Removed the memory leaks directly in execution session #1746
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
Signed-off-by: Alexandre Eichenberger <[email protected]>
chentong319
requested changes
Sep 28, 2022
gongsu832
reviewed
Sep 28, 2022
test/numerical/TestLoop.cpp
Outdated
| auto yInitTensor = OMTensorUniquePtr( | ||
| omTensorCreateEmpty(&yInitShape[0], 1, OM_DATA_TYPE::ONNX_TYPE_INT64), | ||
| omTensorCreateEmpty(&yInitShape, 1, OM_DATA_TYPE::ONNX_TYPE_INT64), |
There was a problem hiding this comment.
Can avoid changing this line by changing the declaration for yInitShape to:
int64_t yInitShape[1] = { 1 };
Same for yRefInitShape below.
There was a problem hiding this comment.
I don't see the difference, but yes we can. done.
There was a problem hiding this comment.
Difference is the array definition serves as a reminder that shape is typically an array.
There was a problem hiding this comment.
sure, I buy that even though there is no difference between a scalar and an array of 1 element.
src/Runtime/OMTensorList.inc
Outdated
| * unpublished call that is mainly used to handle the unique tensor scheme used | ||
| * in Execution Session. | ||
| */ | ||
| void omTensorListDestroyWithoutDestroyingTensors(OMTensorList *list) { |
There was a problem hiding this comment.
Maybe shorter name omTensorListDestroyShallow.
There was a problem hiding this comment.
like it better too
There was a problem hiding this comment.
done
gongsu832
reviewed
Sep 28, 2022
src/Runtime/ExecutionSession.cpp
Outdated
| @@ -26,6 +26,8 @@ | |||
|
|
|||
| #include "ExecutionSession.hpp" | |||
|
|
|||
| void omTensorListDestroyWithoutDestroyingTensors(OMTensorList *list); | |||
There was a problem hiding this comment.
Maybe move into ExecutionSession.hpp?
There was a problem hiding this comment.
I don't want to expose it to anyone doing the ExecutionSession, but created a OMTensorListHelper.hpp file for that.
Signed-off-by: Alexandre Eichenberger <[email protected]>
|
@chentong319 @gongsu832 responded and/or acted on all your requests. |
Signed-off-by: Alexandre Eichenberger <[email protected]>
chentong319
approved these changes
Sep 29, 2022
|
Jenkins Linux s390x Build #7936 [push] Removed the memory leaks... started at 21:10 |
|
Jenkins Linux ppc64le Build #6984 [push] Removed the memory leaks... started at 21:13 |
|
Jenkins Linux amd64 Build #7920 [push] Removed the memory leaks... started at 20:10 |
|
Jenkins Linux amd64 Build #7920 [push] Removed the memory leaks... passed after 1 hr 22 min |
|
Jenkins Linux ppc64le Build #6984 [push] Removed the memory leaks... passed after 1 hr 49 min |
|
Jenkins Linux s390x Build #7936 [push] Removed the memory leaks... passed after 1 hr 55 min |
hamptonm1
added a commit
that referenced
this pull request
Oct 11, 2022
* Update onnx-mlir product version Signed-off-by: Megan Hampton <[email protected]> * fix einsum decomposition bug (#1730) Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Update version to 0.3.1 (#1728) Signed-off-by: Megan Hampton <[email protected]> * Fix issues to compile Yolov3-12. (#1726) Stop conversion to enable onnx.LeakyRelu on NNPA when the stickified layouts of input and output are not the same. Signed-off-by: Yasushi Negishi <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * simplified if lowering (#1717) borrowed ideas from ConvertTrivialIfToSelect from https://github.com/llvm/llvm-project/blob/main/mlir/lib/Dialect/SCF/IR/SCF.cpp Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add definitions Signed-off-by: Megan Hampton <[email protected]> * parse string type from getTypeMap() (#1735) added a parse lit test that passes with this fix Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * fix 2 compiler warnings (#1736) * fix sprintf-is-deprecated clang warning warning from Apple clang version 14.0.0 (clang-1400.0.29.201): ``` .../src/Conversion/ONNXToKrnl/PerfectHash.cpp:47:5: warning: 'sprintf' is deprecated: This function is provided for compatibility reasons only. Due to security concerns inherent in the design of sprintf(3), it is highly recommended that you use snprintf(3) instead. [-Wdeprecated-declarations] sprintf(str, "%lld", (long long)val); /Library/Developer/CommandLineTools/SDKs/MacOSX13.0.sdk/usr/include/stdio.h:188:1: note: 'sprintf' has been explicitly marked deprecated here __deprecated_msg("This function is provided for compatibility reasons only. Due to security concerns inherent in the design of sprintf(3), it is highly recommended that you use snprintf(3) instead.") ``` Signed-off-by: Soren Lassen <[email protected]> * fix copy constructor warning copied solution from InstrumentONNXPass.cpp fixes compiler warning in Linux CI pipelines: ``` .../src/Transform/ONNX/ConvOpt.cpp: In copy constructor '{anonymous}::ConvOptONNXToONNXPass::ConvOptONNXToONNXPass(const {anonymous}::ConvOptONNXToONNXPass&)': .../src/Transform/ONNX/ConvOpt.cpp:211:3: warning: base class 'class mlir::PassWrapper<{anonymous}::ConvOptONNXToONNXPass, mlir::OperationPass<mlir::func::FuncOp> >' should be explicitly initialized in the copy constructor [-Wextra] ConvOptONNXToONNXPass(const ConvOptONNXToONNXPass &pass) {} ``` Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * cleanup some include files (#1733) Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix remaining warnings on Windows and turn on warning-as-error (#1706) This change applies the remaining fixes needed for warnings to be treated as errors on Windows and then turns them on on Windows only. Of particular interest are a couple of changes: * The `check-docs` target needs an update to properly parse the directive even when it contains parenthesis in the path. For example: `C:/Program Files (x86)/Microsoft Visual Studio/Shared/Python39_64/python.exe` * The Windows CI build now builds `onnx` first and then builds all of the targets that are enabled in onnx-mlir. This will make sure that no targets get missed * The remaining warnings (4927) in `ConstProp.cpp` are fixed as well. Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix synax Signed-off-by: Megan Hampton <[email protected]> * Simplify shape-related operations (#1695) * Simplify shape-related operations Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * run-onnx-lib script and docs fixes (#1705) * update documentation for build-run-onnx-lib.sh * clarified Mac issue with running statically linked version in the directory where model library was built * moved usage string up to the top of RunONNXLib.cpp, so it's visible next to the top of file block comment, * allow llvm-project location to be set with LLVM_PROJECT, otherwise read from $MLIR_DIR if defined Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * remove initializedTensors map from ONNX parser (#1739) instead insert a constant for each initializer in the frontend_symbols_ symbol map this is better for several reasons: 1. initializedTensors were incorrectly visible to function bodies (in TryImportFunctionCallNode) 2. nested bindings didn't hide initializedTensors within their scope 3. it's more efficient to create a constant for each initializer once rather than every time an initializer is accessed 4. it's simpler to look up a single symbol mapping Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add decompositions for v11 split, squeeze, and unsqueeze (#1702) * Add decompositions for v11 split, squeeze, and unsqueeze Co-authored-by: Roberto DiCecco <[email protected]> Signed-off-by: Philip Lassen <[email protected]> * Fix lit test Signed-off-by: Philip Lassen <[email protected]> * Add lit tests Signed-off-by: Philip Lassen <[email protected]> * Delete unneccesary decomp for unsqueeze Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * factored input shaping into ModelInputShaper (#1740) * factored input shaping into ModelInputShaper this makes the it easier to read and understand the input shaping logic and also makes the rest of the FrontendDialectTransformer implementation shorter and easier to read Signed-off-by: Soren Lassen <[email protected]> * documented public methods Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Ingest all of Opset 16 (#1704) * Ingest al of opset 16 Signed-off-by: Philip Lassen <[email protected]> * Delete GridSample test which is no longer relevant Signed-off-by: Philip Lassen <[email protected]> * Use NullStringAttr insead of "none" string Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix onnx-mlir.py docker wrapper (#1734) * Requiring input files to have known suffix (.onnx, .json, or .mlir) greatly simplifies the onnx-mlir.py docker wrapper. Signed-off-by: Gong Su <[email protected]> * Return exit code Signed-off-by: Gong Su <[email protected]> * Add copyright notice Signed-off-by: Gong Su <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Update protobuf (#1747) * Update to protobuf 3.20.2 Signed-off-by: Charles Volzka <[email protected]> * Try 3.18.3 Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add the product version and fix formatting Signed-off-by: Megan Hampton <[email protected]> * Make updates to product version Signed-off-by: Megan Hampton <[email protected]> * further updates Signed-off-by: Megan Hampton <[email protected]> * Add product version text file Signed-off-by: Megan Hampton <[email protected]> * Removed the memory leaks directly in execution session (#1746) * removed the memory leaks directly in execution session Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add support for ArgMin (#1737) * add support for ArgMin Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Inclusive terminology update (#1750) * Update `master` reference to `main` in onnx-mlir doc * Update rapidcheck to newer commit with inclusive terminology changes * Update benchmark to 1.6.2 with inclusive terminology changes * Update pybind11 to 2.10 with inclusive terminology changes Signed-off-by: Megan Hampton <[email protected]> * Warning and instructions that --onnx-op-stats needs more than EmitONNXIR (#1719) Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Move product version text file Signed-off-by: Megan Hampton <[email protected]> * Remove comment Signed-off-by: Megan Hampton <[email protected]> * Output the product version for the compiler/version Signed-off-by: Megan Hampton <[email protected]> * Cleaned up commit of the InferTypes change. (#1753) Signed-off-by: Brad Messer <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * setup.py install is deprecated, use pip install instead (#1752) * setup.py install is deprecated, use pip install instead (see https://blog.ganssle.io/articles/2021/10/setup-py-deprecated.html for more details). Signed-off-by: Gong Su <[email protected]> * - pip install doesn't work for protobuf, revert back to setup.py install - use --cpp_implementation for better performance Signed-off-by: Gong Su <[email protected]> * Update Windows CI Signed-off-by: Gong Su <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * adding reference to onnx-mlir-serving (#1745) Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * [TOSA] Update type converter and unary ops (#1553) Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix syntax Signed-off-by: Megan Hampton <[email protected]> * Remove strip Signed-off-by: Megan Hampton <[email protected]> * Revert "Cleaned up commit of the InferTypes change. (#1753)" This reverts commit d08c5ac. Signed-off-by: Megan Hampton <[email protected]> * Cleaned up commit of the InferTypes change. (#1753) Signed-off-by: Brad Messer <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix for GatherND verifier (#1754) Co-authored-by: Hadi Jooybar <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix OnnxBuilder::concat where axis was not used (#1759) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Simplify GatherOp when its inputs are dimensions (#1755) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Assign each question mark a unique negative integer value (#1757) * Assign each question mark a unique negative integer value Signed-off-by: Tung D. Le <[email protected]> * Add a mutex to protect the counter Signed-off-by: Tung D. Le <[email protected]> * Using decrement Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Change the regex Signed-off-by: Megan Hampton <[email protected]> * Fix regex and display product numbers Signed-off-by: Megan Hampton <[email protected]> * normalize axis in ScatterElements verify (#1760) Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * redo ArrayRefOrSmallVector with enable_if (#1758) closes #1729 * redo ArrayRefOrSmallVector with enable_if attempt to address issue #1729: Build error on Linux Signed-off-by: Soren Lassen <[email protected]> * added @qedawkins fix: begin,end -> data,size Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Quinn Dawkins [email protected] Co-authored-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * fix syntax Signed-off-by: Megan Hampton <[email protected]> * Handle scalar tensor tensor<type> in SimplifyShapeRelatesOps pass (#1764) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * namespace cleanup (#1763) * namespace cleanup removed `using namespace mlir` from ConstPropHelper.hpp and added requisite `mlir::` and `llvm::` in header files to get everything to compile again also put ConstPropHelper definitions in namespace onnx_mlir Signed-off-by: Soren Lassen <[email protected]> * added more missing `mlir/llvm::` Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Signed-off-by: Megan Hampton <[email protected]> Fix Clang format error * Make use of the product vendor flag Signed-off-by: Megan Hampton <[email protected]> * Address feedback Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Gong Su <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: Megan Hampton <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Co-authored-by: Charles Volzka <[email protected]> Co-authored-by: Yasushi Negishi <[email protected]> Co-authored-by: chentong319 <[email protected]> Co-authored-by: Stella Stamenova <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Co-authored-by: gongsu832 <[email protected]> Co-authored-by: Meng, Hengyu <[email protected]> Co-authored-by: Brad Messer <[email protected]> Co-authored-by: Philipp Braun <[email protected]> Co-authored-by: Hadi Jooybar <[email protected]>
chentong319
added a commit
to chentong319/onnx-mlir
that referenced
this pull request
Oct 12, 2022
* Update onnx-mlir product version Signed-off-by: Megan Hampton <[email protected]> * fix einsum decomposition bug (onnx#1730) Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Update version to 0.3.1 (onnx#1728) Signed-off-by: Megan Hampton <[email protected]> * Fix issues to compile Yolov3-12. (onnx#1726) Stop conversion to enable onnx.LeakyRelu on NNPA when the stickified layouts of input and output are not the same. Signed-off-by: Yasushi Negishi <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * simplified if lowering (onnx#1717) borrowed ideas from ConvertTrivialIfToSelect from https://github.com/llvm/llvm-project/blob/main/mlir/lib/Dialect/SCF/IR/SCF.cpp Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add definitions Signed-off-by: Megan Hampton <[email protected]> * parse string type from getTypeMap() (onnx#1735) added a parse lit test that passes with this fix Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * fix 2 compiler warnings (onnx#1736) * fix sprintf-is-deprecated clang warning warning from Apple clang version 14.0.0 (clang-1400.0.29.201): ``` .../src/Conversion/ONNXToKrnl/PerfectHash.cpp:47:5: warning: 'sprintf' is deprecated: This function is provided for compatibility reasons only. Due to security concerns inherent in the design of sprintf(3), it is highly recommended that you use snprintf(3) instead. [-Wdeprecated-declarations] sprintf(str, "%lld", (long long)val); /Library/Developer/CommandLineTools/SDKs/MacOSX13.0.sdk/usr/include/stdio.h:188:1: note: 'sprintf' has been explicitly marked deprecated here __deprecated_msg("This function is provided for compatibility reasons only. Due to security concerns inherent in the design of sprintf(3), it is highly recommended that you use snprintf(3) instead.") ``` Signed-off-by: Soren Lassen <[email protected]> * fix copy constructor warning copied solution from InstrumentONNXPass.cpp fixes compiler warning in Linux CI pipelines: ``` .../src/Transform/ONNX/ConvOpt.cpp: In copy constructor '{anonymous}::ConvOptONNXToONNXPass::ConvOptONNXToONNXPass(const {anonymous}::ConvOptONNXToONNXPass&)': .../src/Transform/ONNX/ConvOpt.cpp:211:3: warning: base class 'class mlir::PassWrapper<{anonymous}::ConvOptONNXToONNXPass, mlir::OperationPass<mlir::func::FuncOp> >' should be explicitly initialized in the copy constructor [-Wextra] ConvOptONNXToONNXPass(const ConvOptONNXToONNXPass &pass) {} ``` Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * cleanup some include files (onnx#1733) Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix remaining warnings on Windows and turn on warning-as-error (onnx#1706) This change applies the remaining fixes needed for warnings to be treated as errors on Windows and then turns them on on Windows only. Of particular interest are a couple of changes: * The `check-docs` target needs an update to properly parse the directive even when it contains parenthesis in the path. For example: `C:/Program Files (x86)/Microsoft Visual Studio/Shared/Python39_64/python.exe` * The Windows CI build now builds `onnx` first and then builds all of the targets that are enabled in onnx-mlir. This will make sure that no targets get missed * The remaining warnings (4927) in `ConstProp.cpp` are fixed as well. Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix synax Signed-off-by: Megan Hampton <[email protected]> * Simplify shape-related operations (onnx#1695) * Simplify shape-related operations Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * run-onnx-lib script and docs fixes (onnx#1705) * update documentation for build-run-onnx-lib.sh * clarified Mac issue with running statically linked version in the directory where model library was built * moved usage string up to the top of RunONNXLib.cpp, so it's visible next to the top of file block comment, * allow llvm-project location to be set with LLVM_PROJECT, otherwise read from $MLIR_DIR if defined Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * remove initializedTensors map from ONNX parser (onnx#1739) instead insert a constant for each initializer in the frontend_symbols_ symbol map this is better for several reasons: 1. initializedTensors were incorrectly visible to function bodies (in TryImportFunctionCallNode) 2. nested bindings didn't hide initializedTensors within their scope 3. it's more efficient to create a constant for each initializer once rather than every time an initializer is accessed 4. it's simpler to look up a single symbol mapping Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add decompositions for v11 split, squeeze, and unsqueeze (onnx#1702) * Add decompositions for v11 split, squeeze, and unsqueeze Co-authored-by: Roberto DiCecco <[email protected]> Signed-off-by: Philip Lassen <[email protected]> * Fix lit test Signed-off-by: Philip Lassen <[email protected]> * Add lit tests Signed-off-by: Philip Lassen <[email protected]> * Delete unneccesary decomp for unsqueeze Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * factored input shaping into ModelInputShaper (onnx#1740) * factored input shaping into ModelInputShaper this makes the it easier to read and understand the input shaping logic and also makes the rest of the FrontendDialectTransformer implementation shorter and easier to read Signed-off-by: Soren Lassen <[email protected]> * documented public methods Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Ingest all of Opset 16 (onnx#1704) * Ingest al of opset 16 Signed-off-by: Philip Lassen <[email protected]> * Delete GridSample test which is no longer relevant Signed-off-by: Philip Lassen <[email protected]> * Use NullStringAttr insead of "none" string Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix onnx-mlir.py docker wrapper (onnx#1734) * Requiring input files to have known suffix (.onnx, .json, or .mlir) greatly simplifies the onnx-mlir.py docker wrapper. Signed-off-by: Gong Su <[email protected]> * Return exit code Signed-off-by: Gong Su <[email protected]> * Add copyright notice Signed-off-by: Gong Su <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Update protobuf (onnx#1747) * Update to protobuf 3.20.2 Signed-off-by: Charles Volzka <[email protected]> * Try 3.18.3 Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add the product version and fix formatting Signed-off-by: Megan Hampton <[email protected]> * Make updates to product version Signed-off-by: Megan Hampton <[email protected]> * further updates Signed-off-by: Megan Hampton <[email protected]> * Add product version text file Signed-off-by: Megan Hampton <[email protected]> * Removed the memory leaks directly in execution session (onnx#1746) * removed the memory leaks directly in execution session Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add support for ArgMin (onnx#1737) * add support for ArgMin Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Inclusive terminology update (onnx#1750) * Update `master` reference to `main` in onnx-mlir doc * Update rapidcheck to newer commit with inclusive terminology changes * Update benchmark to 1.6.2 with inclusive terminology changes * Update pybind11 to 2.10 with inclusive terminology changes Signed-off-by: Megan Hampton <[email protected]> * Warning and instructions that --onnx-op-stats needs more than EmitONNXIR (onnx#1719) Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Move product version text file Signed-off-by: Megan Hampton <[email protected]> * Remove comment Signed-off-by: Megan Hampton <[email protected]> * Output the product version for the compiler/version Signed-off-by: Megan Hampton <[email protected]> * Cleaned up commit of the InferTypes change. (onnx#1753) Signed-off-by: Brad Messer <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * setup.py install is deprecated, use pip install instead (onnx#1752) * setup.py install is deprecated, use pip install instead (see https://blog.ganssle.io/articles/2021/10/setup-py-deprecated.html for more details). Signed-off-by: Gong Su <[email protected]> * - pip