[Torch Dialect] add RSub, ScalarImplicit canonicalize (#1899)

* add rsub, scalarimplit canonicalizer

* reformat

* address comments

* fix bug

* fix test

* Update elementwise.py

* resolve merge conflict

* change to 3

* change to 3

* real fix

* fix name

* add torchdynamo fail test

---------

Co-authored-by: zhekun.zhang <zhekun.zhang@bytedance.com>

e2e_testing/xfail_sets.py

@@ -91,6 +91,8 @@ TORCHDYNAMO_XFAIL_SET = {
     "AtenIntTensorByteDtypeModule_basic",
     # ERROR: assert isinstance(e, FakeTensor)
     "ElementwiseAddScalar_NumToTensorFloat_Module_basic",
+    # ERROR: assert isinstance(e, FakeTensor)
+    "RsubInt0d_NumToTensor_Module_basic",
 }
 
 STABLEHLO_PASS_SET = {
@@ -316,6 +318,7 @@ STABLEHLO_PASS_SET = {
     "RsubFloatModule_noalpha_basic",
     "RsubIntModule_basic",
     "RsubIntModule_noalpha_basic",
+    "RsubInt0d_NumToTensor_Module_basic",
     "SliceStaticModule_basic",
     "SliceModule_basic",
     "SliceNegIdxModule_basic",
@@ -513,6 +516,7 @@ TOSA_PASS_SET = {
     "Matmul_3d",
     "RsubFloatModule_basic",
     "RsubFloatModule_noalpha_basic",
+    "RsubInt0d_NumToTensor_Module_basic",
     "ElementwiseBitwiseAndModule_basic",
     "ElementwiseBitwiseAndStaticShapeModule_basic",
     "ElementwiseBitwiseNotInt32Module_basic",

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

@@ -3389,6 +3389,7 @@ def Torch_AtenRsubScalarOp : Torch_Op<"aten.rsub.Scalar", [
       printDefaultTorchOp(printer, *this, 3, 1);
     }
   }];
+  let hasCanonicalizer = 1;
 }
 
 def Torch_AtenGeluOp : Torch_Op<"aten.gelu", [
@@ -10569,6 +10570,7 @@ def Torch_AtenScalarImplicitOp : Torch_Op<"aten.ScalarImplicit", [
       printDefaultTorchOp(printer, *this, 1, 1);
     }
   }];
+  let hasCanonicalizer = 1;
 }
 
 def Torch_Aten_SoftmaxBackwardDataOp : Torch_Op<"aten._softmax_backward_data", [

lib/Dialect/Torch/IR/TorchOps.cpp

@@ -881,13 +881,16 @@ LogicalResult rewrite0DBinaryTensorOp(Operation *op,
     return rewriter.notifyMatchFailure(
         op, "only int scalar lhs or rhs is supported");
   }
-  if (isa<AtenSubTensorOp, AtenSubScalarOp, AtenAddTensorOp, AtenAddScalarOp>(
-          op)) {
+  if (isa<AtenSubTensorOp, AtenSubScalarOp, AtenRsubScalarOp, AtenAddTensorOp,
+          AtenAddScalarOp>(op)) {
     Value alpha = getScalarIntValue(op->getOperand(2), loc, rewriter);
     if (!alpha) {
       return rewriter.notifyMatchFailure(op,
                                          "only int scalar alpha is supported");
     }
+    if (isa<AtenRsubScalarOp>(op))
+      lhs = rewriter.create<AtenMulIntOp>(loc, lhs, alpha);
+    else
       rhs = rewriter.create<AtenMulIntOp>(loc, rhs, alpha);
   }
 
@@ -941,6 +944,8 @@ LogicalResult rewrite0DBinaryTensorOp(Operation *op,
     result = rewriter.create<AtenAddIntOp>(loc, lhs, rhs);
   } else if (isa<AtenSubScalarOp, AtenSubTensorOp>(op)) {
     result = rewriter.create<AtenSubIntOp>(loc, lhs, rhs);
+  } else if (isa<AtenRsubScalarOp>(op)) {
+    result = rewriter.create<AtenSubIntOp>(loc, rhs, lhs);
   } else if (isa<AtenMulScalarOp, AtenMulTensorOp>(op)) {
     result = rewriter.create<AtenMulIntOp>(loc, lhs, rhs);
   }
@@ -988,6 +993,16 @@ void AtenSubScalarOp::getCanonicalizationPatterns(RewritePatternSet &patterns,
   });
 }
 
