mirrors/rocm_jax
1
0
Fork 0
mirror of https://github.com/ROCm/jax.git synced 2025-04-17 20:36:05 +00:00
Code Issues Packages Projects Releases Wiki Activity
rocm_jax/jax/experimental/jax2tf/tests/shape_poly_test.py
George Necula 7d452adfd3 Add support for dynamic shapes to GPU threefry2x32 custom call. 
In presence of dynamic shapes the ThreeFry2x32Descriptor will contain the
value n=-1, and the actual desired output length will be passed as
an additional operand. If the shape is static then the length will be
passed as part of the descriptor.

PiperOrigin-RevId: 497945778
2022-12-27 04:48:26 -08:00

2312 lines
101 KiB
Python
Raw Blame History

# Copyright 2020 The JAX Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Tests for the shape-polymorphic jax2tf conversion."""
import contextlib
import unittest
from absl.testing import absltest, parameterized
from typing import Any, Callable, Dict, List, Optional, Sequence, Tuple, Union
import collections
import functools
from functools import partial
import operator
import re
import jax
from jax import core
from jax.experimental import jax2tf
from jax.experimental.jax2tf import shape_poly
from jax import lax
import jax.numpy as jnp
from jax import random
from jax._src import test_util as jtu
from jax._src import util
from jax._src.lax import lax as lax_internal
from jax._src.lax import control_flow as lax_control_flow
import numpy as np
from jax.experimental.jax2tf.tests import tf_test_util
import tensorflow as tf # type: ignore[import]
from jax.config import config
config.parse_flags_with_absl()
# Import after parsing flags
from jax.experimental.jax2tf.tests import primitive_harness
from jax.experimental.jax2tf.tests.primitive_harness import Harness, CustomArg, RandArg, StaticArg
from jax.experimental.jax2tf.tests.jax2tf_limitations import Jax2TfLimitation
PS = jax2tf.PolyShape
_f32 = np.float32
class DimPolynomialTest(tf_test_util.JaxToTfTestCase):
def test_parse_poly_spec(self):
self.assertEqual((2, 3), shape_poly._parse_spec(None, (2, 3)))
self.assertEqual((2, 3), shape_poly._parse_spec("2, 3", (2, 3)))
self.assertEqual((2, 3), shape_poly._parse_spec("2, _", (2, 3)))
self.assertEqual((2, 3), shape_poly._parse_spec("2, ...", (2, 3)))
self.assertEqual((2, 3), shape_poly._parse_spec("...", (2, 3)))
self.assertEqual((2, 3), shape_poly._parse_spec(" ( 2 , 3 ) ", (2, 3)))
a, b = shape_poly._parse_spec("a, b", (2, 3))
self.assertEqual((a, 3), shape_poly._parse_spec("(a, ...) ", (None, 3)))
tshape = tf.TensorShape([None, 3])
self.assertEqual((a, 3), shape_poly._parse_spec("(a, ...) ", tshape))
a, b = shape_poly._parse_spec("a, b", (2, 3))
@parameterized.named_parameters(
dict(testcase_name=f"_{dim_spec=}",
dim_spec=dim_spec, dim_poly=dim_poly)
for dim_spec, dim_poly in [
("2*a*b", 2 * a * b),
("-2 * a^2 * b + b^2", -2 * a * a * b + b * b),
("-2 * a^2 * b + -1 *b^2*a", -2 * a * a * b - a * b * b),
("3 * a * b * a + -2", 3 * a * b * a - 2),
("a + 1", a + 1),
("a + -1", a - 1),
])
def test_parse_poly_spec_poly(self,
dim_spec="3 * a * b * a + -2",
dim_poly=3 * a * b * a - 2):
# For internal usage only (the polymorphic_shapes of VJP) we need to
# parse polynomials.
self.assertEqual((dim_poly,), shape_poly._parse_spec(dim_spec, (2,)))
self.assertEqual((dim_poly,), shape_poly._parse_spec(str(dim_poly), (2,)))
@parameterized.named_parameters(
dict(testcase_name=f"_{dim_spec=}",
dim_spec=dim_spec, dim_poly=dim_poly)
for dim_spec, dim_poly in [
("2*a*b", 2 * a * b),
("-2 * a^2 * b + b^2", -2 * a * a * b + b * b),
("-2 * a^2 * b + -1 *b^2*a", -2 * a * a * b - a * b * b),
("3 * a * b * a + -2", 3 * a * b * a - 2),
("a + 1", a + 1),
("a + -1", a - 1),
])
def test_parse_poly_spec_shapeenv(self,
dim_spec="3 * a * b * a + -2",
dim_poly=3 * a * b * a - 2):
# For internal usage only (the polymorphic_shapes of VJP) we need to
# parse polynomials.
self.assertEqual((dim_poly,), shape_poly._parse_spec(dim_spec, (2,)))
self.assertEqual((dim_poly,), shape_poly._parse_spec(str(dim_poly), (2,)))
def test_dim_vars(self):
a, b, a1 = shape_poly._parse_spec("a, b, a", (2, 3, 2))
self.assertEqual(True, a == a)
self.assertEqual(True, a == a1)
self.assertEqual(False, a != a)
with self.assertRaisesRegex(
core.InconclusiveDimensionOperation,
"Dimension polynomial comparison 'a' == 'b' is inconclusive"):
a.eq(b)
with self.assertRaisesRegex(
core.InconclusiveDimensionOperation,
"Dimension polynomial comparison 'a' == 'b' is inconclusive"):
a == b
with self.assertRaisesRegex(
core.InconclusiveDimensionOperation,
"Dimension polynomial comparison 'a' == 'b' is inconclusive"):
a != b
self.assertLen({a, a}, 1)
self.assertLen({a, b}, 2)
self.assertIn(a, {a, b})
self.assertIn(b, {a, b})
self.assertIn(a, [a, b])
with self.assertRaisesRegex(core.InconclusiveDimensionOperation,
"Dimension polynomial comparison .* is inconclusive"):
b in [a, b]
def test_get_vars(self):
a, b = shape_poly._parse_spec("a, b", (2, 3))
self.assertEqual({"a"}, a.get_vars())
self.assertEqual({"a", "b"}, (a * b * a).get_vars())
def test_evaluate(self):
a, b = shape_poly._parse_spec("a, b", (2, 3))
self.assertEqual(1, (a * a - b).evaluate(dict(a=2, b=3)))
self.assertEqual(2, (a * a - b + 1).evaluate(dict(a=-2, b=3)))
def test_dim_vars_symbolic_equal(self):
a, b = shape_poly._parse_spec("a, b", (2, 3))
self.assertTrue(core.symbolic_equal_dim(a, a))
self.assertFalse(core.symbolic_equal_dim(a, 1))
self.assertFalse(core.symbolic_equal_dim(a, b))
self.assertTrue(core.symbolic_equal_one_of_dim(a, [2, a]))
self.assertFalse(core.symbolic_equal_one_of_dim(a, [2, b]))
self.assertFalse(core.symbolic_equal_one_of_dim(a, []))
self.assertTrue(core.symbolic_equal_one_of_dim(2, [a, 3, 2]))
self.assertFalse(core.symbolic_equal_one_of_dim(1, [2, b]))
self.assertFalse(core.symbolic_equal_one_of_dim(3, []))
self.assertTrue(core.symbolic_equal_dim(1, jnp.add(0, 1))) # A DeviceArray
with self.assertRaisesRegex(TypeError,
re.escape("Shapes must be 1D sequences of concrete values of integer type, got (1, 'a').")):
self.assertTrue(core.symbolic_equal_dim(1, "a"))
def test_poly_bounds(self):
a, b = shape_poly._parse_spec("a, b", (2, 3))
self.assertEqual(a.bounds(), (1, None))
self.assertEqual((2 * a).bounds(), (2, None))
self.assertEqual((2 * a - 3).bounds(), (-1, None))
self.assertEqual((-2 * a - 3).bounds(), (None, -5))
self.assertEqual((3 * a * b * b + 5 * a - 7).bounds(), (1, None))
self.assertEqual((3 * a * b * b - 5 * a - 7).bounds(), (None, None))
self.assertEqual((a + b - a * b + a * b * a).bounds(), (None, None))
self.assertEqual((a + 2 * b - a).bounds(), (2, None))
def test_poly_equal(self):
a, b = shape_poly._parse_spec("a, b", (2, 3))
poly3 = a + 3 - a
self.assertTrue(poly3 == 3)
self.assertTrue(poly3 == np.array(3, np.int64))
self.assertTrue(poly3 == np.array(3, np.int64)[()])
self.assertFalse((poly3 + 1) == 3)
self.assertFalse(poly3 == poly3 + 1)
self.assertTrue((2 * a * b * a + 3).eq(1 + b * a * a + a * a * b + 2))
self.assertFalse((2 * a * b * a + 3).eq(a * b * a + 3))
self.assertFalse((a * b * a + 3).eq(a * b * a + 4))
self.assertFalse((2 * a * b * a).eq(a * b * a))
self.assertFalse((2 * a * b * a + 1).eq(a * b * a))
self.assertFalse((3 * a * b * a - 1).eq(a * b * a))
with self.assertRaisesRegex(core.InconclusiveDimensionOperation,
re.escape("Dimension polynomial comparison '3*a^2*b + -2' == 'a^2*b' is inconclusive")):
(3 * a * b * a - 2).eq(a * b * a)
def test_poly_compare(self):
a, b = shape_poly._parse_spec("a, b", (2, 3))
poly = 4 * a + b + 3
self.assertTrue(poly.ge(0))
self.assertTrue(poly.ge(8))
self.assertTrue(poly.ge(poly))
self.assertTrue(poly.ge(poly - 1))
with self.assertRaisesRegex(core.InconclusiveDimensionOperation, "inconclusive"):
poly.ge(9)
with self.assertRaisesRegex(core.InconclusiveDimensionOperation, "inconclusive"):
(4 * a - b).ge(0)
def test_poly_compare_overload(self):
a, b = shape_poly._parse_spec("a, b", (2, 3))
poly = 4 * a + b + 3
self.assertTrue(poly >= 0)
self.assertTrue(poly >= 8)
self.assertTrue(poly > 7)
self.assertTrue(poly >= poly)
self.assertTrue(poly >= poly - 1)
with self.assertRaisesRegex(core.InconclusiveDimensionOperation, "inconclusive"):
poly >= 9
with self.assertRaisesRegex(core.InconclusiveDimensionOperation, "inconclusive"):
(4 * a - b) >= 0
def test_core_greater_equal(self):
a, b = shape_poly._parse_spec("a, b", (2, 3))
self.assertTrue(core.greater_equal_dim(a, a))
self.assertTrue(core.greater_equal_dim(a, 0))
self.assertTrue(core.greater_equal_dim(a, 1))
self.assertTrue(core.greater_equal_shape((a, 2), (1, 1)))
with self.assertRaisesRegex(core.InconclusiveDimensionOperation,
"Dimension polynomial comparison .* is inconclusive"):
core.greater_equal_dim(a, 2)
with self.assertRaisesRegex(core.InconclusiveDimensionOperation,
"Dimension polynomial comparison .* is inconclusive"):
core.greater_equal_dim(a, b)
def test_poly_int_results(self):
a, b = shape_poly._parse_spec("a, b", (2, 3))
self.assertEqual(a + 2 - a, 2)
self.assertIsInstance(a + 2 - a, int)
self.assertEqual(a + (2 - a), 2)
self.assertIsInstance(a + (2 - a), int)
self.assertEqual(a * 2 // a, 2)
self.assertIsInstance(a * 2 // a, int)
@parameterized.named_parameters(
dict(testcase_name=f"_D={dividend}_d={divisor}_q={quotient}_r={remainder}",
dividend=dividend, divisor=divisor, quotient=quotient,
remainder=remainder)
for dividend, divisor, quotient, remainder in [
(a, 1, a, 0),
(3 * a, 3, a, 0),
(3 * a + 3, 3, a + 1, 0),
(3 * a + 2, 3, a, 2),
(3 * a + 5, 3, a + 1, 2),
(3 * a - 2, 3, a - 1, 1),
(3 * a * a * b + 2 * b * b * a, a * b, 3 * a + 2 * b, 0),
(a * a - b * b, a + b, a - b, 0),
(a, b, None, None),
(3 * a, 2, None, None),
(2 * a * b + b * b, a + b, None, None),
(3, a, None, None),
])
def test_poly_divmod(self, *, dividend, quotient, divisor, remainder):
if quotient is None:
with self.assertRaisesRegex(core.InconclusiveDimensionOperation,
"Cannot divide .* by .*"):
divmod(dividend, divisor)
else:
self.assertEqual((quotient, remainder), divmod(dividend, divisor))
@parameterized.named_parameters(
dict(testcase_name=f"_D={dividend}_d={divisor}_q={quotient}",
dividend=dividend, divisor=divisor, quotient=quotient)
for dividend, divisor, quotient in [
(a, 1, a),
(3 * a, 3, a),
(3 * a + 3, 3, a + 1),
(3 * a + 2, 3, None),
(3 * a + 5, 3, None),
(3 * a - 2, 3, None),
(3 * a * a * b + 2 * b * b * a, a * b, 3 * a + 2 * b),
(a * a - b * b, a + b, a - b),
(a, b, None),
(3 * a, 2, None),
(2 * a * b + b * b, a + b, None),
])
def test_poly_truediv(self, *, dividend, divisor, quotient):
if quotient is None:
with self.assertRaisesRegex(core.InconclusiveDimensionOperation,
"Cannot divide .* by .*"):
dividend / divisor
else:
self.assertEqual(quotient, dividend / divisor)
def test_poly_truediv_error(self):
a, = shape_poly._parse_spec("a,", (2,))
with self.assertRaisesRegex(core.InconclusiveDimensionOperation,
"Cannot divide .* by .*"):
3 / a
def test_dilate_shape(self):
"""0 if d == 0 else 1 + dilation * (d - 1))"""
a, = shape_poly._parse_spec("a,", (2,))
self.assertEqual((4, 7), core.dilate_shape((2, 3), (3, 3)))
self.assertEqual((0, 7), core.dilate_shape((0, 3), (3, 3)))
self.assertEqual((a, 7), core.dilate_shape((a, 3), (1, 3)))
self.assertEqual((2 * a - 1, 7), core.dilate_shape((a, 3), (2, 3)))
def test_stride_shape(self):
"""(s - window_size) // window_stride + 1"""
a, stride = shape_poly._parse_spec("a, s", (2, 3))
self.assertEqual((8, 9), core.stride_shape((10, 20), window_size=(3, 3), window_stride=(1, 2)))
self.assertEqual((a, 9), core.stride_shape((a, 20), (1, 3), (1, 2)))
self.assertEqual((a - 1, 9), core.stride_shape((a, 20), (2, 3), (1, 2)))
self.assertEqual((a + 1, 9), core.stride_shape((a * stride + 2, 20), (2, 3), (stride, 2)))
with self.assertRaisesRegex(
core.InconclusiveDimensionOperation,
re.escape(
"Cannot compute stride for dimension 'a', window_size '1', stride '2'.\nDetails: Cannot divide 'a + -1' by '2'")):
core.stride_shape((a, 20), (1, 3), (2, 2))
class PolyHarness(Harness):
"""Tests a function with shape polymorphism.
