A while back, I believe back during the last FUDCon in Raleigh I talked with kylem and davej about turning off the "inline-functions-called-once" which inlines functions that are only called once. Turning off this optimization would allow SystemTap to work much better, especially in the NFS code...
Kyle/Dave (or anybody else) remember if anything came out from that conversation?
Is there any reason we should not remove this optimization?
steved.
On Mon, Oct 06, 2008 at 02:56:04PM -0400, Steve Dickson wrote:
A while back, I believe back during the last FUDCon in Raleigh I talked with kylem and davej about turning off the "inline-functions-called-once" which inlines functions that are only called once. Turning off this optimization would allow SystemTap to work much better, especially in the NFS code...
Kyle/Dave (or anybody else) remember if anything came out from that conversation?
Is there any reason we should not remove this optimization?
I don't mind the idea of turning this off in debug builds, but I'm iffy about turning the optimization off in release builds...
regards, Kyle
Kyle McMartin wrote:
On Mon, Oct 06, 2008 at 02:56:04PM -0400, Steve Dickson wrote:
A while back, I believe back during the last FUDCon in Raleigh I talked with kylem and davej about turning off the "inline-functions-called-once" which inlines functions that are only called once. Turning off this optimization would allow SystemTap to work much better, especially in the NFS code...
Kyle/Dave (or anybody else) remember if anything came out from that conversation?
Is there any reason we should not remove this optimization?
I don't mind the idea of turning this off in debug builds, but I'm iffy about turning the optimization off in release builds...
Understandable... but its a catch-22... If its not turned off in release builds then we can't use systemtap [to debug NFS] in production environments which the true value systemtap brings to the table... imho...
steved.
On Tue, Oct 07, 2008 at 09:02:08AM -0400, Steve Dickson wrote:
Kyle McMartin wrote:
On Mon, Oct 06, 2008 at 02:56:04PM -0400, Steve Dickson wrote:
A while back, I believe back during the last FUDCon in Raleigh I talked with kylem and davej about turning off the "inline-functions-called-once" which inlines functions that are only called once. Turning off this optimization would allow SystemTap to work much better, especially in the NFS code...
Kyle/Dave (or anybody else) remember if anything came out from that conversation?
Is there any reason we should not remove this optimization?
I don't mind the idea of turning this off in debug builds, but I'm iffy about turning the optimization off in release builds...
Understandable... but its a catch-22... If its not turned off in release builds then we can't use systemtap [to debug NFS] in production environments which the true value systemtap brings to the table... imho...
Is this liable to actually occur, though?
I mean, I don't have any problem with it per se... I'd just be suprised, is all.
regards, Kyle
On Mon, Oct 06, 2008 at 02:56:04PM -0400, Steve Dickson wrote:
A while back, I believe back during the last FUDCon in Raleigh I talked with kylem and davej about turning off the "inline-functions-called-once" which inlines functions that are only called once. Turning off this optimization would allow SystemTap to work much better, especially in the NFS code...
Kyle/Dave (or anybody else) remember if anything came out from that conversation?
I forgot all about that conversation.
Is there any reason we should not remove this optimization?
I'd be really surprised if it's actually even measurable on modern CPUs.
Dave
kernel@lists.fedoraproject.org