[Torch] Emit rrelu and decompose it (llvm#3250)

as title

include/torch-mlir/Dialect/Torch/IR/GeneratedTorchOps.td

@@ -256,6 +256,106 @@ def Torch_AtenLeakyRelu_Op : Torch_Op<"aten.leaky_relu_", [
   }];
 }
 
+def Torch_AtenRreluOp : Torch_Op<"aten.rrelu", [
+    AllowsTypeRefinement,
+    HasValueSemantics,
+    ReadOnly
+  ]> {
+  let summary = "Generated op for `aten::rrelu : (Tensor, Scalar, Scalar, bool, Generator?) -> (Tensor)`";
+  let arguments = (ins
+    AnyTorchTensorType:$self,
+    AnyTorchScalarType:$lower,
+    AnyTorchScalarType:$upper,
+    Torch_BoolType:$training,
+    AnyTorchOptionalGeneratorType:$generator
+  );
+  let results = (outs
+    AnyTorchOptionalTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenRreluOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 5, 1);
+    }
+    void AtenRreluOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 5, 1);
+    }
+  }];
+}
+
+def Torch_AtenRrelu_Op : Torch_Op<"aten.rrelu_", [
+    IsTrailingUnderscoreInplaceVariant,
+    AllowsTypeRefinement
+  ]> {
+  let summary = "Generated op for `aten::rrelu_ : (Tensor, Scalar, Scalar, bool, Generator?) -> (Tensor)`";
+  let arguments = (ins
+    Torch_NonValueTensorType:$self,
+    AnyTorchScalarType:$lower,
+    AnyTorchScalarType:$upper,
+    Torch_BoolType:$training,
+    AnyTorchOptionalGeneratorType:$generator
+  );
+  let results = (outs
+    AnyTorchOptionalNonValueTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenRrelu_Op::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 5, 1);
+    }
+    void AtenRrelu_Op::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 5, 1);
+    }
+  }];
+}
+
+def Torch_AtenCeluOp : Torch_Op<"aten.celu", [
+    AllowsTypeRefinement,
+    HasValueSemantics,
+    ReadOnly
+  ]> {
+  let summary = "Generated op for `aten::celu : (Tensor, Scalar) -> (Tensor)`";
+  let arguments = (ins
+    AnyTorchTensorType:$self,
+    AnyTorchScalarType:$alpha
+  );
+  let results = (outs
+    AnyTorchOptionalTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenCeluOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 2, 1);
+    }
+    void AtenCeluOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 2, 1);
+    }
+  }];
+}
+
+def Torch_AtenCelu_Op : Torch_Op<"aten.celu_", [
+    IsTrailingUnderscoreInplaceVariant,
+    AllowsTypeRefinement
+  ]> {
+  let summary = "Generated op for `aten::celu_ : (Tensor, Scalar) -> (Tensor)`";
+  let arguments = (ins
+    Torch_NonValueTensorType:$self,
+    AnyTorchScalarType:$alpha
+  );
+  let results = (outs
+    AnyTorchOptionalNonValueTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenCelu_Op::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 2, 1);
+    }
+    void AtenCelu_Op::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 2, 1);
+    }
+  }];
+}
+
 def Torch_AtenSeluOp : Torch_Op<"aten.selu", [
     AllowsTypeRefinement,
     HasValueSemantics,
@@ -4810,53 +4910,6 @@ def Torch_AtenPreluOp : Torch_Op<"aten.prelu", [
   }];
 }
 
-def Torch_AtenCeluOp : Torch_Op<"aten.celu", [
-    AllowsTypeRefinement,
-    HasValueSemantics,
-    ReadOnly
-  ]> {
-  let summary = "Generated op for `aten::celu : (Tensor, Scalar) -> (Tensor)`";
-  let arguments = (ins
-    AnyTorchTensorType:$self,
-    AnyTorchScalarType:$alpha
-  );
-  let results = (outs
-    AnyTorchOptionalTensorType:$result
-  );
-  let hasCustomAssemblyFormat = 1;
-  let extraClassDefinition = [{
-    ParseResult AtenCeluOp::parse(OpAsmParser &parser, OperationState &result) {
-      return parseDefaultTorchOp(parser, result, 2, 1);
-    }
-    void AtenCeluOp::print(OpAsmPrinter &printer) {
-      printDefaultTorchOp(printer, *this, 2, 1);
-    }
-  }];
-}
-
-def Torch_AtenCelu_Op : Torch_Op<"aten.celu_", [
-    IsTrailingUnderscoreInplaceVariant,
-    AllowsTypeRefinement
-  ]> {
-  let summary = "Generated op for `aten::celu_ : (Tensor, Scalar) -> (Tensor)`";
-  let arguments = (ins
-    Torch_NonValueTensorType:$self,
-    AnyTorchScalarType:$alpha
-  );
-  let results = (outs
-    AnyTorchOptionalNonValueTensorType:$result
-  );
-  let hasCustomAssemblyFormat = 1;
-  let extraClassDefinition = [{
-    ParseResult AtenCelu_Op::parse(OpAsmParser &parser, OperationState &result) {
-      return parseDefaultTorchOp(parser, result, 2, 1);
-    }
-    void AtenCelu_Op::print(OpAsmPrinter &printer) {
-      printDefaultTorchOp(printer, *this, 2, 1);
-    }
-  }];
-}
-
 def Torch_AtenRealOp : Torch_Op<"aten.real", [
     AllowsTypeRefinement,
     ReadOnly

