[Torch] Add support for aten.round.decimals op

mmanzoorTT · mmanzoorTT · commit 48005dbbd225 · 2025-05-14T02:32:28.000Z
* Added lowering to Linalg-on-Tensors
* Added test to projects/pt1/python/torch_mlir_e2e_test/test_suite/elementwise.py
diff --git a/include/torch-mlir/Dialect/Torch/IR/GeneratedTorchOps.td b/include/torch-mlir/Dialect/Torch/IR/GeneratedTorchOps.td
@@ -4562,6 +4562,54 @@ def Torch_AtenRound_Op : Torch_Op<"aten.round_", [
   }];
 }
 
+def Torch_AtenRoundDecimalsOp : Torch_Op<"aten.round.decimals", [
+    AllowsTypeRefinement,
+    HasValueSemantics,
+    ReadOnly
+  ]> {
+  let summary = "Generated op for `aten::round.decimals : (Tensor, int) -> (Tensor)`";
+  let arguments = (ins
+    AnyTorchTensorType:$self,
+    Torch_IntType:$decimals
+  );
+  let results = (outs
+    AnyTorchOptionalTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenRoundDecimalsOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 2, 1);
+    }
+    void AtenRoundDecimalsOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 2, 1);
+    }
+  }];
+  let hasFolder = 1;
+}
+
+def Torch_AtenRound_DecimalsOp : Torch_Op<"aten.round_.decimals", [
+    IsTrailingUnderscoreInplaceVariant,
+    AllowsTypeRefinement
+  ]> {
+  let summary = "Generated op for `aten::round_.decimals : (Tensor, int) -> (Tensor)`";
+  let arguments = (ins
+    Torch_NonValueTensorType:$self,
+    Torch_IntType:$decimals
+  );
+  let results = (outs
+    AnyTorchOptionalNonValueTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenRound_DecimalsOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 2, 1);
+    }
+    void AtenRound_DecimalsOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 2, 1);
+    }
+  }];
+}
+
 def Torch_AtenTruncOp : Torch_Op<"aten.trunc", [
     AllowsTypeRefinement,
     HasValueSemantics,
diff --git a/lib/Conversion/TorchToLinalg/Uncategorized.cpp b/lib/Conversion/TorchToLinalg/Uncategorized.cpp
@@ -691,6 +691,43 @@ static Value createLinalgPayloadCalculationForElementwiseOp(
     }
     return b.create<math::RoundEvenOp>(loc, payloadArgs[0]);
   }
+  if (auto round = dyn_cast<AtenRoundDecimalsOp>(op)) {
+    // AtenRoundDecimalsOp is decomposed, if decimals is non-zero, as follow.
+    // scale = 10 ** decimals
+    // return round(x * scale) / scale
+    if (!isa<mlir::FloatType>(
+            cast<ValueTensorType>(round.getType()).getDtype())) {
+      round.emitError("unimplemented: non-floating point dtype");
+      return nullptr;
+    }
+    int64_t decimals;
+    if (!matchPattern(op->getOperand(1), m_TorchConstantInt(&decimals))) {
+      round.emitError("non-constant decimal point is not supported.");
+      return nullptr;
+    }
+
+    Value newOp = payloadArgs[0];
+    Value scale;
+    if (decimals) {
+      auto elementType =
+          cast<RankedTensorType>(
+              converter->convertType(op->getOperand(0).getType()))
+              .getElementType();
+
+      auto scalaVal = static_cast<float>(pow(10, decimals));
+      scale = b.create<arith::ConstantOp>(
+          loc, FloatAttr::get(elementType, scalaVal));
+      newOp = b.create<arith::MulFOp>(loc, newOp, scale);
+    }
+
+    newOp = b.create<math::RoundEvenOp>(loc, newOp);
+
+    if (decimals) {
+      newOp = b.create<arith::DivFOp>(loc, newOp, scale);
+    }
+
+    return newOp;
+  }
   if (auto prelu = dyn_cast<AtenPreluOp>(op)) {
     if (!isa<mlir::FloatType>(
             cast<ValueTensorType>(prelu.getType()).getDtype())) {
@@ -1635,10 +1672,11 @@ class ConvertElementwiseOp : public ConversionPattern {
              AtenNeScalarOp, AtenNegOp, AtenMaskedFillTensorOp, AtenLogicalOrOp,
              AtenLogicalAndOp, AtenLogicalXorOp, AtenLogicalNotOp, AtenIsinfOp,
              AtenTriuOp, AtenTrilOp, AtenBitwiseNotOp, AtenRoundOp,
-             AtenFillScalarOp, AtenFillTensorOp, AtenAtanOp, AtenAcosOp,
-             AtenAtanhOp, AtenAcoshOp, AtenAsinOp, AtenAsinhOp, AtenRealOp,
-             AtenImagOp, AtenDequantizeSelfOp, AtenDequantizeTensorOp,
-             AtenQuantizePerTensorOp, AtenIscloseOp>(op))
+             AtenRoundDecimalsOp, AtenFillScalarOp, AtenFillTensorOp,
+             AtenAtanOp, AtenAcosOp, AtenAtanhOp, AtenAcoshOp, AtenAsinOp,
+             AtenAsinhOp, AtenRealOp, AtenImagOp, AtenDequantizeSelfOp,
+             AtenDequantizeTensorOp, AtenQuantizePerTensorOp, AtenIscloseOp>(
+            op))
       return rewriter.notifyMatchFailure(op, "not a supported elementwise op");
 
