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 d0ea42ae9e [Unity][WEBGPU] Codegen improvements and WebRuntime (#14187)
discard 0881e0b711 [Unity][OP] Add an operator for fused multi head attention
(#14150)
discard a526a0db02 [Unity][Analysis] Restore Python bindings for var analyses
(#14180)
discard 4763de07d1 [Unity][Op] Full support of Relax op `power` (#14171)
discard 3c529ba8be [Unity][BYOC] Add batch matmul support to Relax CUTLASS
BYOC (#14166)
discard a102afb7f9 [Unity][Analysis] Analysis for detecting recursion in Relax
(#14149)
discard ba380fe49e [Unity] Add bind_constants option to FuseOpsByPattern
(#14151)
discard 21da09c779 [Unity][BYOC] Use Relax legalize + CPU build for reference
in tests (#14162)
discard fc627c7f3b [Unity][Analysis] Checking function return struct info in
well-formed check (#14155)
discard d9ad8860ac [Unity][Pass] Support Symbolic Shape Deduction during
BindParam (#14154)
discard cc81d3487f [Unity][Debugging] AST printer (#14152)
discard 8f4eb69594 [Unity][Pass] Enhance constant folding to fold relax ops by
evaluating them. (#14146)
discard 3baa50bd94 [Unity][Legalize] Fix Scalar Constant Legalization (#14127)
discard 8f0f62e841 [Unity] Add callback to FuseOpsByPattern to check match
result is accepted (#14109)
discard eef40997d5 [Unity][BYOC] Assign group to unused bindings and ignroe
PrimFunc (#14139)
discard c02289ad63 [Unity][TVMScript] emit_te sugar (#14123)
discard 8e5712c95e [Unity][BYOC] Add transposed matmul support to Relax
CUTLASS BYOC (#14128)
discard ec0ede82fd [Unity] Add Global info (#14132)
discard b7ecfe9a54 [Unity][WEB] Relax vm on web runtime (#14131)
discard a82bef5820 [Unity][BlockBuilder] Add `name_hint` argument for `emit`
and `emit_output` (#14126)
discard e2ee1af952 [Unity][Fix] Fix bug in MergeCompositeFunctions (#14117)
discard 67a3721be0 [Unity] Update tests again to adapt to latest TVMScript
syntax (#14115)
discard 00be298ed8 [Unity][BYOC]Add relax backend pattern registry (#14106)
discard 356365eb2f [Unity] Remove attributes of relax.print, assert and unique
(#14101)
discard 1b6ccf096b [Unity][Layout] Add layout transformation analysis for
PrimFunc (#14066)
discard 3dabb7089a [Unity] Relax Recursive function (#14092)
discard 1229317707 [Unity] Lower `shape_of` to a builtin (#14093)
discard 9b9fdba6fb [Unity] Fix typo in the comment (#14096)
discard 4b13b7045e [Unity][Relax] Set Shape Function to Be Host Function
(#14090)
discard bc13158d65 [Unity] Refactor Relax Build JIT UX (#14088)
discard e76e94dd9f [Unity][Fix][Pass] FoldConstant with DCE in dataflow block
(#14087)
discard e5340371a3 [Unity][Analysis] TIR pattern kind analysis for
multi-buffer write block (#14075)
discard 4ae6d34957 [Unity][Op] `log_softmax` and `cross_entropy_with_logits`
(#14083)
discard 18fdb29111 [Unity][BYOC] Add DNNL backend (#14082)
discard 2d00e8ebc5 [Unity][BYOC] Add CUTLASS backend (#14081)
discard d953b7659d [Unity] Add testcases for `expr_args_converter` (#14080)
discard 717eb5f5a9 [Unity][Pass] Canonicalize Bindings (#14079)
discard 6ac619dcb3 [Unity][BYOC][Pass] RunCodegen and TensorRT (#14078)
discard 60c59b2e2e [Unity][Transform] Add LiftTransformParams pass (#14069)
discard 1e53a9ce15 [Unity][Frontend] Annotate number of non-static input of FX
