The value of floatToRawIntBits(0.0f/0.0f) is different on x86_64 and aarch64

[theopenlab-ci]

The value of floatToRawIntBits(0.0f/0.0f) is different on x86_64 and aarch64:

on x86_64

root@donotdel-openlab-allinone-l00242678:/home/ubuntu# uname -a
Linux donotdel-openlab-allinone-l00242678 4.4.0-154-generic #181-Ubuntu SMP Tue Jun 25 05:29:03 UTC
2019 x86_64 x86_64 x86_64 GNU/Linux
scala> import java.lang.Float.floatToRawIntBits
import java.lang.Float.floatToRawIntBits
scala> floatToRawIntBits(0.0f/0.0f)
res0: Int = -4194304
scala> floatToRawIntBits(Float.NaN)
res1: Int = 2143289344

on aarch64

[root@arm-huangtianhua spark]# uname -a
Linux arm-huangtianhua 4.14.0-49.el7a.aarch64 #1 SMP Tue Apr 10 17:22:26 UTC 2018 aarch64 aarch64 aarch64 GNU/Linux
scala> import java.lang.Float.floatToRawIntBits
import java.lang.Float.floatToRawIntBits
scala> floatToRawIntBits(0.0f/0.0f)
res1: Int = 2143289344
scala> floatToRawIntBits(Float.NaN)
res2: Int = 2143289344

I test it using java, and the results are same on x86_64 and aarch64, so why scala has different values between these two platforms?

[huangtianhua]

Sorry, missed the info: Scala version 2.12.8

[huangtianhua]

@dwijnand Would you confirm this? Thank you.

[Jasper-M]

I'm confused. The bytecode is the same but the result is different.

scala> class A { def foo = Float.NaN; def bar = java.lang.Float.floatToRawIntBits(foo) }
defined class A

scala> class B { def foo = 0f / 0f; def bar = java.lang.Float.floatToRawIntBits(foo) }
defined class B

scala> new A().bar
res41: Int = 2143289344

scala> new B().bar
res42: Int = -4194304

scala> :javap  -c -filter A
Compiled from "<console>"
public class A {
  public float foo();
    Code:
       0: ldc           #16                 // float NaNf
       2: freturn

  public int bar();
    Code:
       0: aload_0
       1: invokevirtual #22                 // Method foo:()F
       4: invokestatic  #28                 // Method java/lang/Float.floatToRawIntBits:(F)I
       7: ireturn

  public A();
    Code:
       0: aload_0
       1: invokespecial #32                 // Method java/lang/Object."<init>":()V
       4: return
}

scala> :javap  -c -filter B
Compiled from "<console>"
public class B {
  public float foo();
    Code:
       0: ldc           #16                 // float NaNf
       2: freturn

  public int bar();
    Code:
       0: aload_0
       1: invokevirtual #22                 // Method foo:()F
       4: invokestatic  #28                 // Method java/lang/Float.floatToRawIntBits:(F)I
       7: ireturn

  public B();
    Code:
       0: aload_0
       1: invokespecial #32                 // Method java/lang/Object."<init>":()V
       4: return
}

[huangtianhua]

Using java on the two platforms, the results are same:

on aarch64

public class javaTest
{
public static void main(String[] args)
{
System.out.println(java.lang.Float.floatToRawIntBits(0.0f/0.0f));
System.out.println(java.lang.Float.floatToRawIntBits(Float.NaN));
System.out.println(0.0 / 0.0);
}
}
[root@arm-huangtianhua javaTest]# java javaTest
2143289344
2143289344
NaN

on x86_64

root@donotdel-openlab-allinone-l00242678:/hth/javaTest# vi javaTest.java
public class javaTest
{
public static void main(String[] args)
{
System.out.println(java.lang.Float.floatToRawIntBits(0.0f/0.0f));
System.out.println(java.lang.Float.floatToRawIntBits(Float.NaN));
System.out.println(0.0 / 0.0);
}
}
root@donotdel-openlab-allinone-l00242678:/hth/javaTest# java javaTest
2143289344
2143289344
NaN

[plokhotnyuk]

NaN values can have different binary representations:

$ scala
Welcome to Scala 2.12.8 (Java HotSpot(TM) 64-Bit Server VM, Java 1.8.0_212).
Type in expressions for evaluation. Or try :help.

scala> java.lang.Float.isNaN(Float.NaN)
res0: Boolean = true

scala> java.lang.Float.isNaN(0.0f/0.0f)
res1: Boolean = true

[huangtianhua]

My question is why the behaviour is different on x86_64 and aarch64 platforms for java.lang.Float.floatToRawIntBits(0.0f/0.0f)?

[eed3si9n]

https://en.wikipedia.org/wiki/IEEE_754-1985#NaN

sign = either 0 or 1.
biased exponent = all 1 bits.
fraction = anything except all 0 bits (since all 0 bits represents infinity).

Float has 1 bit of sign-bit and 23 bits for fraction, I think there are like 16777215 ways of representing NaN or something like that. I recently read something about someone stuffing information in that space.

[SethTisue]

closing on the assumption this is a JVM thing and not a Scala thing per se

it's not impossible there's some bug here, but in order to reopen, someone would need to demonstrate that there is a bug in Scala and not just a result that surprised you. otherwise what you have is just a question, not a bug report. questions should go on https://users.scala-lang.org, come here once you've become reasonably certain there is a bug in Scala itself

[huangtianhua]

@SethTisue OK, I will open a topic to https://users.scala-lang.org to discuss this, thanks.