install doesn't work for protobuf, revert back to setup.py install - use --cpp_implementation for better performance Signed-off-by: Gong Su <[email protected]> * Update Windows CI Signed-off-by: Gong Su <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * adding reference to onnx-mlir-serving (onnx#1745) Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * [TOSA] Update type converter and unary ops (onnx#1553) Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix syntax Signed-off-by: Megan Hampton <[email protected]> * Remove strip Signed-off-by: Megan Hampton <[email protected]> * Revert "Cleaned up commit of the InferTypes change. (onnx#1753)" This reverts commit d08c5ac. Signed-off-by: Megan Hampton <[email protected]> * Cleaned up commit of the InferTypes change. (onnx#1753) Signed-off-by: Brad Messer <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix for GatherND verifier (onnx#1754) Co-authored-by: Hadi Jooybar <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix OnnxBuilder::concat where axis was not used (onnx#1759) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Simplify GatherOp when its inputs are dimensions (onnx#1755) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Assign each question mark a unique negative integer value (onnx#1757) * Assign each question mark a unique negative integer value Signed-off-by: Tung D. Le <[email protected]> * Add a mutex to protect the counter Signed-off-by: Tung D. Le <[email protected]> * Using decrement Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Change the regex Signed-off-by: Megan Hampton <[email protected]> * Fix regex and display product numbers Signed-off-by: Megan Hampton <[email protected]> * normalize axis in ScatterElements verify (onnx#1760) Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * redo ArrayRefOrSmallVector with enable_if (onnx#1758) closes onnx#1729 * redo ArrayRefOrSmallVector with enable_if attempt to address issue onnx#1729: Build error on Linux Signed-off-by: Soren Lassen <[email protected]> * added @qedawkins fix: begin,end -> data,size Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Quinn Dawkins [email protected] Co-authored-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * fix syntax Signed-off-by: Megan Hampton <[email protected]> * Handle scalar tensor tensor<type> in SimplifyShapeRelatesOps pass (onnx#1764) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * namespace cleanup (onnx#1763) * namespace cleanup removed `using namespace mlir` from ConstPropHelper.hpp and added requisite `mlir::` and `llvm::` in header files to get everything to compile again also put ConstPropHelper definitions in namespace onnx_mlir Signed-off-by: Soren Lassen <[email protected]> * added more missing `mlir/llvm::` Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Signed-off-by: Megan Hampton <[email protected]> Fix Clang format error * Make use of the product vendor flag Signed-off-by: Megan Hampton <[email protected]> * Address feedback Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Gong Su <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: Megan Hampton <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Co-authored-by: Charles Volzka <[email protected]> Co-authored-by: Yasushi Negishi <[email protected]> Co-authored-by: chentong319 <[email protected]> Co-authored-by: Stella Stamenova <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Co-authored-by: gongsu832 <[email protected]> Co-authored-by: Meng, Hengyu <[email protected]> Co-authored-by: Brad Messer <[email protected]> Co-authored-by: Philipp Braun <[email protected]> Co-authored-by: Hadi Jooybar <[email protected]>
chentong319
added a commit
that referenced
this pull request
Oct 13, 2022
* update shape inference and test Signed-off-by: chentong319 <[email protected]> * lowering Signed-off-by: chentong319 <[email protected]> * lowering Signed-off-by: chentong319 <[email protected]> * Update onnx-mlir product version (#1727) * Update onnx-mlir product version Signed-off-by: Megan Hampton <[email protected]> * fix einsum decomposition bug (#1730) Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Update version to 0.3.1 (#1728) Signed-off-by: Megan Hampton <[email protected]> * Fix issues to compile Yolov3-12. (#1726) Stop conversion to enable onnx.LeakyRelu on NNPA when the stickified layouts of input and output are not the same. Signed-off-by: Yasushi Negishi <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * simplified if lowering (#1717) borrowed ideas from ConvertTrivialIfToSelect from https://github.com/llvm/llvm-project/blob/main/mlir/lib/Dialect/SCF/IR/SCF.cpp Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add definitions Signed-off-by: Megan Hampton <[email protected]> * parse string type from getTypeMap() (#1735) added a parse lit test that passes with this fix Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * fix 2 compiler warnings (#1736) * fix sprintf-is-deprecated clang warning warning from Apple clang version 14.0.0 (clang-1400.0.29.201): ``` .../src/Conversion/ONNXToKrnl/PerfectHash.cpp:47:5: warning: 'sprintf' is deprecated: This function is provided for compatibility reasons only. Due to security concerns inherent in the design of sprintf(3), it is highly recommended that you use snprintf(3) instead. [-Wdeprecated-declarations] sprintf(str, "%lld", (long long)val); /Library/Developer/CommandLineTools/SDKs/MacOSX13.0.sdk/usr/include/stdio.h:188:1: note: 'sprintf' has been explicitly marked deprecated here __deprecated_msg("This function is provided for compatibility reasons only. Due to security concerns inherent in the design of sprintf(3), it is highly recommended that you use snprintf(3) instead.") ``` Signed-off-by: Soren Lassen <[email protected]> * fix copy constructor warning copied solution from InstrumentONNXPass.cpp fixes compiler warning in Linux CI pipelines: ``` .../src/Transform/ONNX/ConvOpt.cpp: In copy constructor '{anonymous}::ConvOptONNXToONNXPass::ConvOptONNXToONNXPass(const {anonymous}::ConvOptONNXToONNXPass&)': .../src/Transform/ONNX/ConvOpt.cpp:211:3: warning: base class 'class mlir::PassWrapper<{anonymous}::ConvOptONNXToONNXPass, mlir::OperationPass<mlir::func::FuncOp> >' should be explicitly initialized in the copy constructor [-Wextra] ConvOptONNXToONNXPass(const ConvOptONNXToONNXPass &pass) {} ``` Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * cleanup some include files (#1733) Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix remaining warnings on Windows and turn on warning-as-error (#1706) This change applies the remaining fixes needed for warnings to be treated as errors on Windows and then turns them on on Windows only. Of particular interest are a couple of changes: * The `check-docs` target needs an update to properly parse the directive even when it contains parenthesis in the path. For example: `C:/Program Files (x86)/Microsoft Visual Studio/Shared/Python39_64/python.exe` * The Windows CI build now builds `onnx` first and then builds all of the targets that are enabled in onnx-mlir. This will make sure that no targets get missed * The remaining warnings (4927) in `ConstProp.cpp` are fixed as well. Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix synax Signed-off-by: Megan Hampton <[email protected]> * Simplify shape-related operations (#1695) * Simplify shape-related operations Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * run-onnx-lib script and docs fixes (#1705) * update documentation for build-run-onnx-lib.sh * clarified Mac issue with running statically linked version in the directory where model library was built * moved usage string up to the top of RunONNXLib.cpp, so it's visible next to the top of file block comment, * allow llvm-project location to be set with LLVM_PROJECT, otherwise read from $MLIR_DIR if defined Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * remove initializedTensors map from ONNX parser (#1739) instead insert a constant for each initializer in the frontend_symbols_ symbol map this is better for several reasons: 1. initializedTensors were incorrectly visible to function bodies (in TryImportFunctionCallNode) 2. nested bindings didn't hide initializedTensors within their scope 3. it's more efficient to create a constant for each initializer once rather than every time an initializer is accessed 4. it's simpler to look up a single symbol mapping Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add decompositions for v11 split, squeeze, and unsqueeze (#1702) * Add decompositions for v11 split, squeeze, and unsqueeze Co-authored-by: Roberto DiCecco <[email protected]> Signed-off-by: Philip Lassen <[email protected]> * Fix lit test Signed-off-by: Philip Lassen <[email protected]> * Add lit tests Signed-off-by: Philip Lassen <[email protected]> * Delete unneccesary decomp for unsqueeze Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * factored input shaping into ModelInputShaper (#1740) * factored input shaping into ModelInputShaper this makes the it easier to read and understand the input shaping logic and also makes the rest of the FrontendDialectTransformer implementation shorter and easier to read Signed-off-by: Soren Lassen <[email protected]> * documented public methods Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Ingest all of Opset 16 (#1704) * Ingest al of opset 16 Signed-off-by: Philip Lassen <[email protected]> * Delete GridSample test which is no longer relevant Signed-off-by: Philip Lassen <[email protected]> * Use NullStringAttr insead of "none" string Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix onnx-mlir.py docker wrapper (#1734) * Requiring input files to have known suffix (.onnx, .json, or .mlir) greatly simplifies the onnx-mlir.py docker wrapper. Signed-off-by: Gong Su <[email protected]> * Return exit code Signed-off-by: Gong Su <[email protected]> * Add copyright notice Signed-off-by: Gong Su <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Update protobuf (#1747) * Update to protobuf 3.20.2 Signed-off-by: Charles Volzka <[email protected]> * Try 3.18.3 Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add the product version and fix formatting Signed-off-by: Megan Hampton <[email protected]> * Make updates to product version Signed-off-by: Megan Hampton <[email protected]> * further updates Signed-off-by: Megan Hampton <[email protected]> * Add product version text file Signed-off-by: Megan Hampton <[email protected]> * Removed the memory leaks directly in execution session (#1746) * removed the memory leaks directly in execution session Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add support for ArgMin (#1737) * add support for ArgMin Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Inclusive terminology update (#1750) * Update `master` reference to `main` in onnx-mlir doc * Update rapidcheck to newer commit with inclusive terminology changes * Update benchmark to 1.6.2 with inclusive terminology changes * Update pybind11 to 2.10 with inclusive terminology changes Signed-off-by: Megan Hampton <[email protected]> * Warning and instructions that --onnx-op-stats needs more than EmitONNXIR (#1719) Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Move product version text file Signed-off-by: Megan Hampton <[email protected]> * Remove comment Signed-off-by: Megan Hampton <[email protected]> * Output the product version for the compiler/version Signed-off-by: Megan Hampton <[email protected]> * Cleaned up commit of the InferTypes change. (#1753) Signed-off-by: Brad Messer <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * setup.py install is deprecated, use pip install instead (#1752) * setup.py install is deprecated, use pip install instead (see https://blog.ganssle.io/articles/2021/10/setup-py-deprecated.html for more details). Signed-off-by: Gong Su <[email protected]> * - pip install doesn't work for protobuf, revert back to setup.py install - use --cpp_implementation for better performance Signed-off-by: Gong Su <[email protected]> * Update Windows CI Signed-off-by: Gong Su <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * adding reference to onnx-mlir-serving (#1745) Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * [TOSA] Update type converter and unary ops (#1553) Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix syntax Signed-off-by: Megan Hampton <[email protected]> * Remove strip Signed-off-by: Megan Hampton <[email protected]> * Revert "Cleaned up commit of the InferTypes change. (#1753)" This reverts commit d08c5ac. Signed-off-by: Megan Hampton <[email protected]> * Cleaned up commit of the InferTypes change. (#1753) Signed-off-by: Brad Messer <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix for GatherND verifier (#1754) Co-authored-by: Hadi Jooybar <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix OnnxBuilder::concat where axis was not used (#1759) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Simplify GatherOp when its inputs are dimensions (#1755) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Assign each question mark a unique negative integer value (#1757) * Assign each question mark a unique negative integer value Signed-off-by: Tung D. Le <[email protected]> * Add a mutex to protect the counter Signed-off-by: Tung D. Le <[email protected]> * Using decrement Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Change the regex Signed-off-by: Megan Hampton <[email protected]> * Fix regex and display product numbers Signed-off-by: Megan Hampton <[email protected]> * normalize axis in ScatterElements verify (#1760) Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * redo ArrayRefOrSmallVector with enable_if (#1758) closes #1729 * redo ArrayRefOrSmallVector with enable_if attempt to address issue #1729: Build error on Linux Signed-off-by: Soren Lassen <[email protected]> * added @qedawkins fix: begin,end -> data,size Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Quinn Dawkins [email protected] Co-authored-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * fix syntax Signed-off-by: Megan Hampton <[email protected]> * Handle scalar tensor tensor<type> in SimplifyShapeRelatesOps pass (#1764) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * namespace cleanup (#1763) * namespace cleanup removed `using namespace mlir` from ConstPropHelper.hpp and added requisite `mlir::` and `llvm::` in header files to get everything to compile again also put ConstPropHelper definitions in namespace onnx_mlir Signed-off-by: Soren Lassen <[email protected]> * added more missing `mlir/llvm::` Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Signed-off-by: Megan Hampton <[email protected]> Fix Clang format error * Make use of the product vendor flag Signed-off-by: Megan Hampton <[email protected]> * Address feedback Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Gong Su <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: Megan Hampton <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Co-authored-by: Charles Volzka <[email protected]> Co-authored-by: Yasushi Negishi <[email protected]> Co-authored-by: chentong319 <[email protected]> Co-authored-by: Stella Stamenova <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Co-authored-by: gongsu832 <[email protected]> Co-authored-by: Meng, Hengyu <[email protected]> Co-authored-by: Brad Messer <[email protected]> Co-authored-by: Philipp Braun <[email protected]> Co-authored-by: Hadi Jooybar <[email protected]> * different initial Signed-off-by: chentong319 <[email protected]> Signed-off-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Gong Su <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: hamptonm1 <[email protected]> Co-authored-by: Megan Hampton <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Co-authored-by: Charles Volzka <[email protected]> Co-authored-by: Yasushi Negishi <[email protected]> Co-authored-by: Stella Stamenova <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Co-authored-by: gongsu832 <[email protected]> Co-authored-by: Meng, Hengyu <[email protected]> Co-authored-by: Brad Messer <[email protected]> Co-authored-by: Philipp Braun <[email protected]> Co-authored-by: Hadi Jooybar <[email protected]>
gargaroff
pushed a commit
to ljfitz/onnx-mlir
that referenced
this pull request
Nov 3, 2022
* Update onnx-mlir product version Signed-off-by: Megan Hampton <[email protected]> * fix einsum decomposition bug (onnx#1730) Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Update version to 0.3.1 (onnx#1728) Signed-off-by: Megan Hampton <[email protected]> * Fix issues to compile Yolov3-12. (onnx#1726) Stop conversion to enable onnx.LeakyRelu on NNPA when the stickified layouts of input and output are not the same. Signed-off-by: Yasushi Negishi <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * simplified if lowering (onnx#1717) borrowed ideas from ConvertTrivialIfToSelect from https://github.com/llvm/llvm-project/blob/main/mlir/lib/Dialect/SCF/IR/SCF.cpp Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add definitions Signed-off-by: Megan Hampton <[email protected]> * parse string type from getTypeMap() (onnx#1735) added a parse lit test that passes with this fix Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * fix 2 compiler warnings (onnx#1736) * fix sprintf-is-deprecated clang warning warning from Apple clang version 14.0.0 (clang-1400.0.29.201): ``` .../src/Conversion/ONNXToKrnl/PerfectHash.cpp:47:5: warning: 'sprintf' is deprecated: This function is provided for compatibility reasons only. Due to security concerns inherent in the design of sprintf(3), it is highly recommended that you use snprintf(3) instead. [-Wdeprecated-declarations] sprintf(str, "%lld", (long long)val); /Library/Developer/CommandLineTools/SDKs/MacOSX13.0.sdk/usr/include/stdio.h:188:1: note: 'sprintf' has been explicitly marked deprecated here __deprecated_msg("This function is provided for compatibility reasons only. Due to security concerns inherent in the design of sprintf(3), it is highly recommended that you use snprintf(3) instead.") ``` Signed-off-by: Soren Lassen <[email protected]> * fix copy constructor warning copied solution from InstrumentONNXPass.cpp fixes compiler warning in Linux CI pipelines: ``` .../src/Transform/ONNX/ConvOpt.cpp: In copy constructor '{anonymous}::ConvOptONNXToONNXPass::ConvOptONNXToONNXPass(const {anonymous}::ConvOptONNXToONNXPass&)': .../src/Transform/ONNX/ConvOpt.cpp:211:3: warning: base class 'class mlir::PassWrapper<{anonymous}::ConvOptONNXToONNXPass, mlir::OperationPass<mlir::func::FuncOp> >' should be explicitly initialized in the copy constructor [-Wextra] ConvOptONNXToONNXPass(const ConvOptONNXToONNXPass &pass) {} ``` Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * cleanup some include files (onnx#1733) Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix remaining warnings on Windows and turn on warning-as-error (onnx#1706) This change applies the remaining fixes needed for warnings to be treated as errors on Windows and then turns them on on Windows only. Of particular interest are a couple of changes: * The `check-docs` target needs an update to properly parse the directive even when it contains parenthesis in the path. For example: `C:/Program Files (x86)/Microsoft Visual Studio/Shared/Python39_64/python.exe` * The Windows CI build now builds `onnx` first and then builds all of the targets that are enabled in onnx-mlir. This will make sure that no targets get missed * The remaining warnings (4927) in `ConstProp.cpp` are fixed as well. Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix synax Signed-off-by: Megan Hampton <[email protected]> * Simplify shape-related operations (onnx#1695) * Simplify shape-related operations Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * run-onnx-lib script and docs fixes (onnx#1705) * update documentation for build-run-onnx-lib.sh * clarified Mac issue with running statically linked version in the directory where model library was built * moved usage string up to the top of RunONNXLib.cpp, so it's visible next to the top of file block comment, * allow llvm-project location to be set with LLVM_PROJECT, otherwise read from $MLIR_DIR if defined Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * remove initializedTensors map from ONNX parser (onnx#1739) instead insert a constant for each initializer in the frontend_symbols_ symbol map this is better for several reasons: 1. initializedTensors were incorrectly visible to function bodies (in TryImportFunctionCallNode) 2. nested bindings didn't hide initializedTensors within their scope 3. it's more efficient to create a constant for each initializer once rather than every time an initializer is accessed 4. it's simpler to look up a single symbol mapping Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add decompositions for v11 split, squeeze, and unsqueeze (onnx#1702) * Add decompositions for v11 split, squeeze, and unsqueeze Co-authored-by: Roberto DiCecco <[email protected]> Signed-off-by: Philip Lassen <[email protected]> * Fix lit test Signed-off-by: Philip Lassen <[email protected]> * Add lit tests Signed-off-by: Philip Lassen <[email protected]> * Delete unneccesary decomp for unsqueeze Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * factored input shaping into ModelInputShaper (onnx#1740) * factored input shaping into ModelInputShaper this makes the it easier to read and understand the input shaping logic and also makes the rest of the FrontendDialectTransformer implementation shorter and easier to read Signed-off-by: Soren Lassen <[email protected]> * documented public methods Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Ingest all of Opset 16 (onnx#1704) * Ingest al of opset 16 Signed-off-by: Philip Lassen <[email protected]> * Delete GridSample test which is no longer relevant Signed-off-by: Philip Lassen <[email protected]> * Use NullStringAttr insead of "none" string Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix onnx-mlir.py docker wrapper (onnx#1734) * Requiring input files to have known suffix (.onnx, .json, or .mlir) greatly simplifies the onnx-mlir.py docker wrapper. Signed-off-by: Gong Su <[email protected]> * Return exit code Signed-off-by: Gong Su <[email protected]> * Add copyright notice Signed-off-by: Gong Su <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Update