[Bug tree-optimization/51782] Missing address-space information leads to wrong code
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51782
Georg-Johann Lay gjl at gcc dot gnu.org changed:
What|Removed |Added
CC||mjambor at suse dot cz
--- Comment #9 from Georg-Johann Lay gjl at gcc dot gnu.org 2012-01-12
16:19:14 UTC ---
(In reply to comment #7)
Please figure out where the address-space information is lost.
Is -f[no-]tree-sra enough information to find the bug for someone familiar with
RSA?
Here is an even simpler test case:
struct rgb { char r; };
char read_rgb_bug (const __pgm struct rgb *s)
{
struct rgb t = *s;
return t.r;
}
[Bug tree-optimization/51782] Missing address-space information leads to wrong code
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51782 --- Comment #7 from Richard Guenther rguenth at gcc dot gnu.org 2012-01-09 12:04:30 UTC --- Please figure out where the address-space information is lost.
[Bug tree-optimization/51782] Missing address-space information leads to wrong code
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51782 --- Comment #8 from Georg-Johann Lay gjl at gcc dot gnu.org 2012-01-09 20:30:35 UTC --- It's scalar replacement of aggregates: With -O1 code is wrong. With -O1 -fno-tree-sra code is correct.
[Bug tree-optimization/51782] Missing address-space information leads to wrong code
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51782 Georg-Johann Lay gjl at gcc dot gnu.org changed: What|Removed |Added Status|UNCONFIRMED |NEW Last reconfirmed||2012-01-07 CC||flamorce at videotron dot ||ca Target Milestone|--- |4.7.0 Ever Confirmed|0 |1
[Bug tree-optimization/51782] Missing address-space information leads to wrong code
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51782 --- Comment #1 from Georg-Johann Lay gjl at gcc dot gnu.org 2012-01-07 12:05:20 UTC --- Created attachment 26262 -- http://gcc.gnu.org/bugzilla/attachment.cgi?id=26262 a-bug.c C source file triggering the bug.
[Bug tree-optimization/51782] Missing address-space information leads to wrong code
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51782 --- Comment #2 from Georg-Johann Lay gjl at gcc dot gnu.org 2012-01-07 12:08:25 UTC --- Created attachment 26263 -- http://gcc.gnu.org/bugzilla/attachment.cgi?id=26263 a-bug.s (assembler output) read_rgb_ok uses LPM instructions to read data (okay) read_rgb_bug uses LD/LDD instructions to read data (wrong)
[Bug tree-optimization/51782] Missing address-space information leads to wrong code
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51782 --- Comment #3 from Georg-Johann Lay gjl at gcc dot gnu.org 2012-01-07 12:09:41 UTC --- Created attachment 26264 -- http://gcc.gnu.org/bugzilla/attachment.cgi?id=26264 a-bug.c.003t.original a-bug.c.003t.original tree-dump, FYI
[Bug tree-optimization/51782] Missing address-space information leads to wrong code
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51782 --- Comment #4 from Georg-Johann Lay gjl at gcc dot gnu.org 2012-01-07 12:11:03 UTC --- Created attachment 26265 -- http://gcc.gnu.org/bugzilla/attachment.cgi?id=26265 a-bug.c.004t.gimple a-bug.c.004t.gimple tree-dump, FYI
[Bug tree-optimization/51782] Missing address-space information leads to wrong code
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51782 --- Comment #5 from Georg-Johann Lay gjl at gcc dot gnu.org 2012-01-07 12:11:50 UTC --- Created attachment 26266 -- http://gcc.gnu.org/bugzilla/attachment.cgi?id=26266 a-bug.c.149t.optimized a-bug.c.149t.optimized tree dump, FYI
[Bug tree-optimization/51782] Missing address-space information leads to wrong code
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51782 --- Comment #6 from Georg-Johann Lay gjl at gcc dot gnu.org 2012-01-07 12:17:21 UTC --- Created attachment 26267 -- http://gcc.gnu.org/bugzilla/attachment.cgi?id=26267 a-bug.c.150r.expand a-bug.c.150r.expand RTL dump, FYI As you can see in read_rgb_ok that move insns 8, 9, 13 are reading from AS1 which is __pgm. This code is correct In read_rgb_bug, insns 8, 9 and 13 are reading from generic address space brcause there is no address space information. This code is wrong.
