[MLIR][TORCH] Add decomposition for prims.var and prims.sqrt op

Signed-Off By: Vivek Khandelwal <vivek@nod-labs.com>
pull/1794/head
Vivek Khandelwal 2023-01-11 11:31:45 +05:30
parent b966733e04
commit fd236b2c89
8 changed files with 109 additions and 2 deletions

View File

@ -83,6 +83,8 @@ TORCHDYNAMO_XFAIL_SET = {
# error: unsupported by backend contract: tensor with unknown rank # error: unsupported by backend contract: tensor with unknown rank
# note: see current operation: %1 = "torch.tensor_static_info_cast"(%arg0) : (!torch.vtensor<[5,4,3,2,1],f32>) -> !torch.vtensor<*,f32> # note: see current operation: %1 = "torch.tensor_static_info_cast"(%arg0) : (!torch.vtensor<[5,4,3,2,1],f32>) -> !torch.vtensor<*,f32>
"ElementwisePreluModule_basic", "ElementwisePreluModule_basic",
# error: op lowering missing. Issue: https://github.com/llvm/torch-mlir/issues/1792
"StdCorrectionKeepDimModule_basic",
} }
MHLO_PASS_SET = { MHLO_PASS_SET = {

View File

@ -10881,6 +10881,55 @@ def Torch_PrimsConvertElementTypeOp : Torch_Op<"prims.convert_element_type", [
}]; }];
} }
def Torch_PrimsVarOp : Torch_Op<"prims.var", [
AllowsTypeRefinement,
HasValueSemantics,
ReadOnly
]> {
let summary = "Generated op for `prims::var : (Tensor, int[]?, int, int?) -> (Tensor)`";
let arguments = (ins
AnyTorchTensorType:$inp,
AnyTorchOptionalListOfTorchIntType:$dims,
Torch_IntType:$correction,
AnyTorchOptionalIntType:$output_dtype
);
let results = (outs
AnyTorchTensorType:$result
);
let hasCustomAssemblyFormat = 1;
let extraClassDefinition = [{
ParseResult PrimsVarOp::parse(OpAsmParser &parser, OperationState &result) {
return parseDefaultTorchOp(parser, result, 4, 1);
}
void PrimsVarOp::print(OpAsmPrinter &printer) {
printDefaultTorchOp(printer, *this, 4, 1);
}
}];
}
def Torch_PrimsSqrtOp : Torch_Op<"prims.sqrt", [
AllowsTypeRefinement,
HasValueSemantics,
ReadOnly
]> {
let summary = "Generated op for `prims::sqrt : (Tensor) -> (Tensor)`";
let arguments = (ins
AnyTorchTensorType:$self
);
let results = (outs
AnyTorchTensorType:$result
);
let hasCustomAssemblyFormat = 1;
let extraClassDefinition = [{
ParseResult PrimsSqrtOp::parse(OpAsmParser &parser, OperationState &result) {
return parseDefaultTorchOp(parser, result, 1, 1);
}
void PrimsSqrtOp::print(OpAsmPrinter &printer) {
printDefaultTorchOp(printer, *this, 1, 1);
}
}];
}
def Torch_QuantizedLinearOp : Torch_Op<"quantized.linear", [ def Torch_QuantizedLinearOp : Torch_Op<"quantized.linear", [
HasValueSemantics, HasValueSemantics,
AllowsTypeRefinement, AllowsTypeRefinement,

View File

@ -5768,6 +5768,10 @@ StringRef mlir::torch::Torch::getAbstractInterpLibrary() {
" %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>\n" " %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>\n"
" return %0 : !torch.list<int>\n" " return %0 : !torch.list<int>\n"
" }\n" " }\n"
" func.func @\"__torch_mlir_shape_fn.prims.sqrt\"(%arg0: !torch.list<int>) -> !torch.list<int> {\n"
" %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.neg\"(%arg0: !torch.list<int>) -> !torch.list<int> {\n" " func.func @\"__torch_mlir_shape_fn.aten.neg\"(%arg0: !torch.list<int>) -> !torch.list<int> {\n"
" %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>\n" " %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>\n"
" return %0 : !torch.list<int>\n" " return %0 : !torch.list<int>\n"
@ -6032,6 +6036,13 @@ StringRef mlir::torch::Torch::getAbstractInterpLibrary() {
" %0 = torch.prim.ListConstruct : () -> !torch.list<int>\n" " %0 = torch.prim.ListConstruct : () -> !torch.list<int>\n"
" return %0 : !torch.list<int>\n" " return %0 : !torch.list<int>\n"
" }\n" " }\n"
" func.func @\"__torch_mlir_shape_fn.prims.var\"(%arg0: !torch.list<int>, %arg1: !torch.optional<list<int>>, %arg2: !torch.int, %arg3: !torch.optional<int>) -> !torch.list<int> {\n"
" %none = torch.constant.none\n"
" %false = torch.constant.bool false\n"
" %0 = torch.derefine %none : !torch.none to !torch.any\n"
" %1 = call @__torch__.torch.jit._shape_functions.sum_mean_dim(%arg0, %arg1, %false, %0) : (!torch.list<int>, !torch.optional<list<int>>, !torch.bool, !torch.any) -> !torch.list<int>\n"
" return %1 : !torch.list<int>\n"
" }\n"
" func.func @\"__torch_mlir_shape_fn.aten.var.dim\"(%arg0: !torch.list<int>, %arg1: !torch.optional<list<int>>, %arg2: !torch.bool, %arg3: !torch.bool) -> !torch.list<int> {\n" " func.func @\"__torch_mlir_shape_fn.aten.var.dim\"(%arg0: !torch.list<int>, %arg1: !torch.optional<list<int>>, %arg2: !torch.bool, %arg3: !torch.bool) -> !torch.list<int> {\n"
" %none = torch.constant.none\n" " %none = torch.constant.none\n"
" %0 = torch.derefine %none : !torch.none to !torch.any\n" " %0 = torch.derefine %none : !torch.none to !torch.any\n"