Converts `fun` with shape polymorphism, creates a `tf.ConcreteFunction`
given `input_signature` and checks the inferred output shapes to match
`expected_output_shapes`, then checks that the JAX and the TF functions
produce the same results.
"""
def __init__(self,
group_name: str, name: str,
fun: Callable,
*,
arg_descriptors: Sequence[primitive_harness.ArgDescriptor] = (),
polymorphic_shapes: Optional[Sequence[Any]] = None,
input_signature: Optional[Sequence[tf.TensorSpec]] = None,
poly_axes: Optional[Sequence[Optional[Union[int, Sequence[int]]]]] = None,
expected_output_signature: Optional[tf.TensorSpec] = None,
enable_xla: bool = True,
expect_error: Tuple[Optional[Any], Optional[str]] = (None, None),
skip_jax_run: bool = False,
check_result: bool = True,
tol: Optional[float] = None):
"""Args:
group_name, name: The name for the harness. See `Harness.__init__`.
fun: the function to be converted, possbily after partial application to
static arguments from `arg_descriptors`. See `Harness.__init__`.
arg_descriptors: The argument descriptors. See `Harness.__init__`. May
be missing, in which case `skip_jax_run` should be `True` and
`poly_axes` cannot be used.
polymorphic_shapes: For `jax2tf.convert`. If missing, generated from
`poly_axes`.
input_signature: For `tf.function.get_concrete_function`. If missing,
generated from `poly_axes`.
poly_axes: If present, used to generate `polymorphic_shapes` and
`input_signature`. Must correspond to the non-static arguments, and for
each one it must specify which axes are polymorphic: None, or an int
(for the index of the polymorphic axis), or a tuple of ints
(for multiple polymorphic axes). For each argument, we use its
`poly_axes` entry to generate the polymorphic_shapes specification,
creating dimension variables `b0`, `b1, ..., for each of its polymorphic
axes. This means that separate arguments will share the same dimension
variable names, in the order in which the axes are listed in
`poly_axes`. We also generate the input_signature from `poly_axes`.
expected_output_signature: the expected inferred output shape.
enable_xla: For `jax2tf.convert`.
expect_error: a pair of an Exception type and a regular expression to
match the expected exception string.
skip_jax_run: If True, then neither the JAX nor the TF functions are
executed.
check_result: specifies if we want to check that the result of the shape
polymorphic conversion produces the same result and the JAX function.
tol: the tolerance to use for checking results.
"""
super().__init__(group_name, name, fun, arg_descriptors,
dtype=np.float32)
self.poly_axes = poly_axes
self.polymorphic_shapes = polymorphic_shapes
self.input_signature = input_signature
self.expected_output_signature = expected_output_signature
self.skip_jax_run = skip_jax_run
self.expect_error = expect_error
self.enable_xla = enable_xla
self.tol = tol
self.check_result = check_result
# Replicate the harness for both enable and disable xla
def both_enable_and_disable_xla(self) -> Tuple["PolyHarness", "PolyHarness"]:
assert self.enable_xla
other = PolyHarness(self.group_name,
f"{self.name}_enable_xla=False",
self.fun,
arg_descriptors=self.arg_descriptors,
poly_axes=self.poly_axes,
polymorphic_shapes=self.polymorphic_shapes,
input_signature=self.input_signature,
expected_output_signature=self.expected_output_signature,
expect_error=self.expect_error,
tol=self.tol,
enable_xla=False)
self.name = f"{self.name}_enable_xla=True"
return (self, other)
def run_test(self, tst: tf_test_util.JaxToTfTestCase):
# Make polymorphic_shapes and input_signature from poly_axes.
if self.poly_axes is None:
polymorphic_shapes = self.polymorphic_shapes
input_signature = self.input_signature
assert input_signature is not None
if not self.skip_jax_run:
args = self.dyn_args_maker(tst.rng())
else:
assert isinstance(self.poly_axes, Sequence)
# Make poly_axes: Sequence[Sequence[int]], one top-level element for each argument
poly_axes = tuple(map(lambda pa: pa if isinstance(pa, Sequence) or pa is None else (pa,),
self.poly_axes))
args = self.dyn_args_maker(tst.rng())
assert self.polymorphic_shapes is None
assert self.input_signature is None
assert args is not None and len(args) == len(poly_axes)
# Make the polymorphic_shapes and input_signature
polymorphic_shapes = []
input_signature = []
for arg, poly_axis in zip(args, poly_axes):
if poly_axis is None:
polymorphic_shapes.append(None)
input_signature.append(tf.TensorSpec(np.shape(arg), arg.dtype))
else:
def make_arg_polymorphic_shapes(poly_axis: Sequence[int]) -> Tuple[str, tf.TensorSpec]:
idx = -1
dims = []
tensorspec_dims: List[Optional[int]] = []
for i, d in enumerate(arg.shape):
if i in poly_axis:
idx += 1
dims.append(f"b{idx}")
tensorspec_dims.append(None)
else:
dims.append(str(d))
tensorspec_dims.append(d)
return ", ".join(dims), tf.TensorSpec(tensorspec_dims, arg.dtype)
arg_polymorphic_shapes, arg_tensorspec = make_arg_polymorphic_shapes(poly_axis)
polymorphic_shapes.append(arg_polymorphic_shapes)
input_signature.append(arg_tensorspec)
expect_error_type, expect_error_regex = self.expect_error
if self.skip_jax_run and self.arg_descriptors == ():
f_jax = self.fun
else:
f_jax = self.dyn_fun
with contextlib.ExitStack() as stack:
if expect_error_type is not None:
stack.enter_context(tst.assertRaisesRegex(expect_error_type, expect_error_regex))
f_tf = jax2tf.convert(f_jax, polymorphic_shapes=polymorphic_shapes,
enable_xla=self.enable_xla)
# Run in tf.Eager mode first, because it is friendlier to debuggers
res_tf = f_tf(*args) if not self.skip_jax_run else None
f_tf_func = tf.function(
f_tf, autograph=False, input_signature=input_signature)
# Create tf.ConcreteFunction and check inferred output signature
concrete_f_tf = f_tf_func.get_concrete_function(*input_signature)
if expect_error_type is not None:
return
if self.expected_output_signature:
# Strangely, output_shapes can be a single shape for a function with a
# single result, or a list/tuple of shapes.
expected_output_signature = self.expected_output_signature
concrete_output_tf_shape = concrete_f_tf.output_shapes
if not isinstance(concrete_output_tf_shape, (tuple, list)): # Single result
assert not isinstance(self.expected_output_signature, (tuple, list))
expected_output_signature = [self.expected_output_signature]
concrete_output_tf_shape = [concrete_output_tf_shape]
for expected, found in util.safe_zip(expected_output_signature,
concrete_output_tf_shape):
tst.assertEqual(tuple(expected.shape), tuple(found))
# Run the JAX and the TF functions and compare the results
if not self.skip_jax_run:
res_jax = f_jax(*args)
if self.check_result:
tst.assertAllClose(res_jax, res_tf, atol=self.tol, rtol=self.tol)
def check_shape_poly(tst, f_jax: Callable, *,
arg_descriptors: Sequence[primitive_harness.ArgDescriptor] = (),
skip_jax_run: bool = False,
poly_axes = None,
polymorphic_shapes: Optional[Sequence[Any]] = None,
input_signature: Optional[Sequence[tf.TensorSpec]] = None,
expected_output_signature: Optional[tf.TensorSpec] = None):
# Makes and tests a harness. See PolyHarness documentation.
h = PolyHarness("", "", f_jax,
arg_descriptors=arg_descriptors,
skip_jax_run=skip_jax_run, poly_axes=poly_axes,
polymorphic_shapes=polymorphic_shapes,
input_signature=input_signature,
expected_output_signature=expected_output_signature)
h.run_test(tst)
class ShapePolyTest(tf_test_util.JaxToTfTestCase):
def test_simple_unary(self):
"""Test shape polymorphism for a simple case, unary function."""
def f_jax(x):
return x + jnp.sin(x)
check_shape_poly(self,
f_jax,
arg_descriptors=[RandArg((2, 3), _f32)],
input_signature=[tf.TensorSpec([2, 3])],
polymorphic_shapes=None,
expected_output_signature=tf.TensorSpec([2, 3]))
check_shape_poly(self,
f_jax,
arg_descriptors=[RandArg((2, 3), _f32)],
input_signature=[tf.TensorSpec([2, None])],
polymorphic_shapes=["_, h"],
expected_output_signature=tf.TensorSpec([2, None]))
check_shape_poly(self,
f_jax,
arg_descriptors=[RandArg((3, 3), _f32)],
input_signature=[tf.TensorSpec([None, None])],
polymorphic_shapes=["h, h"],
expected_output_signature=tf.TensorSpec([None, None]))
check_shape_poly(self,
f_jax,
arg_descriptors=[RandArg((3, 3), _f32)],
input_signature=[tf.TensorSpec([None, None])],
polymorphic_shapes="h, h",
expected_output_signature=tf.TensorSpec([None, None]))
def test_simple_binary(self):
"""Test shape polymorphism for a simple case, binary function."""
def f_jax(x, y):
return x + jnp.sin(y)
check_shape_poly(self,
f_jax,
arg_descriptors=[RandArg((2, 3), _f32), RandArg((2, 3), _f32)],
input_signature=[tf.TensorSpec([2, 3]), tf.TensorSpec([2, 3])],
polymorphic_shapes=None,
expected_output_signature=tf.TensorSpec([2, 3]))
check_shape_poly(self,
f_jax,
arg_descriptors=[RandArg((2, 3), _f32), RandArg((2, 3), _f32)],
input_signature=[tf.TensorSpec([2, None]), tf.TensorSpec([2, 3])],
polymorphic_shapes="_, h",
expected_output_signature=(
# for native lowering we cannot refine the inferred shape of the
# output if the input is more specific than polymorphic_shapes.
tf.TensorSpec([2, 3]) if not config.jax2tf_default_experimental_native_lowering
else tf.TensorSpec([2, None])))
check_shape_poly(self,
f_jax,
arg_descriptors=[RandArg((3, 3), _f32), RandArg((3, 3), _f32)],
input_signature=[tf.TensorSpec([None, None]), tf.TensorSpec([None, None])],
polymorphic_shapes=PS("h", "h"),
expected_output_signature=tf.TensorSpec([None, None]))
def test_arange(self):
def f_jax(x):
return x + jnp.arange(x.shape[0], dtype=np.float32)
x = np.ones((3,), dtype=np.float32)
self.assertAllClose(jax2tf.convert(f_jax, polymorphic_shapes="b")(x),
f_jax(x))
def test_argmax(self):
def f_jax(x): # x: f32[b, 4, 5]
return lax.argmax(x, axis=1, index_dtype=np.int32)
x = np.arange(3 * 4 * 5, dtype=np.float32).reshape((3, 4, 5))
self.assertAllClose(jax2tf.convert(f_jax, polymorphic_shapes="(b, _, _)")(x),
f_jax(x))
def test_static_shape_result(self):
"""The result has static shape."""
