[Qemu-devel] SPARC Performance

[Tinnemeyer, Jorn]

Hello,

Just wanted to verify my results with known performance data:  

When compared to native x86 performance, the emulator is working approximately 
47 times slower for memory operations, 44 times slower for integer and 102 
times slower for float.  Full test results are at the end of message. 

I am running this with qemu 0.7.0, using vmlinux-2.6.11+tcx found on the qemu 
website. I cross compiled nbench (2.2.2) for Sparc using Crosstools, loaded 
onto linux.img and executed the program in non graphic mode.

If this differs from other users experience, I would be very interested to see 
what is incorrect with my method.

Thanks,
Joern.


Test results:

Native (x86):
 
 ORIGINAL BYTEMARK RESULTS
 INTEGER INDEX                   : 56.882
 FLOATING-POINT INDEX            : 38.723
 Baseline (MSDOS*)   : Pentium* 90, 256 KB L2-cache, Watcom* compiler 10.0
 
 LINUX DATA BELOW
 CPU             : AuthenticAMD AMD Athlon(tm) XP 2700+ 2163MHz
 L2 Cache        : 256 KB
 OS                      : Linux 2.6.8.1-12mdk
 C compiler      : cc
 libc                    :
 MEMORY INDEX        : 16.509
 INTEGER INDEX       : 12.674
 FLOATING-POINT INDEX: 21.477
 Baseline (LINUX)    : AMD K6/233*, 512 KB L2-cache, gcc 2.7.2.3, libc-5.4.38
 
 Sparc (emulated using qemu 0.7.0):

 ORIGINAL BYTEMARK RESULTS
  INTEGER INDEX       : 1.264
  FLOATING-POINT INDEX: 0.380
 
  Baseline (MSDOS*)   : Pentium* 90, 256 KB L2-cache, Watcom* compiler 10.0
 
  LINUX DATA BELOW:
  CPU                 :
  L2 Cache            :
  OS                  : Linux 2.6.11
  C compiler          : sparc-unknown-linux-gnu-gcc
  libc                : static
  MEMORY INDEX        : 0.352
  INTEGER INDEX       : 0.290
  FLOATING-POINT INDEX: 0.211
  Baseline (LINUX)    : AMD K6/233*, 512 KB L2-cache, gcc 2.7.2.3, libc-5.4.38