Aten dialect #16
Merged
Aten dialect #16
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
stellaraccident
approved these changes
Aug 11, 2020
There was a problem hiding this comment.
Thanks for this contribution - as we've discussed, getting this set of things in-tree/public is the priority. I've made a lot of nit/syntax/etc comments that it would be good to address in a followup/cleanup (I don't want to disrupt the "make it work" mode you are in now).
|
There's a lot more in this PR than just a dialect definition it seems. Can you update the commit messages to make it clearer what functionality is included? |
|
Specifically, I see at least:
|
silvasean
reviewed
Aug 11, 2020
silvasean
reviewed
Aug 11, 2020
Yes, there's some lowering and report passes here. I'll add some better descriptions of this. The short answer is that there is also a bunch of pytorch stuff still to come which intercepts function calls through the pytorch jit and generates MLIR in this dialect, then lowers the MLIR into function calls through AddLayerNames -> aten-to-std -> ReturnElimination -> std-to-llvm. The result is then jitted using the LLVM jit, linked against a runtime library which makes calls back into pytorch to implement all the layers. So, a basic pipecleaning flow that integrates with pytorch, very much in the style of pytorch_xla. The reports are intended to support more interesting optimizations. |
joker-eph
reviewed
Aug 11, 2020
| // The pytorch convolution operator has 9 arguments, but we only have a jit | ||
| // library that supports the first six at the moment. | ||
| def : Pat<(aten_ConvolutionOverrideableOp $a1, $a2, $a3, $a4, $a5, $a6, | ||
| $a7, $a8, $a9), |
There was a problem hiding this comment.
How would you format this?
silvasean
reviewed
Aug 11, 2020
This was referenced Aug 11, 2020
stephenneuendorffer
force-pushed
the
aten_dialect
branch
from
deee347
to
7d87a14
Compare
This patch adds a dialect intended to be used as a frontend dialect to facilitate lowering from "A Tensor Library" in torch/pytorch. This patch includes several passes that are useful in conjuction with the dialect: --aten-layer-name: Generates layer names for each operation, which are not present in the original pytorch. --aten-to-std: Lower the ATen dialect into standard dialect function calls. --return-elimination-pass: convert functions (primarily the toplevel function) to pass return values by reference. This simplifies pytorch integration. --aten-op-report: generate a textual report about the model --liveness-report Future patches will implement actual integration with the pytorch jit to intercept and generates MLIR in this dialect, then lower the resulting MLIR into function calls through aten-layer-name -> aten-to-std -> return-elimination -> std-to-llvm. The result would then jitted using the LLVM jit, linked against a runtime library which makes calls back into pytorch to implement all the layers. Co-authored-by: Jeff Fifield <jeff.fifield@xilinx.com>
stephenneuendorffer
force-pushed
the
aten_dialect
branch
from
7d87a14
to
88c578a
Compare
saeta
added a commit
to saeta/mlir-npcomp
that referenced
this pull request
Dec 31, 2020
Committing this as a snapshot of progress, but this code organization
approach is not scalable.