+//===----------------------------------------------------------------------===//
+// AtenRSubScalarOp
+//===----------------------------------------------------------------------===//
+void AtenRsubScalarOp::getCanonicalizationPatterns(RewritePatternSet &patterns,
+                                                   MLIRContext *context) {
+  patterns.add(+[](AtenRsubScalarOp op, PatternRewriter &rewriter) {
+    return rewrite0DBinaryTensorOp(op, rewriter);
+  });
+}
+
 //===----------------------------------------------------------------------===//
 // AtenMulTensorOp
 //===----------------------------------------------------------------------===//
@@ -1018,6 +1033,23 @@ void AtenDivTensorModeOp::getCanonicalizationPatterns(
   });
 }
 
+//===----------------------------------------------------------------------===//
+// AtenScalarImplicitOp
+//===----------------------------------------------------------------------===//
+void AtenScalarImplicitOp::getCanonicalizationPatterns(
+    RewritePatternSet &patterns, MLIRContext *context) {
+  patterns.add(+[](AtenScalarImplicitOp op, PatternRewriter &rewriter) {
+    Location loc = op.getLoc();
+    Value a = op.getA();
+    auto outType = op.getResult().getType();
+    Value scalarValue = getScalarIntValue(a, loc, rewriter);
+    if (!scalarValue)
+      return failure();
+    rewriter.replaceOpWithNewOp<Torch::DerefineOp>(op, outType, scalarValue);
+    return success();
+  });
+}
+
 //===----------------------------------------------------------------------===//
 // AtenSizeOp
 //===----------------------------------------------------------------------===//

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

@@ -318,7 +318,7 @@ def emit_ops(emitter_td: TextEmitter, registry: Registry):
     emit("aten::maximum : (Tensor, Tensor) -> (Tensor)")
     emit("aten::minimum : (Tensor, Tensor) -> (Tensor)")
     emit("aten::mish : (Tensor) -> (Tensor)")
-    emit("aten::rsub.Scalar : (Tensor, Scalar, Scalar) -> (Tensor)")
+    emit("aten::rsub.Scalar : (Tensor, Scalar, Scalar) -> (Tensor)", has_canonicalizer=True)
     emit("aten::gelu : (Tensor, str) -> (Tensor)")
     emit("aten::pow.Tensor_Scalar : (Tensor, Scalar) -> (Tensor)")
     emit("aten::pow.Tensor_Tensor : (Tensor, Tensor) -> (Tensor)")
@@ -641,7 +641,7 @@ def emit_ops(emitter_td: TextEmitter, registry: Registry):
     emit("aten::eq.device : (Device, Device) -> (bool)")
     emit("aten::ceil.float : (float) -> (int)", has_folder=True)
     emit("aten::narrow : (Tensor, int, int, int) -> (Tensor)")
-    emit("aten::ScalarImplicit : (Tensor) -> (Scalar)")
+    emit("aten::ScalarImplicit : (Tensor) -> (Scalar)", has_canonicalizer=True)
 
     # backprop ops
     emit("aten::_softmax_backward_data : (Tensor, Tensor, int, int) -> (Tensor)")

python/torch_mlir_e2e_test/test_suite/elementwise.py

@@ -774,6 +774,26 @@ class RsubIntModule_noalpha(torch.nn.Module):
 def RsubIntModule_noalpha_basic(module, tu: TestUtils):
     module.forward(tu.randint(3, 4, high=100))
 
+# ==============================================================================
+
+
+class RsubInt0d_NumToTensor_Module(torch.nn.Module):
+
+    def __init__(self):
+        super().__init__()
+
+    @export
+    @annotate_args([
+        None,
+    ])
+    def forward(self):
+        x = torch.ops.prim.NumToTensor(5)
+        return torch.rsub(x, 2)
+
+
+@register_test_case(module_factory=lambda: RsubInt0d_NumToTensor_Module())
+def RsubInt0d_NumToTensor_Module_basic(module, tu: TestUtils):
+    module.forward()
 
