}

TEST_F(CWiseUnaryGradTest, Sigmoid) {
  auto x_fn = [this](const int i) { return RV({0, -1, 1, -2, 2, -3, 3}); };
  TestCWiseGrad<float, float>(SIGMOID, x_fn);
}

TEST_F(CWiseUnaryGradTest, Sigmoid_Complex) {
  auto x_fn = [this](const int i) {
    return CRV({{1, 0}, {0, 0}, {2, -1}, {1, 2}, {3, 4}});
  };
  TestCWiseGrad<complex64, complex64>(SIGMOID, x_fn);
}

TEST_F(CWiseUnaryGradTest, Sign) {

  // inconsistent with Conv 1x1 which always performs per channel quantization.
  bool disable_per_channel_for_dense_layers = false;

  // Whether to use fixed output ranges of the activation ops (tanh, sigmoid,
  // etc.) and not infer weight constants.
  // If this option is set, quantization emulation ops should be placed after
  // the ops in the input graph. This flag should be set to false for
  // post-training quantization.

            (CastValueToI64 $old, $shape),
            MHLO_RngDistributionValue<"NORMAL">),
          [(IsShapedTensor $shape)]>;

//===----------------------------------------------------------------------===//
// Sigmoid grad op.
//===----------------------------------------------------------------------===//

// TODO(hinsu): Handle unranked inputs by broadcasting constant one to the
// shape of $l instead of having it as a constant.

// =============================================================================
// Training OPs
// =============================================================================

// `grad = dy * y * (1 - y)`, where `y = sigmoid(x)`
def LegalizeSigmoidGrad : Pat<(TF_SigmoidGradOp $y, $dy),
  (TFL_MulOp $dy, (TFL_MulOp $y, (TFL_SubOp
        (Arith_ConstantOp ConstantAttr<RankedF32ElementsAttr<[]>, "1.0f">),
        $y, TFL_AF_None),

// be checked first if present.
// TODO: b/323478683: Consider deprecating this.
struct OpQuantScaleSpec {
  // Whether this op has a fixed range requirement (e.g. sigmoid)
  bool has_fixed_output_range = false;
  // Whether this op should have same operand and result scales (e.g. concat)
  bool has_same_scale_requirement = false;

	}

	var (
		err  error
		m, c *bigmod.Nat
		N    *bigmod.Modulus
		t0   = bigmod.NewNat()
	)
	if priv.Precomputed.n == nil {
		N, err = bigmod.NewModulusFromBig(priv.N)
		if err != nil {
			return nil, ErrDecryption
		}
		c, err = bigmod.NewNat().SetBytes(ciphertext, N)
		if err != nil {
			return nil, ErrDecryption
		}
		m = bigmod.NewNat().Exp(c, priv.D.Bytes(), N)
	} else {

	}

	// SEC 1, Version 2.0, Section 4.1.4

	r, err := bigmod.NewNat().SetBytes(rBytes, c.N)
	if err != nil || r.IsZero() == 1 {
		return false
	}
	s, err := bigmod.NewNat().SetBytes(sBytes, c.N)
	if err != nil || s.IsZero() == 1 {
		return false
	}

	e := bigmod.NewNat()
	hashToNat(c, e, hash)

	// w = s⁻¹
	w := bigmod.NewNat()
	inverse(c, w, s)

	// p₁ = [e * s⁻¹]G

// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

package ecdsa

import (
	"bufio"
	"bytes"
	"compress/bzip2"
	"crypto/elliptic"
	"crypto/internal/bigmod"
	"crypto/rand"
	"crypto/sha1"
	"crypto/sha256"
	"crypto/sha512"
	"encoding/hex"
	"hash"
	"io"
	"math/big"
	"os"
	"strings"
	"testing"
)

// Copyright 2021 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

package bigmod

import (
	"errors"
	"internal/byteorder"
	"math/big"
	"math/bits"
)

const (
	// _W is the size in bits of our limbs.
	_W = bits.UintSize
	// _S is the size in bytes of our limbs.
	_S = _W / 8
)

// Copyright 2021 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

package bigmod

import (
	"fmt"
	"math/big"
	"math/bits"
	"math/rand"
	"reflect"
	"strings"
	"testing"
	"testing/quick"
)

func (n *Nat) String() string {
	var limbs []string
	for i := range n.limbs {