[MLIR][TORCH] Add decomposition of aten.numpy_T op

This commit adds the decomposition of `aten.numpy_T` op into
`aten.t` or `aten.permute` op.

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

e2e_testing/torchscript/xfail_sets.py

@@ -160,4 +160,8 @@ TOSA_PASS_SET = {
     "BaddbmmWithBetaModule_basic",
     "BaddbmmBroadcast1DInputModule_basic",
     "BaddbmmBroadcast2DInputModule_basic",
+    "NumpyTRank1Module_basic",
+    "NumpyTRank2Module_basic",
+    "NumpyTRankNStaticModule_basic",
+    "NumpyTRankNDynamicModule_basic",
 }

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

@@ -6002,6 +6002,28 @@ def Torch_AtenTOp : Torch_Op<"aten.t", [
   }];
 }
 
+def Torch_AtenNumpyTOp : Torch_Op<"aten.numpy_T", [
+    AllowsTypeRefinement,
+    ReadOnly
+  ]> {
+  let summary = "Generated op for `aten::numpy_T : (Tensor) -> (Tensor)`";
+  let arguments = (ins
+    AnyTorchTensorType:$self
+  );
+  let results = (outs
+    AnyTorchTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenNumpyTOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 1, 1);
+    }
+    void AtenNumpyTOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 1, 1);
+    }
+  }];
+}
+
 def Torch_AtenFullOp : Torch_Op<"aten.full", [
     AllowsTypeRefinement,
     HasValueSemantics,

lib/Dialect/Torch/Transforms/DecomposeComplexOps.cpp

@@ -1957,6 +1957,29 @@ class DecomposeAtenFloorDivideOp : public OpRewritePattern<AtenFloorDivideOp> {
 };
 } // namespace
 
+namespace {
+// Decompose `aten.numpy_T` op into `aten.permute` op.
+class DecomposeAtenNumpyTOp : public OpRewritePattern<AtenNumpyTOp> {
+  using OpRewritePattern::OpRewritePattern;
+  LogicalResult matchAndRewrite(AtenNumpyTOp op,
+                                PatternRewriter &rewriter) const override {
+    Location loc = op.getLoc();
+    Value self = op.self();
+    int64_t inputRank = getTensorRank(self);
+
+    SmallVector<Value> dimListElements;
+    for (int64_t i = inputRank - 1; i >= 0; i--)
+      dimListElements.push_back(rewriter.create<Torch::ConstantIntOp>(
+          loc, rewriter.getI64IntegerAttr(i)));
+    Value dimList = rewriter.create<PrimListConstructOp>(
+        loc, Torch::ListType::get(Torch::IntType::get(op->getContext())),
+        dimListElements);
+    rewriter.replaceOpWithNewOp<AtenPermuteOp>(op, op.getType(), self, dimList);
+    return success();
+  }
+};
+} // namespace
+
 namespace {
 class DecomposeComplexOpsPass
     : public DecomposeComplexOpsBase<DecomposeComplexOpsPass> {
@@ -2102,6 +2125,8 @@ class DecomposeComplexOpsPass
     target.addIllegalOp<AtenBaddbmmOp>();
     patterns.add<DecomposeAtenFloorDivideOp>(context);
     target.addIllegalOp<AtenFloorDivideOp>();
+    patterns.add<DecomposeAtenNumpyTOp>(context);
+    target.addIllegalOp<AtenNumpyTOp>();
 
     if (failed(applyPartialConversion(getOperation(), target,
                                       std::move(patterns)))) {

lib/Dialect/Torch/Transforms/MaximizeValueSemantics.cpp

@@ -37,7 +37,7 @@ static bool isViewLikeOp(Operation *op) {
              Aten_ReshapeAliasOp, AtenSelectIntOp, AtenSliceTensorOp,
              AtenSqueezeDimOp, AtenSqueezeOp, AtenTOp, AtenToDtypeOp,
              AtenTransposeIntOp, AtenUnsqueezeOp, AtenViewOp,
-             TensorStaticInfoCastOp, AtenToDtypeLayoutOp>(op);
+             TensorStaticInfoCastOp, AtenToDtypeLayoutOp, AtenNumpyTOp>(op);
 }
 
 namespace {

lib/Dialect/Torch/Transforms/RefineTypes.cpp

@@ -642,7 +642,7 @@ ChangeResult TypeAnalyzer::visitOperation(
           AtenZero_Op, AtenIndexTensorOp, ValsemVariantAtenIndexPutImplOp,
           AtenIndexPutOp, ValsemVariantAtenCopyOp, AtenZeroFunctionalOp,
           AtenIndexPutHackedTwinOp, AtenMaskedFillScalarOp, AtenFlipOp,
-          PrimAbsScalarOp>(op)) {
+          PrimAbsScalarOp, AtenNumpyTOp>(op)) {
     return incorporateKnowledge(op->getResult(0), operands[0]->getValue());
   }
 