def f_jax(x):
return jnp.sum(x + jnp.sin(x), axis=0)
check_shape_poly(self,
f_jax,
arg_descriptors=[RandArg((2, 3), _f32)],
input_signature=[tf.TensorSpec([2, 3])],
polymorphic_shapes=None,
expected_output_signature=tf.TensorSpec([3]))
check_shape_poly(self,
f_jax,
arg_descriptors=[RandArg((2, 3), _f32)],
input_signature=[tf.TensorSpec([None, 3])],
polymorphic_shapes="b, _",
expected_output_signature=tf.TensorSpec([3]))
def test_forgot_polymorphic_shapes_error(self):
msg_re = "polymorphic shape None in axis .* must contain a dimension variable for unknown dimension in argument shape .*. Perhaps you forgot to add the polymorphic_shapes"
with self.assertRaisesRegex(ValueError, msg_re):
check_shape_poly(self,
jnp.sin,
arg_descriptors=[RandArg((1, 3,), _f32)],
input_signature=[tf.TensorSpec([1, None])],
polymorphic_shapes=None)
def test_kwargs(self):
"""Test shape polymorphism for a function with kwargs."""
x = np.ones(3, dtype=np.float32)
y = np.ones(1, dtype=np.float32)
def f_jax(x, *, y):
return x + jnp.sin(y)
f_tf: Callable[..., Any] = jax2tf.convert(f_jax, polymorphic_shapes=["b, ..."])
self.assertAllClose(f_jax(x, y=y), f_tf(x, y=y))
def test_arg_avals(self):
"""Test conversion of actual arguments to abstract values."""
def check_avals(*, arg_shapes: Sequence[Sequence[Optional[int]]],
polymorphic_shapes: Sequence[Optional[Union[str, PS]]],
expected_avals: Optional[Sequence[core.ShapedArray]] = None,
expected_shapeenv: Optional[Dict[str, int]] = None,
eager_mode: bool = False):
# Use eager mode only for when all arg_shapes are known, in order to
# check expected_shapeenv.
arg_dtypes = (_f32,) * len(arg_shapes)
def f_tf(*args_tf):
avals = tuple(map(shape_poly.arg_aval, arg_shapes, arg_dtypes, polymorphic_shapes))
dim_vars, get_dim_values_jax = shape_poly.prepare_dim_var_env(avals)
dim_values, _ = jax2tf.jax2tf._interpret_fun_jax(get_dim_values_jax,
args_tf, avals, "")
if expected_avals is not None:
self.assertEqual(expected_avals, avals)
return dict(zip(dim_vars, dim_values))
if eager_mode:
# If we want to check the shape_env then all arg_shapes must be known
assert all(all(d is not None for d in a_s)
for a_s in arg_shapes)
shape_env = f_tf(*[tf.ones(a_s, dtype=_f32) for a_s in arg_shapes])
if expected_shapeenv is not None:
for v, val in expected_shapeenv.items():
self.assertEqual(val, shape_env.get(v))
else:
f_tf = tf.function(autograph=False)(f_tf)
f_tf.get_concrete_function(*[tf.TensorSpec(a_s, _f32)
for a_s in arg_shapes])
assert not expected_shapeenv, "Should use eager_mode=True"
def shaped_array(shape_spec: str, actual_shape: core.Shape):
return core.ShapedArray(
shape_poly._parse_spec(shape_spec, actual_shape), np.float32)
# Known shapes for the arguments
check_avals(
arg_shapes=[(2, 3)],
polymorphic_shapes=[None],
expected_avals=(shaped_array("2, 3", [2, 3]),))
check_avals(
arg_shapes=[(2, 3)],
polymorphic_shapes=["(2, 3)"],
expected_avals=(shaped_array("2, 3", [2, 3]),))
check_avals(
arg_shapes=[(2, 3)],
polymorphic_shapes=["(_, 3)"],
expected_avals=(shaped_array("2, 3", [2, 3]),))
check_avals(
arg_shapes=[(2, 3)],
polymorphic_shapes=[PS("_", 3)],
expected_avals=(shaped_array("2, 3", [2, 3]),))
check_avals(
arg_shapes=[(2, 3)],
polymorphic_shapes=["..."],
expected_avals=(shaped_array("2, 3", [2, 3]),))
check_avals(
arg_shapes=[(2, 3)],
polymorphic_shapes=[PS(...)],
expected_avals=(shaped_array("2, 3", [2, 3]),))
# Partially known shapes for the arguments
check_avals(
arg_shapes=[(None, 3)],
polymorphic_shapes=[PS("b", ...)],
expected_avals=(shaped_array("(b, 3)", (2, 3)),))
check_avals(
arg_shapes=[(None, None)],
polymorphic_shapes=["h, h"],
expected_avals=(shaped_array("(h, h)", (2, 2)),))
check_avals(
arg_shapes=[(2, None)],
polymorphic_shapes=["h, h"],
expected_avals=(shaped_array("(h, h)", (2, 2)),))
check_avals(
arg_shapes=[(None, 3, 4)],
polymorphic_shapes=["(c, b, a)"],
expected_avals=(shaped_array("(c, b, a)", (2, 3, 4)),),
)
# Check cases when the specifications are polynomials
check_avals(
arg_shapes=[(2, 3)],
polymorphic_shapes=[PS("a + 1", "b + 2")],
eager_mode=True,
expected_shapeenv=dict(a=1, b=1))
check_avals(
arg_shapes=[(7, 5)],
polymorphic_shapes=[PS("2 * a + b", "b + 2")],
eager_mode=True,
expected_shapeenv=dict(a=2, b=3))
check_avals(
arg_shapes=[(7, 11, 4)],
polymorphic_shapes=[PS("2 * a + b", "b * b + 2", "b + 1")],
eager_mode=True,
expected_shapeenv=dict(a=2, b=3))
check_avals(
arg_shapes=[(7, 11, 19, 7)],
polymorphic_shapes=[PS("2 * a + b", "b * b + 2", "b + c * c", "2 * c + -1")],
eager_mode=True,
expected_shapeenv=dict(a=2, b=3, c=4))
with self.assertRaisesRegex(ValueError,
"Cannot solve for values of dimension variables {'b'}"):
check_avals(
arg_shapes=[(4, 36, 3)],
polymorphic_shapes=[PS("b * b", "b * d * d", "d")])
with self.assertRaisesRegex(ValueError,
"Dimension variable b must have integer value >= 1"):
check_avals(
arg_shapes=[(5, 36)],
polymorphic_shapes=[PS("3 * b", ...)],
eager_mode=True)
with self.assertRaisesRegex(ValueError,
"Dimension variable b must have integer value >= 1"):
check_avals(
arg_shapes=[(10, 3)],
polymorphic_shapes=[PS("3 * b + 10", ...)],
eager_mode=True)
with self.assertRaisesRegex(ValueError,
"Dimension variable b must have integer value >= 1"):
check_avals(
arg_shapes=[(7, 3)],
polymorphic_shapes=[PS("3 * b + 10", ...)],
eager_mode=True)
for invalid_syntax in [")(", "2a", "a@", "a - 2", "'a'", "('a', ...)"]:
with self.assertRaisesRegex(ValueError,
re.escape("has invalid syntax")):
check_avals(
arg_shapes=[(2,)], polymorphic_shapes=[invalid_syntax])
for invalid_syntax in [5.0, ["a list"], ("a tuple",), re.compile(".")]:
with self.assertRaisesRegex(ValueError,
re.escape("Invalid polymorphic shape element")):
check_avals(
arg_shapes=[(2,)], polymorphic_shapes=[PS([invalid_syntax])])
with self.assertRaisesRegex(
ValueError,
re.escape("polymorphic shape '..., 3' can contain Ellipsis only at the end.")):
check_avals(
arg_shapes=[(2, 3)],
polymorphic_shapes=["..., 3"])
with self.assertRaisesRegex(
ValueError,
re.escape(
"polymorphic shape '2, 3, 4, ...' of rank 3 must match the rank 2 of argument shape (2, 3).")
):
check_avals(
arg_shapes=[(2, 3)],
polymorphic_shapes=["2, 3, 4, ..."])
with self.assertRaisesRegex(
ValueError,
re.escape(
"polymorphic shape (Ellipsis, 3) can contain Ellipsis only at the end.")):
check_avals(
arg_shapes=[(2, 3)],
polymorphic_shapes=[PS(..., 3)])
with self.assertRaisesRegex(
ValueError,
re.escape(
"polymorphic shape None in axis 1 must contain a dimension variable for unknown dimension in argument shape (2, None)"
)):
check_avals(
arg_shapes=[(2, None)],
polymorphic_shapes=[None])
with self.assertRaisesRegex(
ValueError,
re.escape("polymorphic shape '()' of rank 0 must match the rank 2 of argument shape (2, 3)")):
check_avals(
arg_shapes=[(2, 3)], polymorphic_shapes=["()"])
with self.assertRaisesRegex(
ValueError,
re.escape(
"polymorphic shape '(_, _)' in axis 1 must contain a dimension variable "
"for unknown dimension in argument shape (2, None)"
)):
check_avals(
arg_shapes=[(2, None)],
polymorphic_shapes=["(_, _)"])
with self.assertRaisesRegex(
ValueError,
re.escape(
"polymorphic shape '(2, 13)' in axis 1 must match the known dimension size 3 "
"for argument shape (2, 3)"
)):
check_avals(
arg_shapes=[(2, 3)],
polymorphic_shapes=["(2, 13)"])
with self.assertRaisesRegex(
ValueError,
re.escape(
"polymorphic shape '(2, 3)' in axis 1 must contain a dimension variable for "
"unknown dimension in argument shape (2, None)"
)):
check_avals(
arg_shapes=[(2, None)],
polymorphic_shapes=["(2, 3)"])
with self.assertRaisesRegex(
ValueError,
"Found inconsistency when solving.*"):
check_avals(
arg_shapes=[(2, 3)],
polymorphic_shapes=["(a, a)"],
eager_mode=True)
# Same error across multiple arguments
with self.assertRaisesRegex(
ValueError,
"Found inconsistency when solving.*"):
check_avals(
arg_shapes=[(2, 3), (5,)],
polymorphic_shapes=["a, ...", "a"],
eager_mode=True)
def test_pytree(self):
"""Arguments and polymorphic_shapes are pytrees."""
# Arguments are of the form [([x00, x01], [x10]), dict(a=ya, b=yb)]
def add_all_jax(x_pair_of_list, y_dict):
x_list_0, x_list_1 = x_pair_of_list
return functools.reduce(operator.add,
x_list_0 + x_list_1 + [y_dict["a"], y_dict["b"]])
check_shape_poly(self,
add_all_jax,
skip_jax_run=True,
input_signature=[([tf.TensorSpec([None]), tf.TensorSpec([None])],
[tf.TensorSpec([None])]),
dict(a=tf.TensorSpec([None]),
b=tf.TensorSpec([None]))],
polymorphic_shapes=[(["v", "v"], ["v"]),
dict(a="v", b="v")],
expected_output_signature=tf.TensorSpec([None]))
# Now partial polymorphic_shapes; the parts of the polymorphic_shapes that
# are not specified must have full input_signatures.
check_shape_poly(self,
add_all_jax,
skip_jax_run=True,
input_signature=[([tf.TensorSpec([4]), tf.TensorSpec([4])], [tf.TensorSpec([4])]),
dict(a=tf.TensorSpec([4]), b=tf.TensorSpec([4]))],
polymorphic_shapes=[(["(4,)", "(_,)"], [("4,")]),
dict(a="(_,)", b="(4,)")],
expected_output_signature=tf.TensorSpec([4]))
def test_with_nested_jit(self):
def f_jax(x): # x: f32[w, h]
# x + (np.sin(x) + np.broadcast_to(np.arange(x.shape[1]), x.shape))
return jnp.sin(x) + jnp.arange(x.shape[1], dtype=x.dtype)
check_shape_poly(self,
lambda x: x + jax.jit(f_jax)(x),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[(0, 1)])
def test_non_trivial_polynomials(self):
if config.jax_dynamic_shapes:
raise unittest.SkipTest("--jax_dynamic_shapes supports only trivial polynomials")
# We can handle non-trivial polynomials in the input shape,
# as long as all variables also occur in trivial polynoamials
check_shape_poly(self,
lambda x, y: x + y.reshape((-1,)),
arg_descriptors=[RandArg((9,), _f32), RandArg((3, 3), _f32)],
input_signature=[tf.TensorSpec([None]), tf.TensorSpec([None, None])],
polymorphic_shapes=["b * b", "b, b"])
def test_unused_args(self):
