perf(functional): rewrite serval elemwise ops with jit subgraph

GitOrigin-RevId: 26247e21d9

imperative/python/megengine/functional/nn.py

@@ -36,6 +36,7 @@ from ..core.tensor.utils import (
     convert_single_value,
     make_shape_tuple,
     subgraph,
+    subgraph_fn,
 )
 from ..device import get_default_device
 from ..distributed import WORLD, is_distributed
@@ -824,9 +825,37 @@ def sigmoid(x):
     return _elwise(x, mode=Elemwise.Mode.SIGMOID)
 
 
+@lru_cache(maxsize=None)
+def _get_hsigmoid_op(dtype=None, device=None):
+    @subgraph_fn(
+        "Hsigmoid",
+        dtype=dtype,
+        device=device,
+        nr_inputs=1,
+        jit_fusion=True,
+        custom_grad=True,
+    )
+    def hsigmoid(inputs, f, c):
+        (inp,) = inputs[0:1]
+        inp = f("+", inp, c(3))
+        max_0 = f("max", inp, c(0))
+        min_6 = f("min", max_0, c(6))
+        oup = f("/", min_6, c(6))
+        (oup_grad,) = yield (oup,)
+        inp_grad = f("/", oup_grad, c(6))
+        inp_grad = f("cond_leq_mov", max_0, c(6), inp_grad)
+        inp_grad = f("cond_leq_mov", c(0), inp, inp_grad)
+        yield (inp_grad,)
+
+    return hsigmoid
+
+
 def hsigmoid(x):
     r"""Element-wise `relu6(x + 3) / 6`."""
-    return relu6(x + 3) / 6
+    hsigmoid = _get_hsigmoid_op(x.dtype, x.device)
+    (x,) = hsigmoid(x)
+    return x
+    # return relu6(x + 3) / 6
 
 
 def relu(x):
@@ -834,9 +863,60 @@ def relu(x):
     return _elwise(x, mode=Elemwise.Mode.RELU)
 
 
+@lru_cache(maxsize=None)
+def _get_relu6_op(dtype=None, device=None):
+    @subgraph_fn(
+        "ReLU6",
+        dtype=dtype,
+        device=device,
+        nr_inputs=1,
+        jit_fusion=True,
+        custom_grad=True,
+    )
+    def relu6(inputs, f, c):
+        (inp,) = inputs[0:1]
+        max_0 = f("max", inp, c(0))
+        min_6 = f("min", max_0, c(6))
+        oup = min_6
+        (oup_grad,) = yield (oup,)
+        inp_grad = f("cond_leq_mov", max_0, c(6), oup_grad)
+        inp_grad = f("cond_leq_mov", c(0), inp, inp_grad)
+        yield (inp_grad,)
+
+    return relu6
+
+
 def relu6(x):
     r"""Element-wise `min(max(x, 0), 6)`."""
-    return minimum(maximum(x, 0), 6)
+    relu6 = _get_relu6_op(x.dtype, x.device)
+    (x,) = relu6(x)
+    return x
+
+
+@lru_cache(maxsize=None)
+def _get_prelu_op(dtype=None, device=None):
+    @subgraph_fn(
+        "PReLU",
+        dtype=dtype,
+        device=device,
+        nr_inputs=2,
+        jit_fusion=True,
+        custom_grad=True,
+    )
+    def prelu(inputs, f, c):
+        (inp, weight) = inputs[0:2]
+        max_0 = f("max", inp, c(0))
+        min_0 = f("min", inp, c(0))
+        oup = f("fma3", min_0, weight, max_0)
+        (oup_grad,) = yield (oup,)
+        inp_grad_0 = f("cond_leq_mov", inp, c(0), oup_grad)
+        inp_grad_1 = f("*", oup_grad, weight)
+        inp_grad_1 = f("cond_leq_mov", c(0), inp, inp_grad_1)
+        inp_grad = f("+", inp_grad_0, inp_grad_1)
+        weight_grad = f("*", oup_grad, min_0)
+        yield (inp_grad, weight_grad)
+
+    return prelu
 
 
 def prelu(inp: Tensor, weight: Tensor) -> Tensor:
@@ -844,7 +924,34 @@ def prelu(inp: Tensor, weight: Tensor) -> Tensor:
 