     if (failed(verifyLinalgCompatibleTypes(op, rewriter)))
@@ -3988,9 +4026,9 @@ void mlir::torch::torch_to_linalg::populateUncategorizedPatternsAndLegality(
       AtenMaskedFillTensorOp, AtenLogicalOrOp, AtenLogicalAndOp, AtenAtanOp,
       AtenAcosOp, AtenLogicalXorOp, AtenLogicalNotOp, AtenIsinfOp, AtenTriuOp,
       AtenTrilOp, AtenRemainderScalarOp, AtenRemainderTensorOp,
-      AtenBitwiseNotOp, AtenRoundOp, AtenFillScalarOp, AtenFillTensorOp,
-      AtenRealOp, AtenImagOp, AtenDequantizeSelfOp, AtenDequantizeTensorOp,
-      AtenQuantizePerTensorOp, AtenIscloseOp>();
+      AtenBitwiseNotOp, AtenRoundOp, AtenRoundDecimalsOp, AtenFillScalarOp,
+      AtenFillTensorOp, AtenRealOp, AtenImagOp, AtenDequantizeSelfOp,
+      AtenDequantizeTensorOp, AtenQuantizePerTensorOp, AtenIscloseOp>();
   patterns.add<ConvertElementwiseOp>(typeConverter, context);
   target.addIllegalOp<AtenNllLossForwardOp>();
   patterns.add<ConvertAtenDetachOp>(typeConverter, context);
diff --git a/lib/Dialect/Torch/IR/TorchOps.cpp b/lib/Dialect/Torch/IR/TorchOps.cpp
@@ -1992,6 +1992,19 @@ OpFoldResult AtenCeilOp::fold(FoldAdaptor adaptor) {
   return {};
 }
 
+//===----------------------------------------------------------------------===//
+// AtenRoundDecimalsOp
+//===----------------------------------------------------------------------===//
+
+OpFoldResult AtenRoundDecimalsOp::fold(FoldAdaptor adaptor) {
+  auto resultType = dyn_cast<ValueTensorType>(getType());
+  if (resultType && resultType.hasDtype() &&
+      isa<mlir::IntegerType>(resultType.getDtype())) {
+    return getSelf();
+  }
+  return {};
+}
+
 //===----------------------------------------------------------------------===//
 // AtenRoundOp
 //===----------------------------------------------------------------------===//
diff --git a/lib/Dialect/Torch/Transforms/AbstractInterpLibrary.cpp b/lib/Dialect/Torch/Transforms/AbstractInterpLibrary.cpp
@@ -6750,6 +6750,10 @@ 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.round.decimals\"(%arg0: !torch.list<int>, %arg1: !torch.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"
+"  }\n"
 "  func.func @\"__torch_mlir_shape_fn.aten.glu\"(%arg0: !torch.list<int>, %arg1: !torch.int) -> !torch.list<int> {\n"
 "    %none = torch.constant.none\n"
 "    %str = torch.constant.str \"AssertionError: glu's dim size must be multiply of 2\"\n"
@@ -12896,6 +12900,10 @@ 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.round.decimals\"(%arg0: !torch.tuple<int, int>, %arg1: !torch.int) -> !torch.int {\n"
+"    %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.glu\"(%arg0: !torch.tuple<int, int>, %arg1: !torch.int) -> !torch.int {\n"
 "    %0:2 = torch.prim.TupleUnpack %arg0 : !torch.tuple<int, int> -> !torch.int, !torch.int\n"
 "    return %0#1 : !torch.int\n"
diff --git a/projects/pt1/e2e_testing/xfail_sets.py b/projects/pt1/e2e_testing/xfail_sets.py
@@ -547,6 +547,7 @@
     "AtenKthvalueModule_basic",
     "AtenPolarDoubleModule_basic",
     "AtenPolarFloatModule_basic",
+    "AtenRoundFloatDecimalsModule_basic",
     "DiagonalWithStaticShapeModule_basic",
     "EinsumStaticDiagonalDimensionModule_basic",
     "ElementwiseRemainderScalarModule_Bool_NegativeDivisor_basic",
@@ -3408,6 +3409,7 @@
     "AtenSymConstrainRange_basic",
     "Aten_AssertScalar_basic",
     "AvgPool2dSingleIntTupleParamsIncludePadModule_basic",
+    "AtenRoundFloatDecimalsModule_basic",
     "ScatterAddDynamicModule_basic",
     "UniformModule_basic",
     "UniformStaticShapeModule_basic",
diff --git a/projects/pt1/python/torch_mlir/jit_ir_importer/build_tools/abstract_interp_lib_gen.py b/projects/pt1/python/torch_mlir/jit_ir_importer/build_tools/abstract_interp_lib_gen.py
@@ -349,6 +349,9 @@ def aten〇relu6〡shape(self: List[int]) -> List[int]:
 def aten〇round〡shape(self: List[int]) -> List[int]:
     return upstream_shape_functions.unary(self)
 
