SASADA Koichi wrote:
> Hi,
> 
> M. Edward (Ed) Borasky wrote:
>> Much better!! YARV is four times as fast as the stock Ruby, and I have
>> the profiling stuff cleaned up. Thanks!!
>>
>> 1. All tests run on a Pentium III Coppermine 933 MHz 512 megabytes
>> 2. OS is Gentoo Linux 2.6.17 Kernel
>> 3. Compiler is gcc-3.4.6
>>    * non-profiling compiler flags "-O2 -march-pentium3"
>>    * profiling compiler flags "-g -pg"
>>
>> Ruby version				gprof	dim	seconds
>> ruby 1.8.5 (2006-06-22)			no	256	200.635709
>> ruby 1.8.5 (2006-08-19)			yes	256	617.325720
>> YARVCore 0.4.1 Rev: 540 (2006-08-20)	no	256	54.842272
>> YARVCore 0.4.1 Rev: 540 (2006-08-20)	yes	256	243.747870
> 
> Good results :)
> 
> Can you try without a following optimization?
> -> edit "yarv-build/vm_opts.h" and
> 
> #define OPT_SPECIALISED_INSTRUCTION  1
> 
> to
> 
> #define OPT_SPECIALISED_INSTRUCTION  0
> 
> and rebuild.
> 
> This optimization boost simple methods such as Fixnum#+.  Recent YARV
> turns on this option from default.
> 
1. All tests run on a Pentium III Coppermine 933 MHz 512 megabytes
2. OS is Gentoo Linux 2.6.17 Kernel
3. Compiler is gcc-3.4.6
   * non-profiling compiler flags "-O2 -march-pentium3"
   * profiling compiler flags "-g -pg"

Ruby version                            gprof   dim     seconds
ruby 1.8.5 (2006-06-22)                 no      256     200.635709
ruby 1.8.5 (2006-08-19)                 yes     256     617.325720

#define OPT_SPECIALISED_INSTRUCTION  1
YARVCore 0.4.1 Rev: 540 (2006-08-20)    no      256     54.842272
YARVCore 0.4.1 Rev: 540 (2006-08-20)    yes     256     243.747870

#define OPT_SPECIALISED_INSTRUCTION  0
YARVCore 0.4.1 Rev: 540 (2006-08-20)    no      256     89.597011
YARVCore 0.4.1 Rev: 540 (2006-08-20)    yes     256     243.750908



--
ML: yarv-dev-en quickml.atdot.net
Info: http://www.atdot.net/~ko1/quickml