# Tests with functions that do not use their inputs.
# First arg unused, not polymorphic
check_shape_poly(self,
lambda x_unused, y: y * 2.0,
arg_descriptors=[RandArg((2, 3), _f32), RandArg((3,), _f32)],
input_signature=[tf.TensorSpec([]), tf.TensorSpec([None])],
polymorphic_shapes=[None, "b"])
# Some args unused, not polymorphic
check_shape_poly(self,
lambda x_unused, y, z_unused, w: jnp.concatenate([y, w]),
arg_descriptors=[RandArg((3,), _f32), RandArg((4,), _f32),
RandArg((5,), _f32), RandArg((6,), _f32)],
input_signature=[tf.TensorSpec([]), tf.TensorSpec([None]),
tf.TensorSpec([]), tf.TensorSpec([None])],
polymorphic_shapes=[None, "b1", None, "b2"])
# A polymorphic arg is not used, but the dimension var appears
# in a used arg also
check_shape_poly(self,
lambda x_unused, y: y * 2.0,
arg_descriptors=[RandArg((3,), _f32), RandArg((3,), _f32)],
input_signature=[tf.TensorSpec([None]), tf.TensorSpec([None])],
polymorphic_shapes=["b", "b"])
# A polymorphic arg is not used, and the dimension var does not appear
# elsewhere.
check_shape_poly(self,
lambda x_unused, y: y * 2.0,
arg_descriptors=[RandArg((4,), _f32), RandArg((3,), _f32)],
input_signature=[tf.TensorSpec([None]), tf.TensorSpec([None])],
polymorphic_shapes=["b1", "b2"])
# A polymorphic arg is not used, and the dimension var does appear
# elsewhere but not as a trivial monomial.
check_shape_poly(self,
lambda x_unused, y: y * 2.0,
arg_descriptors=[RandArg((3,), _f32), RandArg((9,), _f32)],
input_signature=[tf.TensorSpec([None]), tf.TensorSpec([None])],
polymorphic_shapes=["b1", "b1 * b1"])
def test_with_custom_vjp(self):
"""Shape-polymorphic custom VJP."""
@jax.custom_vjp
def f(x):
# x: [b1, b2, d1, d2] (a batch of matrices)
# res: [b1, b2, d1, d1]
return jnp.matmul(x, jnp.transpose(x, axes=(0, 1, 3, 2)))
# f_fwd: a -> (b, residual)
def f_fwd(x):
# x: [b1, b2, d1, d2]
# b: [b1, b2, d1, d1]
# res: [b1, b2, d1, d1]
# residual: [b1, b2, d1, d2]
return f(x), 3. * x
# f_bwd: (residual, CT b) -> [CT a]
def f_bwd(residual, ct_b):
# residual: [b1, b2, d1, d2]
# ct_b: [b1, b2, d1, d1]
# ct_a: [b1, b2, d1, d2]
return jnp.matmul(ct_b, residual),
f.defvjp(f_fwd, f_bwd)
x = np.ones((2, 3, 4, 5), dtype=np.float32)
res_jax = f(x)
res_jax_grad = jax.grad(lambda x: jnp.sum(f(x)))(x)
f_tf = jax2tf.convert(f, polymorphic_shapes=["(batch1, batch2, d1, d2)"])
self.assertAllClose(res_jax, f_tf(x))
xv = tf.Variable(x, dtype=np.float32)
def tf_value_and_grad(xv):
with tf.GradientTape() as tape:
tape.watch(xv)
res_tf = f_tf(xv)
res_tf_grad = tape.gradient(res_tf, xv)
return res_tf, res_tf_grad
res_tf, res_tf_grad = tf_value_and_grad(xv)
self.assertAllClose(res_jax, res_tf)
self.assertAllClose(res_jax_grad, res_tf_grad)
# Now use TF tracing for the gradient
tf_grad = tf.function(
tf_value_and_grad, autograph=False).get_concrete_function(
tf.TensorSpec([3, 4, 8, 9]))
# for native lowering we cannot refine the inferred shape of the
# output if the input is more specific than polymorphic_shapes.
if config.jax2tf_default_experimental_native_lowering:
self.assertEqual((None, None, None, None), tuple(tf_grad.output_shapes[0]))
self.assertEqual((None, None, None, None), tuple(tf_grad.output_shapes[1]))
else:
self.assertEqual((3, 4, 8, 8), tuple(tf_grad.output_shapes[0]))
self.assertEqual((3, 4, 8, 9), tuple(tf_grad.output_shapes[1]))
def test_gradients_pytree(self):
"""Shape polymorphism with gradients and pytrees for inputs and outputs."""
def f(x):
# x: dict(x=[b, 3, 4])
# res: dict(res=[b, 3, 4])
return dict(res=x["x"] * 2.)
check_shape_poly(self,
f,
skip_jax_run=True,
input_signature=[dict(x=tf.TensorSpec([None, 3, 4]))],
polymorphic_shapes=[dict(x=("b, 3, 4"))])
f_tf = jax2tf.convert(f, polymorphic_shapes=[dict(x=("b, 3, 4"))])
x = dict(x=np.ones((2, 3, 4), dtype=np.float32))
xv = tf.Variable(x["x"], dtype=np.float32)
def tf_value_and_grad(xv):
# xv: [b, 3, 4]
# res_value: dict(res=[b, 3, 4])
# res_grad: dict(grad=[b, 3, 4])
with tf.GradientTape() as tape:
tape.watch(xv)
res_tf = f_tf(dict(x=xv))
res_tf_grad = tape.gradient(res_tf, xv)
return res_tf, dict(grad=res_tf_grad)
res_tf, res_tf_grad = tf_value_and_grad(xv)
# Now use TF tracing for the gradient
tf_grad = tf.function(
tf_value_and_grad,
autograph=False).get_concrete_function(tf.TensorSpec([None, 3, 4]))
# The shape of the value
self.assertEqual((None, 3, 4), tuple(tf_grad.output_shapes[0]["res"]))
# The shape of the gradient should match the input
self.assertEqual((None, 3, 4), tuple(tf_grad.output_shapes[1]["grad"]))
def test_grad_not_var_output(self):
# Output of the function has poly shapes, non-variable
def f_jax(x): # :[b, 3]
return jnp.reshape(x, (-1,)) # : [3b]
x = np.arange(12, dtype=np.float32).reshape((4, 3))
xv = tf.Variable(x)
f_tf = jax2tf.convert(f_jax, with_gradient=True,
polymorphic_shapes=["b, ..."])
with tf.GradientTape() as tape:
res_tf = f_tf(xv)
grad_tf = tape.gradient(res_tf, xv)
self.assertAllClose(np.ones(x.shape, dtype=np.float32), grad_tf.numpy())
def test_cond(self):
# Test the primitive under conditional
def f(x, y):
# x: f32[B, H], y : f32[H]
return lax.cond(
jnp.sum(x) > 0.,
lambda _: x + y,
lambda _: jnp.zeros_like(x),
operand=None)
x = np.ones((2, 3))
y = np.ones((3,))
res_jax = f(x, y)
self.assertAllClose(
res_jax,
jax2tf.convert(f, polymorphic_shapes=["(b, h)", "h"])(x, y))
def test_while(self):
def f(x):
# x: f32[B], iter: i32
return lax.while_loop(lambda x_iter: x_iter[1] < 5,
lambda x_iter: (x_iter[0] + jnp.arange(x_iter[0].shape[0], dtype=np.float32), x_iter[1] + 1),
(x, 0))
x = np.ones((3,), dtype=np.float32)
res_tf = jax2tf.convert(f, polymorphic_shapes=["(b,)"])(x)
self.assertAllClose(f(x), res_tf)
@jtu.sample_product(with_function=[False, True])
def test_grad_int(self, with_function=True):
# https://github.com/google/jax/issues/7093
# Also issue #6975.
x_shape = (2, 3, 4)
xi = np.arange(np.prod(x_shape), dtype=np.int16).reshape(x_shape)
yf = xi.astype(np.float32)
xi_yf = (xi, yf)
zb = np.array([True, False], dtype=np.bool_)
def f_jax(xi_yf, zb): # xi: s16[2, 3, 4], yf: f32[2, 3, 4], zb: bool[2]
xi, yf = xi_yf
# Return a tuple:
# (1) float constant, with 0 tangent;
# (2) a tuple with:
# (2.1) the integer input;
# (2.2) the boolean input;
# (2.3) a float depending on both inputs.
# TODO: there is a problem if we add a None output
return (jnp.zeros(xi.shape, dtype=jnp.float32),
(xi, zb, xi.astype(np.float32) * 2. * yf))
args = (xi_yf, zb)
f_tf = jax2tf.convert(f_jax, polymorphic_shapes=[("b1, b2, 4", "b1, b2, 4"), "b1"])
if with_function:
f_tf = tf.function(f_tf, autograph=False)
res_tf, g_tf = tf_test_util.ComputeTfValueAndGrad(f_tf, args)
self.assertAllClose(g_tf[0][0], np.zeros_like(xi))
self.assertAllClose(g_tf[0][1], (xi * 2).astype(yf.dtype))
self.assertAllClose(g_tf[1], np.zeros_like(zb))
def test_prng(self):
# The PRNG implementation uses opaque types, test shape polymorphism
try:
prev_custom_prng = config.jax_enable_custom_prng
config.update("jax_enable_custom_prng", True)
def f_jax(x): # x: f32[b1, b2]
key = random.PRNGKey(123) # key
# Exercise key operations that have custom lowering rules
broadcast_keys = lax.broadcast_in_dim(key, x.shape, ()) # key[b1, b2]
gather_keys = lax.broadcast_in_dim(broadcast_keys[0], (1, x.shape[1]), (1,)) # : key[1, b2]
slice_keys1 = lax.slice(broadcast_keys, (0, 0), (1, x.shape[1]), (1, 1)) # key[1, b2]
slice_keys2 = lax.dynamic_slice(broadcast_keys, (0, 0), slice_sizes=(1, x.shape[1])) # key[1, b2]
upd1 = lax.dynamic_update_slice(slice_keys2, slice_keys1, start_indices=(0, 0)) # key[1, b2]
_ = lax.dynamic_update_slice(upd1, gather_keys, start_indices=(0, 0))
return x
check_shape_poly(self, f_jax,
arg_descriptors=[RandArg((3, 4), _f32)],
input_signature=[tf.TensorSpec([None, None], dtype=tf.float32)],
polymorphic_shapes=["b1, b2"])
finally:
config.update("jax_enable_custom_prng", prev_custom_prng)
def test_saved_model(self):
f_jax = jnp.sin
f_tf = jax2tf.convert(f_jax, polymorphic_shapes=["(b, ...)"])
x = np.array([0.7, 0.8], dtype=np.float32)
restored_f, _ = tf_test_util.SaveAndLoadFunction(
f_tf, input_signature=[tf.TensorSpec([None], x.dtype)])
self.assertAllClose(f_jax(x), restored_f(x))
# Ensure that restored_f works at other batch size as well
y = np.concatenate([x, x])
self.assertAllClose(f_jax(y), restored_f(y))
def test_saved_model_int_function(self):
def f_jax(x): # x:s32[b, 3, 4]
return jnp.reshape(x, (-1,)) # : s32[b * 12]
f_tf = jax2tf.convert(f_jax, polymorphic_shapes=["(b, ...)"])
f_tf = tf.function(f_tf, autograph=False)
x_shape = (2, 3, 4)
x = np.arange(np.prod(x_shape), dtype=np.int32).reshape(x_shape)
# When saving the model with gradients, we trace the gradient function
# and we used to get an error when creating zeros_like_aval for a
# polymorphic shape
restored_f, _ = tf_test_util.SaveAndLoadFunction(
f_tf, input_signature=[tf.TensorSpec((None,) + x.shape[1:], x.dtype)])
f_jax_rt = jax2tf.call_tf(restored_f)
res_jax_rt = f_jax_rt(x)
self.assertAllClose(f_jax(x), res_jax_rt)
def test_saved_model_constant_gradient(self):
def f_jax(x): # A function whose gradient is a constant
return 3.
f_tf = jax2tf.convert(f_jax, polymorphic_shapes=["(b, ...)"])
x = np.array([0.7, 0.8], dtype=np.float32)
restored_f, _ = tf_test_util.SaveAndLoadFunction(
f_tf, input_signature=[tf.TensorSpec([None], x.dtype)])
self.assertAllClose(3., restored_f(x))
self.assertAllClose(np.array([0., 0.], dtype=np.float32), jax.grad(f_jax)(x))
def test_readme_examples(self):
"""Some of the examples from the README."""