+def aten〇round〇decimals〡shape(self: List[int], decimals: int) -> List[int]:
+    return upstream_shape_functions.unary(self)
+
 def aten〇glu〡shape(self: List[int], dim: int = -1) -> List[int]:
     if dim < 0:
         dim += len(self)
@@ -3616,6 +3619,11 @@ def aten〇round〡dtype(self_rank_dtype: Tuple[int, int]) -> int:
     self_rank, self_dtype = self_rank_dtype
     return self_dtype
 
+@check_dtype_function(_check_tensors_with_the_same_dtype(num_of_tensors=1, decimals=0))
+def aten〇round〇decimals〡dtype(self_rank_dtype: Tuple[int, int], decimals: int) -> int:
+    self_rank, self_dtype = self_rank_dtype
+    return self_dtype
+
 @check_dtype_function(_check_tensors_with_the_same_dtype(tensor_shapes=[(100,)], dim=0))
 def aten〇glu〡dtype(self_rank_dtype: Tuple[int, int], dim: int = -1) -> int:
     self_rank, self_dtype = self_rank_dtype
diff --git a/projects/pt1/python/torch_mlir/jit_ir_importer/build_tools/torch_ods_gen.py b/projects/pt1/python/torch_mlir/jit_ir_importer/build_tools/torch_ods_gen.py
@@ -451,6 +451,9 @@ def emit_with_mutating_variants(key, **kwargs):
     emit_with_mutating_variants("aten::floor : (Tensor) -> (Tensor)", has_folder=True)
     emit_with_mutating_variants("aten::ceil : (Tensor) -> (Tensor)", has_folder=True)
     emit_with_mutating_variants("aten::round : (Tensor) -> (Tensor)", has_folder=True)
+    emit_with_mutating_variants(
+        "aten::round.decimals : (Tensor, int) -> (Tensor)", has_folder=True
+    )
     emit_with_mutating_variants("aten::trunc : (Tensor) -> (Tensor)", has_folder=True)
     emit("aten::special_expm1 : (Tensor) -> (Tensor)")
     emit_with_mutating_variants(
diff --git a/projects/pt1/python/torch_mlir_e2e_test/test_suite/elementwise.py b/projects/pt1/python/torch_mlir_e2e_test/test_suite/elementwise.py
@@ -6503,6 +6503,26 @@ def AtenRoundIntModule_basic(module, tu: TestUtils):
     module.forward(tu.randint(5, 5, low=-10))
 
 
+class AtenRoundFloatDecimalsModule(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.round(x, decimals=2)
+
+
+@register_test_case(module_factory=lambda: AtenRoundFloatDecimalsModule())
+def AtenRoundFloatDecimalsModule_basic(module, tu: TestUtils):
+    module.forward(tu.rand(5, 5, low=-3.0, high=3.0))
+
+
 # ==============================================================================
 
 
