[Torch] Implements TorchToLinalg lowering of torch.ops.aten._weight_norm_interface (#3538)

Resolves https://github.com/nod-ai/SHARK-Turbine/issues/757.

Adds TorchToLinalg lowering for `Aten_WeightNormInterfaceOp`.

---------

Co-authored-by: Ubuntu <rbhowmik@RohanBhowmikVM.judsoscro3wupi0qm4bjlj5m3b.bx.internal.cloudapp.net>

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

@@ -9118,6 +9118,32 @@ def Torch_AtenDiagEmbedOp : Torch_Op<"aten.diag_embed", [
   }];
 }
 
+def Torch_Aten_WeightNormInterfaceOp : Torch_Op<"aten._weight_norm_interface", [
+    AllowsTypeRefinement,
+    HasValueSemantics,
+    ReadOnly
+  ]> {
+  let summary = "Generated op for `aten::_weight_norm_interface : (Tensor, Tensor, int) -> (Tensor, Tensor)`";
+  let arguments = (ins
+    AnyTorchTensorType:$v,
+    AnyTorchTensorType:$g,
+    Torch_IntType:$dim
+  );
+  let results = (outs
+    AnyTorchOptionalTensorType:$result0,
+    AnyTorchOptionalTensorType:$result1
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult Aten_WeightNormInterfaceOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 3, 2);
+    }
+    void Aten_WeightNormInterfaceOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 3, 2);
+    }
+  }];
+}
+
 def Torch_AtenConstantPadNdOp : Torch_Op<"aten.constant_pad_nd", [
     AllowsTypeRefinement,
     HasValueSemantics,