View File

@ -3425,6 +3425,38 @@ public:
}; };
} // namespace } // namespace
namespace {
// Decompose `prims.var` op into `aten.var.correction` op.
class DecomposePrimsVarOp : public OpRewritePattern<PrimsVarOp> {
public:
using OpRewritePattern::OpRewritePattern;
LogicalResult matchAndRewrite(PrimsVarOp op,
PatternRewriter &rewriter) const override {
if (!op.getOutputDtype().getType().isa<Torch::NoneType>())
return rewriter.notifyMatchFailure(
op, "Unimplemented non-None dtype for prims::var op");
Value cstFalse = rewriter.create<Torch::ConstantBoolOp>(op.getLoc(), false);
rewriter.replaceOpWithNewOp<AtenVarCorrectionOp>(
op, op.getType(), op.getInp(), op.getDims(), op.getCorrection(),
/*keepdim=*/cstFalse);
return success();
}
};
} // namespace
namespace {
// Decompose `prims.sqrt` op into `aten.sqrt` op.
class DecomposePrimsSqrtOp : public OpRewritePattern<PrimsSqrtOp> {
public:
using OpRewritePattern::OpRewritePattern;
LogicalResult matchAndRewrite(PrimsSqrtOp op,
PatternRewriter &rewriter) const override {
rewriter.replaceOpWithNewOp<AtenSqrtOp>(op, op.getType(), op.getSelf());
return success();
}
};
} // namespace
namespace { namespace {
// The op is decomposed using the Box-Muller transform. // The op is decomposed using the Box-Muller transform.
// Refer: https://en.wikipedia.org/wiki/Box-Muller_transform // Refer: https://en.wikipedia.org/wiki/Box-Muller_transform
@ -3659,6 +3691,8 @@ public:
addPatternIfTargetOpIsIllegal<DecomposeAtenRandintLowOp>(patterns); addPatternIfTargetOpIsIllegal<DecomposeAtenRandintLowOp>(patterns);
addPatternIfTargetOpIsIllegal<DecomposeAtenVarMeanCorrectionOp>(patterns); addPatternIfTargetOpIsIllegal<DecomposeAtenVarMeanCorrectionOp>(patterns);
addPatternIfTargetOpIsIllegal<DecomposePrimsConvertElementTypeOp>(patterns); addPatternIfTargetOpIsIllegal<DecomposePrimsConvertElementTypeOp>(patterns);
addPatternIfTargetOpIsIllegal<DecomposePrimsVarOp>(patterns);
addPatternIfTargetOpIsIllegal<DecomposePrimsSqrtOp>(patterns);
addPatternIfTargetOpIsIllegal<DecomposeAtenRandnOp>(patterns); addPatternIfTargetOpIsIllegal<DecomposeAtenRandnOp>(patterns);
addPatternIfTargetOpIsIllegal<DecomposeAtenRandnGeneratorOp>(patterns); addPatternIfTargetOpIsIllegal<DecomposeAtenRandnGeneratorOp>(patterns);
addPatternIfTargetOpIsIllegal<DecomposeAtenVarMeanOp>(patterns); addPatternIfTargetOpIsIllegal<DecomposeAtenVarMeanOp>(patterns);

