[Bug target/39429] compiler create bad asm codes.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=39429 Ramana Radhakrishnan ramana at gcc dot gnu.org changed: What|Removed |Added Status|NEW |RESOLVED CC||ramana at gcc dot gnu.org Resolution|--- |FIXED Target Milestone|--- |4.5.0 --- Comment #10 from Ramana Radhakrishnan ramana at gcc dot gnu.org --- Fixed in 4.5.0 but not earlier.
[Bug target/39429] compiler create bad asm codes.
--- Comment #8 from mikpe at it dot uu dot se 2009-07-13 13:05 --- Mystery solved. Buried in revision 146451, which should just fix enum conversions for C++ compatibility, is the following bug fix: --- trunk/gcc/config/arm/arm.c 2009/04/20 19:30:55 146450 +++ trunk/gcc/config/arm/arm.c 2009/04/20 19:35:00 146451 @@ -7408,7 +7410,7 @@ /* Don't accept any offset that will require multiple instructions to handle, since this would cause the arith_adjacentmem pattern to output an overlong sequence. */ - if (!const_ok_for_op (PLUS, val0) || !const_ok_for_op (PLUS, val1)) + if (!const_ok_for_op (val0, PLUS) || !const_ok_for_op (val1, PLUS)) return 0; /* Don't allow an eliminable register: register elimination can make The parameters to const_ok_for_op had been swapped, causing this if statement to not reject offsets that are awkward for ARM. Combined with a non-FL_LDSCHED cpu type this enabled arith_adjacentmem for a bad offset, which forced it to split the LDM into two LDRs, and that code fails to order the LDRs to avoid clobbering the shared base register. With the above patch arith_adjacentmem will not trigger for bad offsets, avoiding the broken LDM splitting code. This patch is needed also for the 4.4 and 4.3 branches, and I've checked that it fixes this test case there too. It seems that there is a bit of redundancy between the adjacent_mem_locations test and the arith_adjacentmem pattern. Both check const_ok_for_arm on the offset and the negated offset. The first attempts to reject bad offsets, while the second attempts to handle them. I'm not sure, but I _think_ that the code in arith_adjacentmem to split an LDM into two LDRs is now dead (after the bug fix above). However, just in case it isn't, I'm attaching a patch to correct it. Unrelated to this PR, buried in revision 146451 is another bug fix: --- trunk/gcc/config/arm/arm.c 2009/04/20 19:30:55 146450 +++ trunk/gcc/config/arm/arm.c 2009/04/20 19:35:00 146451 @@ -5465,7 +5465,7 @@ return true; case ABS: - if (GET_MODE_CLASS (mode == MODE_FLOAT)) + if (GET_MODE_CLASS (mode) == MODE_FLOAT) { if (TARGET_HARD_FLOAT (mode == SFmode || mode == DFmode)) { This one is also needed in 4.4, but not in 4.3. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39429
[Bug target/39429] compiler create bad asm codes.
--- Comment #9 from mikpe at it dot uu dot se 2009-07-13 13:07 --- Created an attachment (id=18186) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=18186action=view) fix arith_adjacentmem LDM splitting code -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39429
[Bug target/39429] compiler create bad asm codes.
--- Comment #4 from mikpe at it dot uu dot se 2009-07-12 11:29 --- Created an attachment (id=18179) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=18179action=view) reduced test case in plain C -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39429
[Bug target/39429] compiler create bad asm codes.
--- Comment #5 from ramana at gcc dot gnu dot org 2009-07-12 20:51 --- (In reply to comment #4) Created an attachment (id=18179) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=18179action=view) [edit] reduced test case in plain C What options did you use ? Did you use -O2 , -O3 or -Os with the testcase you've added here ? I don't see the problem with 4.5.0 trunk 149479 with either -mcpu=arm740t or with arm7tdmi. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39429
[Bug target/39429] compiler create bad asm codes.
--- Comment #6 from mikpe at it dot uu dot se 2009-07-12 21:21 --- (In reply to comment #5) What options did you use ? Did you use -O2 , -O3 or -Os with the testcase you've added here ? I don't see the problem with 4.5.0 trunk 149479 with either -mcpu=arm740t or with arm7tdmi. Either -O2 or -Os plus -mcpu=arm740t will trigger it in gcc-4.3.4 and gcc-4.4.1. After prepping a patch for 4.4.1 I noticed that I couldn't trigger it in 4.5; I'm currently bisecting 4.5 to identify what changed it there. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39429
[Bug target/39429] compiler create bad asm codes.
--- Comment #7 from mikpe at it dot uu dot se 2009-07-12 23:58 --- Revision 146451 on 4.5 changed it from generating broken code to generating not-so-broken code. That's completely unexpected since that revision is a enable-bootstrap-with-c++ thing which isn't supposed to change any behaviour. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39429
[Bug target/39429] compiler create bad asm codes.
--- Comment #3 from mikpe at it dot uu dot se 2009-07-11 20:20 --- It seems that cpu type and tuning options make a difference here. If I compile with -mcpu and -mtune referring to a cpu that does not imply FL_LDSCHED, such as arm740t, then I get the broken code that clobbers r0 before loading r3. Changing cpu and tune types to a cpu that does imply FL_LDSCHED, such as arm8 or xscale, then r3 is loaded before r0 is clobbered and the sub becomes an rsb. -- mikpe at it dot uu dot se changed: What|Removed |Added CC||mikpe at it dot uu dot se http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39429
[Bug target/39429] compiler create bad asm codes.
-- rearnsha at gcc dot gnu dot org changed: What|Removed |Added Keywords||wrong-code Priority|P3 |P2 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39429