llvm
/
torch-mlir
mirror of https://github.com/llvm/torch-mlir
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Revert "E2E support for AtenRemainderScalarOp (#1119)" (#1190) 

This reverts commit 34e207eeb5.
pull/1192/head
powderluv 2022-08-08 22:59:57 -07:00 committed by GitHub
parent bb47c166a0
commit e55fc4deb5
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
8 changed files with 18 additions and 134 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
e2e_testing/torchscript/xfail_sets.py
View File

@ -179,9 +179,6 @@ TOSA_PASS_SET = {
"ArgmaxModule_keepDim",
"ArgmaxModule_with_dim",
"_LogSoftmaxModuleStable_basic",
"ElementwiseRemainderScalarModule_Int_Float_basic",
"ElementwiseRemainderScalarModule_Float_basic",
"ElementwiseRemainderScalarModule_Int_basic",
}
LTC_XFAIL_SET = {

24
include/torch-mlir/Dialect/Torch/IR/GeneratedTorchOps.td
View File

@ -7883,30 +7883,6 @@ def Torch_AtenRemainderIntOp : Torch_Op<"aten.remainder.int", [
let hasFolder = 1;
}
def Torch_AtenRemainderScalarOp : Torch_Op<"aten.remainder.Scalar", [
AllowsTypeRefinement,
HasValueSemantics,
ReadOnly
]> {
let summary = "Generated op for `aten::remainder.Scalar : (Tensor, Scalar) -> (Tensor)`";
let arguments = (ins
AnyTorchTensorType:$self,
AnyTorchScalarType:$other
);
let results = (outs
AnyTorchTensorType:$result
);
let hasCustomAssemblyFormat = 1;
let extraClassDefinition = [{
ParseResult AtenRemainderScalarOp::parse(OpAsmParser &parser, OperationState &result) {
return parseDefaultTorchOp(parser, result, 2, 1);
}
void AtenRemainderScalarOp::print(OpAsmPrinter &printer) {
printDefaultTorchOp(printer, *this, 2, 1);
}
}];
}
def Torch_AtenAddIntOp : Torch_Op<"aten.add.int", [
AllowsTypeRefinement,
HasValueSemantics,

