|
|
|
|
|||||||
| Forums | Portal | Register | Forum Rules | FAQ | Contribute | Members List | Arcade | Search | Today's Posts | Mark Forums Read |
| Linux Benchmarks Download software, run and post benchmarks here! |
More UNIX and Linux Forum Topics You Might Find Helpful
|
||||
| Thread | Thread Starter | Forum | Replies | Last Post |
| Original (Archive) Site for Linux Benchmarks | Neo | Linux Benchmarks | 3 | 02-06-2007 06:38 AM |
| Linux Benchmarks Makes No Sense | philip_38 | Linux Benchmarks | 0 | 07-22-2005 07:29 AM |
| Instructions for Linux Benchmarks | Neo | Linux Benchmarks | 0 | 09-06-2003 04:48 PM |
| Linux Processing Benchmarks ? | lcstephens | UNIX for Advanced & Expert Users | 2 | 07-28-2002 09:13 AM |
| Linux mail servers benchmarks (Qmail, Sendmail,Postfix) | cipango | News, Links, Events and Announcements | 0 | 07-26-2002 10:30 AM |
 |
|
|
Submit Tools | LinkBack | Thread Tools | Search this Thread | Display Modes |
|
#1
12-09-2003
|
|||
|
|||
|
Arithmetic tests needs to be updated for advanced compiler optimization
Hi,
I was trying to build Linux Benchmarks with latest Intel C++ Compiler. When I used -ipo (inter-procedural optimization) option, arithmetic test (arith.c) failed on execution. The problem is Intel compiler's advanced optimization option (-ipo) optimizes much more than expected and this results in elimination of some part of code. /------------------------------------------------- ++iter; /* the loop calls a function to insure that something is done the results of the function are fed back in (just so they they won't be thrown away. A loop with unused assignments may get optimized out of existence */ result = dumb_stuff(result); /-------------------------------------------------------- This part gets eliminated becuase of IPO. Steps to reproduce: $icc -o ./pgms/arithoh -DTIME -O3 -xW -ipo -static -unroll -Darithoh ./src/arith.c $./pgms/arithoh 1 0 loops and this results in divide by zero error. Regards, CP 
|
| Forum Sponsor | ||
|
|
|
#3
12-12-2003
|
|||
|
|||
|
Hi Neo,
I have seen, in this forum, that developers have used advanaced gcc optimization options for benchmarking. I just want to compare the performance of gcc and Intel compiler and thats why trying to use best available optimization options for Intel compiler i.e -ipo (PGO can also be used). If this doesnt answer your question than please let me know why one should not be allowed to use whatever options available with a compiler for these benchmarks? Regards, CP
|
|
#4
12-12-2003
|
||||
|
||||
|
This is a point that has been long debated.... i.e. should be benchmark be changed to reflect new compiler options or left as they are so they relate to all the benchmarks that were ran from the past.
I have always wanted to have the entire benchmark history/ tradition over time and have not been a fan of trying to optimize the marks with compiler flags that were not available in the past. The results of the marks will not be historically related to the beginning..... alpha-to-omega comparisons. There have many many others who strongly dissagee and want to change them on the continuing basis. This 'tension' causes problems in the overall BM community, taking the fun out of it (and one of the reasons I stopped running the Linux BMs many years ago.) Neo
|
|
#5
12-29-2003
|
||||
|
||||
|
CPU/Speed: AMD Duron 900Mhz
Ram: 512 PC133 Motherboard: EliteGroup K7S5A 266 / 200 FSB SiSŪ 735 System Chipset 5 PCI slots, 1 4X AGP slot, 1 AMR slot Disk: 40 Gig Maxtor EIDE Load: 1 user, running pppd and httpd Kernel: Linux 2.4.20-6 Kernel ELF?: yes pgms: gcc 2.5.8 compiled; options Code:
==============================================================
BYTE UNIX Benchmarks (Version 3.11)
System -- Linux Tarball 2.4.20-6 #1 Thu Feb 27 10:01:19 EST 2003 i686 athlon i386 GNU/Linux
Start Benchmark Run: Mon Dec 29 17:04:25 EST 2003
2 interactive users.
Dhrystone 2 without register variables 1738794.9 lps (10 secs, 6 samples)
Dhrystone 2 using register variables 1847801.6 lps (10 secs, 6 samples)
Arithmetic Test (type = arithoh) 3742724.2 lps (10 secs, 6 samples)
Arithmetic Test (type = register) 166336.4 lps (10 secs, 6 samples)
Arithmetic Test (type = short) 157037.3 lps (10 secs, 6 samples)
Arithmetic Test (type = int) 166494.3 lps (10 secs, 6 samples)
Arithmetic Test (type = long) 166528.3 lps (10 secs, 6 samples)
Arithmetic Test (type = float) 335329.0 lps (10 secs, 6 samples)
Arithmetic Test (type = double) 335727.1 lps (10 secs, 6 samples)
System Call Overhead Test 487088.3 lps (10 secs, 6 samples)
Pipe Throughput Test 541380.4 lps (10 secs, 6 samples)
Pipe-based Context Switching Test 185194.6 lps (10 secs, 6 samples)
Process Creation Test 4576.9 lps (10 secs, 6 samples)
Execl Throughput Test 1059.3 lps (9 secs, 6 samples)
File Read (10 seconds) 1642657.0 KBps (10 secs, 6 samples)
File Write (10 seconds) 95026.0 KBps (10 secs, 6 samples)
File Copy (10 seconds) 37314.0 KBps (10 secs, 6 samples)
File Read (30 seconds) 1202209.0 KBps (30 secs, 6 samples)
File Write (30 seconds) 75369.0 KBps (30 secs, 6 samples)
File Copy (30 seconds) 24758.0 KBps (30 secs, 6 samples)
C Compiler Test 215.9 lpm (60 secs, 3 samples)
Shell scripts (1 concurrent) 324.7 lpm (60 secs, 3 samples)
Shell scripts (2 concurrent) 155.0 lpm (60 secs, 3 samples)
Shell scripts (4 concurrent) 81.7 lpm (60 secs, 3 samples)
Shell scripts (8 concurrent) 42.0 lpm (60 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 31901.3 lpm (60 secs, 6 samples)
Recursion Test--Tower of Hanoi 19793.6 lps (10 secs, 6 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Arithmetic Test (type = double) 2541.7 335727.1 132.1
Dhrystone 2 without register variables 22366.3 1738794.9 77.7
Execl Throughput Test 16.5 1059.3 64.2
File Copy (30 seconds) 179.0 24758.0 138.3
Pipe-based Context Switching Test 1318.5 185194.6 140.5
Shell scripts (8 concurrent) 4.0 42.0 10.5
=========
SUM of 6 items 563.3
AVERAGE 93.9
Last edited by google; 12-29-2003 at 05:53 PM.
|
Linear Mode