lib/Dialect/Torch/Transforms/AbstractInterpLibrary.cpp

@@ -7074,6 +7074,10 @@ StringRef mlir::torch::Torch::getAbstractInterpLibrary() {
 "    %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>\n"
 "    return %0 : !torch.list<int>\n"
 "  }\n"
+"  func.func @\"__torch_mlir_shape_fn.aten.rrelu\"(%arg0: !torch.list<int>, %arg1: !torch.float, %arg2: !torch.float, %arg3: !torch.bool, %arg4: !torch.any) -> !torch.list<int> {\n"
+"    %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>\n"
+"    return %0 : !torch.list<int>\n"
+"  }\n"
 "  func.func @\"__torch_mlir_shape_fn.aten.selu\"(%arg0: !torch.list<int>) -> !torch.list<int> {\n"
 "    %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>\n"
 "    return %0 : !torch.list<int>\n"
@@ -10610,6 +10614,26 @@ StringRef mlir::torch::Torch::getAbstractInterpLibrary() {
 "    %0:2 = torch.prim.TupleUnpack %arg0 : !torch.tuple<int, int> -> !torch.int, !torch.int\n"
 "    return %0#1 : !torch.int\n"
 "  }\n"
+"  func.func @\"__torch_mlir_dtype_fn.aten.rrelu\"(%arg0: !torch.tuple<int, int>, %arg1: !torch.number, %arg2: !torch.number, %arg3: !torch.bool, %arg4: !torch.any) -> !torch.int {\n"
+"    %none = torch.constant.none\n"
+"    %str = torch.constant.str \"AssertionError: \"\n"
+"    %true = torch.constant.bool true\n"
+"    %0:2 = torch.prim.TupleUnpack %arg0 : !torch.tuple<int, int> -> !torch.int, !torch.int\n"
+"    %1 = call @__torch__.torch_mlir.jit_ir_importer.build_tools.library_generator.is_float_dtype(%0#1) : (!torch.int) -> !torch.bool\n"
+"    %2 = torch.prim.If %1 -> (!torch.bool) {\n"
+"      torch.prim.If.yield %true : !torch.bool\n"
+"    } else {\n"
+"      %3 = func.call @__torch__.torch_mlir.jit_ir_importer.build_tools.library_generator.is_complex_dtype(%0#1) : (!torch.int) -> !torch.bool\n"
+"      torch.prim.If.yield %3 : !torch.bool\n"
+"    }\n"
+"    torch.prim.If %2 -> () {\n"
+"      torch.prim.If.yield\n"
+"    } else {\n"
+"      torch.prim.RaiseException %str, %none : !torch.str, !torch.none\n"
+"      torch.prim.If.yield\n"
+"    }\n"
+"    return %0#1 : !torch.int\n"
+"  }\n"
 "  func.func @\"__torch_mlir_dtype_fn.aten.relu6\"(%arg0: !torch.tuple<int, int>) -> !torch.int {\n"
 "    %none = torch.constant.none\n"
 "    %str = torch.constant.str \"AssertionError: \"\n"

lib/Dialect/Torch/Transforms/DecomposeComplexOps.cpp

@@ -2520,6 +2520,77 @@ class DecomposeAtenPreluOp : public OpRewritePattern<AtenPreluOp> {
 