Output:
Got a dialect for op %0 = rd.range %c1_i64 to %c3_i64 : (i64, i64) -> !rd.Dataset: rd
walkOp name stringref: 'rd.range'
Made a create fn:
llvm.func internal @__rd_create_foo_fix_me(%arg0: !llvm.ptr<struct<(i64, i64)>>) {
%0 = llvm.mlir.constant(0 : index) : !llvm.i64
%1 = llvm.mlir.constant(1 : index) : !llvm.i64
%2 = llvm.getelementptr %arg0[%0, %0] : (!llvm.ptr<struct<(i64, i64)>>, !llvm.i64, !llvm.i64) -> !llvm.ptr<struct<(i64, i64)>>
%3 = llvm.getelementptr %arg0[%0, %1] : (!llvm.ptr<struct<(i64, i64)>>, !llvm.i64, !llvm.i64) -> !llvm.ptr<struct<(i64, i64)>>
%c1_i64 = constant 1 : i64
%c3_i64 = constant 3 : i64
llvm.store %2, %c1_i64 : i64
llvm.store %3, %c3_i64 : i64
return
}
Made a next function:
llvm.func internal @__rd_next_foo_fix_me(%arg0: !llvm.ptr<struct<(i64, i64)>>) -> !llvm.struct<(i1, i64)> {
%0 = llvm.mlir.constant(0 : index) : !llvm.i64
%1 = llvm.mlir.constant(1 : index) : !llvm.i64
%2 = llvm.getelementptr %arg0[%0, %0] : (!llvm.ptr<struct<(i64, i64)>>, !llvm.i64, !llvm.i64) -> !llvm.ptr<struct<(i64, i64)>>
%3 = llvm.getelementptr %arg0[%0, %1] : (!llvm.ptr<struct<(i64, i64)>>, !llvm.i64, !llvm.i64) -> !llvm.ptr<struct<(i64, i64)>>
%4 = llvm.load %2 : !llvm.ptr<struct<(i64, i64)>>
%5 = llvm.load %3 : !llvm.ptr<struct<(i64, i64)>>
%6 = "llvm.add"(%4, %1) : (!llvm.struct<(i64, i64)>, !llvm.i64) -> !llvm.struct<(i64, i64)>
%7 = llvm.icmp "ne" %4, %5 : !llvm.struct<(i64, i64)>
llvm.store %2, %6 : !llvm.struct<(i64, i64)>
return %7, %6 : !llvm.i1, !llvm.struct<(i64, i64)>
}
Did some sugary! Things now look like:
module {
func @main() {
%c1_i64 = constant 1 : i64
%c3_i64 = constant 3 : i64
%0 = llvm.mlir.constant(1 : index) : !llvm.i64
%1 = llvm.alloca %0 x !llvm.struct<(i64, i64)> : (!llvm.i64) -> !llvm.ptr<struct<(i64, i64)>>
llvm.call @__rd_create_foo_fix_me(%1) : (!llvm.ptr<struct<(i64, i64)>>) -> ()
%valid, %value = rd.iterator_next %1 : (!llvm.ptr<struct<(i64, i64)>>) -> (i1, i64)
"rd.print"(%value) : (i64) -> ()
return
}
}
Walking users.... found: rd.iterator_next... MATCHING!
Walking users.... found: llvm.call... didn't match.
Did some more sugary! Things now look like:
module {
func @main() {
%c1_i64 = constant 1 : i64
%c3_i64 = constant 3 : i64
%0 = llvm.mlir.constant(1 : index) : !llvm.i64
%1 = llvm.alloca %0 x !llvm.struct<(i64, i64)> : (!llvm.i64) -> !llvm.ptr<struct<(i64, i64)>>
llvm.call @__rd_create_foo_fix_me(%1) : (!llvm.ptr<struct<(i64, i64)>>) -> ()
%2 = llvm.call @__rd_next_foo_fix_me(%1) : (!llvm.ptr<struct<(i64, i64)>>) -> !llvm.struct<(i1, i64)>
%3 = llvm.extractvalue %2[0 : i32] : !llvm.struct<(i1, i64)>
%4 = llvm.extractvalue %2[1 : i32] : !llvm.struct<(i1, i64)>
"rd.print"(%4) : (!llvm.struct<(i1, i64)>) -> ()
return
}
}
Stack dump:
0. Program arguments: /usr/local/google/home/saeta/src/mlir-npcomp/build/bin/npcomp-opt basic.mlir -rd-lower-to-llvm
#0 0x00007f607c4110b3 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) /usr/local/google/home/saeta/src/mlir-npcomp/external/llvm-project/llvm/lib/Support/Unix/Signals.inc:563:13
#1 0x00007f607c40f330 llvm::sys::RunSignalHandlers() /usr/local/google/home/saeta/src/mlir-npcomp/external/llvm-project/llvm/lib/Support/Signals.cpp:72:18
llvm#2 0x00007f607c411575 SignalHandler(int) /usr/local/google/home/saeta/src/mlir-npcomp/external/llvm-project/llvm/lib/Support/Unix/Signals.inc:0:3
