mirror of
https://github.com/golang/go
synced 2024-11-26 07:38:00 -07:00
math: increase precision of math.SmallestNonzeroFloat64
The original value was rounded too early, which lead to the surprising behavior that float64(math.SmallestNonzeroFloat64 / 2) wasn't 0. That is, the exact compile-time computation of math.SmallestNonzeroFloat64 / 2 resulted in a value that was rounded up when converting to float64. To address this, added 3 more digits to the mantissa, ending in a 0. While at it, also slightly increased the precision of MaxFloat64 to end in a 0. Computed exact values via https://play.golang.org/p/yt4KTpIx_wP. Added a test to verify expected behavior. In contrast to the other (irrational) constants, expanding these extreme values to more digits is unlikely to be important as they are not going to appear in numeric computations except for tests verifying their correctness (as is the case here). Re-enabled a disabled test in go/types and types2. Updates #44057. Fixes #44058. Change-Id: I8f363155e02331354e929beabe993c8d8de75646 Reviewed-on: https://go-review.googlesource.com/c/go/+/315170 Trust: Robert Griesemer <gri@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
This commit is contained in:
parent
02ab8d1a1d
commit
3498027329
@ -1,4 +1,6 @@
|
||||
pkg encoding/json, method (*RawMessage) MarshalJSON() ([]uint8, error)
|
||||
pkg math, const MaxFloat64 = 1.79769e+308 // 179769313486231570814527423731704356798100000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
|
||||
pkg math, const SmallestNonzeroFloat64 = 4.94066e-324 // 4940656458412465441765687928682213723651/1000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
|
||||
pkg math/big, const MaxBase = 36
|
||||
pkg math/big, type Word uintptr
|
||||
pkg net, func ListenUnixgram(string, *UnixAddr) (*UDPConn, error)
|
||||
|
@ -6,6 +6,8 @@
|
||||
|
||||
package const1
|
||||
|
||||
import "math"
|
||||
|
||||
const(
|
||||
mi = ^int(0)
|
||||
mu = ^uint(0)
|
||||
@ -48,7 +50,7 @@ const (
|
||||
// without overflow. For now we match the compiler.
|
||||
// See also issue #44057.
|
||||
// smallestFloat64 = 1.0 / (1<<(1023 - 1 + 52))
|
||||
smallestFloat64 = 4.940656458412465441765687928682213723651e-324
|
||||
smallestFloat64 = math.SmallestNonzeroFloat64
|
||||
)
|
||||
|
||||
const (
|
||||
@ -63,7 +65,7 @@ const (
|
||||
// without overflow. For now we match the compiler.
|
||||
// See also issue #44057.
|
||||
// maxFloat64 = 1<<1023 * (1<<53 - 1) / (1.0<<52)
|
||||
maxFloat64 = 1.797693134862315708145274237317043567981e+308
|
||||
maxFloat64 = math.MaxFloat64
|
||||
)
|
||||
|
||||
const (
|
||||
@ -281,9 +283,7 @@ const (
|
||||
_ = assert(float64(smallestFloat32) == smallestFloat32)
|
||||
_ = assert(float64(smallestFloat32/2) == smallestFloat32/2)
|
||||
_ = assert(float64(smallestFloat64) == smallestFloat64)
|
||||
// TODO(gri) With the change to the declaration of smallestFloat64
|
||||
// this now fails to be true. See issue #44058.
|
||||
// _ = assert(float64(smallestFloat64/2) == 0)
|
||||
_ = assert(float64(smallestFloat64/2) == 0)
|
||||
)
|
||||
|
||||
const (
|
||||
|
10
src/go/types/testdata/check/const1.src
vendored
10
src/go/types/testdata/check/const1.src
vendored
@ -6,6 +6,8 @@
|
||||
|
||||
package const1
|
||||
|
||||
import "math"
|
||||
|
||||
const(
|
||||
mi = ^int(0)
|
||||
mu = ^uint(0)
|
||||
@ -48,7 +50,7 @@ const (
|
||||
// without overflow. For now we match the compiler.
|
||||
// See also issue #44057.
|
||||
// smallestFloat64 = 1.0 / (1<<(1023 - 1 + 52))
|
||||
smallestFloat64 = 4.940656458412465441765687928682213723651e-324
|
||||
smallestFloat64 = math.SmallestNonzeroFloat64
|
||||
)
|
||||
|
||||
const (
|
||||
@ -63,7 +65,7 @@ const (
|
||||
// without overflow. For now we match the compiler.
|
||||
// See also issue #44057.
|
||||
// maxFloat64 = 1<<1023 * (1<<53 - 1) / (1.0<<52)
|
||||
maxFloat64 = 1.797693134862315708145274237317043567981e+308
|
||||
maxFloat64 = math.MaxFloat64
|
||||
)
|
||||
|
||||
const (
|
||||
@ -281,9 +283,7 @@ const (
|
||||
_ = assert(float64(smallestFloat32) == smallestFloat32)
|
||||
_ = assert(float64(smallestFloat32/2) == smallestFloat32/2)
|
||||
_ = assert(float64(smallestFloat64) == smallestFloat64)
|
||||
// TODO(gri) With the change to the declaration of smallestFloat64
|
||||
// this now fails to be true. See issue #44058.
|
||||
// _ = assert(float64(smallestFloat64/2) == 0)
|
||||
_ = assert(float64(smallestFloat64/2) == 0)
|
||||
)
|
||||
|
||||
const (
|
||||
|
@ -3196,6 +3196,15 @@ func TestFloatMinMax(t *testing.T) {
|
||||
}
|
||||
}
|
||||
|
||||
func TestFloatMinima(t *testing.T) {
|
||||
if q := float32(SmallestNonzeroFloat32 / 2); q != 0 {
|
||||
t.Errorf("float32(SmallestNonzeroFloat32 / 2) = %g, want 0", q)
|
||||
}
|
||||
if q := float64(SmallestNonzeroFloat64 / 2); q != 0 {
|
||||
t.Errorf("float64(SmallestNonzeroFloat64 / 2) = %g, want 0", q)
|
||||
}
|
||||
}
|
||||
|
||||
var indirectSqrt = Sqrt
|
||||
|
||||
// TestFloat32Sqrt checks the correctness of the float32 square root optimization result.
|
||||
|
@ -31,8 +31,8 @@ const (
|
||||
MaxFloat32 = 3.40282346638528859811704183484516925440e+38 // 2**127 * (2**24 - 1) / 2**23
|
||||
SmallestNonzeroFloat32 = 1.401298464324817070923729583289916131280e-45 // 1 / 2**(127 - 1 + 23)
|
||||
|
||||
MaxFloat64 = 1.797693134862315708145274237317043567981e+308 // 2**1023 * (2**53 - 1) / 2**52
|
||||
SmallestNonzeroFloat64 = 4.940656458412465441765687928682213723651e-324 // 1 / 2**(1023 - 1 + 52)
|
||||
MaxFloat64 = 1.79769313486231570814527423731704356798070e+308 // 2**1023 * (2**53 - 1) / 2**52
|
||||
SmallestNonzeroFloat64 = 4.9406564584124654417656879286822137236505980e-324 // 1 / 2**(1023 - 1 + 52)
|
||||
)
|
||||
|
||||
// Integer limit values.
|
||||
|
Loading…
Reference in New Issue
Block a user