Bug 386891 - wish: build separate glibc packages for cpu submodels on powerpc
Summary: wish: build separate glibc packages for cpu submodels on powerpc
Alias: None
Product: Fedora
Classification: Fedora
Component: glibc   
(Show other bugs)
Version: 8
Hardware: powerpc
OS: Linux
Target Milestone: ---
Assignee: Jakub Jelinek
QA Contact: Fedora Extras Quality Assurance
Depends On:
TreeView+ depends on / blocked
Reported: 2007-11-16 14:44 UTC by Marc Dietrich
Modified: 2007-11-30 22:12 UTC (History)
0 users

Fixed In Version:
Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of:
Last Closed: 2007-11-16 14:54:01 UTC
Type: ---
Regression: ---
Mount Type: ---
Documentation: ---
Verified Versions:
Category: ---
oVirt Team: ---
RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: ---

Attachments (Terms of Use)

Description Marc Dietrich 2007-11-16 14:44:34 UTC
Optimized glibc builds can influence system performance greatly. This is 
especially true on powerpc systems, where a lot of cpu variants exist. Up to 
now, the glibc package provides an additional library set for power6, while 
standard libraries are compiled for 604 or common? I'm just guessing... 

While it would be a waste of space to support all varities of cpus in a single 
glibc package, I suggest to split it into several sub-packages, e.g.

- glibc-common (required)
- glibc-power6 (with power6 directory, compiled with -mcpu=power6)
- glibc-cell (with ppc-cell-be directory, compiled with -mcpu=970 
or -mcpu=cell)
- ...

The same holds also for kernels.

I found noticeable performace improvements using my own kernel and glibc 
(build with -mcpu=970 or mcpu=cell using ppu-gcc) on my PS3 system (but not on 
my G3/750).

Comment 1 Jakub Jelinek 2007-11-16 14:54:01 UTC
This would be a maintainance nightmare, plus glibc on ppc and ppc64 already
builds about 3 times longer or more than on other architectures.

Note You need to log in before you can comment on or make changes to this bug.