jax2tf.convert(lambda x: jnp.reshape(x, (x.shape[0] * x.shape[1],)),
polymorphic_shapes=["(b, 4)"])(np.ones((3, 4)))
jax2tf.convert(lambda x: jnp.reshape(x, (np.prod(x.shape),)),
polymorphic_shapes=["(b, 4)"])(np.ones((3, 4)))
jax2tf.convert(lambda x: x + x.shape[0] + jnp.sin(x.shape[0]),
polymorphic_shapes=["b"])(np.ones(3))
jax2tf.convert(lambda x: jnp.sum(x, axis=0) / x.shape[0],
polymorphic_shapes=["(v, _)"])(np.ones((3, 4)))
jax2tf.convert(lambda x: jnp.array(x.shape[0]),
polymorphic_shapes=["(v, _)"])(np.ones((4, 4)))
with self.assertRaisesRegex(TypeError,
"prod requires ndarray or scalar arguments"):
jax2tf.convert(lambda x: jnp.prod(x.shape),
polymorphic_shapes=["(b, 4)"])(np.ones((3, 4)))
jax2tf.convert(lambda x: jnp.prod(jnp.array(x.shape)),
polymorphic_shapes=["(b, 4)"])(np.ones((3, 4)))
four_ones = np.ones((4,))
with self.assertRaisesRegex(
TypeError,
re.escape("add got incompatible shapes for broadcasting: (v,), (4,)")):
jax2tf.convert(lambda x, y: x + y,
polymorphic_shapes=["(v,)", "(4,)"])(four_ones, four_ones)
# We get the error even if we use correct actual arguments
with self.assertRaisesRegex(
TypeError,
re.escape("add got incompatible shapes for broadcasting: (v,), (4,)")):
jax2tf.convert(
lambda x, y: x + y, polymorphic_shapes=["(v,)", "(4,)"])(four_ones,
four_ones)
with self.assertRaisesRegex(TypeError,
re.escape("dot_general requires contracting dimensions to have the same shape, got (4,) and (v,)")):
jax2tf.convert(lambda x: jnp.matmul(x, x),
polymorphic_shapes=["(v, 4)"])(np.ones((4, 4)))
with self.assertRaisesRegex(core.InconclusiveDimensionOperation,
re.compile("Cannot divide evenly the sizes of shapes \\(b, 5, 7\\) and \\(2, -1\\).*Details: Cannot divide '35\\*b' by '-2'",
re.DOTALL)):
jax2tf.convert(lambda x: jnp.reshape(x, (2, -1)),
polymorphic_shapes=["(b, _, _)"])(np.ones((4, 5, 7)))
jax2tf.convert(lambda x: jnp.reshape(x, (2, -1)),
polymorphic_shapes=["(b, _, _)"])(np.ones((4, 5, 6)))
jax2tf.convert(lambda x: jnp.reshape(x, (-1, x.shape[0])),
polymorphic_shapes=["(b1, b2, ...)"])(np.ones((4, 5, 6)))
jax2tf.convert(lambda x: jnp.reshape(x, (2, -1)),
polymorphic_shapes=["(2*b, ...)"])(np.ones((4, 5, 7)))
with self.assertRaisesRegex(
core.InconclusiveDimensionOperation,
re.escape("Dimension polynomial comparison 'a + 1' == 'b' is inconclusive")):
jax2tf.convert(lambda x: 0 if x.shape[0] + 1 == x.shape[1] else 1,
polymorphic_shapes=["(a, b)"])(np.ones((4, 4)))
# Unsoundness: not checking that the shape is 0
def f_jax(x):
return 0 if x.shape[0] == 0 else 1
x0 = np.array([], np.float32)
self.assertEqual(0, f_jax(x0)) # JAX sees that the x.shape[0] == 0
# jax2tf catches the broken assumption b >= 1 if the converted function is executed
# eagerly.
# Raises: ValueError: PolyShape 'b' has dimension variable 'b' corresponding to 0, for argument shape (0,)
with self.assertRaisesRegex(ValueError,
"Dimension variable b must have integer value >= 1. Found value 0 when solving .*"):
jax2tf.convert(f_jax, polymorphic_shapes=["b"])(x0)
# However, if we first trace to a TensorFlow graph, we may miss the broken assumption:
f_tf = tf.function(
jax2tf.convert(f_jax, polymorphic_shapes=["b"])).get_concrete_function(tf.TensorSpec([None], dtype=np.float32))
self.assertEqual(1, f_tf(x0))
# Unsoundness: not checking that the actual dimensions denoted by the same
# dimension variables have equal sizes.
def f_jax(x):
return 0 if x.shape[0] != x.shape[1] else 1
x45 = np.ones((4, 5), dtype=np.float32)
self.assertEqual(0, f_jax(x45)) # JAX seems that x.shape[0] != x.shape[1]
# jax2tf catches the broken assumption x.shape[0] == x.shape[1] if the converted
# function is executed eagerly.
# Raises: ValueError: PolyShape 'b, b' has dimension variable 'b' corresponding to multiple values ([4, 5]), for argument shape (4, 5)
with self.assertRaisesRegex(ValueError,
"Found inconsistency when solving b == .*"):
jax2tf.convert(f_jax, polymorphic_shapes=["b, b"])(x45)
# However, if we first trace to a TensorFlow graph, we may miss the broken assumption.
f_tf = tf.function(
jax2tf.convert(f_jax, polymorphic_shapes=["b, b"])).get_concrete_function(tf.TensorSpec([None, None], dtype=np.float32))
self.assertEqual(1, f_tf(x45))
x = np.ones((5,), dtype=np.float32)
with self.assertRaisesRegex(ValueError,
"Cannot solve for values of dimension variables"):
jax2tf.convert(lambda x: x, polymorphic_shapes=["a + b"])(x)
def test_dynamic_shapes(self):
# Test dim_as_value with dynamic shapes.
def f(x):
return jnp.sum(x, axis=0) * x.shape[0]
x = np.arange(3.)
self.assertAllClose(9., jax2tf.convert(f, polymorphic_shapes=["(b,)"])(x))
self.assertAllClose(
9.,
jax2tf.convert(jax.jit(f), polymorphic_shapes=["(b,)"])(x))
self.assertAllClose(
9.,
tf.function(jax2tf.convert(f, polymorphic_shapes=["(b,)"]))(x))
res_primal, res_tangent = jax2tf.convert(
lambda x, xt: jax.jvp(f, (x,), (xt,)),
polymorphic_shapes=["b", "b"])(x, np.array([0.1, 0.2, 0.3]))
self.assertAllClose((9., 1.8), (res_primal, res_tangent))
self.assertAllClose(
np.array([3., 3., 3.]),
jax2tf.convert(jax.grad(f), polymorphic_shapes=["b"])(x))
xv = np.arange(24.).reshape((2, 3, 4))
res_vmap = jax.vmap(f, in_axes=1)(xv)
# Implement by iteration
res_iter = jnp.stack([f(xv[:, i, :]) for i in range(xv.shape[1])])
self.assertAllClose(res_iter, res_vmap)
res_vmap_tf = jax2tf.convert(jax.vmap(f, in_axes=1),
polymorphic_shapes=["b1, b2, ..."])(xv)
self.assertAllClose(res_iter, res_vmap_tf.numpy())
def test_mean0(self):
def f_jax(x):
return jnp.sum(x, axis=0) / x.shape[0]
check_shape_poly(self,
f_jax,
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0],
expected_output_signature=tf.TensorSpec([4]))
def test_dimension_used_as_value(self):
if config.jax2tf_default_experimental_native_lowering:
raise unittest.SkipTest("TODO(necula): dim_as_value in native mode")
def f_jax(x):
poly = x.shape[0]
x1 = jnp.array(poly) # Try explicit jnp.array on poly
x2 = x1 + poly # And implicit in binary operation
x3 = x2 + jnp.sin(poly) # In jnp operations # A list or tuple of poly in jnp operations
return x3.astype(np.float32)
check_shape_poly(self,
f_jax,
arg_descriptors=[RandArg((3,), np.int32)],
poly_axes=[0],
expected_output_signature=tf.TensorSpec([]))
def test_dimension_used_as_result(self):
if config.jax_enable_x64:
raise unittest.SkipTest("TODO(necula): dim_as_value in x64 mode")
if config.jax2tf_default_experimental_native_lowering:
raise unittest.SkipTest("TODO(necula): dim_as_value in native mode")
def f_jax(x):
return 2 * x.shape[0]
check_shape_poly(self,
f_jax,
arg_descriptors=[RandArg((3,), np.int32)],
poly_axes=[0],
expected_output_signature=tf.TensorSpec([]))
def test_shape_as_array(self):
def f_jax(x):
# The entire x.shape is passed to jnp.array
return jnp.sum(jnp.array(x.shape)).astype(np.int32)
check_shape_poly(self,
f_jax,
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0])
def test_vmap_while(self):
def cond_func(x): # x: f32[3]
return jnp.sum(x) >= 0.
def body_func(x): # x: f32[3]
return x - 1.
def f_jax(x):
return lax.while_loop(cond_func, body_func, x)
check_shape_poly(self,
jax.vmap(f_jax),
arg_descriptors=[RandArg((3,), _f32)],
input_signature=[tf.TensorSpec((None, 3), dtype=tf.float32)],
polymorphic_shapes=["b, ..."],
expected_output_signature=tf.TensorSpec((None, 3), dtype=tf.float32)
)
def test_reshape_compiled(self):
# We compile the result of conversion for two shapes, hence we need to
# involve the TF compiler twice, but we trace only once with shape polymorphism
traced = False
def f_jax(x):
nonlocal traced
traced = True
y = jnp.sin(x)
return y.reshape([x.shape[0], -1])
x = self.rng().rand(4, 2, 3)
res_jax = f_jax(x)
traced = False
# If we get_concrete_function we trace once
f_tf = tf.function(
jax2tf.convert(f_jax, polymorphic_shapes=[PS("b", ...)]),
autograph=False,
jit_compile=True).get_concrete_function(
tf.TensorSpec([None, 2, 3], x.dtype))
self.assertTrue(traced)
traced = False
self.assertAllClose(res_jax, f_tf(x))
self.assertFalse(traced) # We are not tracing again
x = self.rng().rand(6, 2, 3)
res_jax = f_jax(x)
traced = False
self.assertAllClose(res_jax, f_tf(x))
self.assertFalse(traced) # We are not tracing again
# List containing either harnesses, or lists of harnesses
_POLY_SHAPE_TEST_HARNESSES = [
PolyHarness("simple_binary", "",
lambda x, y: x + jnp.sin(y),
arg_descriptors=[RandArg((3, 4), _f32), RandArg((3, 4), _f32)],
input_signature=[tf.TensorSpec([2, None]), tf.TensorSpec([2, 3])],
polymorphic_shapes="_, h",
expected_output_signature=tf.TensorSpec([2, 3])),
PolyHarness("add", "",
jnp.add,
arg_descriptors=[RandArg((3, 4), _f32), RandArg((2, 3, 4), _f32)],
poly_axes=[0, 1]),
PolyHarness("add_transpose", "",
jax.grad(lambda x: jnp.sum(jnp.sum(x, axis=0, keepdims=False) + x)),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("arange", "start",
lambda op: jnp.arange(2 * op.shape[0], dtype=_f32),
arg_descriptors=[RandArg((3,), _f32)],
poly_axes=[0]).both_enable_and_disable_xla(),
PolyHarness("arange", "start_no_dtype",
lambda op: jnp.arange(op.shape[0]),
arg_descriptors=[RandArg((3,), _f32)],
poly_axes=[0]),
PolyHarness("arange", "error1",
lambda op: jnp.arange(op.shape[0], 10),
arg_descriptors=[RandArg((3,), _f32)],
poly_axes=[0],
expect_error=(ValueError, "jax.numpy.arange supports non-constant arguments only in single-argument form")),
PolyHarness("arange", "error2",
lambda op: jnp.arange(1, op.shape[0]),
arg_descriptors=[RandArg((3,), _f32)],
poly_axes=[0],
expect_error=(ValueError, "jax.numpy.arange supports non-constant arguments only in single-argument form")),
PolyHarness("arange", "error3",
lambda op: jnp.arange(1, 5, op.shape[0]),
arg_descriptors=[RandArg((3,), _f32)],
poly_axes=[0],
expect_error=(ValueError, "jax.numpy.arange supports non-constant arguments only in single-argument form")),
# Reduce the poly dimension
PolyHarness("argmax", "0",
lambda op: lax.argmax(op, axis=0, index_dtype=np.int32),
arg_descriptors=[RandArg((3, 4, 5), _f32)],
poly_axes=[0]).both_enable_and_disable_xla(),
# Reduce the non-poly dimension
PolyHarness("argmax", "1",
lambda op: lax.argmax(op, axis=1, index_dtype=np.int32),
arg_descriptors=[RandArg((3, 4, 5), _f32)],
poly_axes=[0]).both_enable_and_disable_xla(),
[
PolyHarness("average",
f"{axis=}_weights=None",
lambda x, axis: jnp.average(x, axis=axis, returned=False, weights=None),
arg_descriptors=[RandArg((7, 8, 4), _f32), StaticArg(axis)],
poly_axes=[0])
for axis in [None, 0, 1]
],
[
PolyHarness("average",
f"{axis=}_weights=Some",
lambda x, weights, axis: jnp.average(x, axis=axis, returned=False, weights=weights),
arg_descriptors=[RandArg((7, 8, 4), _f32), RandArg((7, 8, 4), _f32), StaticArg(axis)],
poly_axes=[0, 0])
for axis in [None, 0, 1]
],
PolyHarness("broadcast_to", "",
lambda x: jnp.broadcast_to(x, [x.shape[0], x.shape[0], 4]),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("broadcast_in_dim", "0",
lambda x: lax.broadcast_in_dim(x, [x.shape[0], 4, 5, 6],
broadcast_dimensions=(0, 2, 3)),
arg_descriptors=[RandArg((3, 1, 6), _f32)],
poly_axes=[0]),
PolyHarness("broadcast_in_dim", "poly",
lambda x: lax.broadcast_in_dim(x, [x.shape[0], x.shape[0] + x.shape[0], 4],
broadcast_dimensions=(0, 1, 2)),
arg_descriptors=[RandArg((3, 1, 4), _f32)],
poly_axes=[0]),
PolyHarness("broadcast_in_dim", "poly2",
lambda x: lax.broadcast_in_dim(x, [x.shape[0], 5, 6, x.shape[2], 4],
broadcast_dimensions=(0, 2, 3)),
arg_descriptors=[RandArg((3, 1, 4), _f32)],
poly_axes=[(0, 2)]),
PolyHarness("broadcast_in_dim", "transpose",
jax.grad(lambda x: jnp.sum(lax.broadcast_in_dim(x, [2, x.shape[0], 5, x.shape[2], 4],