function (#14067)
discard bd15cf149b [Unity][BYOC] Add pass to merge composite functions to
offload large subgraphs (#14062)
discard ae400b5a21 [Unity][Pass] Remove Unused Function (#14061)
discard 23109d7281 [Unity][Fix][Pass] Fix FuseOps for lack graph edges (#14058)
discard 7a7cce60ed [Unity] Relax op: collapse sum (#14059)
discard 7963e6d619 [Unity][BYOC] Add pattern-based partitioning pass (#14054)
discard e7500dbe22 [Unity][VM] Add per-op profiling support (#14053)
discard 138496fabb [Unity][TVMScript] Overload `__neg__` for relax expr
(#14045)
discard 6cec2edcdd [Unity][Pass] FuseOps FuseTIR fixes (#14044)
discard 086d32d6cb [Unity] Statement rewriter for DataflowBlock (#14043)
discard ca293e8957 [Unity] Relax dataflow pattern language (matching) (#14041)
discard 123ede268f [Unity] Update tests to adapt to latest TVMScript syntax
(#14039)
discard 9e0778e35a [Unity] Disallow inline prim_func in relax IR (#14040)
discard f4b05f38ba [Unity][Pass] Block-level static memory planning (#14038)
discard df0c1b00b0 [Unity] Initial PyTorch Frontend (#14037)
discard d0d02a8a04 [Unity][Op] Add ShapeExpr Tests for Reshape Op (#14035)
discard 5a07c5aeed [Unity][Pass] Operator legalization (#14029)
discard 333d42d89f [Unity][TVMScript] Move tir/relax import in script out of
__init__.py (#14033)
discard eb05be6289 [Unity][Pass] Wellformed Analysis (#14032)
discard daf33d5be7 [Unity][BlockBuilder] CallTE convert PrimValue args
(#14028)
discard 195c0ec414 [Unity][Pass] Normalize Pass (#14031)
discard a2a59b0665 [Unity] Relay -> Relax translator (#14026)
discard 4d172093e0 [Unity][Pass][TuningAPI] Introduce TuningAPI and
MetaSchedule pass (#14014)
discard 3cb151f54d [Unity][Pass] BindParams pass, FoldConstant pass (#14016)
discard e4d59f5b4b [Unity][VM] Supporting "compiled" exec mode. (#14015)
discard 534fa748dc [Unity][Pass] LambdaLift pass (#14012)
discard 2ac610a090 [Unity][Pass] Operator Fusion Passes (#14001)
discard b06c02e2a0 [Unity] NestedMsg Support utility (#13995)
discard 9807c1c8eb [Unity] Relax op: manipulation (#13989)
discard 5814dcb926 [Unity] Relax op: search (#13992)
discard 8d25cb5081 [Unity] Relax op: linear algebra (#13988)
discard f5798631bf [Unity] Relax op: creation (#13984)
discard 1219825f12 [Unity] Relax op: neural networks (#13993)
discard d83d57c8c1 [Unity] Relax op: statistical (#13991)
discard 2b55420146 [Unity] Relax op: arithmetic, comparison (#13983)
discard aef2bf4511 [Unity] Relax op: image (#13994)
discard 5d868d895d [Unity] Relax op: set (#13990)
discard c8869b7cbb [Unity] Relax op: datatype (#13986)
discard 7489379ac4 [Unity] Relax op: index (#13987)
discard e784761cdb [Unity][TVMScript] Use explicit `R.shape` in TVMScript
(#13979)
discard e33ed56e41 [Unity] e2e Relax minimum build flow (#13961)
discard ad27005c35 [Unity] Relax VM shape lowering pass (#13956)
discard 78a87e6172 [Unity] Relax VM codegen (#13954)
discard bc6de4b733 [Unity] Relax TVMScript Printer (#13944)
discard f74330451d [Unity] Relax TVMScript Parser. (#13932)
discard 886be6e8b0 [Unity] Relax BlockBuilder and ExprMutator (#13926)
discard 7af2beae0c [Unity] Basic StructInfo Analysis and Expr construction
(#13916)
discard dab5ec4e83 [Unity][CI] Unity specific jenkins setup (do not upstream
