[MLIR][TORCH] Add E2E support for aten.bitwise_not op

This commit adds lowering of `aten.bitwise_not` op.

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

e2e_testing/xfail_sets.py

@@ -254,6 +254,8 @@ TOSA_PASS_SET = {
     "ElementwiseMulScalarModule_float",
     "ElementwiseCeilModule_basic",
     "ElementwiseReciprocalModule_basic",
+    "ElementwiseNotIntegerModule_basic",
+    "ElementwiseNotInt32Module_basic",
     "TypePromotionAlphaWiderModule_basic",
     "Conv2dWithPaddingDilationStrideStaticModule_basic",
     "BatchNorm1DModule_basic",

lib/Conversion/TorchToLinalg/Uncategorized.cpp

@@ -23,6 +23,7 @@
 #include "torch-mlir/Dialect/Torch/IR/TorchOps.h"
 #include "torch-mlir/Dialect/Torch/Utils/TorchUpstream.h"
 #include "torch-mlir/Dialect/Torch/Utils/Utils.h"
+#include "llvm/ADT/APSInt.h"
 
 using namespace mlir;
 using namespace mlir::torch;
@@ -948,6 +949,22 @@ static Value createLinalgPayloadCalculationForElementwiseOp(
     return b.create<arith::SelectOp>(loc, pred, scalar, zero);
   }
 
+  if (auto bitwiseNot = dyn_cast<AtenBitwiseNotOp>(op)) {
+    Type elementType = converter->convertType(bitwiseNot.getType())
+                           .cast<RankedTensorType>()
+                           .getElementType();
+    if (elementType.isa<mlir::FloatType>()) {
+      bitwiseNot.emitError("Bitwise_Not does not support floating point dtype");
+      return nullptr;
+    }
+
+    Value allOnesVal = b.create<arith::ConstantOp>(
+        loc, b.getIntegerAttr(
+                 elementType,
+                 APSInt::getAllOnesValue(elementType.getIntOrFloatBitWidth())));
+    return b.create<arith::XOrIOp>(loc, payloadArgs[0], allOnesVal);
+  }
+
   op->emitError("unimplemented lowering in "
                 "createLinalgPayloadCalculationForElementwiseOp");
   return nullptr;
@@ -995,7 +1012,8 @@ public:
              AtenLtTensorOp, AtenSubScalarOp, AtenAddScalarOp, AtenThresholdOp,
              AtenThresholdBackwardOp, AtenCloneOp, AtenSinOp, AtenCosOp,
              AtenNeScalarOp, AtenNegOp, AtenMaskedFillScalarOp,
-             AtenMaskedFillTensorOp, AtenLogicalOrOp, AtenTriuOp>(op))
+             AtenMaskedFillTensorOp, AtenLogicalOrOp, AtenTriuOp,
+             AtenBitwiseNotOp>(op))
       return rewriter.notifyMatchFailure(op, "not a supported elementwise op");
 
     if (failed(verifyLinalgCompatibleTypes(op, rewriter)))
@@ -1470,7 +1488,7 @@ void mlir::torch::torch_to_linalg::populateUncategorizedPatternsAndLegality(
       AtenLtTensorOp, AtenThresholdOp, AtenThresholdBackwardOp, AtenCloneOp,
       AtenSinOp, AtenCosOp, AtenNeScalarOp, AtenMaskedFillScalarOp,
       AtenMaskedFillTensorOp, AtenLogicalOrOp, AtenTriuOp,
-      AtenRemainderScalarOp>();
+      AtenRemainderScalarOp, AtenBitwiseNotOp>();
   patterns.add<ConvertElementwiseOp>(typeConverter, context);
   target.addIllegalOp<AtenNllLossForwardOp>();
   patterns.add<ConvertAtenDetachOp>(typeConverter, context);

lib/Dialect/Torch/Transforms/ShapeLibrary.cpp

@@ -6369,6 +6369,10 @@ module {
     %0 = call @__torch__.torch.jit._shape_functions.broadcast(%arg0, %arg1) : (!torch.list<int>, !torch.list<int>) -> !torch.list<int>
     return %0 : !torch.list<int>
   }
+  func.func @"__torch_mlir_shape_fn.aten.bitwise_not"(%arg0: !torch.list<int>) -> !torch.list<int> {
+    %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>
+    return %0 : !torch.list<int>
+  }
   func.func @"__torch_mlir_shape_fn.aten.logical_or"(%arg0: !torch.list<int>, %arg1: !torch.list<int>) -> !torch.list<int> {
     %0 = call @__torch__.torch.jit._shape_functions.broadcast(%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

@@ -860,6 +860,9 @@ def aten〇maximum(self: List[int], other: List[int]) -> List[int]:
 def aten〇bitwise_and〇Tensor(self: List[int], other: List[int]) -> List[int]:
     return upstream_shape_functions.broadcast(self, other)
 
+def aten〇bitwise_not(self: List[int]) -> List[int]:
+    return upstream_shape_functions.unary(self)
+
 def aten〇logical_or(self: List[int], other: List[int]) -> List[int]:
     return upstream_shape_functions.broadcast(self, other)
 

python/torch_mlir_e2e_test/test_suite/elementwise.py

@@ -1531,6 +1531,50 @@ def ElementwiseAndIntegerModule_basic(module, tu: TestUtils):
 # ==============================================================================
 
 
+class ElementwiseNotIntegerModule(torch.nn.Module):
+
+    def __init__(self):
+        super().__init__()
+
+    @export
+    @annotate_args([
+        None,
+        ([-1, -1], torch.int64, True),
+    ])
+    def forward(self, x):
+        return torch.bitwise_not(x)
+
+
+@register_test_case(module_factory=lambda: ElementwiseNotIntegerModule())
+def ElementwiseNotIntegerModule_basic(module, tu: TestUtils):
+    module.forward(tu.randint(3, 4, low=-10, high=10))
+
+
+# ==============================================================================
+
+
+class ElementwiseNotInt32Module(torch.nn.Module):
+
+    def __init__(self):
+        super().__init__()
+
+    @export
+    @annotate_args([
+        None,
+        ([-1, -1], torch.int32, True),
+    ])
+    def forward(self, x):
+        return torch.bitwise_not(x)
+
+
+@register_test_case(module_factory=lambda: ElementwiseNotInt32Module())
+def ElementwiseNotInt32Module_basic(module, tu: TestUtils):
+    module.forward(tu.randint(3, 4, low=-10, high=10).to(torch.int32))
+
+
+# ==============================================================================
+
+
 class ElementwiseSubScalarIntModule(torch.nn.Module):
 
     def __init__(self):