[MLIR][TORCH] Add e2e support for aten.var_mean op

Signed-Off By: Vivek Khandelwal<vivek@nod-labs.com>

e2e_testing/xfail_sets.py

@@ -811,4 +811,6 @@ LTC_XFAIL_SET = {
     "NllLossModule_sum_basic",
     "ResNet18Module_basic",
     "ResNet18StaticModule_basic",
+    "VarMeanBiasedModule_basic",
+    "VarMeanUnbiasedModule_basic",
 }

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

@@ -5008,6 +5008,31 @@ def Torch_AtenVarMeanCorrectionOp : Torch_Op<"aten.var_mean.correction", [
   }];
 }
 
+def Torch_AtenVarMeanOp : Torch_Op<"aten.var_mean", [
+    AllowsTypeRefinement,
+    HasValueSemantics,
+    ReadOnly
+  ]> {
+  let summary = "Generated op for `aten::var_mean : (Tensor, bool) -> (Tensor, Tensor)`";
+  let arguments = (ins
+    AnyTorchTensorType:$self,
+    Torch_BoolType:$unbiased
+  );
+  let results = (outs
+    AnyTorchTensorType:$result0,
+    AnyTorchTensorType:$result1
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenVarMeanOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 2, 2);
+    }
+    void AtenVarMeanOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 2, 2);
+    }
+  }];
+}
+
 def Torch_AtenNllLossForwardOp : Torch_Op<"aten.nll_loss_forward", [
     AllowsTypeRefinement,
     HasValueSemantics,

lib/Dialect/Torch/Transforms/DecomposeComplexOps.cpp

@@ -3284,6 +3284,26 @@ public:
 };
 } // namespace
 
+namespace {
+class DecomposeAtenVarMeanOp : public OpRewritePattern<AtenVarMeanOp> {
+public:
+  using OpRewritePattern::OpRewritePattern;
+  LogicalResult matchAndRewrite(AtenVarMeanOp op,
+                                PatternRewriter &rewriter) const override {
+    Location loc = op.getLoc();
+    Value falseVal = rewriter.create<ConstantBoolOp>(loc, false);
+    Value noneVal = rewriter.create<ConstantNoneOp>(loc);
+    Value var = rewriter.create<AtenVarDimOp>(loc, op.getType(0), op.getSelf(),
+                                              /*dim=*/noneVal, op.getUnbiased(),
+                                              /*keepdim=*/falseVal);
+    Value mean = rewriter.create<AtenMeanOp>(loc, op.getType(0), op.getSelf(),
+                                             /*dtype=*/noneVal);
+    rewriter.replaceOp(op, {var, mean});
+    return success();
+  }
+};
+} // namespace
+
 namespace {
 class DecomposeComplexOpsPass
     : public DecomposeComplexOpsBase<DecomposeComplexOpsPass> {
@@ -3428,6 +3448,7 @@ public:
     addPatternIfTargetOpIsIllegal<DecomposePrimsConvertElementTypeOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenRandnOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenRandnGeneratorOp>(patterns);
+    addPatternIfTargetOpIsIllegal<DecomposeAtenVarMeanOp>(patterns);
 
     GreedyRewriteConfig config;
     config.useTopDownTraversal = true;

lib/Dialect/Torch/Transforms/LowerToBackendContract.cpp

@@ -418,6 +418,7 @@ static void markDecomposedOpsAsIllegal(MLIRContext *context,
   target.addIllegalOp<PrimsConvertElementTypeOp>();
   target.addIllegalOp<AtenRandnOp>();
   target.addIllegalOp<AtenRandnGeneratorOp>();
+  target.addIllegalOp<AtenVarMeanOp>();
   for (std::string opName : backendLegalOps) {
     target.addLegalOp(OperationName(opName, context));
   }

lib/Dialect/Torch/Transforms/RefineTypes.cpp

@@ -1182,7 +1182,7 @@ void TypeAnalysis::visitOperation(Operation *op,
     return;
   }
 
-  if (isa<AtenVarMeanCorrectionOp>(op)) {
+  if (isa<AtenVarMeanCorrectionOp, AtenVarMeanOp>(op)) {
     auto input = operands[0]->getValue();
     auto knowledge =
         ValueKnowledge::getTensorPessimisticValueState(op->getContext());