lib/Dialect/Torch/Transforms/AbstractInterpLibrary.cpp

@@ -9739,6 +9739,12 @@ 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._weight_norm_interface\"(%arg0: !torch.list<int>, %arg1: !torch.list<int>, %arg2: !torch.int) -> !torch.tuple<list<int>, list<int>> {\n"
+"    %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>\n"
+"    %1 = call @__torch__.torch.jit._shape_functions.unary(%arg1) : (!torch.list<int>) -> !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.slice.Tensor\"(%arg0: !torch.list<int>, %arg1: !torch.int, %arg2: !torch.optional<int>, %arg3: !torch.optional<int>, %arg4: !torch.int) -> !torch.list<int> {\n"
 "    %0 = call @__torch__.torch.jit._shape_functions.slice(%arg0, %arg1, %arg2, %arg3, %arg4) : (!torch.list<int>, !torch.int, !torch.optional<int>, !torch.optional<int>, !torch.int) -> !torch.list<int>\n"
 "    return %0 : !torch.list<int>\n"
@@ -10934,6 +10940,55 @@ 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._weight_norm_interface\"(%arg0: !torch.tuple<int, int>, %arg1: !torch.tuple<int, int>, %arg2: !torch.int) -> !torch.tuple<int, int> {\n"
+"    %int6 = torch.constant.int 6\n"
+"    %int9 = torch.constant.int 9\n"
+"    %int7 = torch.constant.int 7\n"
+"    %int10 = torch.constant.int 10\n"
+"    %none = torch.constant.none\n"
+"    %str = torch.constant.str \"AssertionError: \"\n"
+"    %false = torch.constant.bool false\n"
+"    %true = torch.constant.bool true\n"
+"    %0 = torch.prim.Uninitialized : !torch.tuple<int, int>\n"
+"    %1:2 = torch.prim.TupleUnpack %arg0 : !torch.tuple<int, int> -> !torch.int, !torch.int\n"
+"    %2:2 = torch.prim.TupleUnpack %arg1 : !torch.tuple<int, int> -> !torch.int, !torch.int\n"
+"    %3 = torch.aten.eq.int %1#1, %2#1 : !torch.int, !torch.int -> !torch.bool\n"
+"    torch.prim.If %3 -> () {\n"
+"      torch.prim.If.yield\n"
+"    } else {\n"
+"      torch.prim.RaiseException %str, %none : !torch.str, !torch.none\n"
+"      torch.prim.If.yield\n"
+"    }\n"
+"    %4 = call @__torch__.torch_mlir.jit_ir_importer.build_tools.library_generator.is_integer_dtype(%2#1) : (!torch.int) -> !torch.bool\n"
+"    %5 = torch.aten.__not__ %4 : !torch.bool -> !torch.bool\n"
+"    torch.prim.If %5 -> () {\n"
+"      torch.prim.If.yield\n"
+"    } else {\n"
+"      torch.prim.RaiseException %str, %none : !torch.str, !torch.none\n"
+"      torch.prim.If.yield\n"
+"    }\n"
+"    %6 = torch.aten.eq.int %2#1, %int10 : !torch.int, !torch.int -> !torch.bool\n"
+"    %7:2 = torch.prim.If %6 -> (!torch.bool, !torch.tuple<int, int>) {\n"
+"      %9 = torch.prim.TupleConstruct %1#1, %int7 : !torch.int, !torch.int -> !torch.tuple<int, int>\n"
+"      torch.prim.If.yield %true, %9 : !torch.bool, !torch.tuple<int, int>\n"
+"    } else {\n"
+"      %9 = torch.aten.eq.int %2#1, %int9 : !torch.int, !torch.int -> !torch.bool\n"
+"      %10:2 = torch.prim.If %9 -> (!torch.bool, !torch.tuple<int, int>) {\n"
+"        %11 = torch.prim.TupleConstruct %1#1, %int6 : !torch.int, !torch.int -> !torch.tuple<int, int>\n"
+"        torch.prim.If.yield %true, %11 : !torch.bool, !torch.tuple<int, int>\n"
+"      } else {\n"
+"        torch.prim.If.yield %false, %0 : !torch.bool, !torch.tuple<int, int>\n"
+"      }\n"
+"      torch.prim.If.yield %10#0, %10#1 : !torch.bool, !torch.tuple<int, int>\n"
+"    }\n"
+"    %8 = torch.prim.If %7#0 -> (!torch.tuple<int, int>) {\n"
+"      torch.prim.If.yield %7#1 : !torch.tuple<int, int>\n"
+"    } else {\n"
+"      %9 = torch.prim.TupleConstruct %1#1, %2#1 : !torch.int, !torch.int -> !torch.tuple<int, int>\n"
+"      torch.prim.If.yield %9 : !torch.tuple<int, int>\n"
+"    }\n"
+"    return %8 : !torch.tuple<int, int>\n"
+"  }\n"
 "  func.func @\"__torch_mlir_dtype_fn.aten.bernoulli_.float\"(%arg0: !torch.tuple<int, int>, %arg1: !torch.float, %arg2: !torch.any) -> !torch.int {\n"
 "    %0:2 = torch.prim.TupleUnpack %arg0 : !torch.tuple<int, int> -> !torch.int, !torch.int\n"
 "    return %0#1 : !torch.int\n"

