiling support (#14053)
new 0f6c4674c1 [Unity][BYOC] Add pattern-based partitioning pass (#14054)
This update added new revisions after undoing existing revisions.
That is to say, some revisions that were in the old version of the
branch are not in the new version. This situation occurs
when a user --force pushes a change and generates a repository
containing something like this:
* -- * -- B -- O -- O -- O (7e4d71cc1e)
\
N -- N -- N refs/heads/unity-staging (0f6c4674c1)
You should already have received notification emails for all of the O
revisions, and so the following emails describe only the N revisions
from the common base, B.
Any revisions marked "omit" are not gone; other references still
refer to them. Any revisions marked "discard" are gone forever.
The 46 revisions listed above as "new" are entirely new to this
repository and will be described in separate emails. The revisions
listed as "add" were already present in the repository and have only
been added to this reference.
Summary of changes:
3rdparty/cutlass | 2 +-
.../work_with_relay/using_pipeline_executor.py | 8 +-
include/tvm/runtime/packed_func.h | 4 +
include/tvm/runtime/relax_vm/vm.h | 5 +
include/tvm/tir/op.h | 9 +
include/tvm/topi/elemwise.h | 49 +-
python/tvm/{arith => contrib/cutlass}/_ffi_api.py | 5 +-
python/tvm/contrib/cutlass/build.py | 65 +-
python/tvm/contrib/cutlass/conv2d_operation.py | 186 +++++
python/tvm/contrib/cutlass/conv2d_profiler.py | 6 +-
python/tvm/contrib/cutlass/gemm_operation.py | 89 +++
python/tvm/contrib/cutlass/gen_gemm.py | 8 +-
python/tvm/contrib/cutlass/gen_tensor_op.py | 232 +++++-
python/tvm/contrib/emcc.py | 13 +-
python/tvm/relax/expr.py | 2 +-
python/tvm/relax/transform/transform.py | 37 +-
python/tvm/relax/vm.py | 33 +-
src/relax/transform/fuse_ops.cc | 253 ++++++-
src/relax/transform/fuse_tir.cc | 15 +-
src/relay/backend/contrib/codegen_c/codegen_c.h | 1 +
src/relay/backend/contrib/cutlass/codegen.cc | 820 ++++-----------------
src/runtime/dso_library.cc | 5 +
src/runtime/library_module.cc | 5 -
src/runtime/relax_vm/executable.cc | 7 +
src/runtime/relax_vm/vm.cc | 129 +++-
src/target/intrin_rule.cc | 16 +
src/target/intrin_rule.h | 3 +
src/target/source/codegen_c.h | 6 +-
src/target/source/codegen_metal.cc | 2 +-
src/target/source/codegen_source_base.cc | 3 +-
src/target/source/codegen_webgpu.cc | 555 ++++++++++++++
src/target/source/codegen_webgpu.h | 92 +++
src/target/source/intrin_rule_metal.cc | 17 -
src/target/source/intrin_rule_webgpu.cc | 118 +++
src/target/spirv/build_vulkan.cc | 16 +-
src/target/spirv/intrin_rule_spirv.cc | 34 -
src/target/spirv/ir_builder.h | 18 +-
src/tir/op/op.cc | 42 ++
tests/lint/flake8.sh | 2 +-
tests/python/contrib/test_cutlass.py | 26 +-
tests/python/relax/test_op_misc.py | 98 +++
tests/python/relax/test_transform_fuse_ops.py | 81 +-
.../relax/test_transform_fuse_ops_by_pattern.py | 464 ++++++++++++
tests/python/relax/test_transform_fuse_tir.py | 141 +++-
tests/python/relax/test_tvmscript_parser.py | 49 ++
tests/python/relax/test_vm_profiler.py | 130 ++++
web/.gitignore | 1 +
web/Makefile | 11 +-
web/emcc/tvmjs_support.cc | 2 +-
web/emcc/webgpu_runtime.cc | 24 +-
web/src/runtime.ts | 4 +-
web/src/webgpu.ts | 19 +-
web/tests/python/webgpu_rpc_test.py | 6 +-
53 files changed, 3042 insertions(+), 926 deletions(-)
copy python/tvm/{arith => contrib/cutlass}/_ffi_api.py (90%)
create mode 100644 src/target/source/codegen_webgpu.cc
create mode 100644 src/target/source/codegen_webgpu.h
create mode 100644 src/target/source/intrin_rule_webgpu.cc
create mode 100644 tests/python/relax/test_op_misc.py
create mode 100644 tests/python/relax/test_transform_fuse_ops_by_pattern.py
create mode 100644 tests/python/relax/test_vm_profiler.py
