[tree-ssa] bootstrap fails on x86_64-linux

Andrew Pinski pinskia@physics.uc.edu
Fri Oct 17 15:21:00 GMT 2003


On Oct 17, 2003, at 5:54 AM, Diego Novillo wrote:

> On Fri, 2003-10-17 at 06:37, Andreas Jaeger wrote:
>
>> Any update on this one?  It still fails...
>>
> genautomata.o is mis-compiled by stage2/cc1.  stage1/cc1 is
> mis-compiling more than a single libbackend file.


I found that applying the following patch (for disabling some of the 
rtl optimizers) helped me bootstrap on powerpc-apple-darwin, I was 
getting a mis-compiling in genautomata.o, gengtypelex.o and others 
(which I did not look into):

Patch:
Index: opts.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/opts.c,v
retrieving revision 1.31.2.11
diff -u -p -r1.31.2.11 opts.c
--- opts.c	28 Sep 2003 06:06:31 -0000	1.31.2.11
+++ opts.c	17 Oct 2003 13:22:42 -0000
@@ -515,6 +515,10 @@ decode_options (unsigned int argc, const
      {
        flag_merge_constants = 0;
      }
+
+
+  /* Right now on PPC-Darwin, branch on count register is causing 
gentype to miscompile. */
+  flag_branch_on_count_reg = 0;

    if (optimize >= 1)
      {
@@ -542,9 +546,12 @@ decode_options (unsigned int argc, const

    if (optimize >= 2)
      {
+/* Disable so that PPC-darwin can bootstrap. */
+#if 0
        flag_optimize_sibling_calls = 1;
        flag_cse_follow_jumps = 1;
        flag_cse_skip_blocks = 1;
+#endif
        flag_gcse = 1;
        flag_expensive_optimizations = 1;
        flag_strength_reduce = 1;



More information about the Gcc mailing list