protobuf (onnx#1747) * Update to protobuf 3.20.2 Signed-off-by: Charles Volzka <[email protected]> * Try 3.18.3 Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add the product version and fix formatting Signed-off-by: Megan Hampton <[email protected]> * Make updates to product version Signed-off-by: Megan Hampton <[email protected]> * further updates Signed-off-by: Megan Hampton <[email protected]> * Add product version text file Signed-off-by: Megan Hampton <[email protected]> * Removed the memory leaks directly in execution session (onnx#1746) * removed the memory leaks directly in execution session Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add support for ArgMin (onnx#1737) * add support for ArgMin Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Inclusive terminology update (onnx#1750) * Update `master` reference to `main` in onnx-mlir doc * Update rapidcheck to newer commit with inclusive terminology changes * Update benchmark to 1.6.2 with inclusive terminology changes * Update pybind11 to 2.10 with inclusive terminology changes Signed-off-by: Megan Hampton <[email protected]> * Warning and instructions that --onnx-op-stats needs more than EmitONNXIR (onnx#1719) Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Move product version text file Signed-off-by: Megan Hampton <[email protected]> * Remove comment Signed-off-by: Megan Hampton <[email protected]> * Output the product version for the compiler/version Signed-off-by: Megan Hampton <[email protected]> * Cleaned up commit of the InferTypes change. (onnx#1753) Signed-off-by: Brad Messer <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * setup.py install is deprecated, use pip install instead (onnx#1752) * setup.py install is deprecated, use pip install instead (see https://blog.ganssle.io/articles/2021/10/setup-py-deprecated.html for more details). Signed-off-by: Gong Su <[email protected]> * - pip install doesn't work for protobuf, revert back to setup.py install - use --cpp_implementation for better performance Signed-off-by: Gong Su <[email protected]> * Update Windows CI Signed-off-by: Gong Su <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * adding reference to onnx-mlir-serving (onnx#1745) Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * [TOSA] Update type converter and unary ops (onnx#1553) Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix syntax Signed-off-by: Megan Hampton <[email protected]> * Remove strip Signed-off-by: Megan Hampton <[email protected]> * Revert "Cleaned up commit of the InferTypes change. (onnx#1753)" This reverts commit d08c5ac. Signed-off-by: Megan Hampton <[email protected]> * Cleaned up commit of the InferTypes change. (onnx#1753) Signed-off-by: Brad Messer <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix for GatherND verifier (onnx#1754) Co-authored-by: Hadi Jooybar <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix OnnxBuilder::concat where axis was not used (onnx#1759) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Simplify GatherOp when its inputs are dimensions (onnx#1755) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Assign each question mark a unique negative integer value (onnx#1757) * Assign each question mark a unique negative integer value Signed-off-by: Tung D. Le <[email protected]> * Add a mutex to protect the counter Signed-off-by: Tung D. Le <[email protected]> * Using decrement Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Change the regex Signed-off-by: Megan Hampton <[email protected]> * Fix regex and display product numbers Signed-off-by: Megan Hampton <[email protected]> * normalize axis in ScatterElements verify (onnx#1760) Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * redo ArrayRefOrSmallVector with enable_if (onnx#1758) closes onnx#1729 * redo ArrayRefOrSmallVector with enable_if attempt to address issue onnx#1729: Build error on Linux Signed-off-by: Soren Lassen <[email protected]> * added @qedawkins fix: begin,end -> data,size Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Quinn Dawkins [email protected] Co-authored-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * fix syntax Signed-off-by: Megan Hampton <[email protected]> * Handle scalar tensor tensor<type> in SimplifyShapeRelatesOps pass (onnx#1764) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * namespace cleanup (onnx#1763) * namespace cleanup removed `using namespace mlir` from ConstPropHelper.hpp and added requisite `mlir::` and `llvm::` in header files to get everything to compile again also put ConstPropHelper definitions in namespace onnx_mlir Signed-off-by: Soren Lassen <[email protected]> * added more missing `mlir/llvm::` Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Signed-off-by: Megan Hampton <[email protected]> Fix Clang format error * Make use of the product vendor flag Signed-off-by: Megan Hampton <[email protected]> * Address feedback Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Gong Su <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: Megan Hampton <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Co-authored-by: Charles Volzka <[email protected]> Co-authored-by: Yasushi Negishi <[email protected]> Co-authored-by: chentong319 <[email protected]> Co-authored-by: Stella Stamenova <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Co-authored-by: gongsu832 <[email protected]> Co-authored-by: Meng, Hengyu <[email protected]> Co-authored-by: Brad Messer <[email protected]> Co-authored-by: Philipp Braun <[email protected]> Co-authored-by: Hadi Jooybar <[email protected]>
gargaroff
pushed a commit
to ljfitz/onnx-mlir
that referenced
this pull request
Nov 3, 2022
* update shape inference and test Signed-off-by: chentong319 <[email protected]> * lowering Signed-off-by: chentong319 <[email protected]> * lowering Signed-off-by: chentong319 <[email protected]> * Update onnx-mlir product version (onnx#1727) * Update onnx-mlir product version Signed-off-by: Megan Hampton <[email protected]> * fix einsum decomposition bug (onnx#1730) Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Update version to 0.3.1 (onnx#1728) Signed-off-by: Megan Hampton <[email protected]> * Fix issues to compile Yolov3-12. (onnx#1726) Stop conversion to enable onnx.LeakyRelu on NNPA when the stickified layouts of input and output are not the same. Signed-off-by: Yasushi Negishi <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * simplified if lowering (onnx#1717) borrowed ideas from ConvertTrivialIfToSelect from https://github.com/llvm/llvm-project/blob/main/mlir/lib/Dialect/SCF/IR/SCF.cpp Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add definitions Signed-off-by: Megan Hampton <[email protected]> * parse string type from getTypeMap() (onnx#1735) added a parse lit test that passes with this fix Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * fix 2 compiler warnings (onnx#1736) * fix sprintf-is-deprecated clang warning warning from Apple clang version 14.0.0 (clang-1400.0.29.201): ``` .../src/Conversion/ONNXToKrnl/PerfectHash.cpp:47:5: warning: 'sprintf' is deprecated: This function is provided for compatibility reasons only. Due to security concerns inherent in the design of sprintf(3), it is highly recommended that you use snprintf(3) instead. [-Wdeprecated-declarations] sprintf(str, "%lld", (long long)val); /Library/Developer/CommandLineTools/SDKs/MacOSX13.0.sdk/usr/include/stdio.h:188:1: note: 'sprintf' has been explicitly marked deprecated here __deprecated_msg("This function is provided for compatibility reasons only. Due to security concerns inherent in the design of sprintf(3), it is highly recommended that you use snprintf(3) instead.") ``` Signed-off-by: Soren Lassen <[email protected]> * fix copy constructor warning copied solution from InstrumentONNXPass.cpp fixes compiler warning in Linux CI pipelines: ``` .../src/Transform/ONNX/ConvOpt.cpp: In copy constructor '{anonymous}::ConvOptONNXToONNXPass::ConvOptONNXToONNXPass(const {anonymous}::ConvOptONNXToONNXPass&)': .../src/Transform/ONNX/ConvOpt.cpp:211:3: warning: base class 'class mlir::PassWrapper<{anonymous}::ConvOptONNXToONNXPass, mlir::OperationPass<mlir::func::FuncOp> >' should be explicitly initialized in the copy constructor [-Wextra] ConvOptONNXToONNXPass(const ConvOptONNXToONNXPass &pass) {} ``` Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * cleanup some include files (onnx#1733) Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix remaining warnings on Windows and turn on warning-as-error (onnx#1706) This change applies the remaining fixes needed for warnings to be treated as errors on Windows and then turns them on on Windows only. Of particular interest are a couple of changes: * The `check-docs` target needs an update to properly parse the directive even when it contains parenthesis in the path. For example: `C:/Program Files (x86)/Microsoft Visual Studio/Shared/Python39_64/python.exe` * The Windows CI build now builds `onnx` first and then builds all of the targets that are enabled in onnx-mlir. This will make sure that no targets get missed * The remaining warnings (4927) in `ConstProp.cpp` are fixed as well. Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix synax Signed-off-by: Megan Hampton <[email protected]> * Simplify shape-related operations (onnx#1695) * Simplify shape-related operations Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * run-onnx-lib script and docs fixes (onnx#1705) * update documentation for build-run-onnx-lib.sh * clarified Mac issue with running statically linked version in the directory where model library was built * moved usage string up to the top of RunONNXLib.cpp, so it's visible next to the top of file block comment, * allow llvm-project location to be set with LLVM_PROJECT, otherwise read from $MLIR_DIR if defined Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * remove initializedTensors map from ONNX parser (onnx#1739) instead insert a constant for each initializer in the frontend_symbols_ symbol map this is better for several reasons: 1. initializedTensors were incorrectly visible to function bodies (in TryImportFunctionCallNode) 2. nested bindings didn't hide initializedTensors within their scope 3. it's more efficient to create a constant for each initializer once rather than every time an initializer is accessed 4. it's simpler to look up a single symbol mapping Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add decompositions for v11 split, squeeze, and unsqueeze (onnx#1702) * Add decompositions for v11 split, squeeze, and unsqueeze Co-authored-by: Roberto DiCecco <[email protected]> Signed-off-by: Philip Lassen <[email protected]> * Fix lit test Signed-off-by: Philip Lassen <[email protected]> * Add lit tests Signed-off-by: Philip Lassen <[email protected]> * Delete unneccesary decomp for unsqueeze Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * factored input shaping into ModelInputShaper (onnx#1740) * factored input shaping into ModelInputShaper this makes the it easier to read and understand the input shaping logic and also makes the rest of the FrontendDialectTransformer implementation shorter and easier to