Re: BZDEV: New snapshot, looptest.cpp (Cray T3E,KCC)

From: Todd Veldhuizen (tveldhui@oonumerics.org)
Date: Sun Apr 12 1998 - 12:55:10 EST


Looptest output for Cray T3E:

Hardware: Cray T3E (mcurie.nersc.gov), single PE, 450 MHz Alpha
Theoretical peak: 900 Mflops/PE
OS: Unicos 2.0.2.24
C++ compiler: KCC 3.2b
Backend: Cray CC 3.0.2.0
Flags: +K3 -O3 --backend -hpipeline3 --backend -hunroll --backend -haggress
       --backend -hscalar2

In-cache:
Mflops/s Description
 332.18 for, indirection, unit stride
 328.41 for, indirection, unit stride, no +=
 305.97 for, indirection, unit stride, backwards loops
  333.9 for, unroll=4, unit stride, constants loaded into temps
  333.2 for, unroll=4, unit stride, constants loaded into temps,
            no +=
 333.26 for, unroll=4, unit stride, constants loaded into temps,
        CSE for index offsets
 325.44 for, unroll=4, unit stride, constants loaded into temps, backwards
 334.85 for, unroll=8, unit stride, constants loaded into temps
 332.27 for, indirection, unit stride, constants into temps
 332.18 for, indirection, non-unit stride
 331.64 for, indirection, non-unit stride, constants loaded into temps
  65.42 while, pointer increment, unit stride
 65.584 while, pointer increment, unit stride,
    constants loaded into temps
  65.43 while, pointer increment, non-unit stride
 69.618 while, pointer increment, unroll=4, non-unit stride,
     constants loaded into temps
 252.13 for, unroll=4, unit stride, constants loaded into temps, prefetching
 335.17 interlaced, for, indirection, unit stride

Out of cache:
Mflops/s Description
 55.845 for, indirection, unit stride
 55.649 for, indirection, unit stride, no +=
 22.098 for, indirection, unit stride, backwards loops
 55.726 for, unroll=4, unit stride, constants loaded into temps
 55.751 for, unroll=4, unit stride, constants loaded into temps,
            no +=
 55.725 for, unroll=4, unit stride, constants loaded into temps,
        CSE for index offsets
 21.608 for, unroll=4, unit stride, constants loaded into temps, backwards
 55.298 for, unroll=8, unit stride, constants loaded into temps
 55.418 for, indirection, unit stride, constants into temps
 54.864 for, indirection, non-unit stride
 55.546 for, indirection, non-unit stride, constants loaded into temps
 36.809 while, pointer increment, unit stride
 37.276 while, pointer increment, unit stride,
    constants loaded into temps
 36.869 while, pointer increment, non-unit stride
 38.936 while, pointer increment, unroll=4, non-unit stride,
     constants loaded into temps
 53.712 for, unroll=4, unit stride, constants loaded into temps, prefetching
 33.151 interlaced, for, indirection, unit stride
--------------------- blitz-dev list --------------------------------
* To subscribe/unsubscribe: mail to majordomo@oonumerics.org, with
"subscribe blitz-dev" or "unsubscribe blitz-dev" in the body of the message
* Blitz++ web page: http://oonumerics.org/blitz/



This archive was generated by hypermail 2b29 : Wed Feb 20 2002 - 04:30:04 EST