 } // namespace
 
+// rrelu = max(0, x) + min(0, alpha * x)
+// if in training mode, the alpha is sampled from uniform distribution (lower,
+// upper) if in testing mode, the alpha is (lower + upper) / 2
+namespace {
+class DecomposeAtenRreluOp : public OpRewritePattern<AtenRreluOp> {
+public:
+  using OpRewritePattern::OpRewritePattern;
+  LogicalResult matchAndRewrite(AtenRreluOp op,
+                                PatternRewriter &rewriter) const override {
+    Location loc = op.getLoc();
+    Value self = op.getSelf();
+    Value lower = op.getLower();
+    Value upper = op.getUpper();
+    auto resType = cast<BaseTensorType>(op.getType());
+    if (!resType.hasDtype()) {
+      return rewriter.notifyMatchFailure(op, "result should have dtype");
+    }
+
+    bool training;
+    if (!matchPattern(op.getTraining(), m_TorchConstantBool(&training))) {
+      return rewriter.notifyMatchFailure(op, "training should be a constant");
+    }
+
+    Value constantZeroFloat =
+        rewriter.create<ConstantFloatOp>(loc, rewriter.getF64FloatAttr(0.0));
+    Value constantOneFloat =
+        rewriter.create<ConstantFloatOp>(loc, rewriter.getF64FloatAttr(1.0));
+    Value constantTwoFloat =
+        rewriter.create<ConstantFloatOp>(loc, rewriter.getF64FloatAttr(2.0));
+
+    Value alpha;
+    if (training) {
+      // Create a uniform random op with low and high set to `lower` and
+      // `upper`, respectively.
+      Value none = rewriter.create<ConstantNoneOp>(loc);
+      Value emptyTensor = rewriter.create<AtenFullLikeOp>(
+          loc, resType, self, constantZeroFloat, /*dtype=*/none,
+          /*layout=*/none,
+          /*device=*/none, /*pin_memoty=*/none, /*memory_format=*/none);
+      alpha = rewriter.create<AtenUniformOp>(loc, resType, emptyTensor,
+                                             /*from=*/lower, /*to=*/upper,
+                                             /*generator=*/none);
+    } else {
+      Value half = rewriter.create<AtenAddOp>(loc, constantTwoFloat.getType(),
+                                              lower, upper);
+      alpha = rewriter.create<AtenDivOp>(loc, constantTwoFloat.getType(), half,
+                                         constantTwoFloat);
+    }
+
+    Value zeroTensor =
+        createRank0Tensor(rewriter, loc, resType, constantZeroFloat);
+    Value positiveOutput =
+        rewriter.create<AtenMaximumOp>(loc, resType, zeroTensor, self);
+
+    Value scaledSelf;
+    if (training) {
+      scaledSelf = rewriter.create<AtenMulTensorOp>(loc, resType, self, alpha);
+    } else {
+      scaledSelf = rewriter.create<AtenMulScalarOp>(loc, resType, self, alpha);
+    }
+
+    Value negativeOutput =
+        rewriter.create<AtenMinimumOp>(loc, resType, zeroTensor, scaledSelf);
+    Value rreluOutput = rewriter.create<AtenAddTensorOp>(
+        loc, resType, positiveOutput, negativeOutput, constantOneFloat);
+    rewriter.replaceOp(op, rreluOutput);
+    return success();
+  }
+};
+} // namespace
+
 // CELU(x)=max(0,x)+min(0,alpha∗(exp(x/alpha)−1))
 namespace {
 class DecomposeAtenCeluOp : public OpRewritePattern<AtenCeluOp> {
@@ -8065,6 +8136,7 @@ class DecomposeComplexOpsPass
     addPatternIfTargetOpIsIllegal<DecomposeAtenHardsigmoidOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenRelu6Op>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenPreluOp>(patterns);
+    addPatternIfTargetOpIsIllegal<DecomposeAtenRreluOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenCeluOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenEinsumOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenTraceOp>(patterns);

lib/Dialect/Torch/Transforms/LowerToBackendContract.cpp

@@ -483,6 +483,7 @@ static void markDecomposedOpsAsIllegal(MLIRContext *context,
   target.addIllegalOp<Aten_UnsafeIndexPutHackedTwinOp>();
   target.addIllegalOp<AtenPadOp>();
   target.addIllegalOp<AtenPreluOp>();
+  target.addIllegalOp<AtenRreluOp>();
   target.addIllegalOp<AtenCeluOp>();
   target.addIllegalOp<AtenToDtypeLayoutOp>();
   target.addIllegalOp<AtenToDeviceOp>();