 # ==============================================================================
 

test/Dialect/Torch/canonicalize.mlir

@@ -1838,3 +1838,52 @@ func.func @torch.aten.slice.tensor$fold_full_domain_slice(%arg0: !torch.vtensor<
   %0 = torch.aten.slice.Tensor %arg0, %int0, %int0, %int-1, %int1 : !torch.vtensor<[4], f32>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[4], f32>
   return %0 : !torch.vtensor<[4],f32>
 }
+
+// CHECK-LABEL:   func.func @torch.aten.rsub.Scalar$canonicalize_literal_0d() -> !torch.vtensor<[],si64> {
+// CHECK:             %int-1 = torch.constant.int -1
+// CHECK:             %[[VAL_0:.*]] = torch.prim.NumToTensor.Scalar %int-1 : !torch.int -> !torch.vtensor<[],si64>
+// CHECK:             return %[[VAL_0]] : !torch.vtensor<[],si64>
+func.func @torch.aten.rsub.Scalar$canonicalize_literal_0d() -> !torch.vtensor<[],si64> {
+    %int2 = torch.constant.int 2
+    %int3 = torch.constant.int 3
+    %0 = torch.vtensor.literal(dense<1> : tensor<si64>) : !torch.vtensor<[],si64>
+    %2 = torch.aten.rsub.Scalar %0, %int2, %int3 : !torch.vtensor<[],si64>, !torch.int, !torch.int -> !torch.vtensor<[],si64>
+    return %2 : !torch.vtensor<[],si64>
+}
+
+// CHECK-LABEL:   func.func @torch.aten.rsub.Scalar$canonicalize_numtotensor_0d() -> !torch.vtensor<[],si64> {
+// CHECK:             %int-1 = torch.constant.int -1
+// CHECK:             %int1 = torch.constant.int 1
+// CHECK:             %[[VAL_0:.*]] = torch.prim.NumToTensor.Scalar %int1 : !torch.int -> !torch.vtensor<[],si64>
+// CHECK:             %[[VAL_1:.*]] = torch.prim.NumToTensor.Scalar %int-1 : !torch.int -> !torch.vtensor<[],si64>
+// CHECK:             return %[[VAL_1]] : !torch.vtensor<[],si64>
+func.func @torch.aten.rsub.Scalar$canonicalize_numtotensor_0d() -> !torch.vtensor<[],si64> {
+    %int1 = torch.constant.int 1
+    %int2 = torch.constant.int 2
+    %int3 = torch.constant.int 3
+    %0 = torch.prim.NumToTensor.Scalar %int1 : !torch.int -> !torch.vtensor<[],si64>
+    %2 = torch.aten.rsub.Scalar %0, %int2, %int3 : !torch.vtensor<[],si64>, !torch.int, !torch.int -> !torch.vtensor<[],si64>
+    return %2 : !torch.vtensor<[],si64>
+}
+
+// CHECK-LABEL:   func.func @torch.aten.ScalarImplicit$canonicalize_numtotensor_0d() -> !torch.number {
+// CHECK:             %int1 = torch.constant.int 1
+// CHECK:             %[[VAL_0:.*]] = torch.prim.NumToTensor.Scalar %int1 : !torch.int -> !torch.vtensor<[],si64>
+// CHECK:             %[[VAL_1:.*]] = torch.derefine %int1 : !torch.int to !torch.number
+// CHECK:             return %[[VAL_1]] : !torch.number
+func.func @torch.aten.ScalarImplicit$canonicalize_numtotensor_0d() -> !torch.number {
+    %int1 = torch.constant.int 1
+    %0 = torch.prim.NumToTensor.Scalar %int1 : !torch.int -> !torch.vtensor<[],si64>
+    %1 = torch.aten.ScalarImplicit %0 : !torch.vtensor<[],si64> -> !torch.number
+    return %1 : !torch.number
+}
+
+// CHECK-LABEL:   func.func @torch.aten.ScalarImplicit$canonicalize_literal_0d() -> !torch.number {
+// CHECK:             %int1 = torch.constant.int 1
+// CHECK:             %[[VAL_0:.*]] = torch.derefine %int1 : !torch.int to !torch.number
+// CHECK:             return %[[VAL_0]] : !torch.number
+func.func @torch.aten.ScalarImplicit$canonicalize_literal_0d() -> !torch.number {
+    %0 = torch.vtensor.literal(dense<1> : tensor<si64>) : !torch.vtensor<[],si64>
+    %1 = torch.aten.ScalarImplicit %0 : !torch.vtensor<[],si64> -> !torch.number
+    return %1 : !torch.number
+}