Text archives Help

---

I was under the impression that -march=pentium4 turned on -msse and -msse2. Adding these flags showed no performance change.

I also tried to disable SCI Malloc by adding -DDISABLE_SCI_MALLOC to the command line. This had little effect probably because we aren't allocating memory during runtime.

I also set the SCI_ASSERTION_LEVEL to 0 in SCIRun/include/sci_defs/error_defs.h, but that also had no discernible effect.

James

Steven Parker wrote:

Try the ones from the Uintah web page (-msse -msse2 and some others)

Steve

On May 27, 2005, at 11:12 AM, James Bigler wrote:

I tried a few of the flags Steve put in there for GCC on Mac and I saw a noticible improvement on my Pentium 4 Xeon 2 GHz machine running linux.

With GCC 3.3.2 I ran:

bin/manta -bench 20 10

-O3
Benchmark completed in 10.8703 seconds (20 frames, 1.83987 frames per second)

-O3 -fgcse-sm -funroll-loops -fstrict-aliasing -fsched-interblock -falign-functions=16 -ffast-math -freorder-blocks

Benchmark completed in 8.71735 seconds (20 frames, 2.29427 frames per second)

added -march=pentium4
Benchmark completed in 8.52678 seconds (20 frames, 2.34555 frames per second)

So, just by adding a few flags I improved the framerate on the default scene by about 25%. Not too bad.

What other types of optimizations are people using for linux based systems and GCC?

James

---

• [MANTA] gcc compiler optimizations, James Bigler, 05/27/2005
  • Re: [MANTA] gcc compiler optimizations, Steven Parker, 05/27/2005
    • Re: [MANTA] gcc compiler optimizations, James Bigler, 05/31/2005

---

Archive powered by MHonArc 2.6.16.