sablevm-developer

[Sablevm-developer] problem with divisions, float, double

[Fabien R. <re...@ne...>]

Hello,

With Arith.java I have some problem with numbers :
[verbose class: loading "Arith"]
iadd
[verbose class: loading "java/lang/Integer"]
[verbose class: loading "java/lang/Number"]
        = 4
isub
        = 2
imul
        = 3
idiv
        = 3
irem
        = 0
ineg
        = -3
ladd
[verbose class: loading "java/lang/Long"]
        = 4
lsub
        = 2
lmul
        = 3
ldiv
        = 3
lrem
        = 0
lneg
        = -3
fadd
[verbose class: loading "java/lang/Float"]
[verbose class: loading "java/lang/Double"]
        = 3.0
fsub
        = 3.0
fmul
        = 0.0
fdiv
[verbose gc: previously allocated 16777180 bytes, surviving 56040 bytes, 
new heap is 16777216 bytes, gc time = 0 sec 9971 usec]
[verbose gc: previously allocated 16777208 bytes, surviving 55972 bytes, 
new heap is 16777216 bytes, gc time = 0 sec 8790 usec]
[verbose gc: previously allocated 16777192 bytes, surviving 55980 bytes, 
new heap is 16777216 bytes, gc time = 0 sec 9381 usec]


Kaffe too has problems (crashs when there is a double)




This is the source :

public class Arith
{
   static int add(int a, int b)
   {
       return a + b;
   }

   static int sub(int a, int b)
   {
       return a - b;
   }

   static int mul(int a, int b)
   {
       return a * b;
   }

   static int div(int a, int b)
   {
       return a / b;
   }

   static int rem(int a, int b)
   {
       return a % b;
   }

   static int neg(int a)
   {
       return -a;
   }

   static long add(long a, long b)
   {
       return a + b;
   }

   static long sub(long a, long b)
   {
       return a - b;
   }

   static long mul(long a, long b)
   {
       return a * b;
   }

   static long div(long a, long b)
   {
       return a / b;
   }

   static long rem(long a, long b)
   {
       return a % b;
   }

   static long neg(long a)
   {
       return -a;
   }

   static float add(float a, float b)
   {
       return a + b;
   }

   static float sub(float a, float b)
   {
       return a - b;
   }

   static float mul(float a, float b)
   {
       return a * b;
   }

   static float div(float a, float b)
   {
       return a / b;
   }

   static float rem(float a, float b)
   {
       return a % b;
   }

   static float neg(float a)
   {
       return -a;
   }

   static double add(double a, double b)
   {
       return a + b;
   }

   static double sub(double a, double b)
   {
       return a - b;
   }

   static double mul(double a, double b)
   {
       return a * b;
   }

   static double div(double a, double b)
   {
       return a / b;
   }

   static double rem(double a, double b)
   {
       return a % b;
   }

   static double neg(double a)
   {
       return -a;
   }

   public static void main (String[] args)
   {
       test_int (3, 1);
       test_long (3, 1);
       test_float (3, 0);
       test_double (3, 0);
       System.out.println("Done!");
   }

   static void test_int(int a, int b)
   {
       char prefix = 'i';

       System.out.println(prefix + "add");
       System.out.println("\t= " + add(a,b));

       System.out.println(prefix + "sub");
       System.out.println("\t= " + sub(a,b));

       System.out.println(prefix + "mul");
       System.out.println("\t= " + mul(a,b));

       System.out.println(prefix + "div");
       System.out.println("\t= " + div(a,b));

       System.out.println(prefix + "rem");
       System.out.println("\t= " + rem(a,b));

       System.out.println(prefix + "neg");
       System.out.println("\t= " + neg(a));
   }

   static void test_long(long a, long b)
   {
       char prefix = 'l';

       System.out.println(prefix + "add");
       System.out.println("\t= " + add(a,b));

       System.out.println(prefix + "sub");
       System.out.println("\t= " + sub(a,b));

       System.out.println(prefix + "mul");
       System.out.println("\t= " + mul(a,b));

       System.out.println(prefix + "div");
       System.out.println("\t= " + div(a,b));

       System.out.println(prefix + "rem");
       System.out.println("\t= " + rem(a,b));

       System.out.println(prefix + "neg");
       System.out.println("\t= " + neg(a));
   }

   static void test_float(float a, float b)
   {
       char prefix = 'f';

       System.out.println(prefix + "add");
       System.out.println("\t= " + add(a,b));

       System.out.println(prefix + "sub");
       System.out.println("\t= " + sub(a,b));

       System.out.println(prefix + "mul");
       System.out.println("\t= " + mul(a,b));

       System.out.println(prefix + "div");
       System.out.println("\t= " + div(a,b));

       System.out.println(prefix + "rem");
       System.out.println("\t= " + rem(a,b));

       System.out.println(prefix + "neg");
       System.out.println("\t= " + neg(a));
   }

   static void test_double(double a, double b)
   {
       char prefix = 'd';

       System.out.println(prefix + "add");
       System.out.println("\t= " + add(a,b));

       System.out.println(prefix + "sub");
       System.out.println("\t= " + sub(a,b));

       System.out.println(prefix + "mul");
       System.out.println("\t= " + mul(a,b));

       System.out.println(prefix + "div");
       System.out.println("\t= " + div(a,b));

       System.out.println(prefix + "rem");
       System.out.println("\t= " + rem(a,b));

       System.out.println(prefix + "neg");
       System.out.println("\t= " + neg(a));
   }
}


If someone has an idea ...