broadcast_dimensions=(1, 2, 3)))),
arg_descriptors=[RandArg((3, 1, 4), _f32)],
poly_axes=[(0, 2)]),
PolyHarness("clamp", "",
lax.clamp,
arg_descriptors=[RandArg((3, 4, 5), _f32), RandArg((3, 4, 5), _f32),
RandArg((3, 4, 5), _f32)],
poly_axes=[0, 0, 0]),
PolyHarness("collapse", "",
lambda x: lax.collapse(x, 1, 4),
arg_descriptors=[RandArg((3, 4, 5, 6, 7), _f32)],
poly_axes=[(0, 1, 3)]),
PolyHarness("concatenate", "",
lambda x: jnp.concatenate([x, x], axis=0),
arg_descriptors=[RandArg((3, 4, 5), _f32)],
poly_axes=[(0, 1)]),
PolyHarness("concatenate", "grad",
jax.grad(lambda x: jnp.sum(jnp.concatenate([x, x], axis=0))),
arg_descriptors=[RandArg((3, 4, 5), _f32)],
poly_axes=[(0, 1)]),
# Issue #11402
# We play a trick here. Since the stride is 2, when we compute the padding
# for "SAME" we need to divide by 2. We cannot do this in general, so we
# write the test with the assumption that the dimension is a multiple of 2.
# We pass the lhs as (1, b, 2, 16) and then we
# reshape it as (1, 2*b, 16), so that we know that the lhs's dimension 1
# is a multiple of 2.
PolyHarness("conv_general_dilated", "1d_1",
lambda lhs, rhs: lax.conv_general_dilated(
jnp.reshape(lhs, (1, -1, 16)), rhs,
window_strides=(2,),
padding="SAME",
rhs_dilation=None,
dimension_numbers=lax.ConvDimensionNumbers(lhs_spec=(0, 2, 1),
rhs_spec=(2, 1, 0),
out_spec=(0, 2, 1))),
arg_descriptors=[RandArg((1, 6, 2, 16), _f32), RandArg((4, 16, 16), _f32)],
poly_axes=[1, None]).both_enable_and_disable_xla(),
# The same example from above, but without the reshape trick.
PolyHarness("conv_general_dilated", "1d_1err",
lambda lhs, rhs: lax.conv_general_dilated(
lhs, rhs,
window_strides=(2,),
padding="SAME",
rhs_dilation=None,
dimension_numbers=lax.ConvDimensionNumbers(lhs_spec=(0, 2, 1),
rhs_spec=(2, 1, 0),
out_spec=(0, 2, 1))),
arg_descriptors=[RandArg((1, 12, 16), _f32), RandArg((4, 16, 16), _f32)],
poly_axes=[1, None],
expect_error=(core.InconclusiveDimensionOperation,
"Cannot divide .* by '2'")
).both_enable_and_disable_xla(),
# Issue #11402
PolyHarness("conv_general_dilated", "1d_2",
lambda lhs, rhs: lax.conv_transpose(lhs, rhs,
strides=(2,),
padding="SAME",
rhs_dilation=None,
transpose_kernel=False),
arg_descriptors=[RandArg((5, 12, 16), _f32), RandArg((4, 16, 16), _f32)],
poly_axes=[0, None],
tol=1e-5).both_enable_and_disable_xla(),
# Issue #11402
PolyHarness("conv_general_dilated", "1d_3",
lambda lhs, rhs: lax.conv_transpose(lhs, rhs,
strides=(2,),
padding="SAME",
rhs_dilation=None,
transpose_kernel=False),
arg_descriptors=[RandArg((5, 12, 16), _f32), RandArg((4, 16, 16), _f32)],
poly_axes=[1, None],
tol=1e-5).both_enable_and_disable_xla(),
PolyHarness("conv_general_dilated", "",
lambda lhs, rhs: lax.conv_general_dilated(
lhs, rhs,
window_strides=(2, 3),
padding=((0, 0), (0, 0)),
lhs_dilation=(1, 1),
rhs_dilation=(1, 2),
dimension_numbers=("NCHW", "OIHW", "NCHW"),
feature_group_count=1,
batch_group_count=1,
precision=None),
arg_descriptors=[RandArg((7, 3, 9, 10), _f32), RandArg((3, 3, 4, 5), _f32)],
poly_axes=[0, None]).both_enable_and_disable_xla(),
PolyHarness("cummax", "",
lambda x: lax_control_flow.cummax(x, axis=1, reverse=False),
arg_descriptors=[RandArg((3, 4, 5), _f32)],
poly_axes=[0]),
PolyHarness("delta", "0",
lambda x: lax_internal._delta(_f32, x.shape, axes=(0, 1)),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("dot_general", "",
lambda lhs, rhs: lax.dot_general(lhs, rhs,
dimension_numbers=(((2,), (1,)), ((0,), (0,)))),
arg_descriptors=[RandArg((3, 4, 4), _f32), RandArg((3, 4), _f32)],
poly_axes=[0, 0]),
PolyHarness("dynamic_slice", "idx=tuple_int",
# x:shape: (b, 4)
lambda x: lax.dynamic_slice(x, (0, 1), (x.shape[0], 2)),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]).both_enable_and_disable_xla(),
PolyHarness("dynamic_slice", "idx=tuple_arg",
# x:shape: (b, 4)
lambda x, i0: lax.dynamic_slice(x, (i0, np.int32(1)), (x.shape[0], 2)),
arg_descriptors=[RandArg((3, 4), _f32), np.array(-2, dtype=np.int32)],
poly_axes=[0, None]).both_enable_and_disable_xla(),
PolyHarness("dynamic_slice", "idx=array",
# x:shape: (b, 4)
lambda x, idx: lax.dynamic_slice(x, idx, (x.shape[0], 2)),
arg_descriptors=[RandArg((3, 4), _f32), np.array([-2, -1], dtype=np.int32)],
poly_axes=[0, None]).both_enable_and_disable_xla(),
PolyHarness("dynamic_slice_in_dim", "idx=0",
# x:shape: (b, 4)
lambda x: lax.dynamic_slice_in_dim(x, 0, x.shape[0], axis=0),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]).both_enable_and_disable_xla(),
PolyHarness("dynamic_update_slice", "idx=tuple_int",
# x:shape: (b, 4)
lambda x: lax.dynamic_update_slice(x, x, (0, 0)),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]).both_enable_and_disable_xla(),
PolyHarness("dynamic_update_slice", "idx=tuple_arg",
# x:shape: (b, 4)
lambda x, i0: lax.dynamic_update_slice(x, x, (i0, np.int32(0))),
arg_descriptors=[RandArg((3, 4), _f32), np.array(-2, dtype=np.int32)],
poly_axes=[0, None]).both_enable_and_disable_xla(),
PolyHarness("dynamic_update_slice", "idx=array",
# x:shape: (b, 4)
lambda x, idx: lax.dynamic_update_slice(x, x, idx),
arg_descriptors=[RandArg((3, 4), _f32), np.array([-2, -1], dtype=np.int32)],
poly_axes=[0, None]).both_enable_and_disable_xla(),
PolyHarness("einsum", "0",
lambda x: jnp.einsum("...i->...", x),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("einsum", "0_alt",
lambda x: jnp.einsum(x, (..., 1), [...]),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("einsum", "1",
lambda x, y: jnp.einsum("...ij,...jk->...ik", x, y),
arg_descriptors=[RandArg((3, 4, 5), _f32), RandArg((3, 5, 6), _f32)],
poly_axes=[0, 0]),
PolyHarness("einsum", "1_alt",
lambda x, y: jnp.einsum(x, [..., 0, 1], y, (..., 1, 2), [..., 0, 2]),
arg_descriptors=[RandArg((3, 4, 5), _f32), RandArg((3, 5, 6), _f32)],
poly_axes=[0, 0]),
PolyHarness("einsum", "2",
lambda x, y: jnp.einsum("...ij,jk->...ik", x, y),
arg_descriptors=[RandArg((3, 4, 5), _f32), RandArg((5, 6), _f32)],
poly_axes=[0, None]),
PolyHarness("einsum", "2_alt",
lambda x, y: jnp.einsum(x, [..., 0, 1], y, [1, 2], [..., 0, 2]),
arg_descriptors=[RandArg((3, 4, 5), _f32), RandArg((5, 6), _f32)],
poly_axes=[0, None]),
PolyHarness("einsum", "3",
# Reduced dimension is polymorphic
lambda x, y: jnp.einsum("ij,jk->ik", x, y),
arg_descriptors=[RandArg((3, 4), _f32), RandArg((4, 5), _f32)],
poly_axes=[1, 0]),
PolyHarness("einsum", "3_alt",
# Reduced dimension is polymorphic
lambda x, y: jnp.einsum(x, [0, 1], y, [1, 2], [0, 2]),
arg_descriptors=[RandArg((3, 4), _f32), RandArg((4, 5), _f32)],
poly_axes=[1, 0]),
PolyHarness("einsum", "4",
# Reduced dimension is polymorphic, and is 2*b
lambda x, y: jnp.einsum("ij,jk->ik",
jnp.concatenate([x, x], axis=1),
jnp.concatenate([y, y], axis=0)),
arg_descriptors=[RandArg((3, 4), _f32), RandArg((4, 5), _f32)],
poly_axes=[1, 0]),
PolyHarness("einsum", "4_alt",
# Reduced dimension is polymorphic, and is 2*b
lambda x, y: jnp.einsum(jnp.concatenate([x, x], axis=1), [0, 1],
jnp.concatenate([y, y], axis=0), [1, 2],
[0, 2]),
arg_descriptors=[RandArg((3, 4), _f32), RandArg((4, 5), _f32)],
poly_axes=[1, 0]),
PolyHarness("einsum", "multiple_contractions",
lambda x, y, z: jnp.einsum("ab,bc,cd->ad", x, y, z),
arg_descriptors=[RandArg((3, 2), _f32), RandArg((2, 3), _f32), RandArg((3, 4), _f32)],
poly_axes=[0, None, None]),
PolyHarness("einsum", "incompatible_contractions_error",
lambda x, y: jnp.einsum("ab,cb->ac", x, y),
arg_descriptors=[RandArg((2, 3), _f32), RandArg((2, 3), _f32)],
polymorphic_shapes=["(2, b0)", "(2, b1)"],
input_signature=[tf.TensorSpec((2, None)), tf.TensorSpec((2, None))],
expect_error=(core.InconclusiveDimensionOperation,
"Dimension polynomial comparison 'b1' == 'b0' is inconclusive")),
PolyHarness("eye", "N=poly_M=None",
lambda x: jnp.eye(x.shape[0]),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("eye", "N=poly_M=poly",
lambda x: jnp.eye(x.shape[0], M=x.shape[0] + 2),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("full", "",
lambda x: lax.full((x.shape[0], 2), 3.),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
# operand is non-poly, index is poly
PolyHarness("getitem", "op=static_idx=poly",
lambda a, i: a[i],
arg_descriptors=[RandArg((3, 4), _f32), np.array([2, 2], np.int32)],
poly_axes=[None, 0]).both_enable_and_disable_xla(),
# operand is poly, index is integer
PolyHarness("getitem", "op=poly_idx=const",
lambda a: a[1],
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]).both_enable_and_disable_xla(),
# operand is poly, index is dim poly
PolyHarness("getitem", "op=poly_idx=dim",
lambda a: a[jnp.array(a.shape[0] - 2)],
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]).both_enable_and_disable_xla(),
# Both the operand and the index are poly
PolyHarness("getitem", "op=poly_idx=poly",
lambda a, i: a[i],
arg_descriptors=[RandArg((3, 4), _f32), np.array([1, 2, 0], np.int32)],
poly_axes=[0, 0]).both_enable_and_disable_xla(),
# op is poly and index is an entire slice
PolyHarness("getitem", "op=poly_idx=slice-all",
lambda a: a[:],
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]).both_enable_and_disable_xla(),
# op is poly and index is a partial slice
PolyHarness("getitem", "op=poly_idx=slice-ct-1",
lambda a: a[:2],
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0],
expect_error=(IndexError, "Cannot use NumPy slice indexing on an array dimension")
).both_enable_and_disable_xla(),
PolyHarness("getitem", "op=poly_idx=slice-ct-2",
lambda a: a[:, :2],
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]).both_enable_and_disable_xla(),
PolyHarness("getitem", "op=poly_idx=slice-None-1",
lambda a: a[:a.shape[0]],
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]).both_enable_and_disable_xla(),
PolyHarness("getitem", "op=poly_idx=slice-poly",
lambda a: a[:a.shape[0] - 1],
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0],
