This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: disappearing labels fix

At 20:53 11.05.2001, Franz Sirl wrote:
>At 02:23 11.05.2001, Jeffrey A Law wrote:
>>This patch should fix the regressions with -fPIC on the PPC that were
>>caused by my disappearing labels patch.  It should also fix the problems
>>in libjava on the PPC.
>>It's been 3-staged on the PPC (linux & aix), PA32 hpux, x86 (linux).
>>The fr30 still works with this patch (which is the port which originally
>>caused me to change the code in cse a year or two ago).
>>And it just feels right.  I like to be able to delete/simplify code to
>>fix bugs :-)
>>Installed on the branch as well as the mainline sources.
>>         * cse.c (cse_insn): Mostly remove first hunk of code which collapsed
>>         a computed jump into an unconditional jump.
>>         In second hunk of code for collapsing computed and conditional jumps
>>         into unconditional jumps, reemit the jump in additional cases.
>this works fine on PPC, but it may be responsible for these x86 RTL 
>checking regressions on the 3.0 branch:
>FAIL: gcc.c-torture/compile/20001221-1.c,  -O2
>FAIL: gcc.c-torture/compile/20001221-1.c,  -O3 -fomit-frame-pointer
>FAIL: gcc.c-torture/compile/20001221-1.c,  -O3 -g
>FAIL: gcc.c-torture/compile/20001221-1.c,  -Os
>spawn /home/gnu/obj/gcc3/gcc/xgcc -B/home/gnu/obj/gcc3/gcc/ -w -O2 -c -O2 
>-o /home/gnu/obj/gcc3/gcc/testsuite/20001221-1.o 
>/home/gnu/gcc3/gcc/testsuite/gcc.c-torture/compile/20001221-1.c: In 
>function `foo':
>/home/gnu/gcc3/gcc/testsuite/gcc.c-torture/compile/20001221-1.c:70: RTL 
>check: expected code `code_label', have `note' in 
>propagate_block_delete_insn, at flow.c:3722
>Please submit a full bug report, with preprocessed source if appropriate.
>See <URL:> for instructions.
>compiler exited with status 1
>Unfortunately I bootstrapped the branch last on May 3, so I'm not 100% 
>it's your patch causing this. But since the testcase is full of case 
>constructs... I'll have no time to track it down to a certain patch during 
>the next days, so I thought I better report it now.

Jeff, I just confirmed this. Reverting your last patch to cse.c (which 
fixes the PPC problems with jumptable labels) makes this testcase pass 
again on x86-linux-gnu with RTL checking enabled.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]