Skip to content
GitLab
Explore
Sign in
Primary navigation
Search or go to…
Project
O
osv
Manage
Activity
Members
Labels
Plan
Issues
Issue boards
Milestones
Wiki
Code
Merge requests
Repository
Branches
Commits
Tags
Repository graph
Compare revisions
Snippets
Build
Pipelines
Jobs
Pipeline schedules
Artifacts
Deploy
Releases
Package registry
Container Registry
Model registry
Operate
Environments
Terraform modules
Monitor
Incidents
Service Desk
Analyze
Value stream analytics
Contributor analytics
CI/CD analytics
Repository analytics
Model experiments
Help
Help
Support
GitLab documentation
Compare GitLab plans
Community forum
Contribute to GitLab
Provide feedback
Keyboard shortcuts
?
Snippets
Groups
Projects
Show more breadcrumbs
Verlässliche Systemsoftware
projects
osv
Commits
b1ec4adc
Commit
b1ec4adc
authored
11 years ago
by
Nadav Har'El
Browse files
Options
Downloads
Patches
Plain Diff
todo: add todo/fpu
A todo item to avoid unnecessary fpu state saving
parent
682e57fd
No related branches found
Branches containing commit
No related tags found
No related merge requests found
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
todo/fpu
+21
-0
21 additions, 0 deletions
todo/fpu
with
21 additions
and
0 deletions
todo/fpu
0 → 100644
+
21
−
0
View file @
b1ec4adc
Avoiding unnecessary FPU state saving
=====================================
1. Understand why the "sunflow" benchmark fails unless we also save FPU
state on reschedule_from_interrupt with !preempt:
1. One guess is that Java has a bug where it doesn't follow the calling
conventions and doesn't save its fpu state before calling some library
function. Avi tried to reproduce this in Linux (with a layer that
ruins the FPU state before calling library functions) and couldn't
so this may not be the correct explanation.
2. Another guess (by Avi) is that a signal handler is involved and
we don't save the FPU on signals! Need to fix that.
2. Even if we don't need to save the whole fpu state on !preempt, we probably
need to save the fcw and mxcsr.
3. If for some reason (?) this problem is unfixable and we must (why?) save
FPU on every context switch, we may be forced to consider lazy FPU
switching - which we tried to avoid.
This diff is collapsed.
Click to expand it.
Preview
0%
Loading
Try again
or
attach a new file
.
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Save comment
Cancel
Please
register
or
sign in
to comment