lib/Dialect/Torch/Transforms/DecomposeComplexOps.cpp

@@ -5553,6 +5553,63 @@ class DecomposeAtenInstanceNormOp
 };
 } // namespace
 
+namespace {
+class DecomposeAten_WeightNormInterfaceOp
+    : public OpRewritePattern<Aten_WeightNormInterfaceOp> {
+  using OpRewritePattern<Aten_WeightNormInterfaceOp>::OpRewritePattern;
+  LogicalResult matchAndRewrite(Aten_WeightNormInterfaceOp op,
+                                PatternRewriter &rewriter) const override {
+
+    Location loc = op.getLoc();
+    Value v = op.getV();
+    Value g = op.getG();
+    Value dim = op.getDim();
+
+    auto inputType = cast<BaseTensorType>(v.getType());
+    if (!inputType.hasSizes())
+      return rewriter.notifyMatchFailure(op, "expected input to have sizes");
+
+    if (!cast<ConstantIntOp>(dim.getDefiningOp()))
+      return rewriter.notifyMatchFailure(op, "dim is not a ConstantIntOp");
+
+    auto sizes = inputType.getSizes();
+    SmallVector<Value> keepDims;
+    for (int64_t i = 0; i < static_cast<int64_t>(sizes.size()); ++i) {
+      if (i !=
+          static_cast<int64_t>(dim.getDefiningOp<ConstantIntOp>().getValue()))
+        keepDims.push_back(
+            rewriter.create<ConstantIntOp>(loc, rewriter.getI64IntegerAttr(i)));
+    }
+
+    Value ord =
+        rewriter.create<ConstantIntOp>(loc, rewriter.getI64IntegerAttr(2));
+    Value keepdim =
+        rewriter.create<ConstantBoolOp>(loc, rewriter.getBoolAttr(true));
+    Value dtypeNone = rewriter.create<ConstantNoneOp>(loc);
+
+    Value dimList = rewriter.create<PrimListConstructOp>(
+        loc, Torch::ListType::get(Torch::IntType::get(op->getContext())),
+        keepDims);
+
+    Value norm = rewriter.create<AtenLinalgVectorNormOp>(
+        loc, v.getType(), v, ord, dimList, keepdim, dtypeNone);
+
+    auto vShape = rewriter.create<AtenSizeOp>(
+        loc, Torch::ListType::get(rewriter.getI64Type()), v);
+
+    Value gDivNorm =
+        rewriter.create<AtenDivTensorOp>(loc, g.getType(), g, norm);
+    Value broadcastedGDivNorm =
+        rewriter.create<AtenBroadcastToOp>(loc, v.getType(), gDivNorm, vShape);
+    Value vMulBroadcastedGDivNorm = rewriter.create<AtenMulTensorOp>(
+        loc, v.getType(), v, broadcastedGDivNorm);
+
+    rewriter.replaceOp(op, ArrayRef<Value>{vMulBroadcastedGDivNorm, norm});
+    return success();
+  }
+};
+} // namespace
+
 namespace {
 class DecomposeAtenNativeLayerNormOp
     : public OpRewritePattern<AtenNativeLayerNormOp> {
@@ -7194,7 +7251,6 @@ public:
     rewriter.replaceOpWithNewOp<AtenEmbeddingBagPaddingIdxOp>(
         op, returnTypes, weight, indices, offsets, scaleGradByFreq, mode,
         sparse, perSampleWeights, includeLastOffset, paddingIdx);
-
     return success();
   }
 };
@@ -8704,6 +8760,8 @@ public:
     legalOpsSet.clear();
     legalOpsSet.insert(legalOps.begin(), legalOps.end());
 
+    addPatternIfTargetOpIsIllegal<DecomposeAten_WeightNormInterfaceOp>(
+        patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenSoftmaxIntOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAten_SoftmaxOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAten_LogSoftmaxOp>(patterns);

lib/Dialect/Torch/Transforms/LowerToBackendContract.cpp

@@ -418,6 +418,7 @@ static void markDecomposedOpsAsIllegal(MLIRContext *context,
   });
   target.addIllegalOp<AtenAddcmulOp>();
   target.addIllegalOp<AtenAddcdivOp>();
+  target.addIllegalOp<Aten_WeightNormInterfaceOp>();
   target.addIllegalOp<AtenInstanceNormOp>();
   target.addIllegalOp<AtenLayerNormOp>();
   target.addIllegalOp<AtenNativeLayerNormOp>();

projects/pt1/e2e_testing/xfail_sets.py

@@ -469,6 +469,7 @@ FX_IMPORTER_XFAIL_SET = {
     "UpSampleNearest2dDynamicFactor_basic",
     "ViewCollapseDynamicWithAtenSizeIntModule_basic",
     "ViewSizeFromOtherTensor_basic",
+    "WeightNormInterfaceModule_basic",
 }
 
 FX_IMPORTER_CRASHING_SET = {
@@ -2629,6 +2630,7 @@ ONNX_XFAIL_SET = {
     "ViewNoChange1dModule_basic",
     "ViewNoChange2dModule_basic",
     "ViewNoChange3dModule_basic",
+    "WeightNormInterfaceModule_basic",
     "_Convolution2DAllFalseModule_basic",
     "_Convolution2DBenchmarkModule_basic",
     "_Convolution2DCudnnModule_basic",