llvm#3 0x00007f608108e140 __restore_rt (/lib/x86_64-linux-gnu/libpthread.so.0+0x14140)
llvm#4 0x00007f60804fe420 llvm::ilist_node_base<true>::isSentinel() const /usr/local/google/home/saeta/src/mlir-npcomp/external/llvm-project/llvm/include/llvm/ADT/ilist_node_base.h:45:36
llvm#5 0x00007f60804fe420 llvm::ilist_node_base<true>::isKnownSentinel() const /usr/local/google/home/saeta/src/mlir-npcomp/external/llvm-project/llvm/include/llvm/ADT/ilist_node_base.h:46:41
llvm#6 0x00007f60804fe420 llvm::ilist_iterator<llvm::ilist_detail::node_options<mlir::Operation, true, false, void>, false, false>::operator*() const /usr/local/google/home/saeta/src/mlir-npcom
p/external/llvm-project/llvm/include/llvm/ADT/ilist_iterator.h:138:5
llvm#7 0x00007f60804fe420 llvm::early_inc_iterator_impl<llvm::ilist_iterator<llvm::ilist_detail::node_options<mlir::Operation, true, false, void>, false, false> >::operator*() /usr/local/google
/home/saeta/src/mlir-npcomp/external/llvm-project/llvm/include/llvm/ADT/STLExtras.h:546:12
llvm#8 0x00007f60804fe420 mlir::detail::walk(mlir::Operation*, llvm::function_ref<void (mlir::Operation*)>) /usr/local/google/home/saeta/src/mlir-npcomp/external/llvm-project/mlir/lib/IR/Visito
rs.cpp:41:27
llvm#9 0x00007f60804fe43c mlir::detail::walk(mlir::Operation*, llvm::function_ref<void (mlir::Operation*)>) /usr/local/google/home/saeta/src/mlir-npcomp/external/llvm-project/mlir/lib/IR/Visito
rs.cpp:0:9
llvm#10 0x00007f6080e7fbf8 std::enable_if<(!(llvm::is_one_of<mlir::NPCOMP::rd::MakeIteratorOp, mlir::Operation*, mlir::Region*, mlir::Block*>::value)) && (std::is_same<void, void>::value), void>
::type mlir::detail::walk<(anonymous namespace)::LowerToRuntimePass::runOnOperation()::'lambda'(mlir::NPCOMP::rd::MakeIteratorOp), mlir::NPCOMP::rd::MakeIteratorOp, void>(mlir::Operation*, (
anonymous namespace)::LowerToRuntimePass::runOnOperation()::'lambda'(mlir::NPCOMP::rd::MakeIteratorOp)&&) /usr/local/google/home/saeta/src/mlir-npcomp/build/install-mlir/include/mlir/IR/Visi
tors.h:119:3
llvm#11 0x00007f6080e7fb90 void mlir::Operation::walk<(anonymous namespace)::LowerToRuntimePass::runOnOperation()::'lambda'(mlir::NPCOMP::rd::MakeIteratorOp), void>((anonymous namespace)::LowerT
oRuntimePass::runOnOperation()::'lambda'(mlir::NPCOMP::rd::MakeIteratorOp)&&) /usr/local/google/home/saeta/src/mlir-npcomp/build/install-mlir/include/mlir/IR/Operation.h:527:5
llvm#12 0x00007f6080e7fb03 void mlir::OpState::walk<(anonymous namespace)::LowerToRuntimePass::runOnOperation()::'lambda'(mlir::NPCOMP::rd::MakeIteratorOp), void>((anonymous namespace)::LowerToR
untimePass::runOnOperation()::'lambda'(mlir::NPCOMP::rd::MakeIteratorOp)&&) /usr/local/google/home/saeta/src/mlir-npcomp/build/install-mlir/include/mlir/IR/OpDefinition.h:178:5
llvm#13 0x00007f6080e7f876 (anonymous namespace)::LowerToRuntimePass::runOnOperation() /usr/local/google/home/saeta/src/mlir-npcomp/build/../lib/Dialect/RD/Transforms/LowerToLLVM.cpp:189:33
llvm#14 0x00007f6080522617 mlir::detail::OpToOpPassAdaptor::run(mlir::Pass*, mlir::Operation*, mlir::AnalysisManager, bool) /usr/local/google/home/saeta/src/mlir-npcomp/external/llvm-project/mli
r/lib/Pass/Pass.cpp:0:11
