Add lowering to linalg for softplus and log1p

Follows existing conventions for unary operators.

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

@@ -2069,6 +2069,51 @@ def Torch_AtenSqrt_Op : Torch_Op<"aten.sqrt_", [
   }];
 }
 
+def Torch_AtenLog1pOp : Torch_Op<"aten.log1p", [
+    AllowsTypeRefinement,
+    HasValueSemantics,
+    ReadOnly
+  ]> {
+  let summary = "Generated op for `aten::log1p : (Tensor) -> (Tensor)`";
+  let arguments = (ins
+    AnyTorchTensorType:$self
+  );
+  let results = (outs
+    AnyTorchTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenLog1pOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 1, 1);
+    }
+    void AtenLog1pOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 1, 1);
+    }
+  }];
+}
+
+def Torch_AtenLog1p_Op : Torch_Op<"aten.log1p_", [
+    IsTrailingUnderscoreInplaceVariant,
+    AllowsTypeRefinement
+  ]> {
+  let summary = "Generated op for `aten::log1p_ : (Tensor) -> (Tensor)`";
+  let arguments = (ins
+    AnyTorchTensorType:$self
+  );
+  let results = (outs
+    AnyTorchTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenLog1p_Op::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 1, 1);
+    }
+    void AtenLog1p_Op::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 1, 1);
+    }
+  }];
+}
+
 def Torch_AtenRsqrtOp : Torch_Op<"aten.rsqrt", [
     AllowsTypeRefinement,
     HasValueSemantics,
@@ -2708,6 +2753,31 @@ def Torch_AtenFloorDivideOp : Torch_Op<"aten.floor_divide", [
   }];
 }
 
+def Torch_AtenSoftplusOp : Torch_Op<"aten.softplus", [
+    AllowsTypeRefinement,
+    HasValueSemantics,
+    ReadOnly
+  ]> {
+  let summary = "Generated op for `aten::softplus : (Tensor, Scalar, Scalar) -> (Tensor)`";
+  let arguments = (ins
+    AnyTorchTensorType:$self,
+    AnyTorchScalarType:$beta,
+    AnyTorchScalarType:$threshold
+  );
+  let results = (outs
+    AnyTorchTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenSoftplusOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 3, 1);
+    }
+    void AtenSoftplusOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 3, 1);
+    }
+  }];
+}
+
 def Torch_AtenFill_ScalarOp : Torch_Op<"aten.fill_.Scalar", [
     AllowsTypeRefinement
   ]> {

lib/Conversion/TorchToLinalg/Uncategorized.cpp

@@ -139,6 +139,10 @@ static Value createLinalgPayloadCalculationForElementwiseOp(
     return createCalculationForMathOpWithDtypeConversion<math::Log2Op>(
         b, converter, payloadArgs[0], op);
   }
+  if (isa<AtenLog1pOp>(op)) {
+    return createCalculationForMathOpWithDtypeConversion<math::Log1pOp>(
+        b, converter, payloadArgs[0], op);
+  }
   if (isa<AtenErfOp>(op)) {
     return createCalculationForMathOpWithDtypeConversion<math::ErfOp>(
         b, converter, payloadArgs[0], op);
@@ -922,14 +926,15 @@ public:
              AtenLerpTensorOp, AtenSigmoidOp, AtenExpOp, AtenMinimumOp,
              AtenMaximumOp, AtenToDtypeOp, AtenClampOp, AtenRsubScalarOp,
              AtenMulScalarOp, AtenLogOp, AtenErfOp, AtenSqrtOp, AtenFloorOp,
-             AtenPowTensorScalarOp, AtenLog2Op, AtenRsqrtOp, AtenDivScalarOp,
-             AtenAbsOp, AtenReciprocalOp, AtenBitwiseAndTensorOp,
-             AtenGtScalarOp, AtenGeScalarOp, AtenEqScalarOp, AtenLtScalarOp,
-             AtenLeScalarOp, AtenWhereSelfOp, AtenCeilOp, AtenGtTensorOp,
-             AtenEqTensorOp, AtenLtTensorOp, AtenSubScalarOp, AtenAddScalarOp,
-             AtenThresholdOp, AtenThresholdBackwardOp, AtenCloneOp, AtenSinOp,
-             AtenCosOp, AtenNeScalarOp, AtenNegOp, AtenMaskedFillScalarOp,
-             AtenLogicalOrOp, AtenTriuOp>(op))
+             AtenPowTensorScalarOp, AtenLog2Op, AtenLog1pOp, AtenRsqrtOp,
+             AtenDivScalarOp, AtenAbsOp, AtenReciprocalOp,
+             AtenBitwiseAndTensorOp, AtenGtScalarOp, AtenGeScalarOp,
+             AtenEqScalarOp, AtenLtScalarOp, AtenLeScalarOp, AtenWhereSelfOp,
+             AtenCeilOp, AtenGtTensorOp, AtenEqTensorOp, AtenLtTensorOp,
+             AtenSubScalarOp, AtenAddScalarOp, AtenThresholdOp,
+             AtenThresholdBackwardOp, AtenCloneOp, AtenSinOp, AtenCosOp,
+             AtenNeScalarOp, AtenNegOp, AtenMaskedFillScalarOp, AtenLogicalOrOp,
+             AtenTriuOp>(op))
       return rewriter.notifyMatchFailure(op, "not a supported elementwise op");
 
