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

From: Todd Veldhuizen (tveldhui@oonumerics.org)
Date: Sun Apr 12 1998 - 13:05:20 EST


Looptest output for RS/6000 (powerPC), using KCC:

Hardware: RS/6000 PowerPC Model 43P, 100 MHz (oonumerics.org)
Theoretical peak: 200 Mflops
OS: AIX 4.2
C++ compiler: KCC 3.2e
Backend: XL C++ 3.1.4.4
Flags: -O3

In-cache:
Mflops/s Description
 42.622 for, indirection, unit stride
 44.101 for, indirection, unit stride, no +=
 44.357 for, indirection, unit stride, backwards loops
 56.936 for, unroll=4, unit stride, constants loaded into temps
 55.689 for, unroll=4, unit stride, constants loaded into temps,
                4 read then 4 write
 57.798 for, unroll=4, unit stride, constants loaded into temps,
            no +=
 59.143 for, unroll=4, unit stride, constants loaded into temps,
        CSE for index offsets
 57.364 for, unroll=4, unit stride, constants loaded into temps, backwards
 58.688 for, unroll=8, unit stride, constants loaded into temps
 57.364 for, indirection, unit stride, constants into temps
 44.616 for, indirection, non-unit stride
 51.901 for, indirection, non-unit stride, constants loaded into temps
 40.799 while, pointer increment, unit stride
 46.806 while, pointer increment, unit stride,
    constants loaded into temps
 38.925 while, pointer increment, non-unit stride
  51.55 while, pointer increment, unroll=4, non-unit stride,
     constants loaded into temps
 43.349 for, unroll=4, unit stride, constants loaded into temps, prefetching
 43.847 interlaced, for, indirection, unit stride

Out of cache:
Mflops/s Description
 7.0226 for, indirection, unit stride
 7.0278 for, indirection, unit stride, no +=
 7.0643 for, indirection, unit stride, backwards loops
 6.2475 for, unroll=4, unit stride, constants loaded into temps
 7.2248 for, unroll=4, unit stride, constants loaded into temps,
                4 read then 4 write
 6.2068 for, unroll=4, unit stride, constants loaded into temps,
            no +=
 6.2332 for, unroll=4, unit stride, constants loaded into temps,
        CSE for index offsets
 7.2111 for, unroll=4, unit stride, constants loaded into temps, backwards
 6.1927 for, unroll=8, unit stride, constants loaded into temps
 7.0252 for, indirection, unit stride, constants into temps
 6.6878 for, indirection, non-unit stride
 6.7042 for, indirection, non-unit stride, constants loaded into temps
 6.6667 while, pointer increment, unit stride
 6.6621 while, pointer increment, unit stride,
    constants loaded into temps
  6.532 while, pointer increment, non-unit stride
 6.9434 while, pointer increment, unroll=4, non-unit stride,
     constants loaded into temps
 7.1975 for, unroll=4, unit stride, constants loaded into temps, prefetching
 5.4402 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