llvm#15 0x00007f6080525917 mlir::failed(mlir::LogicalResult) /usr/local/google/home/saeta/src/mlir-npcomp/external/llvm-project/mlir/include/mlir/Support/LogicalResult.h:47:23
llvm#16 0x00007f6080525917 mlir::detail::OpToOpPassAdaptor::runPipeline(llvm::iterator_range<llvm::pointee_iterator<std::unique_ptr<mlir::Pass, std::default_delete<mlir::Pass> >*, mlir::Pass> >,
mlir::Operation*, mlir::AnalysisManager, bool) /usr/local/google/home/saeta/src/mlir-npcomp/external/llvm-project/mlir/lib/Pass/Pass.cpp:402:9
llvm#17 0x00007f6080525917 mlir::PassManager::run(mlir::Operation*) /usr/local/google/home/saeta/src/mlir-npcomp/external/llvm-project/mlir/lib/Pass/Pass.cpp:817:13
llvm#18 0x00007f608055b69f mlir::failed(mlir::LogicalResult) /usr/local/google/home/saeta/src/mlir-npcomp/external/llvm-project/mlir/include/mlir/Support/LogicalResult.h:47:23
llvm#19 0x00007f608055b69f performActions(llvm::raw_ostream&, bool, bool, llvm::SourceMgr&, mlir::MLIRContext*, mlir::PassPipelineCLParser const&) /usr/local/google/home/saeta/src/mlir-npcomp/ex
ternal/llvm-project/mlir/lib/Support/MlirOptMain.cpp:75:7
llvm#20 0x00007f608055a26d processBuffer(llvm::raw_ostream&, std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer> >, bool, bool, bool, bool, mlir::PassPipelineCLParser con
st&, mlir::DialectRegistry&) /usr/local/google/home/saeta/src/mlir-npcomp/external/llvm-project/mlir/lib/Support/MlirOptMain.cpp:109:12
llvm#21 0x00007f6080559ff5 mlir::MlirOptMain(llvm::raw_ostream&, std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer> >, mlir::PassPipelineCLParser const&, mlir::DialectRe
gistry&, bool, bool, bool, bool, bool) /usr/local/google/home/saeta/src/mlir-npcomp/external/llvm-project/mlir/lib/Support/MlirOptMain.cpp:146:10
llvm#22 0x000000000040d2ef main /usr/local/google/home/saeta/src/mlir-npcomp/build/../tools/npcomp-opt/npcomp-opt.cpp:91:14
llvm#23 0x00007f607b688d0a __libc_start_main ./csu/../csu/libc-start.c:308:16
llvm#24 0x000000000040ceca _start (/usr/local/google/home/saeta/src/mlir-npcomp/build/bin/npcomp-opt+0x40ceca)
Segmentation fault
qedawkins
pushed a commit
to nod-ai/torch-mlir
that referenced
this pull request
Oct 3, 2022
* Allow importing variadic inputs/outputs of onnx operators * Enable testcases for variadic ops * Modify gen_doc.py
JianzheXiao
pushed a commit
to JianzheXiao/torch-mlir
that referenced
this pull request
Aug 4, 2023
# This is the 1st commit message: [Stablehlo] Add converter to stablehlo for aten.(Int,Float,Bool).Tensor op (llvm#2340) [Stablehlo] Add converter to stablehlo for aten.(Int,Float,Bool).Tensor op and configure crashing e2e sets for stablehlo backend. # This is the commit message llvm#2: update PyTorch version to 2.1.0.dev20230729 (llvm#2354) - torch version: 2.1.0.dev20230729 - torch commit hash: b638df0afb83572724032c824c64e481bb4499a0 - torchvision version: 0.16.0.dev20230729 Co-authored-by: Roll PyTorch Action <torch-mlir@users.noreply.github.com> # This is the commit message llvm#3: update PyTorch version to 2.1.0.dev20230730 (llvm#2356) - torch version: 2.1.0.dev20230730 - torch commit hash: 0ff243ff350268cc98fe03fa6364375ee2824742 - torchvision version: 0.16.0.dev20230730 Co-authored-by: Roll PyTorch Action <torch-mlir@users.noreply.github.com> # This is the commit message llvm#4: update PyTorch version to 2.1.0.dev20230731 (llvm#2359) - torch