     if (failed(verifyLinalgCompatibleTypes(op, rewriter)))
@@ -1662,7 +1667,7 @@ void mlir::torch::torch_to_linalg::populateUncategorizedPatternsAndLegality(
       AtenSubTensorOp, AtenLerpTensorOp, AtenSigmoidOp, AtenMinimumOp,
       AtenMaximumOp, AtenToDtypeOp, AtenClampOp, AtenRsubScalarOp, AtenLogOp,
       AtenErfOp, AtenSqrtOp, AtenFloorOp, AtenCeilOp, AtenPowTensorScalarOp,
-      AtenLog2Op, AtenRsqrtOp, AtenAbsOp, AtenReciprocalOp,
+      AtenLog2Op, AtenLog1pOp, AtenRsqrtOp, AtenAbsOp, AtenReciprocalOp,
       AtenBitwiseAndTensorOp, AtenGtScalarOp, AtenGeScalarOp, AtenEqScalarOp,
       AtenLtScalarOp, AtenLeScalarOp, AtenWhereSelfOp, AtenGtTensorOp,
       AtenEqTensorOp, AtenLtTensorOp, AtenThresholdOp, AtenThresholdBackwardOp,

lib/Dialect/Torch/Transforms/DecomposeComplexOps.cpp

@@ -1172,6 +1172,40 @@ public:
 };
 } // namespace
 
+// Softplus(x, beta, threshold) =
+//   x * beta > threshold ? x : log(1 + exp(x * beta)) / beta
+namespace {
+class DecomposeAtenSoftplusOp : public OpRewritePattern<AtenSoftplusOp> {
+public:
+  using OpRewritePattern::OpRewritePattern;
+  LogicalResult matchAndRewrite(AtenSoftplusOp op,
+                                PatternRewriter &rewriter) const override {
+    Location loc = op.getLoc();
+    Value input = op.self();
+    BaseTensorType inputType = input.getType().cast<BaseTensorType>();
+
+    Value inputTimesBeta = rewriter.create<AtenMulScalarOp>(
+        loc, inputType, input, op.beta());
+
+    // out = log1p(exp(input * beta)) / beta
+    Value exp = rewriter.create<AtenExpOp>(loc, inputType, inputTimesBeta);
+    Value log1p = rewriter.create<AtenLog1pOp>(loc, inputType, exp);
+    Value out = rewriter.create<AtenDivScalarOp>(
+        loc, inputType, log1p, op.beta());
+
+    // Select where x * beta > threshold
+    auto boolResType = inputType.getWithSizesAndDtype(inputType.getSizes(),
+                                                      rewriter.getI1Type());
+    Value condition = rewriter.create<AtenGtScalarOp>(
+        loc, boolResType, inputTimesBeta, op.threshold());
+
+    rewriter.replaceOpWithNewOp<AtenWhereSelfOp>(
+        op, op.getType(), condition, input, out);
+    return success();
+  }
+};
+} // namespace
+
 // Hardsigmoid(x) = max(0, min(1, (x+3)/6))
 namespace {
 class DecomposeAtenHardsigmoidOp : public OpRewritePattern<AtenHardsigmoidOp> {
@@ -2344,6 +2378,8 @@ class DecomposeComplexOpsPass
     target.addIllegalOp<AtenHardsigmoidOp>();
     patterns.add<DecomposeAtenHardswishOp>(context);
     target.addIllegalOp<AtenHardswishOp>();
+    patterns.add<DecomposeAtenSoftplusOp>(context);
+    target.addIllegalOp<AtenSoftplusOp>();
     patterns.add<DecomposeAtenSiluOp>(context);
     target.addIllegalOp<AtenSiluOp>();
     patterns.add<DecomposeConstantTensorNewLikeOp<AtenNewZerosOp, AtenZerosOp>>(