lib/Dialect/Torch/Transforms/ShapeLibrary.cpp

@@ -5628,6 +5628,19 @@ module {
     %0 = call @__torch__.torch.jit._shape_functions.transpose(%arg0, %int0, %int1) : (!torch.list<int>, !torch.int, !torch.int) -> !torch.list<int>
     return %0 : !torch.list<int>
   }
+  func.func @"__torch_mlir_shape_fn.aten.numpy_T"(%arg0: !torch.list<int>) -> !torch.list<int> {
+    %int0 = torch.constant.int 0
+    %true = torch.constant.bool true
+    %0 = torch.prim.ListConstruct  : () -> !torch.list<int>
+    %1 = torch.aten.len.t %arg0 : !torch.list<int> -> !torch.int
+    torch.prim.Loop %1, %true, init() {
+    ^bb0(%arg1: !torch.int):
+      %2 = torch.aten.__getitem__.t %arg0, %arg1 : !torch.list<int>, !torch.int -> !torch.int
+      torch.aten.insert.t %0, %int0, %2 : !torch.list<int>, !torch.int, !torch.int
+      torch.prim.Loop.condition %true, iter()
+    } : (!torch.int, !torch.bool) -> ()
+    return %0 : !torch.list<int>
+  }
   func.func @"__torch_mlir_shape_fn.aten.matmul"(%arg0: !torch.list<int>, %arg1: !torch.list<int>) -> !torch.list<int> {
     %0 = call @__torch__.torch.jit._shape_functions.matmul(%arg0, %arg1) : (!torch.list<int>, !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

@@ -529,6 +529,12 @@ def aten〇transpose〇int(self: List[int], dim0: int, dim1: int) -> List[int]:
 def aten〇t(self: List[int]) -> List[int]:
     return upstream_shape_functions.transpose(self, 0, 1)
 
+def aten〇numpy_T(self: List[int]) -> List[int]:
+    result_shape: List[int] = []
+    for i in self:
+        result_shape.insert(0, i)
+    return result_shape
+
 def aten〇matmul(self: List[int], other: List[int]) -> List[int]:
     return upstream_shape_functions.matmul(self, other)
 

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

@@ -465,6 +465,7 @@ def emit_ops(emitter_td: TextEmitter, registry: Registry):
     emit_with_mutating_variants("aten::dropout : (Tensor, float, bool) -> (Tensor)")
     emit("aten::native_dropout : (Tensor, float, bool?) -> (Tensor, Tensor)")
     emit("aten::t : (Tensor) -> (Tensor)")
+    emit("aten::numpy_T : (Tensor) -> (Tensor)")
     emit("aten::full : (int[], Scalar, int?, int?, Device?, bool?) -> (Tensor)")
     emit("aten::full_like : (Tensor, Scalar, int?, int?, Device?, bool?, int?) -> (Tensor)")
     emit_with_mutating_variants("aten::baddbmm : (Tensor, Tensor, Tensor, Scalar, Scalar) -> (Tensor)")

