http://codereview.appspot.com/4798045/diff/1/ipa.c
File ipa.c (right):
http://codereview.appspot.com/4798045/diff/1/ipa.c#newcode1034
ipa.c:1034: {
Has varpool node linking happened at this point? If not, the new code
here is not excersised.
On Thu, Jul 21, 2011 at 11:09 AM, davi...@google.com wrote:
http://codereview.appspot.com/4798045/diff/1/ipa.c
File ipa.c (right):
http://codereview.appspot.com/4798045/diff/1/ipa.c#newcode1034
ipa.c:1034: {
Has varpool node linking happened at this point? If not, the new code
here is not
On Thu, Jul 21, 2011 at 11:32 AM, Rong Xu x...@google.com wrote:
On Thu, Jul 21, 2011 at 11:09 AM, davi...@google.com wrote:
http://codereview.appspot.com/4798045/diff/1/ipa.c
File ipa.c (right):
http://codereview.appspot.com/4798045/diff/1/ipa.c#newcode1034
ipa.c:1034: {
Has varpool node
this is a good point. ipa_discover_readonly_nonaddressable_vars() is
called in two passes. whole-program (whole program visibility
analysis) and static-var. The one in whole-program is ok here as it is
bundled together with the analysis. the invocation in static-var can
go wrong.
should we add a
Please review the new patch attached to this email.
thanks,
-Rong
On Thu, Jul 21, 2011 at 12:44 PM, Rong Xu x...@google.com wrote:
wait a second. this still won't work if we disable the whole-program
pass (like my original change, the visibility analysis change won't
kick in). we also need
Ok.
David
On Thu, Jul 21, 2011 at 1:45 PM, Rong Xu x...@google.com wrote:
Please review the new patch attached to this email.
thanks,
-Rong
On Thu, Jul 21, 2011 at 12:44 PM, Rong Xu x...@google.com wrote:
wait a second. this still won't work if we disable the whole-program
pass (like my