lib/Dialect/Torch/Transforms/RefineTypes.cpp

@@ -664,8 +664,8 @@ void TypeAnalysis::visitOperation(Operation *op,
 
   // Dtype is always float32, except for bfloat16, float64 and nullptr.
   if (isa<AtenTanhOp, AtenExpOp, AtenSinOp, AtenCosOp, AtenSigmoidOp,
-          AtenReciprocalOp, AtenLogOp, AtenSqrtOp, AtenLog2Op, AtenRsqrtOp,
-          AtenErfOp>(op)) {
+          AtenReciprocalOp, AtenLogOp, AtenSqrtOp, AtenLog2Op, AtenLog1pOp,
+          AtenRsqrtOp, AtenErfOp, AtenSoftplusOp>(op)) {
     ValueKnowledge knowledge =
         ValueKnowledge::getTensorPessimisticValueState(op->getContext());
     Type dtype = operands[0]->getValue().dtype;

lib/Dialect/Torch/Transforms/ShapeLibrary.cpp

@@ -5317,6 +5317,10 @@ module {
     %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>
     return %0 : !torch.list<int>
   }
+  func.func @"__torch_mlir_shape_fn.aten.softplus"(%arg0: !torch.list<int>, %arg1: !torch.float, %arg2: !torch.float) -> !torch.list<int> {
+    %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>
+    return %0 : !torch.list<int>
+  }
   func.func @"__torch_mlir_shape_fn.aten.square"(%arg0: !torch.list<int>) -> !torch.list<int> {
     %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>
     return %0 : !torch.list<int>
@@ -5365,6 +5369,10 @@ module {
     %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>
     return %0 : !torch.list<int>
   }
+  func.func @"__torch_mlir_shape_fn.aten.log1p"(%arg0: !torch.list<int>) -> !torch.list<int> {
+    %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>
+    return %0 : !torch.list<int>
+  }
   func.func @"__torch_mlir_shape_fn.aten.rsqrt"(%arg0: !torch.list<int>) -> !torch.list<int> {
     %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>
     return %0 : !torch.list<int>

python/torch_mlir/dialects/torch/importer/jit_ir/build_tools/shape_lib_gen.py

@@ -300,6 +300,9 @@ def aten〇sigmoid(self: List[int]) -> List[int]:
 def aten〇hardsigmoid(self: List[int]) -> List[int]:
     return upstream_shape_functions.unary(self)
 
+def aten〇softplus(self: List[int], beta: float = 1, threshold: float = 20) -> List[int]:
+    return upstream_shape_functions.unary(self)
+
 def aten〇square(self: List[int]) -> List[int]:
     return upstream_shape_functions.unary(self)
 
@@ -336,6 +339,9 @@ def aten〇detach(self: List[int]) -> List[int]:
 def aten〇log2(self: List[int]) -> List[int]:
     return upstream_shape_functions.unary(self)
 
+def aten〇log1p(self: List[int]) -> List[int]:
+    return upstream_shape_functions.unary(self)
+
 def aten〇rsqrt(self: List[int]) -> List[int]:
     return upstream_shape_functions.unary(self)
 

python/torch_mlir/dialects/torch/importer/jit_ir/build_tools/torch_ods_gen.py

@@ -282,6 +282,7 @@ def emit_ops(emitter_td: TextEmitter, registry: Registry):
             "aten::clamp_max : (Tensor, Scalar) -> (Tensor)",
             "aten::log2 : (Tensor) -> (Tensor)",
             "aten::sqrt : (Tensor) -> (Tensor)",
+            "aten::log1p : (Tensor) -> (Tensor)",
             "aten::rsqrt : (Tensor) -> (Tensor)",
             "aten::abs : (Tensor) -> (Tensor)",
             "aten::reciprocal : (Tensor) -> (Tensor)",
@@ -304,6 +305,7 @@ def emit_ops(emitter_td: TextEmitter, registry: Registry):
     emit("aten::pow.Tensor_Scalar : (Tensor, Scalar) -> (Tensor)")
     emit("aten::threshold_backward : (Tensor, Tensor, Scalar) -> (Tensor)")
     emit("aten::floor_divide : (Tensor, Tensor) -> (Tensor)")
+    emit("aten::softplus : (Tensor, Scalar, Scalar) -> (Tensor)")
 
     # Ops without value semantics but the corresponding without trailing
     # underscore variant doesn't exist.

python/torch_mlir_e2e_test/test_suite/basic.py

@@ -959,6 +959,28 @@ def _LogSoftmaxModuleStable_basic(module, tu: TestUtils):
 # ==============================================================================
 