to main) (#13910)
discard 051e1f1474 [Unity][IR] First-class StructInfo (#13907)
discard 8eb3d7c495 [Unity] Relax expressions and types (#13901)
discard 079abf1949 [Unity] Relax VM (#13878)
add 22c47ee6de [TOPI] Batch Norm Training Mode (#14190)
add baedf7f04d [TOPI] Group normalization (#14193)
new f79850db8b [Unity] Relax VM (#13878)
new 543503d2a7 [Unity] Relax expressions and types (#13901)
new 1e5110d557 [Unity][IR] First-class StructInfo (#13907)
new 2b2a8bfede [Unity][CI] Unity specific jenkins setup (do not upstream
to main) (#13910)
new 39e1571752 [Unity] Basic StructInfo Analysis and Expr construction
(#13916)
new 12f38f5d46 [Unity] Relax BlockBuilder and ExprMutator (#13926)
new 23b31ddcab [Unity] Relax TVMScript Parser. (#13932)
new 6425cd1b41 [Unity] Relax TVMScript Printer (#13944)
new 5a13e0d4ae [Unity] Relax VM codegen (#13954)
new 4ba6f1d175 [Unity] Relax VM shape lowering pass (#13956)
new c50df572c5 [Unity] e2e Relax minimum build flow (#13961)
new b9807a62d4 [Unity][TVMScript] Use explicit `R.shape` in TVMScript
(#13979)
new a3e275df76 [Unity] Relax op: index (#13987)
new 0ed6b9aebf [Unity] Relax op: datatype (#13986)
new a0ecbe955a [Unity] Relax op: set (#13990)
new 4d0485cd20 [Unity] Relax op: image (#13994)
new 379c8d28a7 [Unity] Relax op: arithmetic, comparison (#13983)
new b52113a0a9 [Unity] Relax op: statistical (#13991)
new f8867390a6 [Unity] Relax op: neural networks (#13993)
new c11338702d [Unity] Relax op: creation (#13984)
new 202a6e6cf5 [Unity] Relax op: linear algebra (#13988)
new f9d1dccf7a [Unity] Relax op: search (#13992)
new 47b43f1a92 [Unity] Relax op: manipulation (#13989)
new 900776db89 [Unity] NestedMsg Support utility (#13995)
new bf7e8eaf3a [Unity][Pass] Operator Fusion Passes (#14001)
new c1be350206 [Unity][Pass] LambdaLift pass (#14012)
new 6419af94cc [Unity][VM] Supporting "compiled" exec mode. (#14015)
new a1e237358a [Unity][Pass] BindParams pass, FoldConstant pass (#14016)
new 3200f9237a [Unity][Pass][TuningAPI] Introduce TuningAPI and
MetaSchedule pass (#14014)
new be61b78370 [Unity] Relay -> Relax translator (#14026)
new 9ddbc488f1 [Unity][Pass] Normalize Pass (#14031)
new 9a876efecc [Unity][BlockBuilder] CallTE convert PrimValue args
(#14028)
new 28c12d2a6f [Unity][Pass] Wellformed Analysis (#14032)
new 2a64d3ea69 [Unity][TVMScript] Move tir/relax import in script out of
__init__.py (#14033)
new 5e5e994ffc [Unity][Pass] Operator legalization (#14029)
new 29fb528fb0 [Unity][Op] Add ShapeExpr Tests for Reshape Op (#14035)
new 1d4118dec0 [Unity] Initial PyTorch Frontend (#14037)
new 111f71f928 [Unity][Pass] Block-level static memory planning (#14038)
new fddc43d835 [Unity] Disallow inline prim_func in relax IR (#14040)
new 582efa6b02 [Unity] Update tests to adapt to latest TVMScript syntax
(#14039)
new 167f09a8cf [Unity] Relax dataflow pattern language (matching) (#14041)
new c0158714dc [Unity] Statement rewriter for DataflowBlock (#14043)
new 341ef4239b [Unity][Pass] FuseOps FuseTIR fixes (#14044)
new 45afc08856 [Unity][TVMScript] Overload `__neg__` for relax expr
(#14045)
new 53b5fab032 [Unity][VM] Add per-op profiling support (#14053)
new 1660159905 [Unity][BYOC] Add pattern-based partitioning pass (#14054)
