From: Matan Ziv-Av (matan@svgalib.org)
Date: Tue 16 Apr 2002 - 11:06:50 IDT
In svgalib-1.9.13 I modified linearspeed to test two memory copy to
video ram methods. One is libc's memcpy, and the other is as simple
for loop.
On my main system (Athlon, Matrox G450 AGP, MSK8.1) the results are
somewhat surprising. The libc's optimized memcpy routine is very slow
when copying to video ram:
Video memory timing: 2.1s, 152380K/s
To Video memcpy timing: 8.1s, 39506K/s
To Video mem copy timing: 2.1s, 152380K/s
System memory timing: 0.9s, 355555K/s
To System memcpy timing: 1.3s, 246153K/s
To System mem copy timing: 2.6s, 123076K/s
On another system I can test (Celeron, nVidia GF2, Slackware 8):
Video memory timing: 1.3s, 246153K/s
To Video memcpy timing: 1.3s, 246153K/s
To Video mem copy timing: 1.5s, 213333K/s
System memory timing: 1.8s, 177777K/s
To System memcpy timing: 1.8s, 177777K/s
To System mem copy timing: 4.8s, 66666K/s
I saw a behaviour like the athlon on a duron system, but not on another
celeron or on a K6, so I suspect it might have something to do with
Athlon architecture, and memcpy assembler optimisations.
Please try this and report the results to me, including processor type,
video card used and distribution (or libc version, if you compiled your
own).
--
Matan Ziv-Av. matan@svgalib.org
------------------------------------------------------------------
Unsubscribe: To: listbot@svgalib.org
Body: unsubscribe linux-svgalib
This archive was generated by hypermail 2.1.4 : Wed 21 Jan 2004 - 22:10:24 IST