 
+class SoftplusModule(torch.nn.Module):
+
+    def __init__(self):
+        super().__init__()
+
+    @export
+    @annotate_args([
+        None,
+        ([-1, -1], torch.float32, True),
+    ])
+    def forward(self, x):
+        return torch.ops.aten.softplus(x)
+
+
+@register_test_case(module_factory=lambda: SoftplusModule())
+def SoftplusModule_basic(module, tu: TestUtils):
+    module.forward(tu.rand(3, 3))
+
+
+# ==============================================================================
+
+
 class HardsigmoidModule(torch.nn.Module):
 
     def __init__(self):

python/torch_mlir_e2e_test/test_suite/elementwise.py

@@ -844,6 +844,27 @@ class ElementwiseLogIntModule(torch.nn.Module):
 def ElementwiseLogIntModule_basic(module, tu: TestUtils):
     module.forward(torch.randint(1, 10, (3, 4), dtype=torch.int32))
 
+# ==============================================================================
+
+
+class ElementwiseLog1pModule(torch.nn.Module):
+
+    def __init__(self):
+        super().__init__()
+
+    @export
+    @annotate_args([
+        None,
+        ([-1, -1], torch.float32, True),
+    ])
+    def forward(self, a):
+        return torch.log1p(a)
+
+
+@register_test_case(module_factory=lambda: ElementwiseLog1pModule())
+def ElementwiseLog1pModule_basic(module, tu: TestUtils):
+    module.forward(tu.rand(3, 4))
+
 
 # ==============================================================================
 

test/Dialect/Torch/decompose-complex-ops.mlir

@@ -1189,3 +1189,22 @@ func.func @torch.aten.var.dim(%arg0: !torch.vtensor<[3,4,7],f32>) -> !torch.vten
   %0 = torch.aten.var.dim %arg0, %dims, %unbiased, %keepdim: !torch.vtensor<[3,4,7],f32>, !torch.list<int>, !torch.bool, !torch.bool -> !torch.vtensor<[3,4,1],f32>
   return %0 : !torch.vtensor<[3,4,1],f32>
 }
+
+// -----
+// CHECK-LABEL:   func.func @torch.aten.softplus(
+// CHECK-SAME:                                   %[[VAL_0:.*]]: !torch.tensor<[2,3],f32>,
+// CHECK-SAME:                                   %[[VAL_1:.*]]: !torch.int) -> !torch.tensor<[2,3],f32> {
+// CHECK:           %[[VAL_2:.*]] = torch.constant.int 0
+// CHECK:           %[[VAL_3:.*]] = torch.aten.mul.Scalar %[[VAL_0]], %[[VAL_1]] : !torch.tensor<[2,3],f32>, !torch.int -> !torch.tensor<[2,3],f32>
+// CHECK:           %[[VAL_4:.*]] = torch.aten.exp %[[VAL_3]] : !torch.tensor<[2,3],f32> -> !torch.tensor<[2,3],f32>
+// CHECK:           %[[VAL_5:.*]] = torch.aten.log1p %[[VAL_4]] : !torch.tensor<[2,3],f32> -> !torch.tensor<[2,3],f32>
+// CHECK:           %[[VAL_6:.*]] = torch.aten.div.Scalar %[[VAL_5]], %[[VAL_1]] : !torch.tensor<[2,3],f32>, !torch.int -> !torch.tensor<[2,3],f32>
+// CHECK:           %[[VAL_7:.*]] = torch.aten.gt.Scalar %[[VAL_3]], %[[VAL_2]] : !torch.tensor<[2,3],f32>, !torch.int -> !torch.tensor<[2,3],i1>
+// CHECK:           %[[VAL_8:.*]] = torch.aten.where.self %[[VAL_7]], %[[VAL_0]], %[[VAL_6]] : !torch.tensor<[2,3],i1>, !torch.tensor<[2,3],f32>, !torch.tensor<[2,3],f32> -> !torch.tensor<[2,3],f32>
+// CHECK:           return %[[VAL_8]] : !torch.tensor<[2,3],f32>
+// CHECK:         }
+func.func @torch.aten.softplus(%t: !torch.tensor<[2,3],f32>, %dim: !torch.int) -> !torch.tensor<[2,3],f32> {
+  %int0 = torch.constant.int 0
+  %ret = torch.aten.softplus %t, %dim, %int0: !torch.tensor<[2,3],f32>, !torch.int, !torch.int -> !torch.tensor<[2,3],f32>
+  return %ret : !torch.tensor<[2,3],f32>
+}