|
| 1 | +# (c) Meta Platforms, Inc. and affiliates. Confidential and proprietary. |
| 2 | + |
| 3 | +# pyre-strict |
| 4 | + |
| 5 | +import copy |
| 6 | +from typing import List, OrderedDict, Tuple |
| 7 | + |
| 8 | +import torch |
| 9 | +from inputgen.argtuple.gen import ArgumentTupleGenerator |
| 10 | +from inputgen.specs.model import ConstraintProducer as cp |
| 11 | +from inputgen.utils.random_manager import random_manager |
| 12 | +from inputgen.variable.type import ScalarDtype |
| 13 | +from specdb.db import SpecDictDB |
| 14 | + |
| 15 | +# seed to generate identical cases every run to reproduce from bisect |
| 16 | +random_manager.seed(1729) |
| 17 | + |
| 18 | + |
| 19 | +def apply_tensor_contraints(op_name: str, tensor_constraints: list[object]) -> None: |
| 20 | + match op_name: |
| 21 | + case ( |
| 22 | + "sigmoid.default" |
| 23 | + | "_softmax.default" |
| 24 | + | "rsqrt.default" |
| 25 | + | "exp.default" |
| 26 | + | "mul.Tensor" |
| 27 | + | "div.Tensor" |
| 28 | + ): |
| 29 | + tensor_constraints.append( |
| 30 | + cp.Dtype.In(lambda deps: [torch.float]), |
| 31 | + ) |
| 32 | + case ( |
| 33 | + "add.Tensor" |
| 34 | + | "sub.Tensor" |
| 35 | + | "add.Scalar" |
| 36 | + | "sub.Scalar" |
| 37 | + | "mul.Scalar" |
| 38 | + | "div.Scalar" |
| 39 | + ): |
| 40 | + tensor_constraints.append( |
| 41 | + cp.Dtype.In(lambda deps: [torch.float, torch.int]), |
| 42 | + ) |
| 43 | + case _: |
| 44 | + tensor_constraints.append( |
| 45 | + cp.Dtype.In(lambda deps: [torch.float, torch.int]), |
| 46 | + ) |
| 47 | + tensor_constraints.extend( |
| 48 | + [ |
| 49 | + cp.Value.Ge(lambda deps, dtype, struct: -(2**8)), |
| 50 | + cp.Value.Le(lambda deps, dtype, struct: 2**8), |
| 51 | + cp.Rank.Ge(lambda deps: 1), |
| 52 | + cp.Rank.Le(lambda deps: 2**2), |
| 53 | + cp.Size.Ge(lambda deps, r, d: 1), |
| 54 | + cp.Size.Le(lambda deps, r, d: 2**2), |
| 55 | + ] |
| 56 | + ) |
| 57 | + |
| 58 | + |
| 59 | +def facto_testcase_gen(op_name: str) -> List[Tuple[List[str], OrderedDict[str, str]]]: |
| 60 | + # minimal example to test add.Tensor using FACTO |
| 61 | + spec = SpecDictDB[op_name] |
| 62 | + |
| 63 | + for index, in_spec in enumerate(copy.deepcopy(spec.inspec)): |
| 64 | + if in_spec.type.is_scalar(): |
| 65 | + if in_spec.name != "alpha": |
| 66 | + spec.inspec[index].constraints.extend( |
| 67 | + [ |
| 68 | + cp.Dtype.In(lambda deps: [ScalarDtype.float, ScalarDtype.int]), |
| 69 | + cp.Value.Ge(lambda deps, dtype: -(2**8)), |
| 70 | + cp.Value.Le(lambda deps, dtype: 2**2), |
| 71 | + cp.Size.Ge(lambda deps, r, d: 1), |
| 72 | + cp.Size.Le(lambda deps, r, d: 2**2), |
| 73 | + ] |
| 74 | + ) |
| 75 | + else: |
| 76 | + spec.inspec[index].constraints.extend( |
| 77 | + [ |
| 78 | + cp.Value.Gt(lambda deps, dtype: 0), |
| 79 | + cp.Value.Le(lambda deps, dtype: 2), |
| 80 | + ] |
| 81 | + ) |
| 82 | + elif in_spec.type.is_tensor(): |
| 83 | + tensor_constraints = [] |
| 84 | + # common tensor constraints |
| 85 | + apply_tensor_contraints(op_name, tensor_constraints) |
| 86 | + spec.inspec[index].constraints.extend(tensor_constraints) |
| 87 | + |
| 88 | + return [ |
| 89 | + (posargs, inkwargs) |
| 90 | + for posargs, inkwargs, _ in ArgumentTupleGenerator(spec).gen() |
| 91 | + ] |
0 commit comments