gdimageline_aa failing on some arches due to fpu implementations (ia64, powerpc, s390x, ...)
Issue #52
resolved
Or it could be something else... but occam's razor says it's big endian problem :)
gdimageline/gdimageline_aa.c:38: Total pixels changed: 244 with a maximum channel difference of 0.
./gdimageline/gdimageline_aa_d_1_exp.png failed
Comments (13)
-
-
Account Deleted reporter
Yep, I do have an access to Debian porters machines.
-
Account Deleted reporter
- attached gdimageline_aa.c_38_out.png
- attached gdimageline_aa.c_38_diff.png
-
Account Deleted reporter
- attached gdimageline_aa_d_1_exp.png
-
Account Deleted reporter
- attached gdimageline_aa_amd64.out
- attached gdimageline_aa_ia64.out
-
Account Deleted reporter
I think we have found a bug in GCC :)
-
ia64 is not big endian :)
-
Account Deleted reporter
It's not the bug in GCC, but bug in FPU implementation, see PR323.
We should really rewrite the part to use fixed point arithmetics.
-
Account Deleted reporter
See the (now wrong) reimplementation in branch clip_1d_fixed_point
-
What is done in the version I implemented for php. Should drop it in maybe.
-
what is the status for this issue? Do we need more tweaks or can we (finally :) fire the release? :)
-
Account Deleted reporter
- changed status to resolved
fixed in 3be2f89
-
- changed title to gdimageline_aa failing on some arches due to fpu implementations (ia64, powerpc, s390x, ...)
- Log in to comment
Sadly I don't have any of these CPUs at hand to debug. Do you?