49
lib/Conversion/TorchToLinalg/Uncategorized.cpp
View File

@ -803,23 +803,6 @@ static Value createLinalgPayloadCalculationForElementwiseOp(
Value other = convertScalarToDtype(b, loc, operands[1], dtype);
return b.create<arith::DivFOp>(loc, self, other);
}
if (auto remScalar = dyn_cast<AtenRemainderScalarOp>(op)) {
Type newResultType = converter->convertType(remScalar.getType())
.cast<RankedTensorType>()
.getElementType();
Value self = convertScalarToDtype(b, loc, payloadArgs[0], newResultType);
Value other = convertScalarToDtype(b, loc, operands[1], newResultType);
Value result;
if (newResultType.isa<mlir::FloatType>()) {
result = b.create<arith::RemFOp>(loc, self, other);
} else {
result = b.create<arith::RemSIOp>(loc, self, other);
}
return result;
}
if (auto reciprocal = dyn_cast<AtenReciprocalOp>(op)) {
Type dtype = converter->convertType(reciprocal.getType())
.cast<RankedTensorType>()
@ -960,14 +943,14 @@ public:
AtenMinimumOp, AtenMaximumOp, AtenToDtypeOp, AtenClampOp,
AtenRsubScalarOp, AtenMulScalarOp, AtenLogOp, AtenErfOp,
AtenSqrtOp, AtenFloorOp, AtenPowTensorScalarOp, AtenLog2Op,
AtenLog1pOp, AtenRsqrtOp, AtenDivScalarOp, AtenRemainderScalarOp,
AtenAbsOp, AtenReciprocalOp, AtenBitwiseAndTensorOp,
AtenGtScalarOp, AtenGeScalarOp, AtenEqScalarOp, AtenLtScalarOp,
AtenLeScalarOp, AtenWhereSelfOp, AtenCeilOp, AtenGtTensorOp,
AtenEqTensorOp, AtenLtTensorOp, AtenSubScalarOp, AtenAddScalarOp,
AtenThresholdOp, AtenThresholdBackwardOp, AtenCloneOp, AtenSinOp,
AtenCosOp, AtenNeScalarOp, AtenNegOp, AtenMaskedFillScalarOp,
AtenLogicalOrOp, AtenTriuOp>(op))
AtenLog1pOp, AtenRsqrtOp, AtenDivScalarOp, AtenAbsOp,
AtenReciprocalOp, AtenBitwiseAndTensorOp, AtenGtScalarOp,
AtenGeScalarOp, AtenEqScalarOp, AtenLtScalarOp, AtenLeScalarOp,
AtenWhereSelfOp, AtenCeilOp, AtenGtTensorOp, AtenEqTensorOp,
AtenLtTensorOp, AtenSubScalarOp, AtenAddScalarOp, AtenThresholdOp,
AtenThresholdBackwardOp, AtenCloneOp, AtenSinOp, AtenCosOp,
AtenNeScalarOp, AtenNegOp, AtenMaskedFillScalarOp, AtenLogicalOrOp,
AtenTriuOp>(op))
return rewriter.notifyMatchFailure(op, "not a supported elementwise op");
if (failed(verifyLinalgCompatibleTypes(op, rewriter)))
@ -1698,14 +1681,14 @@ void mlir::torch::torch_to_linalg::populateUncategorizedPatternsAndLegality(
AtenTanhOp, AtenReluOp, AtenLeakyReluOp, AtenGeluOp, AtenGeluBackwardOp,
AtenAddTensorOp, AtenMulTensorOp, AtenDivTensorOp, AtenDivTensorModeOp,
AtenSubTensorOp, AtenLerpTensorOp, AtenSigmoidOp, AtenMinimumOp,
AtenMaximumOp, AtenToDtypeOp, AtenClampOp, AtenRsubScalarOp, AtenLogOp,
AtenErfOp, AtenSqrtOp, AtenFloorOp, AtenCeilOp, AtenPowTensorScalarOp,
AtenLog2Op, AtenLog1pOp, AtenRsqrtOp, AtenAbsOp, AtenReciprocalOp,
AtenBitwiseAndTensorOp, AtenGtScalarOp, AtenGeScalarOp, AtenEqScalarOp,
AtenLtScalarOp, AtenLeScalarOp, AtenWhereSelfOp, AtenGtTensorOp,
AtenEqTensorOp, AtenLtTensorOp, AtenThresholdOp, AtenThresholdBackwardOp,
AtenCloneOp, AtenSinOp, AtenCosOp, AtenNeScalarOp, AtenMaskedFillScalarOp,
AtenLogicalOrOp, AtenTriuOp, AtenRemainderScalarOp>();
AtenAtan2Op, AtenMaximumOp, AtenToDtypeOp, AtenClampOp, AtenRsubScalarOp,
AtenLogOp, AtenErfOp, AtenSqrtOp, AtenFloorOp, AtenCeilOp,
AtenPowTensorScalarOp, AtenLog2Op, AtenLog1pOp, AtenRsqrtOp, AtenAbsOp,
AtenReciprocalOp, AtenBitwiseAndTensorOp, AtenGtScalarOp, AtenGeScalarOp,
AtenEqScalarOp, AtenLtScalarOp, AtenLeScalarOp, AtenWhereSelfOp,
AtenGtTensorOp, AtenEqTensorOp, AtenLtTensorOp, AtenThresholdOp,
AtenThresholdBackwardOp, AtenCloneOp, AtenSinOp, AtenCosOp,
AtenNeScalarOp, AtenMaskedFillScalarOp, AtenLogicalOrOp, AtenTriuOp>();
patterns.add<ConvertElementwiseOp>(typeConverter, context);
target.addIllegalOp<AtenNllLossForwardOp>();
patterns.add<ConvertAtenDetachOp>(typeConverter, context);