View File

@ -438,6 +438,8 @@ static void markDecomposedOpsAsIllegal(MLIRContext *context,
target.addIllegalOp<AtenRandintLowOp>(); target.addIllegalOp<AtenRandintLowOp>();
target.addIllegalOp<AtenVarMeanCorrectionOp>(); target.addIllegalOp<AtenVarMeanCorrectionOp>();
target.addIllegalOp<PrimsConvertElementTypeOp>(); target.addIllegalOp<PrimsConvertElementTypeOp>();
target.addIllegalOp<PrimsVarOp>();
target.addIllegalOp<PrimsSqrtOp>();
target.addIllegalOp<AtenRandnOp>(); target.addIllegalOp<AtenRandnOp>();
target.addIllegalOp<AtenRandnGeneratorOp>(); target.addIllegalOp<AtenRandnGeneratorOp>();
target.addIllegalOp<AtenVarMeanOp>(); target.addIllegalOp<AtenVarMeanOp>();

View File

@ -675,7 +675,8 @@ void TypeAnalysis::visitOperation(Operation *op,
// Dtype is always float32, except for bfloat16, float16, float64 and nullptr. // Dtype is always float32, except for bfloat16, float16, float64 and nullptr.
if (isa<AtenTanhOp, AtenExpOp, AtenSinOp, AtenCosOp, AtenSigmoidOp, if (isa<AtenTanhOp, AtenExpOp, AtenSinOp, AtenCosOp, AtenSigmoidOp,
AtenReciprocalOp, AtenLogOp, AtenSqrtOp, AtenLog2Op, AtenLog1pOp, AtenReciprocalOp, AtenLogOp, AtenSqrtOp, AtenLog2Op, AtenLog1pOp,
AtenRsqrtOp, AtenErfOp, AtenSoftplusOp, AtenFrobeniusNormDimOp>(op)) { AtenRsqrtOp, AtenErfOp, AtenSoftplusOp, AtenFrobeniusNormDimOp,
PrimsSqrtOp>(op)) {
ValueKnowledge knowledge = ValueKnowledge knowledge =
ValueKnowledge::getTensorPessimisticValueState(op->getContext()); ValueKnowledge::getTensorPessimisticValueState(op->getContext());
Type dtype = operands[0]->getValue().dtype; Type dtype = operands[0]->getValue().dtype;
@ -978,7 +979,7 @@ void TypeAnalysis::visitOperation(Operation *op,
visitReductionAlongAllDimsOp(op, dtype, operands); visitReductionAlongAllDimsOp(op, dtype, operands);
return; return;
} else if (isa<AtenStdOp, AtenStdDimOp, AtenStdCorrectionOp, AtenVarOp, } else if (isa<AtenStdOp, AtenStdDimOp, AtenStdCorrectionOp, AtenVarOp,
AtenVarDimOp, AtenVarCorrectionOp>(op)) { AtenVarDimOp, AtenVarCorrectionOp, PrimsVarOp>(op)) {
auto input = operands[0]->getValue(); auto input = operands[0]->getValue();
visitReductionAlongAllDimsOp(op, input.dtype, operands); visitReductionAlongAllDimsOp(op, input.dtype, operands);
return; return;

View File

@ -107,6 +107,9 @@ def atenhardtanh〡shape(self: List[int], min_val: float = -1, max_val: float
def atensqrt〡shape(self: List[int]) -> List[int]: def atensqrt〡shape(self: List[int]) -> List[int]:
return upstream_shape_functions.unary(self) return upstream_shape_functions.unary(self)
def primssqrt〡shape(self: List[int]) -> List[int]:
return upstream_shape_functions.unary(self)
def atenneg〡shape(self: List[int]) -> List[int]: def atenneg〡shape(self: List[int]) -> List[int]:
return upstream_shape_functions.unary(self) return upstream_shape_functions.unary(self)
@ -307,6 +310,9 @@ def atenmean〡shape(self: List[int], dtype: Optional[int] = None) -> List[in
def atenvar〡shape(self: List[int], unbiased: bool = True) -> List[int]: def atenvar〡shape(self: List[int], unbiased: bool = True) -> List[int]:
return [] return []
def primsvar〡shape(inp: List[int], dims: Optional[List[int]], correction: int, output_dtype: Optional[int] = None) -> List[int]:
return upstream_shape_functions.sum_mean_dim(inp, dims, False, None)
def atenvardim〡shape(self: List[int], dim: Optional[List[int]], unbiased: bool = True, keepdim: bool = False) -> List[int]: def atenvardim〡shape(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) return upstream_shape_functions.sum_mean_dim(self, dim, keepdim, None)

View File

@ -674,6 +674,8 @@ def emit_ops(emitter_td: TextEmitter, registry: Registry):
# ========================================================================== # ==========================================================================
emit("prims::convert_element_type : (Tensor, int) -> (Tensor)") emit("prims::convert_element_type : (Tensor, int) -> (Tensor)")
emit("prims::var : (Tensor, int[]?, int, int?) -> (Tensor)")
emit("prims::sqrt : (Tensor) -> (Tensor)")
# ========================================================================== # ==========================================================================
# `quantized::` namespace. # `quantized::` namespace.