Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

feat(osmomath): log2 approximation #2788

Merged
merged 16 commits into from
Oct 24, 2022
Merged

feat(osmomath): log2 approximation #2788

Show file tree
Hide file tree
Changes from 14 commits
Commits
Show all changes
16 commits
Select commit Hold shift + click to select a range
dd0c5a7
feat: osmomath log2 approximation
p0mvn Sep 20, 2022
07d8ce6
lint
p0mvn Sep 20, 2022
c50fd9d
fix comment
p0mvn Sep 20, 2022
e4d196f
comment
p0mvn Sep 20, 2022
5512017
Update osmomath/decimal.go
p0mvn Sep 20, 2022
add2e78
Merge branch 'main' into roman/log2-approx
p0mvn Oct 1, 2022
0f8b563
Merge branch 'main' into roman/log2-approx
p0mvn Oct 22, 2022
5dbf722
improve accuracy with narrower range
p0mvn Oct 22, 2022
e7eb2b3
comment
p0mvn Oct 22, 2022
66a6c53
implement and test precise log for x >= 1
p0mvn Oct 23, 2022
bfc5c16
tests for negative values
p0mvn Oct 23, 2022
d6324f4
make non-mutative and test
p0mvn Oct 23, 2022
003b0c1
bench
p0mvn Oct 23, 2022
d09e451
changelog
p0mvn Oct 23, 2022
cdbc726
remove redundant assignments
p0mvn Oct 24, 2022
5aaa278
improve comments
p0mvn Oct 24, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,6 +44,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0

### Features

* [#2788](https://github.com/osmosis-labs/osmosis/pull/2788) Add logarithm base 2 implementation.
* [#2739](https://github.com/osmosis-labs/osmosis/pull/2739) Add pool type query

### Bug fixes
Expand Down
65 changes: 64 additions & 1 deletion osmomath/decimal.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,6 +30,9 @@ const (

// max number of iterations in ApproxRoot function
maxApproxRootIterations = 100

// max number of iterations in Log2 function
maxLog2Iterations = 300
)

var (
Expand All @@ -39,6 +42,10 @@ var (
zeroInt = big.NewInt(0)
oneInt = big.NewInt(1)
tenInt = big.NewInt(10)

// initialized in init() since requires
// precision to be defined.
twoBigDec BigDec
)

// Decimal errors
Expand All @@ -54,6 +61,8 @@ func init() {
for i := 0; i <= Precision; i++ {
precisionMultipliers[i] = calcPrecisionMultiplier(int64(i))
}

twoBigDec = NewBigDec(2)
}

func precisionInt() *big.Int {
Expand Down Expand Up @@ -211,7 +220,8 @@ func (d BigDec) GTE(d2 BigDec) bool { return (d.i).Cmp(d2.i) >= 0 } /
func (d BigDec) LT(d2 BigDec) bool { return (d.i).Cmp(d2.i) < 0 } // less than
func (d BigDec) LTE(d2 BigDec) bool { return (d.i).Cmp(d2.i) <= 0 } // less than or equal
func (d BigDec) Neg() BigDec { return BigDec{new(big.Int).Neg(d.i)} } // reverse the decimal sign
func (d BigDec) Abs() BigDec { return BigDec{new(big.Int).Abs(d.i)} } // absolute value
// nolint: stylecheck
func (d BigDec) Abs() BigDec { return BigDec{new(big.Int).Abs(d.i)} } // absolute value

// BigInt returns a copy of the underlying big.Int.
func (d BigDec) BigInt() *big.Int {
Expand Down Expand Up @@ -856,3 +866,56 @@ func DecApproxEq(t *testing.T, d1 BigDec, d2 BigDec, tol BigDec) (*testing.T, bo
diff := d1.Sub(d2).Abs()
return t, diff.LTE(tol), "expected |d1 - d2| <:\t%v\ngot |d1 - d2| = \t\t%v", tol.String(), diff.String()
}