2
lib/Dialect/Torch/Transforms/RefineTypes.cpp
View File

@ -770,7 +770,7 @@ void TypeAnalysis::visitOperation(Operation *op,
// Promote LHS with scalar RHS.
if (isa<AtenAddScalarOp, AtenSubScalarOp, AtenMulScalarOp, AtenDivScalarOp,
AtenFmodScalarOp, AtenFloorDivideScalarOp, AtenPowTensorScalarOp,
AtenRsubScalarOp, AtenLeakyReluOp, AtenRemainderScalarOp>(op)) {
AtenRsubScalarOp, AtenLeakyReluOp>(op)) {
auto lhs = operands[0]->getValue();
Value scalar = op->getOperand(1);
auto knowledge =

4
lib/Dialect/Torch/Transforms/ShapeLibrary.cpp
View File

@ -5441,10 +5441,6 @@ module {
%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.remainder.Scalar"(%arg0: !torch.list<int>, %arg1: !torch.float) -> !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.to.dtype"(%arg0: !torch.list<int>, %arg1: !torch.int, %arg2: !torch.bool, %arg3: !torch.bool, %arg4: !torch.optional<int>) -> !torch.list<int> {
%0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>
return %0 : !torch.list<int>

3
python/torch_mlir/dialects/torch/importer/jit_ir/build_tools/shape_lib_gen.py
View File

@ -478,9 +478,6 @@ def aten〇mul〇Scalar(self: List[int], other: float) -> List[int]:
def aten〇div〇Scalar(self: List[int], other: float) -> List[int]:
return upstream_shape_functions.unary(self)
def aten〇remainder〇Scalar(self: List[int], other: float) -> List[int]:
return upstream_shape_functions.unary(self)
def aten〇floor_divide〇Scalar(self: List[int], other: float) -> List[int]:
return upstream_shape_functions.unary(self)

1
python/torch_mlir/dialects/torch/importer/jit_ir/build_tools/torch_ods_gen.py
View File

@ -549,7 +549,6 @@ def emit_ops(emitter_td: TextEmitter, registry: Registry):
emit("aten::eq.int : (int, int) -> (bool)", has_folder=True)
emit("aten::floordiv.int : (int, int) -> (int)", has_folder=True)
emit("aten::remainder.int : (int, int) -> (int)", has_folder=True)
emit("aten::remainder.Scalar : (Tensor, Scalar) -> (Tensor)")
emit("aten::add.int : (int, int) -> (int)", has_folder=True)
emit("aten::sub.int : (int, int) -> (int)", has_folder=True)
emit("aten::mul.int : (int, int) -> (int)", has_folder=True)

64
python/torch_mlir_e2e_test/test_suite/elementwise.py
View File

@ -1304,70 +1304,6 @@ class ElementwiseDivScalarModule(torch.nn.Module):
def ElementwiseDivScalarModule_basic(module, tu: TestUtils):
module.forward(tu.rand(3, 4))
# ==============================================================================
class ElementwiseRemainderScalarModule_Int_Float(torch.nn.Module):
def __init__(self):
super().__init__()
@export
@annotate_args([
None,
([-1], torch.int64, True),
])
def forward(self, x):
return torch.remainder(x, 2.0)
@register_test_case(module_factory=lambda: ElementwiseRemainderScalarModule_Int_Float())
def ElementwiseRemainderScalarModule_Int_Float_basic(module, tu: TestUtils):
module.forward(torch.randint(10, (3,)))
# ==============================================================================
class ElementwiseRemainderScalarModule_Float(torch.nn.Module):
def __init__(self):
super().__init__()
@export
@annotate_args([
None,
([-1, -1], torch.float, True),
])
def forward(self, x):
return torch.remainder(x, 2.0)
@register_test_case(module_factory=lambda: ElementwiseRemainderScalarModule_Float())
def ElementwiseRemainderScalarModule_Float_basic(module, tu: TestUtils):
module.forward(torch.rand(10, 3))
# ==============================================================================
class ElementwiseRemainderScalarModule_Int(torch.nn.Module):
def __init__(self):
super().__init__()
@export
@annotate_args([
None,
([-1, -1], torch.int64, True),
])
def forward(self, x):
return torch.remainder(x, 2)
@register_test_case(module_factory=lambda: ElementwiseRemainderScalarModule_Int())
def ElementwiseRemainderScalarModule_Int_basic(module, tu: TestUtils):
module.forward(torch.randint(10, (3, 2)))
# ==============================================================================