expect_error=(IndexError, "Array slice indices must have static")).both_enable_and_disable_xla(),
PolyHarness("image_resize", "linear_0",
lambda x: jax.image.resize(x, (x.shape[0], 2 * x.shape[1], 2 * x.shape[2], x.shape[3]),
method="linear"),
arg_descriptors=[RandArg((3, 16, 32, 3), _f32)],
poly_axes=[(1, 2)]),
PolyHarness("image_resize", "linear_to_fixed_dim",
lambda x: jax.image.resize(x, (x.shape[0], 64, 64, x.shape[3]),
method="linear"),
arg_descriptors=[RandArg((3, 16, 32, 3), _f32)],
poly_axes=[(1, 2)]),
PolyHarness("image_resize", "nearest_0",
lambda x: jax.image.resize(x, (x.shape[0], 2 * x.shape[1], 2 * x.shape[2], x.shape[3]),
method="nearest"),
arg_descriptors=[RandArg((3, 5, 7, 3), _f32)],
poly_axes=[(1, 2)]),
PolyHarness("index_in_dim", "0",
lambda x: lax.index_in_dim(x, -1, axis=0, keepdims=False),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("index_in_dim", "idx=neg",
lambda x: lax.index_in_dim(x, -1, axis=0, keepdims=False),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("index_in_dim", "idx=last",
lambda x: lax.index_in_dim(x, x.shape[0] - 1, axis=0, keepdims=False),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("iota", "",
lambda x: x + lax.iota(_f32, x.shape[0]),
arg_descriptors=[RandArg((3,), _f32)],
poly_axes=[0]),
PolyHarness("matmul", "0",
jnp.matmul,
arg_descriptors=[RandArg((7, 8, 4), _f32), RandArg((7, 4, 5), _f32)],
poly_axes=[0, 0],
tol=1e-5),
PolyHarness("matmul", "1",
jnp.matmul,
arg_descriptors=[RandArg((7, 8, 4), _f32), RandArg((4, 5), _f32)],
poly_axes=[0, None],
tol=1e-5),
[
PolyHarness("mean",
f"{axis=}_{keepdims=}_where=None",
lambda x, axis, keepdims: jnp.mean(x, axis=axis, keepdims=keepdims, where=None),
arg_descriptors=[RandArg((7, 8, 4), _f32), StaticArg(axis), StaticArg(keepdims)],
poly_axes=[0])
for keepdims in [False, True]
for axis in [None, (0,), (0, 1), (1,)]
],
[
PolyHarness("mean",
f"{axis=}_{keepdims=}_where=Some",
lambda x, where, axis, keepdims: jnp.mean(x, axis=axis, keepdims=keepdims, where=where),
arg_descriptors=[RandArg((7, 8, 4), _f32), RandArg((7, 8, 4), np.bool_),
StaticArg(axis), StaticArg(keepdims)],
poly_axes=[0, 0])
for keepdims in [False, True]
for axis in [None, (0,), (0, 1), (1,)]
],
PolyHarness("ones", "",
lambda x: jnp.ones(x.shape, dtype=_f32),
arg_descriptors=[RandArg((3, 2, 4), _f32)],
poly_axes=[0]),
PolyHarness("pad", "",
lax.pad,
arg_descriptors=[RandArg((3, 2, 5), _f32), np.float32(5.),
StaticArg(((0, 0, 0), (0, 0, 0), (1, 1, 1)))],
poly_axes=[0, None]),
PolyHarness("random_gamma", "",
lambda key, a: jax.random.gamma(key, a),
arg_descriptors=[RandArg((3, 2), np.uint32), RandArg((3, 3), _f32)],
poly_axes=[0, 0]),
# The known dimensions product must be even.
PolyHarness("random_categorical", "axis=0",
lambda key, a: jax.random.categorical(key, a, axis=0),
arg_descriptors=[RandArg((2,), np.uint32), RandArg((3, 8), _f32)],
poly_axes=[None, 0]),
PolyHarness("random_categorical", "axis=1",
lambda key, a: jax.random.categorical(key, a, axis=1),
arg_descriptors=[RandArg((2,), np.uint32), RandArg((3, 5, 8), _f32)],
poly_axes=[None, (0, 1)]),
PolyHarness("random_categorical", "axis=1_then_reshape",
lambda key, a: jax.random.categorical(key, a, axis=1).reshape((-1)),
arg_descriptors=[RandArg((2,), np.uint32), RandArg((3, 5, 8), _f32)],
poly_axes=[None, (0, 1)]),
PolyHarness("random_categorical", "0_dim", # One axis has 0 size
lambda key, a: jax.random.categorical(key, a, axis=1),
arg_descriptors=[RandArg((2,), np.uint32), RandArg((3, 5, 0), _f32)],
poly_axes=[None, (0, 1)]),
# Works when the known dimensions are known to be even or odd.
PolyHarness("random_uniform", "even_1",
lambda key, a: jax.random.uniform(key, a.shape, dtype=_f32),
arg_descriptors=[RandArg((2,), np.uint32), RandArg((3, 4), _f32)],
poly_axes=[None, 0]),
PolyHarness("random_uniform", "even_2",
lambda key, a: jax.random.uniform(key, (2 * a.shape[0], a.shape[1]),
dtype=_f32),
arg_descriptors=[RandArg((2,), np.uint32), RandArg((3, 5), _f32)],
poly_axes=[None, 0]),
PolyHarness("random_uniform", "error_not_even",
lambda key, a: jax.random.uniform(key, a.shape, dtype=_f32),
arg_descriptors=[RandArg((2,), np.uint32), RandArg((3, 5), _f32)],
poly_axes=[None, 0],
expect_error=(core.InconclusiveDimensionOperation,
"the product of the known dimensions must be even")),
PolyHarness("reduce_window", "min",
# x.shape = (b, 8)
lambda x: lax.reduce_window(x, np.array(1., _f32), lax.min,
(2, 2), (1, 1), "VALID"),
arg_descriptors=[RandArg((3, 8), _f32)],
poly_axes=[0]).both_enable_and_disable_xla(),
PolyHarness("reduce_window", "add_0",
# x.shape = (b, 8)
lambda x: lax.reduce_window(x, 0, lax.add, (2, 2), (1, 1),
"VALID"),
arg_descriptors=[RandArg((3, 8), _f32)],
poly_axes=[0]).both_enable_and_disable_xla(),
# https://github.com/google/jax/issues/11804
# Use the reshape trick to simulate a polymorphic dimension of 16*b.
# (See test "conv_general_dilated.1d_1" above for more details.)
PolyHarness("reduce_window", "add_1",
# x.shape = (1, 16*b, 1)
lambda x: lax.reduce_window(
jnp.reshape(x, (1, -1, 1)),
0., lax.add, (1, 4, 1), (1, 2, 1), "SAME"),
arg_descriptors=[RandArg((1, 128, 16), _f32)],
poly_axes=[1]).both_enable_and_disable_xla(),
# TODO(necula): not yet supported, but also unlikely to come up.
# PolyHarness("random_uniform", "odd",
# lambda key, a: jax.random.uniform(key, (2 * a.shape[0] + 1, a.shape[1]),
# dtype=_f32),
# [RandArg((2,), np.uint32), RandArg((3, 5), _f32)],
# poly_axes=[None, 0]),
[
PolyHarness("reduce", reduce_op.__name__,
lambda x: reduce_op(x, axis=-1, keepdims=True), # type: ignore
arg_descriptors=[RandArg((3, 5), _f32)],
poly_axes=[0])
for reduce_op in [jnp.all, jnp.any, jnp.max, jnp.min, jnp.prod, jnp.sum]
],
# Repeat f32[b, 2] * 3
PolyHarness("repeat", "repeats=int_axis=0",
lambda x: jnp.repeat(x, repeats=3, axis=0),
arg_descriptors=[RandArg((3, 2), _f32)],
poly_axes=[0]),
# Repeat f32[b, 2] * b
PolyHarness("repeat", "repeats=poly_axis=0",
lambda x: jnp.repeat(x, repeats=x.shape[0], axis=0),
arg_descriptors=[RandArg((3, 2), _f32)],
poly_axes=[0]),
# Repeat f32[b, 2] * b
PolyHarness("repeat", "repeats=poly_axis=None",
lambda x: jnp.repeat(x, repeats=x.shape[0], axis=None),
arg_descriptors=[RandArg((3, 2), _f32)],
poly_axes=[0]),
# Repeat f32 * b
PolyHarness("repeat", "repeats=poly_axis=None_scalar",
lambda x, y: jnp.repeat(x, repeats=y.shape[0], axis=None),
arg_descriptors=[RandArg((), _f32), RandArg((3, 2), _f32)],
poly_axes=[None, 0]),
PolyHarness("repeat", "repeats=poly_axis=None_total_repeat_length1",
lambda x: jnp.repeat(x, repeats=x.shape[0], axis=None, total_repeat_length=8),
arg_descriptors=[RandArg((3, 2), _f32)],
poly_axes=[0],
expect_error=(ValueError, "jnp.repeat with a DimPolynomial `repeats` is supported only .*")),
PolyHarness("reshape", "0",
lambda x: x.reshape([x.shape[0], -1]),
arg_descriptors=[RandArg((3, 2, 3), _f32)],
poly_axes=[0]),
PolyHarness("reshape", "1",
lambda x: x.reshape([x.shape[0], -1]),
arg_descriptors=[RandArg((3, 2, 3), _f32)],
poly_axes=[(0, 1)]),
PolyHarness("reshape", "2",
lambda x: x.reshape([x.shape[0], -1, x.shape[3], x.shape[2]]),
arg_descriptors=[RandArg((3, 4, 5, 6, 7), _f32)],
poly_axes=[(0, 2, 3)]),
PolyHarness("reshape", "3",
lambda x: jnp.reshape(x, [2, -1]),
arg_descriptors=[RandArg((3, 4, 5, 6, 7), _f32)],
poly_axes=[(0, 2)]),
PolyHarness("reshape", "_issue_9975",
# The newshape is a scalar
lambda x: jnp.reshape(x, x.shape[0] * x.shape[1]),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("reshape", "error",
lambda x: x.reshape([x.shape[0], -1, 3]),
arg_descriptors=[RandArg((3, 2, 4), _f32)],
poly_axes=[0],
skip_jax_run=True,
expect_error=(core.InconclusiveDimensionOperation,
re.escape(
"Cannot divide evenly the sizes of shapes (b0, 2, 4) and (b0, -1, 3)"))),
PolyHarness("roll", "axis=0",
lambda x: jnp.roll(x, 2, axis=0),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("roll", "axis=None",
lambda x: jnp.roll(x, 2),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("scatter_add", "",
partial(lax.scatter_add, indices_are_sorted=False, unique_indices=True),
arg_descriptors=[RandArg((7, 4), _f32),
np.array([[1], [2]], np.int32), # indices: [2, 1]
RandArg((7, 2), _f32), # updates: [7, 2]
StaticArg(lax.ScatterDimensionNumbers((0,), (1,), (1,)))],
poly_axes=[0, None, 0]),
PolyHarness("scatter_add", "clip0",
partial(lax.scatter_add, indices_are_sorted=False, unique_indices=True, mode=lax.GatherScatterMode.CLIP),
arg_descriptors=[RandArg((7, 4), _f32), # [b, 4]
np.array([[1], [2]], np.int32), # indices: [2, 1]
RandArg((7, 2), _f32), # updates: [b, 2]
StaticArg(lax.ScatterDimensionNumbers((0,), (1,), (1,)))],
poly_axes=[0, None, 0]),
PolyHarness("scatter_add", "clip1",
partial(lax.scatter_add, indices_are_sorted=False, unique_indices=True, mode=lax.GatherScatterMode.CLIP),
arg_descriptors=[RandArg((7, 4), _f32), # [b, 4]
np.array([[1, 2], [-2, 0], [6, 4], [7, -1], [1, 0], [3, 0], [0, 5]], np.int32), # indices: [b, 2]
RandArg((7, 1), _f32), # updates: [b, 1]
StaticArg(lax.ScatterDimensionNumbers((1,), (0,), (0, 1,)))],
poly_axes=[0, 0, 0]),
PolyHarness("select", "0",
# x.shape = (b, 3)
lambda x: lax.select(x > 5., x, x),
arg_descriptors=[RandArg((7, 3), _f32)],
poly_axes=[0]),
PolyHarness("select", "1",
# x.shape = (b, 3); y.shape = (3,)
jax.vmap(lambda x, y: lax.select(x > 5., x, y), in_axes=[0, None]),
arg_descriptors=[RandArg((7, 3), _f32), RandArg((3,), _f32)],
poly_axes=[0, None]),
PolyHarness("slice", "entire_axis",
lambda x: lax.slice(x, start_indices=(0, 1), limit_indices=(x.shape[0], 3)),
arg_descriptors=[RandArg((7, 3), _f32)],
poly_axes=[0]),
PolyHarness("slice_in_dim", "entire_axis",
lambda x: lax.slice_in_dim(x, 0, x.shape[0], stride=1, axis=0),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("slice_in_dim", "start=neg",