// LogBase2 returns log_2 {x}.
// Rounds down by truncations during division and right shifting.
// Accurate up to 32 precision digits.
// Implementation is based on:
// https://stm32duinoforum.com/forum/dsp/BinaryLogarithm.pdf
func (x BigDec) LogBase2() BigDec {
// create a new decimal to avoid mutating
// the receiver's int buffer.
xCopy := ZeroDec()
p0mvn marked this conversation as resolved.
Show resolved Hide resolved
xCopy.i = new(big.Int).Set(x.i)
if xCopy.LTE(ZeroDec()) {
panic(fmt.Sprintf("log is not defined at <= 0, given (%s)", xCopy))
}

// Normalize x to be 1 <= x < 2

// y is the exponent that results in a whole multiple of 2.
y := ZeroDec()

// invariant: x >= 1
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

this is just a wording comment, but doesn't invariant imply that x>=1 is true throughout the iteration? I think you mean that after the iteration x>=1. Are we using these invariants for formal verification somewhere?

Copy link
Member Author

@p0mvn p0mvn Oct 24, 2022 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think I was misusing the term to imply that we are aiming to normalize the value to be >= 1 after the iteration is complete. Going to change this wording

// while x < 1
for xCopy.LT(OneDec()) {
xCopy.i = xCopy.i.Lsh(xCopy.i, 1)
p0mvn marked this conversation as resolved.
Show resolved Hide resolved
y = y.Sub(OneDec())
}

// invariant: x < 2
// while x >= 2
for xCopy.GTE(twoBigDec) {
xCopy.i = xCopy.i.Rsh(xCopy.i, 1)
y = y.Add(OneDec())
}

b := OneDec().Quo(twoBigDec)

// N.B. At this point x is a positive real number representing
// mantissa of the log. We estimate it using the following
// algorithm:
// https://stm32duinoforum.com/forum/dsp/BinaryLogarithm.pdf
// This has shown precision of 32 digits relative
// to Wolfram Alpha in tests
for i := 0; i < maxLog2Iterations; i++ {
xCopy = xCopy.Mul(xCopy)
if xCopy.GTE(twoBigDec) {
xCopy.i = xCopy.i.Rsh(xCopy.i, 1)
y = y.Add(b)
}
b.i = b.i.Rsh(b.i, 1)
nicolaslara marked this conversation as resolved.
Show resolved Hide resolved
}

return y
}
147 changes: 118 additions & 29 deletions osmomath/decimal_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,6 +11,8 @@ import (
"github.com/stretchr/testify/require"
"github.com/stretchr/testify/suite"
"gopkg.in/yaml.v2"

"github.com/osmosis-labs/osmosis/v12/app/apptesting/osmoassert"
)

type decimalTestSuite struct {
Expand Down Expand Up @@ -152,8 +154,8 @@ func (s *decimalTestSuite) TestDecFloat64() {

func (s *decimalTestSuite) TestSdkDec() {
tests := []struct {
d BigDec
want sdk.Dec
d BigDec
want sdk.Dec
expPanic bool
}{
{NewBigDec(0), sdk.MustNewDecFromStr("0.000000000000000000"), false},
Expand All @@ -177,8 +179,8 @@ func (s *decimalTestSuite) TestSdkDec() {

func (s *decimalTestSuite) TestBigDecFromSdkDec() {
tests := []struct {
d sdk.Dec
want BigDec
d sdk.Dec
want BigDec
expPanic bool
}{
{sdk.MustNewDecFromStr("0.000000000000000000"), NewBigDec(0), false},
Expand All @@ -202,8 +204,8 @@ func (s *decimalTestSuite) TestBigDecFromSdkDec() {