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

@@ -1771,6 +1771,9 @@ def aten〇native_group_norm〡shape(input: List[int], weight: Optional[List[int
 def aten〇instance_norm〡shape(input: List[int], weight: Optional[List[int]], bias: Optional[List[int]], running_mean: Optional[List[int]], running_var: Optional[List[int]], use_input_stats: bool, momentum: float, eps: float, cudnn_enabled: bool) -> List[int]:
     return upstream_shape_functions.unary(input)
 
+def aten〇_weight_norm_interface〡shape(v: List[int], g: List[int], dim: int = 0) -> Tuple[List[int], List[int]]:
+    return upstream_shape_functions.unary(v), upstream_shape_functions.unary(g)
+
 def aten〇slice〇Tensor〡shape(self: List[int], dim: int = 0, start: Optional[int] = None, end: Optional[int] = None, step: int = 1) -> List[int]:
     return upstream_shape_functions.slice(self, dim, start, end, step)
 
@@ -2544,6 +2547,18 @@ def aten〇instance_norm〡dtype(input_rank_dtype: Tuple[int, int], weight_rank_
     input_rank, input_dtype = input_rank_dtype
     return input_dtype
 
+@check_dtype_function(_check_tensors_with_the_same_dtype(num_of_tensors=2, error_types={*all_integer_dtypes()}))
+def aten〇_weight_norm_interface〡dtype(v_rank_dtype: Tuple[int, int], g_rank_dtype: Tuple[int, int], dim: int = 0) -> Tuple[int, int]:
+    v_rank, v_dtype = v_rank_dtype
+    g_rank, g_dtype = g_rank_dtype
+    assert v_dtype == g_dtype
+    assert not is_integer_dtype(g_dtype)
+    if g_dtype == torch.complex128:
+        return v_dtype, torch.float64
+    elif g_dtype == torch.complex64:
+        return v_dtype, torch.float32
+    return v_dtype, g_dtype
+
 @check_dtype_function(_check_tensors_with_the_same_dtype(num_of_tensors=1))
 def aten〇bernoulli_〇float〡dtype(self_rank_dtype: Tuple[int, int], p: float = 0.5, generator: Any = None) -> int:
     self_rank, self_dtype = self_rank_dtype

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

@@ -732,6 +732,7 @@ def emit_ops(emitter_td: TextEmitter, registry: Registry):
         "aten::cosine_embedding_loss : (Tensor, Tensor, Tensor, float, int) -> (Tensor)"
     )
     emit("aten::diag_embed : (Tensor, int, int, int) -> (Tensor)")
+    emit("aten::_weight_norm_interface : (Tensor, Tensor, int) -> (Tensor, Tensor)")
 
     # Misc tensor ops.
     emit("aten::constant_pad_nd : (Tensor, int[], Scalar) -> (Tensor)")

projects/pt1/python/torch_mlir_e2e_test/test_suite/norm_like.py

@@ -726,3 +726,28 @@ class RenormModuleFloat32DynamicDims(torch.nn.Module):
 @register_test_case(module_factory=lambda: RenormModuleFloat32DynamicDims())
 def RenormModuleFloat32DynamicDims_basic(module, tu: TestUtils):
     module.forward(tu.rand(3, 4, 3))
+
+
+# ==============================================================================
+class WeightNormInterfaceModule(torch.nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.dim = 2
+
+    @export
+    @annotate_args(
+        [
+            None,
+            ([-1, -1, -1], torch.float32, True),
+            ([-1, -1, -1], torch.float32, True),
+        ]
+    )
+    def forward(self, v, g):
+        return torch.ops.aten._weight_norm_interface(v, g, self.dim)
+
+
+@register_test_case(module_factory=lambda: WeightNormInterfaceModule())
+def WeightNormInterfaceModule_basic(module, tu: TestUtils):
+    g = tu.rand(3, 10, 10)
+    v = tu.rand(1, 1, 10)
+    module.forward(g, v)