new f804b8832a [Unity] Relax op: collapse sum (#14059)
new d353855d57 [Unity][Fix][Pass] Fix FuseOps for lack graph edges (#14058)
new 2a8333f932 [Unity][Pass] Remove Unused Function (#14061)
new bdae3d6b66 [Unity][BYOC] Add pass to merge composite functions to
offload large subgraphs (#14062)
new 93e3b96dd2 [Unity][Frontend] Annotate number of non-static input of FX
function (#14067)
new b27326c552 [Unity][Transform] Add LiftTransformParams pass (#14069)
new 5e7005d967 [Unity][BYOC][Pass] RunCodegen and TensorRT (#14078)
new 4575409c7d [Unity][Pass] Canonicalize Bindings (#14079)
new 989f645f4e [Unity] Add testcases for `expr_args_converter` (#14080)
new 30349b73ff [Unity][BYOC] Add CUTLASS backend (#14081)
new 2bc5cd495f [Unity][BYOC] Add DNNL backend (#14082)
new 7f08122274 [Unity][Op] `log_softmax` and `cross_entropy_with_logits`
(#14083)
new 15642fa9bb [Unity][Analysis] TIR pattern kind analysis for
multi-buffer write block (#14075)
new df290df975 [Unity][Fix][Pass] FoldConstant with DCE in dataflow block
(#14087)
new f355a29055 [Unity] Refactor Relax Build JIT UX (#14088)
new 846edd58e2 [Unity][Relax] Set Shape Function to Be Host Function
(#14090)
new d3e844bb51 [Unity] Fix typo in the comment (#14096)
new b482a149d4 [Unity] Lower `shape_of` to a builtin (#14093)
new ec50992045 [Unity] Relax Recursive function (#14092)
new 4e8ddc81e7 [Unity][Layout] Add layout transformation analysis for
PrimFunc (#14066)
new 2af3ec9204 [Unity] Remove attributes of relax.print, assert and unique
(#14101)
new 7a6a6a2393 [Unity][BYOC]Add relax backend pattern registry (#14106)
new 8849dcb47d [Unity] Update tests again to adapt to latest TVMScript
syntax (#14115)
new d41a2be7f2 [Unity][Fix] Fix bug in MergeCompositeFunctions (#14117)
new a110a23d62 [Unity][BlockBuilder] Add `name_hint` argument for `emit`
and `emit_output` (#14126)
new 834d626cc6 [Unity][WEB] Relax vm on web runtime (#14131)
new 89c2170228 [Unity] Add Global info (#14132)
new ec5e478c1f [Unity][BYOC] Add transposed matmul support to Relax
CUTLASS BYOC (#14128)
new 08a0c63837 [Unity][TVMScript] emit_te sugar (#14123)
new 6077887837 [Unity][BYOC] Assign group to unused bindings and ignroe
PrimFunc (#14139)
new 620da6f8e9 [Unity] Add callback to FuseOpsByPattern to check match
result is accepted (#14109)
new c1b8f8a7fb [Unity][Legalize] Fix Scalar Constant Legalization (#14127)
new 6b86111283 [Unity][Pass] Enhance constant folding to fold relax ops by
evaluating them. (#14146)
new c008f7d084 [Unity][Debugging] AST printer (#14152)
new de9c84127a [Unity][Pass] Support Symbolic Shape Deduction during
BindParam (#14154)
new 5304964616 [Unity][Analysis] Checking function return struct info in
well-formed check (#14155)
new b098ae0e37 [Unity][BYOC] Use Relax legalize + CPU build for reference
in tests (#14162)
new d2d0bafbab [Unity] Add bind_constants option to FuseOpsByPattern
(#14151)
new 3d7b55c66b [Unity][Analysis] Analysis for detecting recursion in Relax
(#14149)
new 321412beb1 [Unity][BYOC] Add batch matmul support to Relax CUTLASS
BYOC (#14166)
new a6411eecd3 [Unity][Op] Full support of Relax op `power` (#14171)
new 5e58b7e457 [Unity][Analysis] Restore Python bindings for var analyses
(#14180)