func (s *decimalTestSuite) TestBigDecFromSdkDecSlice() {
tests := []struct {
d []sdk.Dec
want []BigDec
d []sdk.Dec
want []BigDec
expPanic bool
}{
{[]sdk.Dec{sdk.MustNewDecFromStr("0.000000000000000000")}, []BigDec{NewBigDec(0)}, false},
Expand Down Expand Up @@ -441,14 +443,14 @@ func (s *decimalTestSuite) TestDecCeil() {
input BigDec
expected BigDec
}{
{MustNewDecFromStr("0.001"), NewBigDec(1)}, // 0.001 => 1.0
{MustNewDecFromStr("-0.001"), ZeroDec()}, // -0.001 => 0.0
{ZeroDec(), ZeroDec()}, // 0.0 => 0.0
{MustNewDecFromStr("0.9"), NewBigDec(1)}, // 0.9 => 1.0
{MustNewDecFromStr("0.001"), NewBigDec(1)}, // 0.001 => 1.0
{MustNewDecFromStr("-0.001"), ZeroDec()}, // -0.001 => 0.0
{ZeroDec(), ZeroDec()}, // 0.0 => 0.0
{MustNewDecFromStr("0.9"), NewBigDec(1)}, // 0.9 => 1.0
{MustNewDecFromStr("4.001"), NewBigDec(5)}, // 4.001 => 5.0
{MustNewDecFromStr("-4.001"), NewBigDec(-4)}, // -4.001 => -4.0
{MustNewDecFromStr("4.7"), NewBigDec(5)}, // 4.7 => 5.0
{MustNewDecFromStr("-4.7"), NewBigDec(-4)}, // -4.7 => -4.0
{MustNewDecFromStr("4.7"), NewBigDec(5)}, // 4.7 => 5.0
{MustNewDecFromStr("-4.7"), NewBigDec(-4)}, // -4.7 => -4.0
}

for i, tc := range testCases {
Expand All @@ -463,11 +465,11 @@ func (s *decimalTestSuite) TestPower() {
power uint64
expected BigDec
}{
{OneDec(), 10, OneDec()}, // 1.0 ^ (10) => 1.0
{NewDecWithPrec(5, 1), 2, NewDecWithPrec(25, 2)}, // 0.5 ^ 2 => 0.25
{NewDecWithPrec(2, 1), 2, NewDecWithPrec(4, 2)}, // 0.2 ^ 2 => 0.04
{NewDecFromInt(NewInt(3)), 3, NewDecFromInt(NewInt(27))}, // 3 ^ 3 => 27
{NewDecFromInt(NewInt(-3)), 4, NewDecFromInt(NewInt(81))}, // -3 ^ 4 = 81
{OneDec(), 10, OneDec()}, // 1.0 ^ (10) => 1.0
{NewDecWithPrec(5, 1), 2, NewDecWithPrec(25, 2)}, // 0.5 ^ 2 => 0.25
{NewDecWithPrec(2, 1), 2, NewDecWithPrec(4, 2)}, // 0.2 ^ 2 => 0.04
{NewDecFromInt(NewInt(3)), 3, NewDecFromInt(NewInt(27))}, // 3 ^ 3 => 27
{NewDecFromInt(NewInt(-3)), 4, NewDecFromInt(NewInt(81))}, // -3 ^ 4 = 81
{MustNewDecFromStr("1.414213562373095048801688724209698079"), 2, NewDecFromInt(NewInt(2))}, // 1.414213562373095048801688724209698079 ^ 2 = 2
}