     Refer to :class:`~.PReLU` for more information.
     """
-    return maximum(inp, 0) + weight * minimum(inp, 0)
+    prelu = _get_prelu_op(dtype=inp.dtype, device=inp.device)
+    (oup,) = prelu(inp, weight)
+    return oup
+
+
+@lru_cache(maxsize=None)
+def _get_leagk_relu_op(negative_slope, *, dtype=None, device=None):
+    @subgraph_fn(
+        "LeakyReLU",
+        dtype=dtype,
+        device=device,
+        nr_inputs=1,
+        jit_fusion=True,
+        custom_grad=True,
+    )
+    def leakyReLU(inputs, f, c):
+        (inp,) = inputs[0:1]
+        max_0 = f("max", inp, c(0))
+        min_0 = f("min", inp, c(0))
+        oup = f("+", max_0, f("*", min_0, c(negative_slope)))
+        (oup_grad,) = yield (oup,)
+        inp_grad_0 = f("cond_leq_mov", c(0), inp, oup_grad)
+        inp_grad_1 = f("*", oup_grad, c(negative_slope))
+        inp_grad_1 = f("cond_leq_mov", inp, c(negative_slope), inp_grad_1)
+        inp_grad = f("+", inp_grad_0, inp_grad_1)
+        yield (inp_grad,)
+
+    return leakyReLU
 
 
 def leaky_relu(inp: Tensor, negative_slope: float = 0.01) -> Tensor:
@@ -852,7 +959,9 @@ def leaky_relu(inp: Tensor, negative_slope: float = 0.01) -> Tensor:
 
     Refer to :class:`~.LeakyReLU` for more information.
     """
-    return maximum(inp, 0) + negative_slope * minimum(inp, 0)
+    leakyReLU = _get_leagk_relu_op(negative_slope, dtype=inp.dtype, device=inp.device)
+    (oup,) = leakyReLU(inp)
+    return oup
 
 
 def silu(x):
@@ -871,6 +980,36 @@ def gelu(x):
     return _elwise(x, mode=Elemwise.Mode.GELU)
 
 
+@lru_cache(maxsize=None)
+def _get_softplus_op(dtype=None, device=None):
+    @subgraph_fn(
+        "Softplus",
+        dtype=dtype,
+        device=device,
+        nr_inputs=1,
+        jit_fusion=True,
+        # gopt_level=0,
+        custom_grad=True,
+    )
+    def softplus(inputs, f, c):
+        (inp,) = inputs[0:1]
+        neg_abs = f("-", f("abs", inp))
+        exp = f("exp", neg_abs)
+        oup = f("log1p", exp)
+        oup = f("+", oup, f("relu", inp))
+        (oup_grad,) = yield (oup,)
+        inp_grad_0 = f("switch_gt0", inp, oup_grad)
+        inp_grad_1 = oup_grad
+        inp_grad_1 = f("/", oup_grad, f("+", exp, c(1)))
+        inp_grad_1 = f("*", oup_grad, exp)
+        inp_grad_1 = f("-", inp_grad_1)
+        inp_grad_1 = f("abs_grad", inp, inp_grad_1)
+        inp_grad = f("+", inp_grad_0, inp_grad_1)
+        yield (inp_grad,)
+
+    return softplus
+
+
 def softplus(inp: Tensor) -> Tensor:
     r"""Applies the element-wise function:
 
@@ -904,7 +1043,9 @@ def softplus(inp: Tensor) -> Tensor:
 
             [0.0486 0.1269 0.3133 0.6931 1.3133 2.1269]
     """
-    return log1p(exp(-abs(inp))) + relu(inp)
+    softplus = _get_softplus_op(inp.dtype, inp.device)
+    (oup,) = softplus(inp)
+    return oup
 
 
 def logsoftmax(inp: Tensor, axis: Union[int, Sequence[int]]) -> Tensor:
@@ -944,6 +1085,38 @@ def logsoftmax(inp: Tensor, axis: Union[int, Sequence[int]]) -> Tensor:
     return inp - logsumexp(inp, axis, keepdims=True)
 
 
+@lru_cache(maxsize=None)
+def _get_logsigmoid_op(dtype=None, device=None):
+    @subgraph_fn(
+        "LogSigmoid",
+        dtype=dtype,
+        device=device,
+        nr_inputs=1,
+        jit_fusion=True,
+        custom_grad=True,
+    )
+    def logsigmoid(inputs, f, c):
+        (inp,) = inputs[0:1]
+        neg_abs = f("-", f("abs", inp))
+        exp = f("exp", neg_abs)
+        oup = f("log1p", exp)
+        oup = f("+", oup, f("relu", f("-", inp)))
+        oup = f("-", oup)
+        (oup_grad,) = yield (oup,)
+        oup_grad = f("-", oup_grad)
+        inp_grad_0 = f("switch_gt0", inp, oup_grad)
+        inp_grad_0 = f("-", inp_grad_0)
+        inp_grad_1 = oup_grad
+        inp_grad_1 = f("/", oup_grad, f("+", exp, c(1)))
+        inp_grad_1 = f("*", oup_grad, exp)
+        inp_grad_1 = f("-", inp_grad_1)
+        inp_grad_1 = f("abs_grad", inp, inp_grad_1)
+        inp_grad = f("+", inp_grad_0, inp_grad_1)
+        yield (inp_grad,)
+
+    return logsigmoid
+
+
 def logsigmoid(inp: Tensor) -> Tensor:
     r"""Applies the element-wise function:
 