read Signed-off-by: Soren Lassen <[email protected]> * documented public methods Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Ingest all of Opset 16 (onnx#1704) * Ingest al of opset 16 Signed-off-by: Philip Lassen <[email protected]> * Delete GridSample test which is no longer relevant Signed-off-by: Philip Lassen <[email protected]> * Use NullStringAttr insead of "none" string Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix onnx-mlir.py docker wrapper (onnx#1734) * Requiring input files to have known suffix (.onnx, .json, or .mlir) greatly simplifies the onnx-mlir.py docker wrapper. Signed-off-by: Gong Su <[email protected]> * Return exit code Signed-off-by: Gong Su <[email protected]> * Add copyright notice Signed-off-by: Gong Su <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Update protobuf (onnx#1747) * Update to protobuf 3.20.2 Signed-off-by: Charles Volzka <[email protected]> * Try 3.18.3 Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add the product version and fix formatting Signed-off-by: Megan Hampton <[email protected]> * Make updates to product version Signed-off-by: Megan Hampton <[email protected]> * further updates Signed-off-by: Megan Hampton <[email protected]> * Add product version text file Signed-off-by: Megan Hampton <[email protected]> * Removed the memory leaks directly in execution session (onnx#1746) * removed the memory leaks directly in execution session Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add support for ArgMin (onnx#1737) * add support for ArgMin Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Inclusive terminology update (onnx#1750) * Update `master` reference to `main` in onnx-mlir doc * Update rapidcheck to newer commit with inclusive terminology changes * Update benchmark to 1.6.2 with inclusive terminology changes * Update pybind11 to 2.10 with inclusive terminology changes Signed-off-by: Megan Hampton <[email protected]> * Warning and instructions that --onnx-op-stats needs more than EmitONNXIR (onnx#1719) Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Move product version text file Signed-off-by: Megan Hampton <[email protected]> * Remove comment Signed-off-by: Megan Hampton <[email protected]> * Output the product version for the compiler/version Signed-off-by: Megan Hampton <[email protected]> * Cleaned up commit of the InferTypes change. (onnx#1753) Signed-off-by: Brad Messer <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * setup.py install is deprecated, use pip install instead (onnx#1752) * setup.py install is deprecated, use pip install instead (see https://blog.ganssle.io/articles/2021/10/setup-py-deprecated.html for more details). Signed-off-by: Gong Su <[email protected]> * - pip install doesn't work for protobuf, revert back to setup.py install - use --cpp_implementation for better performance Signed-off-by: Gong Su <[email protected]> * Update Windows CI Signed-off-by: Gong Su <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * adding reference to onnx-mlir-serving (onnx#1745) Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * [TOSA] Update type converter and unary ops (onnx#1553) Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix syntax Signed-off-by: Megan Hampton <[email protected]> * Remove strip Signed-off-by: Megan Hampton <[email protected]> * Revert "Cleaned up commit of the InferTypes change. (onnx#1753)" This reverts commit d08c5ac. Signed-off-by: Megan Hampton <[email protected]> * Cleaned up commit of the InferTypes change. (onnx#1753) Signed-off-by: Brad Messer <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix for GatherND verifier (onnx#1754) Co-authored-by: Hadi Jooybar <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix OnnxBuilder::concat where axis was not used (onnx#1759) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Simplify GatherOp when its inputs are dimensions (onnx#1755) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Assign each question mark a unique negative integer value (onnx#1757) * Assign each question mark a unique negative integer value Signed-off-by: Tung D. Le <[email protected]> * Add a mutex to protect the counter Signed-off-by: Tung D. Le <[email protected]> * Using decrement Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Change the regex Signed-off-by: Megan Hampton <[email protected]> * Fix regex and display product numbers Signed-off-by: Megan Hampton <[email protected]> * normalize axis in ScatterElements verify (onnx#1760) Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * redo ArrayRefOrSmallVector with enable_if (onnx#1758) closes onnx#1729 * redo ArrayRefOrSmallVector with enable_if attempt to address issue onnx#1729: Build error on Linux Signed-off-by: Soren Lassen <[email protected]> * added @qedawkins fix: begin,end -> data,size Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Quinn Dawkins [email protected] Co-authored-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * fix syntax Signed-off-by: Megan Hampton <[email protected]> * Handle scalar tensor tensor<type> in SimplifyShapeRelatesOps pass (onnx#1764) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * namespace cleanup (onnx#1763) * namespace cleanup removed `using namespace mlir` from ConstPropHelper.hpp and added requisite `mlir::` and `llvm::` in header files to get everything to compile again also put ConstPropHelper definitions in namespace onnx_mlir Signed-off-by: Soren Lassen <[email protected]> * added more missing `mlir/llvm::` Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Signed-off-by: Megan Hampton <[email protected]> Fix Clang format error * Make use of the product vendor flag Signed-off-by: Megan Hampton <[email protected]> * Address feedback Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Gong Su <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: Megan Hampton <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Co-authored-by: Charles Volzka <[email protected]> Co-authored-by: Yasushi Negishi <[email protected]> Co-authored-by: chentong319 <[email protected]> Co-authored-by: Stella Stamenova <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Co-authored-by: gongsu832 <[email protected]> Co-authored-by: Meng, Hengyu <[email protected]> Co-authored-by: Brad Messer <[email protected]> Co-authored-by: Philipp Braun <[email protected]> Co-authored-by: Hadi Jooybar <[email protected]> * different initial Signed-off-by: chentong319 <[email protected]> Signed-off-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Gong Su <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: hamptonm1 <[email protected]> Co-authored-by: Megan Hampton <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Co-authored-by: Charles Volzka <[email protected]> Co-authored-by: Yasushi Negishi <[email protected]> Co-authored-by: Stella Stamenova <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Co-authored-by: gongsu832 <[email protected]> Co-authored-by: Meng, Hengyu <[email protected]> Co-authored-by: Brad Messer <[email protected]> Co-authored-by: Philipp Braun <[email protected]> Co-authored-by: Hadi Jooybar <[email protected]>
gargaroff
pushed a commit
to ljfitz/onnx-mlir
that referenced
this pull request
Nov 8, 2022
* Update onnx-mlir product version Signed-off-by: Megan Hampton <[email protected]> * fix einsum decomposition bug (onnx#1730) Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Update version to 0.3.1 (onnx#1728) Signed-off-by: Megan Hampton <[email protected]> * Fix issues to compile Yolov3-12. (onnx#1726) Stop conversion to enable onnx.LeakyRelu on NNPA when the stickified layouts of input and output are not the same. Signed-off-by: Yasushi Negishi <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * simplified if lowering (onnx#1717) borrowed ideas from ConvertTrivialIfToSelect from https://github.com/llvm/llvm-project/blob/main/mlir/lib/Dialect/SCF/IR/SCF.cpp Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add definitions Signed-off-by: Megan Hampton <[email protected]> * parse string type from getTypeMap() (onnx#1735) added a parse lit test that passes with this fix Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * fix 2 compiler warnings (onnx#1736) * fix sprintf-is-deprecated clang warning warning from Apple clang version 14.0.0 (clang-1400.0.29.201): ``` .../src/Conversion/ONNXToKrnl/PerfectHash.cpp:47:5: warning: 'sprintf' is deprecated: This function is provided for compatibility reasons only. Due to security concerns inherent in the design of sprintf(3), it is highly recommended that you use snprintf(3) instead. [-Wdeprecated-declarations] sprintf(str, "%lld", (long long)val); /Library/Developer/CommandLineTools/SDKs/MacOSX13.0.sdk/usr/include/stdio.h:188:1: note: 'sprintf' has been explicitly marked deprecated here __deprecated_msg("This function is provided for compatibility reasons only. Due to security concerns inherent in the design of sprintf(3), it is highly recommended that you use snprintf(3) instead.") ``` Signed-off-by: Soren Lassen <[email protected]> * fix copy constructor warning copied solution from InstrumentONNXPass.cpp fixes compiler warning in Linux CI pipelines: ``` .../src/Transform/ONNX/ConvOpt.cpp: In copy constructor '{anonymous}::ConvOptONNXToONNXPass::ConvOptONNXToONNXPass(const {anonymous}::ConvOptONNXToONNXPass&)': .../src/Transform/ONNX/ConvOpt.cpp:211:3: warning: base class 'class mlir::PassWrapper<{anonymous}::ConvOptONNXToONNXPass, mlir::OperationPass<mlir::func::FuncOp> >' should be explicitly initialized in the copy constructor [-Wextra] ConvOptONNXToONNXPass(const ConvOptONNXToONNXPass &pass) {} ``` Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * cleanup some include files (onnx#1733) Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix remaining warnings on Windows and turn on warning-as-error (onnx#1706) This change applies the remaining fixes needed for warnings to be treated as errors on Windows and then turns them on on Windows only. Of particular interest are a couple of changes: * The `check-docs` target needs an update to properly parse the directive even when it contains parenthesis in the path. For example: `C:/Program Files (x86)/Microsoft Visual Studio/Shared/Python39_64/python.exe` * The Windows CI build now builds `onnx` first and then builds all of the targets that are enabled in onnx-mlir. This will make sure that no targets get missed * The remaining warnings (4927) in `ConstProp.cpp` are fixed as well. Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix synax Signed-off-by: Megan Hampton <[email protected]> * Simplify shape-related operations (onnx#1695) * Simplify shape-related operations Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * run-onnx-lib script and docs fixes (onnx#1705) * update documentation for build-run-onnx-lib.sh * clarified Mac issue with running statically linked version in the directory where model library was built * moved usage string up to the top of RunONNXLib.cpp, so it's visible next to the top of file block comment, * allow llvm-project location to be set with LLVM_PROJECT, otherwise read from $MLIR_DIR if defined Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * remove initializedTensors map from ONNX parser (onnx#1739) instead insert a constant for each initializer in the frontend_symbols_ symbol map this is better for several reasons: 1. initializedTensors were incorrectly visible to function bodies (in TryImportFunctionCallNode) 2. nested bindings didn't hide initializedTensors within their scope 3. it's more efficient to create a constant for each initializer once rather than every time an initializer is accessed 4. it's simpler to look up a single symbol mapping Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add decompositions for v11 split, squeeze, and unsqueeze (onnx#1702) * Add decompositions for v11 split, squeeze, and unsqueeze Co-authored-by: Roberto DiCecco <[email protected]> Signed-off-by: Philip Lassen <[email protected]> * Fix lit test Signed-off-by: Philip Lassen <[email protected]> * Add lit tests Signed-off-by: Philip Lassen <[email protected]> * Delete unneccesary decomp for unsqueeze Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * factored input shaping into ModelInputShaper (onnx#1740) * factored input shaping into ModelInputShaper this makes the it easier to read and understand the input shaping logic and also makes the rest of the FrontendDialectTransformer implementation shorter and easier to read Signed-off-by: Soren Lassen <[email protected]> * documented public methods Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Ingest all of Opset 16 (onnx#1704) * Ingest al of opset 16 Signed-off-by: Philip Lassen <[email protected]> * Delete GridSample test which is no longer relevant Signed-off-by: Philip Lassen <[email protected]> * Use NullStringAttr insead of "none" string Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix onnx-mlir.py docker wrapper (onnx#1734) * Requiring input files to have known suffix (.onnx, .json, or .mlir) greatly simplifies the onnx-mlir.py docker wrapper. Signed-off-by: Gong Su <[email protected]> * Return exit code Signed-off-by: Gong Su <[email protected]> * Add copyright notice Signed-off-by: Gong Su <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Update protobuf (onnx#1747) * Update to protobuf 3.20.2 Signed-off-by: Charles Volzka <[email protected]> * Try 3.18.3 Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add the product version and fix formatting Signed-off-by: Megan Hampton <[email protected]> * Make updates to product version Signed-off-by: Megan Hampton <[email protected]> * further updates Signed-off-by: Megan Hampton <[email protected]> * Add product version text file Signed-off-by: Megan Hampton <[email protected]> * Removed the memory leaks directly in execution session (onnx#1746) * removed the memory leaks directly in execution session Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add support for ArgMin (onnx#1737) * add support for ArgMin Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Inclusive terminology update (onnx#1750) * Update `master` reference to `main` in onnx-mlir doc * Update rapidcheck to newer commit with inclusive terminology changes * Update benchmark to 1.6.2 with inclusive terminology changes * Update pybind11 to 2.10 with inclusive terminology changes Signed-off-by: Megan Hampton <[email protected]> * Warning and instructions that --onnx-op-stats needs more than EmitONNXIR (onnx#1719) Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Move product version text file Signed-off-by: Megan Hampton <[email protected]> * Remove comment Signed-off-by: Megan Hampton <[email protected]> * Output the product version for the compiler/version Signed-off-by: Megan Hampton <[email protected]> * Cleaned up commit of the InferTypes change. (onnx#1753) Signed-off-by: Brad Messer <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * setup.py install is deprecated, use pip install instead (onnx#1752) * setup.py install is deprecated, use pip install instead (see https://blog.ganssle.io/articles/2021/10/setup-py-deprecated.html for more details). Signed-off-by: Gong Su <[email protected]> * - pip install doesn't work for protobuf, revert back to setup.py install - use --cpp_implementation for better performance Signed-off-by: Gong Su <[email protected]> * Update Windows CI Signed-off-by: Gong Su <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * adding reference to onnx-mlir-serving (onnx#1745) Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * [TOSA] Update type converter and unary ops (onnx#1553) Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix syntax Signed-off-by: Megan Hampton <[email protected]> * Remove strip Signed-off-by: Megan Hampton <[email protected]> * Revert "Cleaned up commit of the InferTypes change. (onnx#1753)" This reverts commit d08c5ac. Signed-off-by: Megan Hampton <[email protected]> * Cleaned up commit of the InferTypes change. (onnx#1753) Signed-off-by: Brad Messer <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix for GatherND verifier (onnx#1754) Co-authored-by: Hadi Jooybar <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix OnnxBuilder::concat where axis was not used (onnx#1759) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Simplify GatherOp when its inputs are dimensions (onnx#1755) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Assign each question mark a unique negative integer value (onnx#1757) * Assign each question mark a unique negative integer value Signed-off-by: Tung D. Le <[email protected]> * Add a mutex to protect the counter Signed-off-by: Tung D. Le <[email protected]> * Using decrement Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Change the regex Signed-off-by: Megan Hampton <[email protected]> * Fix regex and display product numbers Signed-off-by: Megan Hampton <[email protected]> * normalize axis in ScatterElements verify (onnx#1760) Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * redo ArrayRefOrSmallVector with enable_if (onnx#1758) closes onnx#1729 * redo ArrayRefOrSmallVector with enable_if attempt to address issue onnx#1729: Build error on Linux Signed-off-by: Soren Lassen <[email protected]> * added @qedawkins fix: begin,end -> data,size Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Quinn Dawkins [email protected] Co-authored-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * fix syntax Signed-off-by: Megan Hampton <[email protected]> * Handle scalar tensor tensor<type> in SimplifyShapeRelatesOps pass (onnx#1764) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * namespace cleanup (onnx#1763) * namespace cleanup removed `using namespace mlir` from ConstPropHelper.hpp and added requisite `mlir::` and `llvm::` in header files to get everything to compile again also put ConstPropHelper definitions in namespace onnx_mlir Signed-off-by: Soren Lassen <[email protected]> * added more missing `mlir/llvm::` Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Signed-off-by: Megan Hampton <[email protected]> Fix Clang format error * Make use of the product vendor flag Signed-off-by: Megan Hampton <[email protected]> * Address feedback Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Gong Su <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: Megan Hampton <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Co-authored-by: Charles Volzka <[email protected]> Co-authored-by: Yasushi Negishi <[email protected]> Co-authored-by: chentong319 <[email protected]> Co-authored-by: Stella Stamenova <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Co-authored-by: gongsu832 <[email protected]> Co-authored-by: Meng, Hengyu <[email protected]> Co-authored-by: Brad Messer <[email protected]> Co-authored-by: Philipp Braun <[email protected]> Co-authored-by: Hadi Jooybar <[email protected]>
gargaroff
pushed a commit
to ljfitz/onnx-mlir
that referenced
this pull request
Nov 8, 2022
* update shape inference and test Signed-off-by: chentong319 <[email protected]> * lowering Signed-off-by: chentong319 <[email protected]> * lowering Signed-off-by: chentong319 <[email protected]> * Update onnx-mlir product version (onnx#1727) * Update onnx-mlir product version Signed-off-by: Megan Hampton <[email protected]> * fix einsum decomposition bug (onnx#1730) Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Update version to 0.3.1 (onnx#1728) Signed-off-by: Megan Hampton <[email protected]> * Fix issues to compile Yolov3-12. (onnx#1726) Stop conversion to enable onnx.LeakyRelu on NNPA when the stickified layouts of input and output are not the same. Signed-off-by: Yasushi Negishi <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * simplified if lowering (onnx#1717) borrowed ideas from ConvertTrivialIfToSelect from https://github.com/llvm/llvm-project/blob/main/mlir/lib/Dialect/SCF/IR/SCF.cpp Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add definitions Signed-off-by: Megan Hampton <[email protected]> * parse string type from getTypeMap() (onnx#1735) added a parse lit test that passes with this fix Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * fix 2 compiler warnings (onnx#1736) * fix sprintf-is-deprecated clang warning warning from Apple clang version 14.0.0 (clang-1400.0.29.201): ``` .../src/Conversion/ONNXToKrnl/PerfectHash.cpp:47:5: warning: 'sprintf' is deprecated: This function is provided for compatibility reasons only. Due to security concerns inherent in the design of sprintf(3), it is highly recommended that you use snprintf(3) instead. [-Wdeprecated-declarations] sprintf(str, "%lld", (long long)val); /Library/Developer/CommandLineTools/SDKs/MacOSX13.0.sdk/usr/include/stdio.h:188:1: note: 'sprintf' has been explicitly marked deprecated here __deprecated_msg("This function is provided for compatibility reasons only. Due to security concerns inherent in the design of sprintf(3), it is highly recommended that you use snprintf(3) instead.") ``` Signed-off-by: Soren Lassen <[email protected]> * fix copy constructor warning copied solution from InstrumentONNXPass.cpp fixes compiler warning in Linux CI pipelines: ``` .../src/Transform/ONNX/ConvOpt.cpp: In copy constructor '{anonymous}::ConvOptONNXToONNXPass::ConvOptONNXToONNXPass(const {anonymous}::ConvOptONNXToONNXPass&)': .../src/Transform/ONNX/ConvOpt.cpp:211:3: warning: base class 'class mlir::PassWrapper<{anonymous}::ConvOptONNXToONNXPass, mlir::OperationPass<mlir::func::FuncOp> >' should be explicitly initialized in the copy constructor [-Wextra] ConvOptONNXToONNXPass(const ConvOptONNXToONNXPass &pass) {} ``` Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * cleanup some include files (onnx#1733) Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix remaining warnings on Windows and turn on warning-as-error (onnx#1706) This change applies the remaining fixes needed for warnings to be treated as errors on Windows and then turns them on on Windows only. Of particular interest are a couple of changes: * The `check-docs` target needs an update to properly parse the directive even when it contains parenthesis in the path. For example: `C:/Program Files (x86)/Microsoft Visual Studio/Shared/Python39_64/python.exe` * The Windows CI build now builds `onnx` first and then builds all of the targets that are enabled in onnx-mlir. This will make sure that no targets get missed * The remaining warnings (4927) in `ConstProp.cpp` are fixed as well. Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix synax Signed-off-by: Megan Hampton <[email protected]> * Simplify shape-related operations (onnx#1695) * Simplify shape-related operations Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * run-onnx-lib script and docs fixes (onnx#1705) * update documentation for build-run-onnx-lib.sh * clarified Mac issue with running statically linked version in the directory where model library was built * moved usage string up to the top of RunONNXLib.cpp, so it's visible next to the top of file block comment, * allow llvm-project location to be set with LLVM_PROJECT, otherwise read from $MLIR_DIR if defined Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * remove initializedTensors map from ONNX parser (onnx#1739) instead insert a constant for each initializer in the frontend_symbols_ symbol map this is better for several reasons: 1. initializedTensors were incorrectly visible to function bodies (in TryImportFunctionCallNode) 2. nested bindings didn't hide initializedTensors within their scope 3. it's more efficient to create a constant for each initializer once rather than every time an initializer is accessed 4. it's simpler to look up a single symbol mapping Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add decompositions for v11 split, squeeze, and unsqueeze (onnx#1702) * Add decompositions for v11 split, squeeze, and unsqueeze Co-authored-by: Roberto DiCecco <[email protected]> Signed-off-by: Philip Lassen <[email protected]> * Fix lit test Signed-off-by: Philip Lassen <[email protected]> * Add lit tests Signed-off-by: Philip Lassen <[email protected]> * Delete unneccesary decomp for unsqueeze Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * factored input shaping into ModelInputShaper (onnx#1740) * factored input shaping into ModelInputShaper this makes the it easier to read and understand the input shaping logic and also makes the rest of the FrontendDialectTransformer implementation shorter and easier to read Signed-off-by: Soren Lassen <[email protected]> * documented public methods Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Ingest all of Opset 16 (onnx#1704) * Ingest al of opset 16 Signed-off-by: Philip Lassen <[email protected]> * Delete GridSample test which is no longer relevant Signed-off-by: Philip Lassen <[email protected]> * Use NullStringAttr insead of "none" string Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix onnx-mlir.py docker wrapper (onnx#1734) * Requiring input files to have known suffix (.onnx, .json, or .mlir) greatly simplifies the onnx-mlir.py docker wrapper. Signed-off-by: Gong Su <[email protected]> * Return exit code Signed-off-by: Gong Su <[email protected]> * Add copyright notice Signed-off-by: Gong Su <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Update protobuf (onnx#1747) * Update to protobuf 3.20.2 Signed-off-by: Charles Volzka <[email protected]> * Try 3.18.3 Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add the product version and fix formatting Signed-off-by: Megan Hampton <[email protected]> * Make updates to product version Signed-off-by: Megan Hampton <[email protected]> * further updates Signed-off-by: Megan Hampton <[email protected]> * Add product version text file Signed-off-by: Megan Hampton <[email protected]> * Removed the memory leaks directly in execution session (onnx#1746) * removed the memory leaks directly in execution session Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Add support for ArgMin (onnx#1737) * add support for ArgMin Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Inclusive terminology update (onnx#1750) * Update `master` reference to `main` in onnx-mlir doc * Update rapidcheck to newer commit with inclusive terminology changes * Update benchmark to 1.6.2 with inclusive terminology changes * Update pybind11 to 2.10 with inclusive terminology changes Signed-off-by: Megan Hampton <[email protected]> * Warning and instructions that --onnx-op-stats needs more than EmitONNXIR (onnx#1719) Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Move product version text file Signed-off-by: Megan Hampton <[email protected]> * Remove comment Signed-off-by: Megan Hampton <[email protected]> * Output the product version for the compiler/version Signed-off-by: Megan Hampton <[email protected]> * Cleaned up commit of the InferTypes change. (onnx#1753) Signed-off-by: Brad Messer <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * setup.py install is deprecated, use pip install instead (onnx#1752) * setup.py install is deprecated, use pip install instead (see https://blog.ganssle.io/articles/2021/10/setup-py-deprecated.html for more details). Signed-off-by: Gong Su <[email protected]> * - pip install doesn't work for protobuf, revert back to setup.py install - use --cpp_implementation for better performance Signed-off-by: Gong Su <[email protected]> * Update Windows CI Signed-off-by: Gong Su <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * adding reference to onnx-mlir-serving (onnx#1745) Signed-off-by: Alexandre Eichenberger <[email protected]> Co-authored-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * [TOSA] Update type converter and unary ops (onnx#1553) Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix syntax Signed-off-by: Megan Hampton <[email protected]> * Remove strip Signed-off-by: Megan Hampton <[email protected]> * Revert "Cleaned up commit of the InferTypes change. (onnx#1753)" This reverts commit d08c5ac. Signed-off-by: Megan Hampton <[email protected]> * Cleaned up commit of the InferTypes change. (onnx#1753) Signed-off-by: Brad Messer <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix for GatherND verifier (onnx#1754) Co-authored-by: Hadi Jooybar <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Fix OnnxBuilder::concat where axis was not used (onnx#1759) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Simplify GatherOp when its inputs are dimensions (onnx#1755) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Assign each question mark a unique negative integer value (onnx#1757) * Assign each question mark a unique negative integer value Signed-off-by: Tung D. Le <[email protected]> * Add a mutex to protect the counter Signed-off-by: Tung D. Le <[email protected]> * Using decrement Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Change the regex Signed-off-by: Megan Hampton <[email protected]> * Fix regex and display product numbers Signed-off-by: Megan Hampton <[email protected]> * normalize axis in ScatterElements verify (onnx#1760) Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * redo ArrayRefOrSmallVector with enable_if (onnx#1758) closes onnx#1729 * redo ArrayRefOrSmallVector with enable_if attempt to address issue onnx#1729: Build error on Linux Signed-off-by: Soren Lassen <[email protected]> * added @qedawkins fix: begin,end -> data,size Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Quinn Dawkins [email protected] Co-authored-by: Philip Lassen <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * fix syntax Signed-off-by: Megan Hampton <[email protected]> * Handle scalar tensor tensor<type> in SimplifyShapeRelatesOps pass (onnx#1764) Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * namespace cleanup (onnx#1763) * namespace cleanup removed `using namespace mlir` from ConstPropHelper.hpp and added requisite `mlir::` and `llvm::` in header files to get everything to compile again also put ConstPropHelper definitions in namespace onnx_mlir Signed-off-by: Soren Lassen <[email protected]> * added more missing `mlir/llvm::` Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Signed-off-by: Megan Hampton <[email protected]> * Signed-off-by: Megan Hampton <[email protected]> Fix Clang format error * Make use of the product vendor flag Signed-off-by: Megan Hampton <[email protected]> * Address feedback Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Gong Su <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: Megan Hampton <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Co-authored-by: Charles Volzka <[email protected]> Co-authored-by: Yasushi Negishi <[email protected]> Co-authored-by: chentong319 <[email protected]> Co-authored-by: Stella Stamenova <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Co-authored-by: gongsu832 <[email protected]> Co-authored-by: Meng, Hengyu <[email protected]> Co-authored-by: Brad Messer <[email protected]> Co-authored-by: Philipp Braun <[email protected]> Co-authored-by: Hadi Jooybar <[email protected]> * different initial Signed-off-by: chentong319 <[email protected]> Signed-off-by: chentong319 <[email protected]> Signed-off-by: Megan Hampton <[email protected]> Signed-off-by: Soren Lassen <[email protected]> Signed-off-by: Stella Stamenova <[email protected]> Signed-off-by: Tung D. Le <[email protected]> Signed-off-by: Philip Lassen <[email protected]> Signed-off-by: Gong Su <[email protected]> Signed-off-by: Charles Volzka <[email protected]> Signed-off-by: Hengyu Meng <[email protected]> Signed-off-by: Alexandre Eichenberger <[email protected]> Signed-off-by: Philipp Braun <[email protected]> Co-authored-by: hamptonm1 <[email protected]> Co-authored-by: Megan Hampton <[email protected]> Co-authored-by: Soren Lassen <[email protected]> Co-authored-by: Charles Volzka <[email protected]> Co-authored-by: Yasushi Negishi <[email protected]> Co-authored-by: Stella Stamenova <[email protected]> Co-authored-by: Tung D. Le <[email protected]> Co-authored-by: Alexandre Eichenberger <[email protected]> Co-authored-by: Philip Lassen <[email protected]> Co-authored-by: Roberto DiCecco <[email protected]> Co-authored-by: gongsu832 <[email protected]> Co-authored-by: Meng, Hengyu <[email protected]> Co-authored-by: Brad Messer <[email protected]> Co-authored-by: Philipp Braun <[email protected]> Co-authored-by: Hadi Jooybar <[email protected]>
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.
There were some issues with respect to memory leaks in the ExecussionSession interfaces that use unique pointers.
These leaks were addressed in this issue. Namely, the following were removed.
Now there remain two types of remaining memory leaks. First there is some memory allocated in the compiled model that is not freed.
As I believed it is shown as
==67421== by 0x5CCA559: ???... namely a call site in the code that we generated from the model for which the loader has to direct location as it comes from ONNX.The second leak is due to issues with the ONNX-MLIR compiler, as this model compiles and run the model. Up to now, the above leaks were related to running an inference. Now this is the compiler part of the program. While these should be eliminated, there are less of a concerns as we compile less often than perform inferences.
some leaks might be related to the handling of loops, others may be just other generic compiler stuff.
In summary: before this PR:
With fixes of this PR
A step in the right direction, and important as it target losses during the inference time.
Signed-off-by: Alexandre Eichenberger [email protected]