Unable to compile IcedTea with CACAO under Fedora 17

[Robert Lougher]

On 18 December 2012 21:00, Robert Lougher <rob.lougher at gmail.com> wrote:
> Hi Greg,
>
> On 18 December 2012 18:02, Greg Johnson <gregjo at yahoo.com> wrote:
>>
>> Under Debian, I have this:
>> # java -version
>> java version "1.6.0_18"
>> OpenJDK Runtime Environment (IcedTea6 1.8.13) (6b18-1.8.13-0+squeeze1)
>> OpenJDK Zero VM (build 14.0-b16, mixed mode)
>>
>> Under Fedora, I have this:
>> # java -version
>> java version "1.7.0_b147-icedtea"
>> OpenJDK Runtime Environment (fedora-2.1.fc17.6-arm)
>> OpenJDK Zero VM (build 22.0-b10, interpreted mode)
>>
>> Both are using Zero VM, but with different modes. Then I received this
>> explanation from the Raspberry Pi forums:
>> --------------------------------------
>> The main reason why you see a magnitude difference in the Debian vs Fedora
>> results really boils down to that there exist two variants of Zero.
>>
>> In Fedora you have installed OpenJDK Zero VM (build 22.0-b10, interpreted
>> mode)
>> This is using the C++ interpreter only.
>>
>> In Debian you have installed OpenJDK Zero VM (build 14.0-b16, mixed mode)
>> The "mixed mode" version contains two addons, designed by Edward Nevill, on
>> top of the generic Zero C++ interpreter, a rewritten interpreter main loop
>> in ARM assembly this speeds up the jvm about a factor x3 this do benefit the
>> Pi.
>> The "mixed mode" also include a Thumb2 JIT that improves speed for ARMv7
>> cores and later, the Pi could benefit from this JIT if it got rewritten from
>> Thumb2 to target generic ARM or Thumb1 assembly.
>> -----------------------------------------
>>
>
> Performance-wise JamVM is faster than the assembly interpreter but
> slower than the Thumb2 JIT.  However this doesn't work on the Pi, so
> JamVM should be faster than what you used on Debian (as long as it was
> on the Pi).
>

Some numbers (SciMark is far from my favourite but it runs quickly):

rob at terminus ~ $ java -version
java version "1.7.0_07"
OpenJDK Runtime Environment (IcedTea7 2.3.2) (7u7-2.3.2a-1+rpi1)
OpenJDK Zero VM (build 22.0-b10, mixed mode)
rob at terminus ~ $
rob at terminus ~ $ java -jamvm -version
java version "1.7.0_07"
OpenJDK Runtime Environment (IcedTea7 2.3.2) (7u7-2.3.2a-1+rpi1)
JamVM (build 1.6.0-devel, inline-threaded interpreter with stack-caching)
rob at terminus ~ $
rob at terminus ~ $ java -jamvm -cp scimark2lib.jar jnt.scimark2.commandline

SciMark 2.0a

Composite Score: 5.5537918178428125
FFT (1024): 3.5766089775997916
SOR (100x100):   10.826562572167525
Monte Carlo : 1.0230009994051166
Sparse matmult (N=1000, nz=5000): 5.1887509149849
LU (100x100): 7.154035625056732

java.vendor: Oracle Corporation
java.version: 1.7.0_07
os.arch: arm
os.name: Linux
os.version: 3.2.27+
rob at terminus ~ $ java -cp scimark2lib.jar jnt.scimark2.commandline

SciMark 2.0a

Composite Score: 3.025386835322935
FFT (1024): 1.769265203055126
SOR (100x100):   6.539676672735603
Monte Carlo : 0.627889791674138
Sparse matmult (N=1000, nz=5000): 2.660431296672497
LU (100x100): 3.5296712124773117

java.vendor: Oracle Corporation
java.version: 1.7.0_07
os.arch: arm
os.name: Linux
os.version: 3.2.27+
rob at terminus ~ $

>> The Thumb2 info is out of my league. However, the disparity on performance
>> was drastic, thus I searched out other avenues. Cacao was suggested, so I
>> thought I'd try it out. Since it won't compile on 32bit x86, I have no hope
>> of compiling it on ARM. I guess my search continues.... :-(  Stefan
>> suggested JamVM as a possibility. Off I go. :-)
>