@@ -972,7 +1145,9 @@ def logsigmoid(inp: Tensor) -> Tensor:
             [-5.0067 -4.0182 -3.0486 -2.1269 -1.3133 -0.6931 -0.3133 -0.1269 -0.0486
             -0.0181]
     """
-    return -softplus(-inp)
+    logsigmoid = _get_logsigmoid_op(inp.dtype, inp.device)
+    (oup,) = logsigmoid(inp)
+    return oup
 
 
 def logsumexp(

imperative/python/megengine/functional/tensor.py

@@ -6,6 +6,7 @@
 # Unless required by applicable law or agreed to in writing,
 # software distributed under the License is distributed on an
 # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+from functools import lru_cache
 from typing import Iterable, Optional, Sequence, Tuple, Union
 
 import numpy as np
@@ -17,7 +18,14 @@ from ..core.ops import builtin
 from ..core.ops.builtin import Copy, Identity
 from ..core.ops.special import Const
 from ..core.tensor.array_method import _broadcast, _remove_axis
-from ..core.tensor.utils import astensor1d, convert_inputs, get_device
+from ..core.tensor.utils import (
+    astensor1d,
+    convert_inputs,
+    get_device,
+    isscalar,
+    setscalar,
+    subgraph_fn,
+)
 from ..device import get_default_device
 from ..tensor import Tensor
 from .elemwise import ceil
@@ -731,6 +739,29 @@ def scatter(inp: Tensor, axis: int, index: Tensor, source: Tensor) -> Tensor:
     return inp
 
 
+@lru_cache(maxsize=None)
+def _get_where_op(dtype=None, device=None):
+    @subgraph_fn(
+        "Where",
+        dtype=dtype,
+        device=device,
+        nr_inputs=3,
+        jit_fusion=True,
+        custom_grad=True,
+    )
+    def where(inputs, f, c):
+        (mask, x, y) = inputs[0:3]
+        oup = f("switch_gt0", mask, x)
+        ksam = f("-", c(1), mask)
+        oup = f("+", oup, f("switch_gt0", ksam, y))
+        (oup_grad,) = yield (oup,)
+        x_grad = f("switch_gt0", mask, oup_grad)
+        y_grad = f("switch_gt0", ksam, oup_grad)
+        yield (None, x_grad, y_grad)
+
+    return where
+
+
 def where(mask: Tensor, x: Tensor, y: Tensor) -> Tensor:
     r"""Selects elements either from Tensor x or Tensor y, according to mask.
 
@@ -780,20 +811,19 @@ def where(mask: Tensor, x: Tensor, y: Tensor) -> Tensor:
         raise ValueError("ambiguous device: {} vs {}".format(x.device, mask.device))
 
     dtype = dtype_promotion(x, y)
+    device = x.device
+
     if x.dtype != dtype:
         x = x.astype(dtype)
     if y.dtype != dtype:
         y = y.astype(dtype)
+    mask = mask.astype(dtype)
 
-    v0, index0 = cond_take(mask, x)
-    v1, index1 = cond_take(~mask, y)
-
-    out = concat([v0, v1])
-
-    out[index0] = v0
-    out[index1] = v1
-    out = out.reshape(x.shape)
-    return out
+    where = _get_where_op(dtype=dtype, device=device)
+    (oup,) = where(mask, x, y)
+    if isscalar(mask):
+        setscalar(oup)
+    return oup
 
 
 def cond_take(mask: Tensor, x: Tensor) -> Tensor:

imperative/python/test/unit/functional/test_elemwise.py

@@ -166,7 +166,7 @@ def test_hsigmoid():
     x = np.random.randn(100).astype("float32")
     y_np = np.minimum(np.maximum(x + 3, 0), 6) / 6
     y_mge = F.hsigmoid(tensor(x)).numpy()
-    np.testing.assert_equal(y_np, y_mge)
+    np.testing.assert_almost_equal(y_np, y_mge, decimal=6)
 
 
 def test_logical_oprs():

imperative/python/test/unit/functional/test_functional.py

@@ -27,6 +27,8 @@ from megengine.core.tensor.utils import make_shape_tuple
 from megengine.device import get_device_count
 from megengine.module import LayerNorm
 
+_assert_allclose = partial(np.testing.assert_allclose, atol=5e-6, rtol=5e-6)
+
 
 def test_where():
     maskv0 = np.array([[1, 0], [0, 1]], dtype=np.bool_)
@@ -627,6 +629,7 @@ def test_binary_cross_entropy():
         {"input": [data1, label1], "output": expect1,},
         {"input": [data2, label2], "output": expect2,},
     ]
+
     opr_test(cases, F.nn.binary_cross_entropy, compare_fn=compare_fn)
 
     cases = [