Expand All @@ -483,14 +485,14 @@ func (s *decimalTestSuite) TestApproxRoot() {
root uint64
expected BigDec
}{
{OneDec(), 10, OneDec()}, // 1.0 ^ (0.1) => 1.0
{NewDecWithPrec(25, 2), 2, NewDecWithPrec(5, 1)}, // 0.25 ^ (0.5) => 0.5
{NewDecWithPrec(4, 2), 2, NewDecWithPrec(2, 1)}, // 0.04 ^ (0.5) => 0.2
{NewDecFromInt(NewInt(27)), 3, NewDecFromInt(NewInt(3))}, // 27 ^ (1/3) => 3
{NewDecFromInt(NewInt(-81)), 4, NewDecFromInt(NewInt(-3))}, // -81 ^ (0.25) => -3
{NewDecFromInt(NewInt(2)), 2, MustNewDecFromStr("1.414213562373095048801688724209698079")}, // 2 ^ (0.5) => 1.414213562373095048801688724209698079
{OneDec(), 10, OneDec()}, // 1.0 ^ (0.1) => 1.0
{NewDecWithPrec(25, 2), 2, NewDecWithPrec(5, 1)}, // 0.25 ^ (0.5) => 0.5
{NewDecWithPrec(4, 2), 2, NewDecWithPrec(2, 1)}, // 0.04 ^ (0.5) => 0.2
{NewDecFromInt(NewInt(27)), 3, NewDecFromInt(NewInt(3))}, // 27 ^ (1/3) => 3
{NewDecFromInt(NewInt(-81)), 4, NewDecFromInt(NewInt(-3))}, // -81 ^ (0.25) => -3
{NewDecFromInt(NewInt(2)), 2, MustNewDecFromStr("1.414213562373095048801688724209698079")}, // 2 ^ (0.5) => 1.414213562373095048801688724209698079
{NewDecWithPrec(1005, 3), 31536000, MustNewDecFromStr("1.000000000158153903837946258002096839")}, // 1.005 ^ (1/31536000) ≈ 1.000000000158153903837946258002096839
{SmallestDec(), 2, NewDecWithPrec(1, 18)}, // 1e-36 ^ (0.5) => 1e-18
{SmallestDec(), 2, NewDecWithPrec(1, 18)}, // 1e-36 ^ (0.5) => 1e-18
{SmallestDec(), 3, MustNewDecFromStr("0.000000000001000000000000000002431786")}, // 1e-36 ^ (1/3) => 1e-12
{NewDecWithPrec(1, 8), 3, MustNewDecFromStr("0.002154434690031883721759293566519280")}, // 1e-8 ^ (1/3) ≈ 0.002154434690031883721759293566519
}
Expand All @@ -511,11 +513,11 @@ func (s *decimalTestSuite) TestApproxSqrt() {
input BigDec
expected BigDec
}{
{OneDec(), OneDec()}, // 1.0 => 1.0
{NewDecWithPrec(25, 2), NewDecWithPrec(5, 1)}, // 0.25 => 0.5
{NewDecWithPrec(4, 2), NewDecWithPrec(2, 1)}, // 0.09 => 0.3
{NewDecFromInt(NewInt(9)), NewDecFromInt(NewInt(3))}, // 9 => 3
{NewDecFromInt(NewInt(-9)), NewDecFromInt(NewInt(-3))}, // -9 => -3
{OneDec(), OneDec()}, // 1.0 => 1.0
{NewDecWithPrec(25, 2), NewDecWithPrec(5, 1)}, // 0.25 => 0.5
{NewDecWithPrec(4, 2), NewDecWithPrec(2, 1)}, // 0.09 => 0.3
{NewDecFromInt(NewInt(9)), NewDecFromInt(NewInt(3))}, // 9 => 3
{NewDecFromInt(NewInt(-9)), NewDecFromInt(NewInt(-3))}, // -9 => -3
{NewDecFromInt(NewInt(2)), MustNewDecFromStr("1.414213562373095048801688724209698079")}, // 2 => 1.414213562373095048801688724209698079
}

Expand Down Expand Up @@ -650,3 +652,90 @@ func BenchmarkMarshalTo(b *testing.B) {
}
}
}

func (s *decimalTestSuite) TestLog2() {
var expectedErrTolerance = MustNewDecFromStr("0.000000000000000000000000000000000100")