new c9ae7941d0 [Unity][OP] Add an operator for fused multi head attention
(#14150)
new 2cef9d4377 [Unity][WEBGPU] Codegen improvements and WebRuntime (#14187)
new c293969aaa [Unity][Transform] LiftTransformParams handling multiple
functions (#14192)
new a29ec38a07 [Unity][Op] Group normalization (#14194)
new a1b9261f3e [Unity][Op] Argmax and argmin (#14195)
new b858986f4b [Unity][Op] Legalize `round`, `floor`, `ceil`, `sign`
(#14198)
new a76332a8f2 [Unity][Frontend] FX translator supporting more ops (#14196)
new 169448b8a9 [Unity][Frontend] FX translator returning weights with
`keep_params_as_input` (#14197)
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 (d0ea42ae9e)
\
N -- N -- N refs/heads/unity-staging (169448b8a9)
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 96 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:
include/tvm/relax/attrs/nn.h | 21 +
.../tvm/relax/attrs/{statistical.h => search.h} | 24 +-
include/tvm/topi/nn/group_norm.h | 151 +++++++
python/tvm/relax/frontend/__init__.py | 2 +
python/tvm/relax/frontend/common.py | 48 ++
python/tvm/relax/frontend/torch/dynamo.py | 28 +-
python/tvm/relax/frontend/torch/fx_translator.py | 269 +++++++++---
python/tvm/relax/op/nn/nn.py | 58 +++
python/tvm/relax/op/op_attrs.py | 5 +
python/tvm/relax/op/search.py | 54 +++
python/tvm/relax/transform/legalize_ops/nn.py | 14 +
python/tvm/relax/transform/legalize_ops/search.py | 20 +
python/tvm/relax/transform/legalize_ops/unary.py | 4 +
python/tvm/script/ir_builder/relax/ir.py | 4 +
python/tvm/topi/nn/__init__.py | 1 +
python/tvm/topi/nn/batch_norm.py | 46 +-
.../tvm/topi/nn/{layer_norm.py => group_norm.py} | 20 +-
python/tvm/topi/testing/__init__.py | 1 +
python/tvm/topi/testing/batch_norm.py | 33 +-
python/tvm/topi/testing/group_norm_python.py | 82 ++++
src/relax/op/nn/nn.cc | 83 ++++
src/relax/op/nn/nn.h | 4 +
src/relax/op/tensor/search.cc | 77 ++++
src/relax/op/tensor/search.h | 8 +
src/relax/transform/lift_transform_params.cc | 37 +-
src/topi/nn.cc | 7 +
tests/python/relax/test_ast_printer.py | 4 +-
tests/python/relax/test_frontend_dynamo.py | 4 +-
tests/python/relax/test_frontend_from_fx.py | 481 +++++++++++++++++++--
tests/python/relax/test_op_nn.py | 238 ++++++++++
tests/python/relax/test_op_search.py | 158 +++++++
.../python/relax/test_transform_legalize_ops_nn.py | 162 +++++++
...st_transform_legalize_ops_search_statistical.py | 197 +++++++++
.../relax/test_transform_legalize_ops_unary.py | 405 +++++++++++++++++
.../relax/test_transform_lift_transform_params.py | 105 ++++-
tests/python/relax/test_tvmscript_parser_op_nn.py | 25 ++
.../relax/test_tvmscript_parser_op_search.py | 45 ++
tests/python/topi/python/test_topi_batch_norm.py | 46 +-
..._topi_layer_norm.py => test_topi_group_norm.py} | 30 +-
tests/python/topi/python/test_topi_layer_norm.py | 2 +-
40 files changed, 2841 insertions(+), 162 deletions(-)
copy include/tvm/relax/attrs/{statistical.h => search.h} (60%)
create mode 100644 include/tvm/topi/nn/group_norm.h
create mode 100644 python/tvm/relax/frontend/common.py
copy python/tvm/topi/nn/{layer_norm.py => group_norm.py} (69%)
create mode 100644 python/tvm/topi/testing/group_norm_python.py
copy tests/python/topi/python/{test_topi_layer_norm.py =>
test_topi_group_norm.py} (62%)