python/torch_mlir_e2e_test/test_suite/basic.py

@@ -2167,3 +2167,101 @@ class BaddbmmBroadcast2DInputModule(torch.nn.Module):
 @register_test_case(module_factory=lambda: BaddbmmBroadcast2DInputModule())
 def BaddbmmBroadcast2DInputModule_basic(module, tu: TestUtils):
     module.forward(tu.rand(2, 7), tu.rand(5, 2, 9), tu.rand(5, 9, 7))
+
+
+# ==============================================================================
+
+
+class NumpyTRankNStaticModule(torch.nn.Module):
+
+    def __init__(self):
+        super().__init__()
+
+    @export
+    @annotate_args([
+        None,
+        ([3, 4, 5, 6], torch.float32, True),
+    ])
+    def forward(self, lhs):
+        return torch.ops.aten.numpy_T(lhs)
+
+
+@register_test_case(module_factory=lambda: NumpyTRankNStaticModule())
+def NumpyTRankNStaticModule_basic(module, tu: TestUtils):
+    module.forward(tu.rand(3, 4, 5, 6))
+
+
+class NumpyTRankNDynamicModule(torch.nn.Module):
+
+    def __init__(self):
+        super().__init__()
+
+    @export
+    @annotate_args([
+        None,
+        ([-1, -1, -1, -1, -1], torch.float32, True),
+    ])
+    def forward(self, lhs):
+        return torch.ops.aten.numpy_T(lhs)
+
+
+@register_test_case(module_factory=lambda: NumpyTRankNDynamicModule())
+def NumpyTRankNDynamicModule_basic(module, tu: TestUtils):
+    module.forward(tu.rand(3, 4, 5, 6, 2))
+
+
+class NumpyTRank2Module(torch.nn.Module):
+
+    def __init__(self):
+        super().__init__()
+
+    @export
+    @annotate_args([
+        None,
+        ([-1, -1], torch.float32, True),
+    ])
+    def forward(self, lhs):
+        return torch.ops.aten.numpy_T(lhs)
+
+
+@register_test_case(module_factory=lambda: NumpyTRank2Module())
+def NumpyTRank2Module_basic(module, tu: TestUtils):
+    module.forward(tu.rand(3, 4))
+
+
+class NumpyTRank1Module(torch.nn.Module):
+
+    def __init__(self):
+        super().__init__()
+
+    @export
+    @annotate_args([
+        None,
+        ([-1], torch.float32, True),
+    ])
+    def forward(self, lhs):
+        return torch.ops.aten.numpy_T(lhs)
+
+
+@register_test_case(module_factory=lambda: NumpyTRank1Module())
+def NumpyTRank1Module_basic(module, tu: TestUtils):
+    module.forward(tu.rand(3))
+
+
+class NumpyTRank0Module(torch.nn.Module):
+
+    def __init__(self):
+        super().__init__()
+
+    @export
+    @annotate_args([
+        None,
+        ([], torch.float32, True),
+    ])
+    def forward(self, lhs):
+        return torch.ops.aten.numpy_T(lhs)
+
+
+@register_test_case(module_factory=lambda: NumpyTRank0Module())
+def NumpyTRank0Module_basic(module, tu: TestUtils):
+    module.forward(torch.tensor(7, dtype=torch.float32))

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

@@ -1036,3 +1036,30 @@ func.func @torch.aten.floor_divide(%arg0: !torch.vtensor<[?,?],f32>, %arg1: !tor
   %0 = torch.aten.floor_divide %arg0, %arg1 : !torch.vtensor<[?,?],f32>, !torch.vtensor<[?,?],f32> -> !torch.vtensor<[?,?],f32>
   return %0 : !torch.vtensor<[?,?],f32>
 }
+
+// -----
+// CHECK-LABEL:   func @torch.aten.numpy_T$rank_two(
+// CHECK-SAME:                  %[[SELF:.*]]: !torch.vtensor<[5,4],f32>) -> !torch.vtensor<[4,5],f32> {
+// CHECK:           %[[CST1:.*]] = torch.constant.int 1
+// CHECK:           %[[CST0:.*]] = torch.constant.int 0
+// CHECK:           %[[DIMS:.*]] = torch.prim.ListConstruct %[[CST1]], %[[CST0]] : (!torch.int, !torch.int) -> !torch.list<int>
+// CHECK:           %[[OUT:.*]] = torch.aten.permute %[[SELF]], %[[DIMS]] : !torch.vtensor<[5,4],f32>, !torch.list<int> -> !torch.vtensor<[4,5],f32>
+// CHECK:           return %[[OUT]] : !torch.vtensor<[4,5],f32>
+func.func @torch.aten.numpy_T$rank_two(%arg0: !torch.vtensor<[5,4],f32>) -> !torch.vtensor<[4,5],f32> {
+  %0 = torch.aten.numpy_T %arg0 : !torch.vtensor<[5,4],f32> -> !torch.vtensor<[4,5],f32>
+  return %0 : !torch.vtensor<[4,5],f32>
+}
+
+// -----
+// CHECK-LABEL:   func @torch.aten.numpy_T$rank_three(
+// CHECK-SAME:                  %[[SELF:.*]]: !torch.vtensor<[5,4,3],f32>) -> !torch.vtensor<[3,4,5],f32> {
+// CHECK:           %[[CST2:.*]] = torch.constant.int 2
+// CHECK:           %[[CST1:.*]] = torch.constant.int 1
+// CHECK:           %[[CST0:.*]] = torch.constant.int 0
+// CHECK:           %[[DIMS:.*]] = torch.prim.ListConstruct %[[CST2]], %[[CST1]], %[[CST0]] : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
+// CHECK:           %[[OUT:.*]] = torch.aten.permute %[[SELF]], %[[DIMS]] : !torch.vtensor<[5,4,3],f32>, !torch.list<int> -> !torch.vtensor<[3,4,5],f32>
+// CHECK:           return %[[OUT]] : !torch.vtensor<[3,4,5],f32>
+func.func @torch.aten.numpy_T$rank_three(%arg0: !torch.vtensor<[5,4,3],f32>) -> !torch.vtensor<[3,4,5],f32> {
+  %0 = torch.aten.numpy_T %arg0 : !torch.vtensor<[5,4,3],f32> -> !torch.vtensor<[3,4,5],f32>
+  return %0 : !torch.vtensor<[3,4,5],f32>
+}