lib/Dialect/Torch/Transforms/ShapeLibrary.cpp

@@ -5780,6 +5780,12 @@ StringRef mlir::torch::Torch::getShapeLibrary() {
 "    %2 = torch.prim.TupleConstruct %1, %1 : !torch.list<int>, !torch.list<int> -> !torch.tuple<list<int>, list<int>>\n"
 "    return %2 : !torch.tuple<list<int>, list<int>>\n"
 "  }\n"
+"  func.func @\"__torch_mlir_shape_fn.aten.var_mean\"(%arg0: !torch.list<int>, %arg1: !torch.bool) -> !torch.tuple<list<int>, list<int>> {\n"
+"    %0 = torch.prim.ListConstruct  : () -> !torch.list<int>\n"
+"    %1 = torch.prim.ListConstruct  : () -> !torch.list<int>\n"
+"    %2 = torch.prim.TupleConstruct %0, %1 : !torch.list<int>, !torch.list<int> -> !torch.tuple<list<int>, list<int>>\n"
+"    return %2 : !torch.tuple<list<int>, list<int>>\n"
+"  }\n"
 "  func.func @\"__torch_mlir_shape_fn.aten.std\"(%arg0: !torch.list<int>, %arg1: !torch.bool) -> !torch.list<int> {\n"
 "    %0 = torch.prim.ListConstruct  : () -> !torch.list<int>\n"
 "    return %0 : !torch.list<int>\n"

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

@@ -544,13 +544,16 @@ def aten〇var(self: List[int], unbiased: bool = True) -> List[int]:
 def aten〇var〇dim(self: List[int], dim: Optional[List[int]], unbiased: bool = True, keepdim: bool = False) -> List[int]:
     return upstream_shape_functions.sum_mean_dim(self, dim, keepdim, None)
 
-def aten〇var〇correction(self: List[int], dim: Optional[List[int]], correction: Optional[int], keepdim: bool = False) -> List[int]:
+def aten〇var〇correction(self: List[int], dim: Optional[List[int]] = None, correction: Optional[int] = None, keepdim: bool = False) -> List[int]:
     return upstream_shape_functions.sum_mean_dim(self, dim, keepdim, None)
 
-def aten〇var_mean〇correction(self: List[int], dim: Optional[List[int]], correction: Optional[int], keepdim: bool = False) -> Tuple[List[int], List[int]]:
+def aten〇var_mean〇correction(self: List[int], dim: Optional[List[int]] = None, correction: Optional[int] = None, keepdim: bool = False) -> Tuple[List[int], List[int]]:
     out = upstream_shape_functions.sum_mean_dim(self, dim, keepdim, None)
     return out, out
 
+def aten〇var_mean(self: List[int], unbiased: bool = True) -> Tuple[List[int], List[int]]:
+    return [], []
+
 def aten〇std(self: List[int], unbiased: bool = True) -> List[int]:
     return []
 

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

@@ -409,6 +409,7 @@ def emit_ops(emitter_td: TextEmitter, registry: Registry):
     emit("aten::var.dim : (Tensor, int[]?, bool, bool) -> (Tensor)")
     emit("aten::var.correction : (Tensor, int[]?, int?, bool) -> (Tensor)")
     emit("aten::var_mean.correction : (Tensor, int[]?, int?, bool) -> (Tensor, Tensor)")
+    emit("aten::var_mean : (Tensor, bool) -> (Tensor, Tensor)")
     emit("aten::nll_loss_forward : (Tensor, Tensor, Tensor?, int, int) -> (Tensor, Tensor)")
     emit("aten::nll_loss_backward : (Tensor, Tensor, Tensor, Tensor?, int, int, Tensor) -> (Tensor)")
     emit("aten::bincount : (Tensor, Tensor?, int) -> (Tensor)")

python/torch_mlir_e2e_test/test_suite/stats.py

@@ -799,3 +799,47 @@ class VarMeanCorrectionNoneModule(torch.nn.Module):
 @register_test_case(module_factory=lambda: VarMeanCorrectionNoneModule())
 def VarMeanCorrectionNoneModule_basic(module, tu: TestUtils):
     module.forward(tu.rand(3, 4, 7))
+
+
+# ==============================================================================
+
+
+class VarMeanUnbiasedModule(torch.nn.Module):
+
+    def __init__(self):
+        super().__init__()
+
+    @export
+    @annotate_args([
+        None,
+        ([-1, -1, -1], torch.float32, True),
+    ])
+    def forward(self, x):
+        return torch.ops.aten.var_mean(x)
+
+
+@register_test_case(module_factory=lambda: VarMeanUnbiasedModule())
+def VarMeanUnbiasedModule_basic(module, tu: TestUtils):
+    module.forward(tu.rand(3, 4, 7))
+
+
+# ==============================================================================
+
+
+class VarMeanBiasedModule(torch.nn.Module):
+
+    def __init__(self):
+        super().__init__()
+
+    @export
+    @annotate_args([
+        None,
+        ([-1, -1, -1], torch.float32, True),
+    ])
+    def forward(self, x):
+        return torch.ops.aten.var_mean(x, unbiased=False)
+
+
+@register_test_case(module_factory=lambda: VarMeanBiasedModule())
+def VarMeanBiasedModule_basic(module, tu: TestUtils):
+    module.forward(tu.rand(3, 4, 7))