tests := map[string]struct {
initialValue BigDec
expected BigDec

expectedPanic bool
}{
"log_2{-1}; invalid; panic": {
initialValue: OneDec().Neg(),
expectedPanic: true,
},
"log_2{0}; invalid; panic": {
initialValue: ZeroDec(),
expectedPanic: true,
},
"log_2{0.001} = -9.965784284662087043610958288468170528": {
initialValue: MustNewDecFromStr("0.001"),
// From: https://www.wolframalpha.com/input?i=log+base+2+of+0.999912345+with+33+digits
expected: MustNewDecFromStr("-9.965784284662087043610958288468170528"),
},
"log_2{0.56171821941421412902170941} = -0.832081497183140708984033250637831402": {
initialValue: MustNewDecFromStr("0.56171821941421412902170941"),
// From: https://www.wolframalpha.com/input?i=log+base+2+of+0.56171821941421412902170941+with+36+digits
expected: MustNewDecFromStr("-0.832081497183140708984033250637831402"),
},
"log_2{0.999912345} = -0.000126464976533858080645902722235833": {
initialValue: MustNewDecFromStr("0.999912345"),
// From: https://www.wolframalpha.com/input?i=log+base+2+of+0.999912345+with+37+digits
expected: MustNewDecFromStr("-0.000126464976533858080645902722235833"),
},
"log_2{1} = 0": {
initialValue: NewBigDec(1),
expected: NewBigDec(0),
},
"log_2{2} = 1": {
initialValue: NewBigDec(2),
expected: NewBigDec(1),
},
"log_2{7} = 2.807354922057604107441969317231830809": {
initialValue: NewBigDec(7),
// From: https://www.wolframalpha.com/input?i=log+base+2+of+7+37+digits
expected: MustNewDecFromStr("2.807354922057604107441969317231830809"),
},
"log_2{512} = 9": {
initialValue: NewBigDec(512),
expected: NewBigDec(9),
},
"log_2{580} = 9.179909090014934468590092754117374938": {
initialValue: NewBigDec(580),
// From: https://www.wolframalpha.com/input?i=log+base+2+of+600+37+digits
expected: MustNewDecFromStr("9.179909090014934468590092754117374938"),
},
"log_2{1024} = 10": {
initialValue: NewBigDec(1024),
expected: NewBigDec(10),
},
"log_2{1024.987654321} = 10.001390817654141324352719749259888355": {
initialValue: NewDecWithPrec(1024987654321, 9),
// From: https://www.wolframalpha.com/input?i=log+base+2+of+1024.987654321+38+digits
expected: MustNewDecFromStr("10.001390817654141324352719749259888355"),
},
"log_2{912648174127941279170121098210.92821920190204131121} = 99.525973560175362367047484597337715868": {
initialValue: MustNewDecFromStr("912648174127941279170121098210.92821920190204131121"),
// From: https://www.wolframalpha.com/input?i=log+base+2+of+912648174127941279170121098210.92821920190204131121+38+digits
expected: MustNewDecFromStr("99.525973560175362367047484597337715868"),
},
}

for name, tc := range tests {
s.Run(name, func() {
osmoassert.ConditionalPanic(s.T(), tc.expectedPanic, func() {
// Create a copy to test that the original was not modified.
// That is, that LogbBase2() is non-mutative.
initialCopy := ZeroDec()
initialCopy.i.Set(tc.initialValue.i)

// system under test.
res := tc.initialValue.LogBase2()
require.True(DecApproxEq(s.T(), tc.expected, res, expectedErrTolerance))
require.Equal(s.T(), initialCopy, tc.initialValue)
})
})
}
}
24 changes: 24 additions & 0 deletions osmomath/log2_bench_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,24 @@
package osmomath

import (
"math/rand"
"testing"
)

func BenchmarkLog2(b *testing.B) {
tests := []BigDec{
MustNewDecFromStr("1.2"),
MustNewDecFromStr("1.234"),
MustNewDecFromStr("1024"),
NewBigDec(2048 * 2048 * 2048 * 2048 * 2048),
MustNewDecFromStr("999999999999999999999999999999999999999999999999999999.9122181273612911"),
MustNewDecFromStr("0.563289239121902491248219047129047129"),
}

for i := 0; i < b.N; i++ {
b.StopTimer()
test := tests[rand.Int63n(int64(len(tests)))]
b.StartTimer()
_ = test.LogBase2()
}
}
1 change: 0 additions & 1 deletion osmomath/rounding_direction_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,6 @@ func TestDivIntByU64ToBigDec(t *testing.T) {
addTCForAllRoundingModes("odd divided by 2", sdk.NewInt(5), 2, NewDecWithPrec(25, 1))

for name, tt := range tests {
fmt.Println("start")
t.Run(name, func(t *testing.T) {
got, err := DivIntByU64ToBigDec(tt.i, tt.u, tt.round)
require.Equal(t, tt.want, got)
Expand Down