- Sort Score
- Result 10 results
- Languages All
Results 1 - 10 of 23 for Hinsu (0.06 sec)
-
tensorflow/compiler/mlir/tf2xla/transforms/legalize_tf_patterns.td
// Relu op patterns. //===----------------------------------------------------------------------===// // TODO(hinsu): Make these patterns to TF to TF lowering. Relu6 lowering will // require HLO canonicalization of min and max on a tensor to ClampOp. // TODO(hinsu): Lower quantized types after supporting them in GetScalarOfType. def : Pat<(TF_ReluOp AnyTensor:$input), (CHLO_BroadcastMaxOp
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Mon May 06 18:46:23 UTC 2024 - 34.8K bytes - Viewed (0) -
tensorflow/compiler/mlir/tensorflow/transforms/prepare_tpu_computation_for_tf_export.cc
if (auto sharding = func.getArgAttrOfType<mlir::StringAttr>(i, kShardingAttr)) { if (!sharding.getValue().empty()) { BlockArgument arg = func.getArgument(i); // TODO(hinsu): Instead of setting both 'sharding' and '_XlaSharding' // attributes, only set the 'sharding' attribute. Both attributes are // currently required as the XlaSharding xla op kernel doesn't use the
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Apr 25 16:01:03 UTC 2024 - 11.8K bytes - Viewed (0) -
tensorflow/compiler/mlir/tensorflow/transforms/constant_fold.cc
// to empty elements attribute. This is restricted to the numerical element // types as the DenseElementsAttr only supports numerical and string types. // TODO(hinsu): Handle ops that have one of the results empty for constant // propagation. bool has_empty_numerical_results = llvm::all_of(inst->getResultTypes(), [](Type ty) {
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Apr 25 16:01:03 UTC 2024 - 5.3K bytes - Viewed (0) -
tensorflow/compiler/mlir/tf2xla/transforms/passes.h
/// list. void PopulateLegalizeTfPatterns(MLIRContext* context, RewritePatternSet* patterns); // Populates TF to MHLO legalization for some of the quantization ops. // // TODO(hinsu): Remove this once we combine quantized and non quantized op // legalization in the ODML conversion pipeline. void PopulateLegalizeTfQuantizationPatterns(MLIRContext* context,
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Tue May 28 21:49:50 UTC 2024 - 4.8K bytes - Viewed (0) -
tensorflow/compiler/mlir/tensorflow/transforms/lower_tf.td
// // Computation of the reduction axis for the Sum op depends on whether the // input is a scalar or not. Restrict pattern to ranked inputs so that input to // the Sum op is also ranked. // TODO(hinsu): Support scalar inputs by introducing reshape to 1D. def NonScalarType : Type<Neg<HasAnyRankOfPred<[0]>>, "Non scalar type">; def LowerSoftmaxCrossEntropyWithLogitsOp : Pattern<
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Tue Jun 04 13:30:42 UTC 2024 - 24.7K bytes - Viewed (0) -
tensorflow/compiler/mlir/lite/transforms/optimize_functional_ops.cc
return success(); } // Extract the constant cond value. DenseElementsAttr cond; if (!matchPattern(op.getCond(), m_Constant(&cond))) return failure(); // TODO(hinsu): Handle constants that are not scalar booleans. auto cond_type = mlir::dyn_cast<RankedTensorType>(cond.getType()); if (!cond_type || !cond_type.getShape().equals({}) ||
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Apr 25 16:01:03 UTC 2024 - 6.6K bytes - Viewed (0) -
tensorflow/compiler/mlir/tf2xla/transforms/legalization_op_config.cc
return *ops; } bool IsOpTypeAllowedTf2XlaFallback(const TypeID& type_id) { // Allowlisted TensorFlow ops are known to have well behaved tf2xla kernels // building valid MLIR using MlirHloBuilder. // TODO(hinsu): Drop explicit allowlist when MLIR based bridge is enabled for // all tf2xla kernels. // Use a pointer for the static set, so the set is not destructed upon thread // end, which would not be thread safe.
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Wed Apr 24 04:08:35 UTC 2024 - 21.7K bytes - Viewed (0) -
tensorflow/compiler/mlir/lite/converter_gen.cc
os << " return b.Finish();\n}\n"; } } // For each TFLite op, emits a builder function that packs the TFLite op into // the corresponding FlatBuffer object. // // TODO(hinsu): Revisit if only builtin_options and mutating_variable_inputs // arguments that depend on op definitions should be auto-generated and then // operator should be built by the caller because it does not require // auto-generation.
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Tue Dec 19 15:05:28 UTC 2023 - 23.7K bytes - Viewed (0) -
tensorflow/compiler/mlir/tf2xla/api/v1/compile_mlir_util.cc
pm.addNestedPass<mlir::func::FuncOp>(mlir::createCanonicalizerPass()); // The SCCP pass performs constant propagation across the IR, which, for // example, propagates constant arguments into callee functions. // TODO(hinsu): Investigate if we really need SCCP pass before shape inference // and can do with just one pass after the shape inference. pm.addPass(mlir::createSCCPPass());
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Tue May 21 17:24:39 UTC 2024 - 45.3K bytes - Viewed (0) -
tensorflow/compiler/mlir/tensorflow/utils/convert_tensor.cc
case DT_FLOAT8_E4M3FN: return ConvertTensorOfCustomFloatType(input_tensor, type); case DT_STRING: return ConvertStringTensor(input_tensor, type); default: // TODO(hinsu): Remove mangling now that there is a special attribute. return ElementsAttr( mlir::TF::TensorProtoAttr::get(type, MangleTensor(input_tensor))); } #undef CONVERT_FLAT }
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Fri Apr 26 09:37:10 UTC 2024 - 20.5K bytes - Viewed (0)