Cheers,
Fabien

Re: [Sablevm-developer] problem with divisions, float, double

[Chris P. <chr...@ma...>]

Hi Fabien,

We're able to help you much better if you include any details that might 
possibly be relevant.  It's also helpful if you provide minimal test 
cases (i.e. remove all the working stuff from a test case).  I take it 
that divide-by-zero is meant to print out "Infinity" for floating point 
numbers, but that sablevm hangs.

Did you try it with the --with-threading=switch and 
--enable-debugging-features options passed to configure?  Or just the 
default (which would be the inlined-engine with signalling for exceptions)?

What architectures did you try it on?

Does it work with Sun's java?

The attached Arith.java works for me on x86 with inlined and 
switch-debug engines.

Cheers,
Chris

Re: [Sablevm-developer] problem with divisions, float, double

[Fabien R. <re...@ne...>]

I use --with-thread=direct on ARM and that doesn´t wotk. On my x86, that 
works with default options.
I´ll try with --enable-debugging-features.

Cheers,
Fabien

Chris Pickett wrote:

> Hi Fabien,
>
> We're able to help you much better if you include any details that 
> might possibly be relevant.  It's also helpful if you provide minimal 
> test cases (i.e. remove all the working stuff from a test case).  I 
> take it that divide-by-zero is meant to print out "Infinity" for 
> floating point numbers, but that sablevm hangs.
>
> Did you try it with the --with-threading=switch and 
> --enable-debugging-features options passed to configure?  Or just the 
> default (which would be the inlined-engine with signalling for 
> exceptions)?
>
> What architectures did you try it on?
>
> Does it work with Sun's java?
>
> The attached Arith.java works for me on x86 with inlined and 
> switch-debug engines.
>
> Cheers,
> Chris
>
>------------------------------------------------------------------------
>
>public class Arith
>{
>    public static void main(String[] args)
>    {
>	float a = 3;
>	float b = 0;
>	System.out.println(a / b);
>    }
>}
>  
>

Re: [Sablevm-developer] problem with divisions, float, double

[Grzegorz B. P. <ga...@de...>]

On (13/05/04 13:42), Fabien Renaud wrote:
> I use --with-thread=direct on ARM and that doesn?t wotk. On my x86, that 
> works with default options.
> I?ll try with --enable-debugging-features.

This will not help.  ARM is the last architecture that needs ironing out
of these arithmetic CPU-specific quirks.  Basically ARM doesn't catch any
special cases, so you have to check for them yourself, by hand.

The practical problem here is that all Debian ARM machines are
unaccessible since some weeks, so I don't have access to a system where
I could finish the port.  I tried iPAQ cluster at handhelds.org, but
software there is much too old and installing newer versions of a dozen
of tools in home directory would be a lot of work.

So, if you knew of an up-to-date ARM system where I could get an account
- let me know.

You could also try to fix it yourself.  Compile SableVM in testing mode by
passing --enable-inlinability-testing [*] to ./configure and run BTF tests
ex. from http://devel.sablevm.org/arm/btf.tgz ('sablevm -Y
BytecodeVerificationTest') and then look into instructions.m4.c
to fix the instructions that give us problems on ARM.  It's not terribly
complicated but it practically impossible to do it w/o access to hardware
(unless you want to spend time reading StrongARM specs to find out which
cases you have to catch yourself).

I wish I could help you more,

				Grzegorz B. Prokopski

[*] This is not ideal, as this testing mode was designed to help
porting *inlined engine* and not to track portability issues like
arithmetic problems.  It is a side effect, but this will help, becaue it'll
catch pretty much all unexpected signals and will try to recover from them
- thus allowing the tests to run further.  The only problem would be that
you'd have mixed "inlining failures" with "bytecode implementation failures".

PS: Regardless of the impression you can get from the above description,
finishing ARM port is *not* a problem per se.  Given proper access to
hardware, this should be an hour or a few hours at most of not very
challenging work.

-- 
Grzegorz B. Prokopski <ga...@de...>
Debian GNU/Linux      http://www.debian.org
SableVM - LGPLed JVM  http://www.sablevm.org
Why SableVM ?!?       http://devel.sablevm.org/wiki/WhySableVM

Re: [Sablevm-developer] problem with divisions, float, double

[Grzegorz B. P. <ga...@de...>]

Hi again,

Small update.  With help from our Debian Project Leader, Martin
Michlmayr I got the output of running BTF on an ARM machine (attached).
It looks like all DADD, DDIV, DMUL, DREM, operations involving NaN or
Infinity return wrong results.  At the end we also see that an integer
operation causes CPU to generate signal which should not be there...

I did some tests running "equivalent" C code on ARM machnes from
handhelds.org's iPAQ cluster, but everything seems to work just fine,
so I have no clue yet as to what exactly is the problem.

You might want to look at instructions.m4.c and instrucions*.c files
generated for it for the engine you're using (switch/direct/inlined).
After all it's just a plain C code so it shouldn't be anything magical
to just get it to work.  Ex. you can easily take gdb/ddd and put a
breakpoint on DADD and see what exactly it exectues.  You want to
configure --enable-debugging-features to change optimiation level from
-O2 to -O0, which will make debugging much easier.

Let me know if you have questions,

HTH

			Grzegorz B. Prokopski

-- 
Grzegorz B. Prokopski <ga...@de...>
Debian GNU/Linux      http://www.debian.org
SableVM - LGPLed JVM  http://www.sablevm.org
Why SableVM ?!?       http://devel.sablevm.org/wiki/WhySableVM