projects/pt1/e2e_testing/xfail_sets.py

@@ -387,6 +387,10 @@
     "ElementwiseDequantizePerTensorModule_basic",
     "ElementwiseQuantizePerTensorModule_basic",
     "ElementwiseQuantizePerTensorUIntModule_basic",
+    "ElementwiseRreluEvalModule_basic",
+    "ElementwiseRreluEvalStaticModule_basic",
+    "ElementwiseRreluTrainModule_basic",
+    "ElementwiseRreluTrainStaticModule_basic",
     "ElementwiseToDtypeI64ToUI8Module_basic",
     "EqIntModule_basic",
     "FakeQuantizePerTensorAffineDynamicShapeModule_basic",
@@ -1014,6 +1018,8 @@
     "ElementwiseRemainderTensorModule_Float_basic",
     "ElementwiseRemainderTensorModule_Int_Float_basic",
     "ElementwiseRemainderTensorModule_Int_basic",
+    "ElementwiseRreluEvalStaticModule_basic",
+    "ElementwiseRreluTrainStaticModule_basic",
     "ElementwiseRsqrtModule_basic",
     "ElementwiseSigmoidModule_basic",
     "ElementwiseSinModule_basic",
@@ -1692,6 +1698,8 @@
     "ElementwiseRemainderScalarModule_Int_Float_basic",
     "ElementwiseRemainderScalarModule_Int_basic",
     "ElementwiseRemainderScalarModule_Int_basic",
+    "ElementwiseRreluEvalModule_basic",
+    "ElementwiseRreluEvalStaticModule_basic",
     "ElementwiseRsqrtModule_basic",
     "ElementwiseSeluModule_basic",
     "ElementwiseSigmoidModule_basic",
@@ -1978,6 +1986,9 @@
     "ElementwisePreluModule_basic",
     "ElementwisePreluStaticModule_basic",
     "ElementwiseLogSigmoidModule_basic",
+    # failed to legalize operation 'torch.aten.rrelu_with_noise'
+    "ElementwiseRreluEvalModule_basic",
+    "ElementwiseRreluEvalStaticModule_basic",
     # Shape Related failures
     "PrimListUnpackNumMismatchModule_basic",
     "ReshapeExpandModule_basic",

projects/pt1/python/torch_mlir/jit_ir_importer/build_tools/abstract_interp_lib_gen.py

@@ -555,6 +555,9 @@ def aten〇prelu〡shape(self: List[int], weight: List[int]) -> List[int]:
 def aten〇celu〡shape(self: List[int], alpha: float = 1.) -> List[int]:
     return upstream_shape_functions.unary(self)
 
+def aten〇rrelu〡shape(self: List[int], lower: float = 0.125, upper: float = 0.33333333333333331, training: bool = False, generator: Any = None) -> List[int]:
+    return upstream_shape_functions.unary(self)
+
 def aten〇selu〡shape(self: List[int]) -> List[int]:
     return upstream_shape_functions.unary(self)
 
@@ -2723,6 +2726,12 @@ def aten〇celu〡dtype(self_rank_dtype: Tuple[int, int], alpha: Union[int, floa
     self_rank, self_dtype = self_rank_dtype
     return self_dtype
 
+@check_dtype_function(_check_tensors_with_the_same_dtype(num_of_tensors=1, error_types={torch.bool, *all_integer_dtypes()}))
+def aten〇rrelu〡dtype(self_rank_dtype: Tuple[int, int], lower: Union[int, float, complex] = 0.125, upper: Union[int, float, complex] = 0.33333333333333331, training: bool = False, generator: Any = None) -> int:
+    self_rank, self_dtype = self_rank_dtype
+    assert is_float_dtype(self_dtype) or is_complex_dtype(self_dtype)
+    return self_dtype
+
 @check_dtype_function(_check_tensors_with_the_same_dtype(num_of_tensors=1, error_types={torch.bool}))
 def aten〇relu6〡dtype(self_rank_dtype: Tuple[int, int]) -> int:
     self_rank, self_dtype = self_rank_dtype

projects/pt1/python/torch_mlir/jit_ir_importer/build_tools/torch_ods_gen.py

@@ -301,6 +301,8 @@ def emit_with_mutating_variants(key, **kwargs):
         "aten::relu : (Tensor) -> (Tensor)",
         "aten::relu6 : (Tensor) -> (Tensor)",
         "aten::leaky_relu : (Tensor, Scalar) -> (Tensor)",
+        "aten::rrelu : (Tensor, Scalar, Scalar, bool, Generator?) -> (Tensor)",
+        "aten::celu : (Tensor, Scalar) -> (Tensor)",
         "aten::selu : (Tensor) -> (Tensor)",
         "aten::sigmoid : (Tensor) -> (Tensor)",
         "aten::sinh : (Tensor) -> (Tensor)",
@@ -472,7 +474,6 @@ def emit_with_mutating_variants(key, **kwargs):
     emit("aten::floor_divide : (Tensor, Tensor) -> (Tensor)")
     emit("aten::softplus : (Tensor, Scalar, Scalar) -> (Tensor)")
     emit("aten::prelu : (Tensor, Tensor) -> (Tensor)")
-    emit_with_mutating_variants("aten::celu : (Tensor, Scalar) -> (Tensor)")
     emit("aten::real : (Tensor) -> (Tensor)")
     emit("aten::imag : (Tensor) -> (Tensor)")
     emit("aten::view_as_complex : (Tensor) -> (Tensor)")