version: 2.1.0.dev20230731 - torch commit hash: 6298ac688f8caafe30d71ff2ea2e20fbb32065c7 - torchvision version: 0.16.0.dev20230731 Co-authored-by: Roll PyTorch Action <torch-mlir@users.noreply.github.com> # This is the commit message llvm#5: LTC->MLIR Debug Info support (llvm#1922) * LTC->MLIR Debug Info support * SW-95317 Propagate Lazy->Jit->MLIR scope name. * Enhance location information based on op names Currently, the location information attached to the ops just considers the filename, line number and column number. Attaching operation name would help identify the type of computation by just looking at the profile of execution. * Update locations logic; updated debug-info.py test * Use {scope}/{op_name} format to track names by default --------- Co-authored-by: Gleb Kazantaev <gleb.kazantaev@cerebras.net> Co-authored-by: Mark Browning <mark@cerebras.net> Co-authored-by: Vimal Patel <vimal@polymagelabs.com> # This is the commit message llvm#6: build: update llvm tag to 41895843 Summary of changes: - Update tags llvm: 41895843b5915bb78e9d02aa711fa10f7174db43 mhlo: 4726d31f7025da66de0dea709bd56c462edb83c2 Signed-Off By: Vivek Khandelwal <vivek@nod-labs.com> # This is the commit message llvm#7: update PyTorch version to 2.1.0.dev20230802 (llvm#2366) - torch version: 2.1.0.dev20230802 - torch commit hash: c89b16917755c2abbef7b6420e340baf9ae8089e - torchvision version: 0.16.0.dev20230802 Co-authored-by: Roll PyTorch Action <torch-mlir@users.noreply.github.com> # This is the commit message llvm#8: Change Python version from 3.10 to 3.11 in installation instructions (llvm#2370) # This is the commit message llvm#9: Add CITATION file (llvm#2371) # This is the commit message llvm#10: Add packaging as an install dependency (llvm#2369) Needed by `torch_mlir._version`. Resolves llvm#2368. # This is the commit message llvm#11: [Torch Dialect] emit aten.masked_scatter and aten.masked_scatter_ op (llvm#2358) * [Torch Dialect] emit aten.masked_scatter and aten.masked_scatter_ op # This is the commit message llvm#12: update PyTorch version to 2.1.0.dev20230803 (llvm#2372) - torch version: 2.1.0.dev20230803 - torch commit hash: f89c73be3a3e8274d025ac46a33a780853841c9e - torchvision version: 0.16.0.dev20230803 Co-authored-by: Roll PyTorch Action <torch-mlir@users.noreply.github.com> # This is the commit message llvm#13: Prevent failed stable CI job from cancelling nightly jobs (llvm#2373) The CI jobs that use stable PyTorch are currently not required to pass in order for a patch to get merged in `main`. This commit makes sure that if a CI job for stable PyTorch fails, it does not cancel the other required jobs. # This is the commit message llvm#14: [Torch Dialect] emit aten.tile op and decompose it into aten.repeat (llvm#2355) # This is the commit message llvm#15: update # This is the commit message llvm#16: update xfail sets # This is the commit message llvm#17: update xfail_sets # This is the commit message llvm#18: update # This is the commit message llvm#19: fix xfail_sets # This is the commit message llvm#20: update: # This is the commit message llvm#21: update # This is the commit message llvm#22: update:
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Add the Aten Dialect.
Co-authored-by: Jeff Fifield jefff@xilinx.com