colin
/
onnx-mlir
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
onnx-mlir/src/Builder/OpBuildTable.inc

319 lines
21 KiB
PHP
Raw Normal View History

clean up gen_doc.py (#59) Co-authored-by: Gheorghe-Teodor Bercea <gt.bercea@gmail.com>
2020-01-30 02:54:46 +08:00
//********************************************************
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
// Do not modify this file directly.
// This file is automatically generated via script.
Setup Documentation Page using Github Pages (#76) * Setup documentation server, move doc files from /doc to /docs as per Github Pages convention. * Include deleted files in patch. * /doc -> /docs * /doc -> /docs * Update documentation on importing ONNX spec into ONNX Dialect; provide documentation on how to add new documentation pages.
2020-04-09 23:37:04 +08:00
// Details can be found in docs/readonnxdefs.md .
clean up gen_doc.py (#59) Co-authored-by: Gheorghe-Teodor Bercea <gt.bercea@gmail.com>
2020-01-30 02:54:46 +08:00
//********************************************************
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Abs")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXAbsOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Acos")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXAcosOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Acosh")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXAcoshOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Add")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXAddOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "And")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXAndOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ArgMax")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXArgMaxOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ArgMin")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXArgMinOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Asin")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXAsinOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Asinh")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXAsinhOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Atan")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXAtanOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Atanh")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXAtanhOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "AveragePool")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXAveragePoolOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "BatchNormalization")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
ImportNodeBatchNormalization(node, /* expected_num_operands = */ 5, /* expected_num_results = */ 5);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "BitShift")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXBitShiftOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Cast")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXCastOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Ceil")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXCeilOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Clip")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXClipOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Compress")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXCompressOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Concat")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXConcatOp>(node, /* expected_num_operands = */ -1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ConcatFromSequence")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXConcatFromSequenceOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Constant")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXConstantOp>(node, /* expected_num_operands = */ 0, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ConstantOfShape")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXConstantOfShapeOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Conv")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXConvOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ConvInteger")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXConvIntegerOp>(node, /* expected_num_operands = */ 4, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ConvTranspose")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXConvTransposeOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Cos")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXCosOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Cosh")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXCoshOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "CumSum")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXCumSumOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "DepthToSpace")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXDepthToSpaceOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "DequantizeLinear")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXDequantizeLinearOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Det")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXDetOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Div")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXDivOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Dropout")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXDropoutOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 2);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "DynamicQuantizeLinear")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXDynamicQuantizeLinearOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 3);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Elu")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXEluOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Equal")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXEqualOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Erf")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXErfOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Exp")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXExpOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Expand")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXExpandOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "EyeLike")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXEyeLikeOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Flatten")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXFlattenOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Floor")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXFloorOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "GRU")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXGRUOp>(node, /* expected_num_operands = */ 6, /* expected_num_results = */ 2);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Gather")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXGatherOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "GatherElements")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXGatherElementsOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "GatherND")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXGatherNDOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Gemm")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXGemmOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "GlobalAveragePool")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXGlobalAveragePoolOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "GlobalLpPool")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXGlobalLpPoolOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "GlobalMaxPool")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXGlobalMaxPoolOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Greater")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXGreaterOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "HardSigmoid")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXHardSigmoidOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Hardmax")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXHardmaxOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Identity")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXIdentityOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "If")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXIfOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ -1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "InstanceNormalization")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXInstanceNormalizationOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "IsInf")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXIsInfOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "IsNaN")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXIsNaNOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "LRN")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXLRNOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "LSTM")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXLSTMOp>(node, /* expected_num_operands = */ 8, /* expected_num_results = */ 3);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "LeakyRelu")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXLeakyReluOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Less")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXLessOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Log")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXLogOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "LogSoftmax")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXLogSoftmaxOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Loop")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXLoopOp>(node);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "LpNormalization")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXLpNormalizationOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "LpPool")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXLpPoolOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "MatMul")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXMatMulOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "MatMulInteger")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXMatMulIntegerOp>(node, /* expected_num_operands = */ 4, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Max")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXMaxOp>(node, /* expected_num_operands = */ -1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "MaxPool")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
ImportNodeMaxPool(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 2);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "MaxRoiPool")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXMaxRoiPoolOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "MaxUnpool")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXMaxUnpoolOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Mean")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXMeanOp>(node, /* expected_num_operands = */ -1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "MeanVarianceNormalization")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXMeanVarianceNormalizationOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Min")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXMinOp>(node, /* expected_num_operands = */ -1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Mod")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXModOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Mul")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXMulOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Multinomial")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXMultinomialOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Neg")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXNegOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "NonMaxSuppression")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXNonMaxSuppressionOp>(node, /* expected_num_operands = */ 5, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "NonZero")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXNonZeroOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Not")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXNotOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "OneHot")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXOneHotOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Or")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXOrOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "PRelu")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXPReluOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Pad")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