lambda x: lax.slice_in_dim(x, -1, x.shape[0], stride=1, axis=0),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("slice_in_dim", "limit=neg",
lambda x: lax.slice_in_dim(x, 0, -1, stride=1, axis=0),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("squeeze", "axis=empty",
jnp.squeeze,
arg_descriptors=[RandArg((5,), _f32), StaticArg(())],
poly_axes=[0]),
PolyHarness("squeeze", "axis=None",
jnp.squeeze,
arg_descriptors=[RandArg((5,), _f32), StaticArg(None)],
poly_axes=[0],
expect_error=(ValueError, "jnp.squeeze with axis=None is not supported with shape polymorphism")),
PolyHarness("squeeze", "axis=1",
jnp.squeeze,
arg_descriptors=[RandArg((4, 1), _f32), StaticArg((1,))],
poly_axes=[0]),
PolyHarness("squeeze", "axis=1_2",
jnp.squeeze,
arg_descriptors=[RandArg((4, 1, 1), _f32), StaticArg((1, 2))],
poly_axes=[0]),
PolyHarness("squeeze", "error",
jnp.squeeze,
arg_descriptors=[RandArg((3, 33), _f32), StaticArg(-1)],
poly_axes=[(0, 1)],
skip_jax_run=True,
expect_error=(ValueError,
re.escape(
"cannot select an axis to squeeze out which has size not equal to one, got shape=(b0, b1) and dimensions=(1,)"))
),
PolyHarness("take", "",
lambda a, i: jnp.take(a, i, axis=1),
arg_descriptors=[RandArg((3, 4, 5), _f32), np.array([1, 2], np.int32)],
poly_axes=[0, None]).both_enable_and_disable_xla(),
PolyHarness("take_along_axis", "0",
lambda x, y: jnp.take_along_axis(x, y, axis=0),
arg_descriptors=[RandArg((5, 2), _f32), RandArg((5, 1), np.int32)],
poly_axes=[0, 0]),
PolyHarness("take_along_axis", "1",
lambda x, y: jnp.take_along_axis(x, y, axis=1),
arg_descriptors=[RandArg((5, 2), _f32), RandArg((5, 1), np.int32)],
poly_axes=[0, 0]),
PolyHarness("tile", "0",
lambda x: jnp.tile(x, (1, 2)),
arg_descriptors=[RandArg((4, 3), _f32)],
poly_axes=[0]),
PolyHarness("tile", "1",
# The repetitions are polys
lambda x: jnp.tile(x, (1, x.shape[0])),
arg_descriptors=[RandArg((4, 2), _f32)],
poly_axes=[0]),
PolyHarness("tri", "N=poly_M=None",
lambda x: jnp.tri(x.shape[0]),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
PolyHarness("tri", "N=poly_M=poly",
lambda x: jnp.tri(x.shape[0], M=x.shape[0] + 2),
arg_descriptors=[RandArg((3, 4), _f32)],
poly_axes=[0]),
[
PolyHarness("var",
f"{axis=}_{keepdims=}_where=None",
lambda x, axis, keepdims: jnp.var(x, axis=axis, keepdims=keepdims, where=None),
arg_descriptors=[RandArg((7, 8, 4), _f32), StaticArg(axis), StaticArg(keepdims)],
poly_axes=[0])
for keepdims in [False, True]
for axis in [None, (0,), (0, 1), (1,)]
],
[
PolyHarness("var",
f"{axis=}_{keepdims=}_where=Some",
lambda x, where, axis, keepdims: jnp.var(x, axis=axis, keepdims=keepdims, where=where),
arg_descriptors=[RandArg((7, 8, 4), _f32), RandArg((7, 8, 4), np.bool_), StaticArg(axis), StaticArg(keepdims)],
poly_axes=[0, 0])
for keepdims in [False, True]
for axis in [None, (0,), (0, 1), (1,)]
],
PolyHarness("where", "",
jnp.where,
arg_descriptors=[RandArg((2,), np.bool_), RandArg((), _f32), RandArg((2,), _f32)],
poly_axes=[0, None, 0]),
]
def _get_jax2tf_limitations(
device, h: primitive_harness.Harness) -> Sequence[Jax2TfLimitation]:
# And the jax2tf limitations
def applicable_jax2tf_limitation(l: Jax2TfLimitation) -> bool:
# The CheckShapePolymorphism uses tf.function, so we care about "graph"
return l.filter(device=device, dtype=h.dtype, mode="graph")
limitations = Jax2TfLimitation.limitations_for_harness(h)
return tuple(filter(applicable_jax2tf_limitation, limitations))
### We add to the test harnesses some that are obtained from the
### primitive harnesses by applying vmap to the function and then asserting
### that we can convert shape polymorphically the result.
def _make_vmap_primitive_harnesses() -> Sequence[PolyHarness]:
"""For each harness group, pick a single dtype.
See PolyHarness for documentation.
Ignore harnesses that fail in graph mode in jax2tf.
"""
all_h = primitive_harness.all_harnesses
res = []
# Index by group
harness_groups: Dict[
str, Sequence[primitive_harness.Harness]] = collections.defaultdict(list)
device = jtu.device_under_test()
for h in all_h:
# Drop the JAX limitations
if not h.filter(device_under_test=device, include_jax_unimpl=False):
continue
# And the jax2tf limitations that are known to result in TF error.
if any(l.expect_tf_error for l in _get_jax2tf_limitations(device, h)):
continue
# TODO(marcvanzee): We currently exclude tests with enable_xla=False because
# this doesn't work with vmap due to a call to lax.gather. We should include
# them once vmap works with enable_xla=False.
if not h.params.get("enable_xla", True):
continue
harness_groups[h.group_name].append(h)
selected_harnesses = []
for group_name, hlist in harness_groups.items():
# Pick the dtype with the most harnesses in this group. Some harness
# groups only test different use cases at a few dtypes.
c = collections.Counter([h.dtype for h in hlist])
(dtype, _), = c.most_common(1)
selected_harnesses.extend([h for h in hlist if h.dtype == dtype])
# We do not yet support shape polymorphism for vmap for some primitives
_NOT_SUPPORTED_YET = frozenset([
# In linalg._lu_python we do reshape(-1, ...)
"lu",
"custom_linear_solve",
# We do *= shapes in the batching rule for conv_general_dilated
"conv_general_dilated",
"tridiagonal_solve", # batching not implemented in JAX
"iota", # vmap does not make sense for 0-argument functions
"rng_bit_generator", # vmap not implemented
])
batch_size = 3
for h in selected_harnesses:
if h.group_name in _NOT_SUPPORTED_YET:
continue
def make_batched_arg_descriptor(
ad: primitive_harness.ArgDescriptor) -> Optional[primitive_harness.ArgDescriptor]:
if isinstance(ad, RandArg):
return RandArg((batch_size,) + ad.shape, ad.dtype)
elif isinstance(ad, CustomArg):
def wrap_custom(rng):
arg = ad.make(rng)
return np.stack([arg] * batch_size)
return CustomArg(wrap_custom)
else:
assert isinstance(ad, np.ndarray), ad
return np.stack([ad] * batch_size)
new_args = [make_batched_arg_descriptor(ad)
for ad in h.arg_descriptors
if not isinstance(ad, StaticArg)]
# This test does not make sense for nullary functions
if not new_args:
continue
# We do not check the result of harnesses that require custom assertions.
check_result = all(not l.custom_assert and not l.skip_comparison and l.tol is None
for l in _get_jax2tf_limitations(device, h))
if h.group_name == "cumsum":
# TODO(necula): why do we need to adjust the cumsum tolerance?
tol = 1e-5
else:
tol = None
vmap_harness = PolyHarness("vmap_" + h.group_name, h.name,
jax.vmap(h.dyn_fun, in_axes=0, out_axes=0),
arg_descriptors=new_args,
poly_axes=[0] * len(new_args),
check_result=check_result,
tol=tol)
vmap_harness.original_harness = h
res.append(vmap_harness)
return res
_POLY_SHAPE_TEST_HARNESSES.append(_make_vmap_primitive_harnesses())
def _flatten_harnesses(harnesses):
res = []
for h in harnesses:
if isinstance(h, Sequence):
res.extend(h)
else:
res.append(h)
return res
class ShapePolyPrimitivesTest(tf_test_util.JaxToTfTestCase):
"""Tests for primitives that take shape values as parameters."""
# This test runs for all _POLY_SHAPE_PRIMITIVE_HARNESSES.
# For each primitive "xxx" the test will be called "test_harness_xxx_...".
# If you want to run this test for only one harness that includes "foo"
# in the name (after test_harness), add parameter `one_containing="foo"`
# to parameterized below.
@primitive_harness.parameterized(
_flatten_harnesses(_POLY_SHAPE_TEST_HARNESSES),
#one_containing="",
)
def test_harness(self, harness: PolyHarness):
# Exclude some harnesses that are known to fail for native serialization
if config.jax2tf_default_experimental_native_lowering:
if not harness.enable_xla:
raise unittest.SkipTest("disabled for experimental_native_lowering and enable_xla=False")
# Set of harness.group_name:platform that are implemented with custom call
custom_call_harnesses = {
"vmap_cholesky:cpu", "vmap_cholesky:gpu", "vmap_eig:cpu",
"vmap_eigh:cpu", "vmap_eigh:gpu", "vmap_fft:cpu",
"householder_product:cpu", "householder_product:gpu",
"vmap_geqrf:cpu", "vmap_geqrf:gpu",
"vmap_lu:cpu", "vmap_lu:gpu", "vmap_qr:cpu", "vmap_qr:gpu",
"vmap_svd:cpu", "vmap_svd:gpu"}
if f"{harness.group_name}:{jtu.device_under_test()}" in custom_call_harnesses:
raise unittest.SkipTest("native lowering with shape polymorphism not implemented for custom calls; b/261671778")
# Set of harness.group_name or harness.group_name:platform that are implemented with HLO fallback lowering rules
fallback_lowering_harnesses = {
"vmap_approx_top_k:cpu", "vmap_bessel_i0e", "vmap_eigh:tpu",
"vmap_erf_inv", "vmap_igamma", "vmap_igammac", "vmap_lu",
"vmap_regularized_incomplete_beta", "vmap_qr:tpu",
"vmap_random_gamma:cpu", "random_gamma:tpu",
"vmap_random_gamma:tpu", "vmap_svd:tpu"}
if (harness.group_name in fallback_lowering_harnesses or
f"{harness.group_name}:{jtu.device_under_test()}" in fallback_lowering_harnesses):
raise unittest.SkipTest(
"native lowering with shape polymorphism not implemented for JAX primitives still using HLO fallback lowering; b/261682623")
# Disable these tests because they crash in MHLO
_dynamic_gather_failures = [
# In mlir::mhlo::simplifyDynamicGatherToGather(): LLVM ERROR: Failed to infer result type(s)
"getitem_op=poly_idx=slice-ct-2",
"image_resize_nearest_0",
"take__enable_xla=True",
"vmap_gather_from_slicing",
"vmap_random_randint",
"vmap_random_uniform",
]
for s in _dynamic_gather_failures:
if harness.fullname.find(s) != -1:
raise unittest.SkipTest("TODO(necula): crashes in simplifyDynamicGatherToGather")
if harness.fullname.find("vmap_dynamic_update_slice_shapes_operand=float32[3]_update=float32[1]_start_indices=(array(1, dtype=uint32),)_enable_xla=True") != -1:
# Python int 4294967295 too large to convert to int32
raise unittest.SkipTest("TODO(necula): re-land cl/496854361")
harness.run_test(self)
if __name__ == "__main__":
absltest.main(testLoader=jtu.JaxTestLoader())
Reference in New Issue View Git Blame Copy Permalink