ImportNodePad(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Pow")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXPowOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "QLinearConv")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXQLinearConvOp>(node, /* expected_num_operands = */ 9, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "QLinearMatMul")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXQLinearMatMulOp>(node, /* expected_num_operands = */ 8, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "QuantizeLinear")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXQuantizeLinearOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "RNN")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXRNNOp>(node, /* expected_num_operands = */ 6, /* expected_num_results = */ 2);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "RandomNormal")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXRandomNormalOp>(node, /* expected_num_operands = */ 0, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "RandomNormalLike")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXRandomNormalLikeOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "RandomUniform")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXRandomUniformOp>(node, /* expected_num_operands = */ 0, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "RandomUniformLike")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXRandomUniformLikeOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Range")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXRangeOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Reciprocal")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXReciprocalOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ReduceL1")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXReduceL1Op>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ReduceL2")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXReduceL2Op>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ReduceLogSum")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXReduceLogSumOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ReduceLogSumExp")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXReduceLogSumExpOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ReduceMax")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXReduceMaxOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ReduceMean")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXReduceMeanOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ReduceMin")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXReduceMinOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ReduceProd")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXReduceProdOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ReduceSum")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXReduceSumOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ReduceSumSquare")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXReduceSumSquareOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Relu")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXReluOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Reshape")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
ImportNodeReshape(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Resize")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXResizeOp>(node, /* expected_num_operands = */ 4, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ReverseSequence")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXReverseSequenceOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "RoiAlign")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXRoiAlignOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Round")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXRoundOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Scan")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXScanOp>(node);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Scatter")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXScatterOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ScatterElements")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXScatterElementsOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ScatterND")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXScatterNDOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Selu")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSeluOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "SequenceAt")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSequenceAtOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "SequenceConstruct")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSequenceConstructOp>(node, /* expected_num_operands = */ -1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "SequenceEmpty")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSequenceEmptyOp>(node, /* expected_num_operands = */ 0, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "SequenceErase")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSequenceEraseOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "SequenceInsert")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSequenceInsertOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "SequenceLength")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSequenceLengthOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Shape")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXShapeOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Shrink")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXShrinkOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Sigmoid")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSigmoidOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Sign")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSignOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Sin")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSinOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Sinh")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSinhOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Size")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSizeOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Slice")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSliceOp>(node, /* expected_num_operands = */ 5, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Softmax")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSoftmaxOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Softplus")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSoftplusOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Softsign")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSoftsignOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "SpaceToDepth")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSpaceToDepthOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Split")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSplitOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ -1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "SplitToSequence")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSplitToSequenceOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Sqrt")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSqrtOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Squeeze")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSqueezeOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "StringNormalizer")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXStringNormalizerOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Sub")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSubOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Sum")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXSumOp>(node, /* expected_num_operands = */ -1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Tan")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXTanOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Tanh")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXTanhOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "TfIdfVectorizer")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXTfIdfVectorizerOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "ThresholdedRelu")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXThresholdedReluOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Tile")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXTileOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "TopK")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXTopKOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 2);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Transpose")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXTransposeOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Unique")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXUniqueOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 4);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Unsqueeze")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXUnsqueezeOp>(node, /* expected_num_operands = */ 1, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Upsample")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXUpsampleOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Where")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXWhereOp>(node, /* expected_num_operands = */ 3, /* expected_num_results = */ 1);
Support Optional Inputs (#94) * 1. Combine variadicIn/Out with expectedNumOperands/Results to simplify import function arguments. 2. Generic improvements to code readability in gen_doc.py. * Update ONNX Dialect doc. * Remove redundant code in ImportNode. * Prettify op_build_table.inc. * 1. Remove irrelevant code in gen_doc.py * Refactor code to be more readable. * Further refactoring for readability improvements. * Allow gemm to have an optional operand (bias term), and include an example of declarative optimization pattern targeting gemm with bias term ommitted. * Make shape inference/lowering of gemm op compatible with optional operand declaration. * Apply canonicalization again after lowering from onnx -> std dialects. * Make hasBias compatible with the situation of GemmNoBias op. * Update doc. * Add a canonicalization test. * Remove special handler for importing Gemm op, as it's redundant now.
2020-02-24 23:46:48 +08:00
if (opName == "Xor")
Implement PadOp based on attribute promotion (#71) * enable promote attr for pad * use optional arguments for pad * shape infereance for pad * Lowering Pad * format file * use DenseTensor for the attribute * use Pad in ONNXRewrite * fix the merge conflict * fix the attr given to constantOp * handle ONNXConstantOp in attribute promotion * Fix bug when AttributePromotion is called more than once * update ONNXOps.td.inc with correct version of onnx * update onnx.md * responses to review * fix the build error * change the implementation of Pad * delete commented out code * clang format Co-authored-by: Tian Jin <tjingrant@gmail.com>
2020-05-15 13:19:28 +08:00
buildOperation<mlir::ONNXXorOp>(node, /* expected_num_operands = */ 2, /* expected_num_results = */ 1);