[tree-ssa] Mainline merge as of 2004-03-24

[Diego Novillo]

Bootstrapped and tested x86, x86-64 and alpha.  We have new c-torture
regressions that are not present in mainline:

         FAIL: gcc.c-torture/execute/20040313-1.c execution
         FAIL: gcc.c-torture/execute/20040313-1.c execution
         FAIL: gcc.c-torture/execute/20040313-1.c execution
         FAIL: gcc.c-torture/execute/20040313-1.c execution
         FAIL: gcc.c-torture/execute/20040313-1.c execution
         FAIL: gcc.c-torture/execute/20040313-1.c execution
         FAIL: gcc.c-torture/execute/loop-2c.c execution
         FAIL: gcc.c-torture/execute/loop-2d.c execution
         FAIL: gcc.dg/torture/builtin-nonneg-1.c (test for excess errors)
         FAIL: gcc.dg/torture/builtin-nonneg-1.c (test for excess errors)
         FAIL: gcc.dg/torture/builtin-nonneg-1.c (test for excess errors)
         FAIL: gcc.dg/torture/builtin-nonneg-1.c (test for excess errors)
         FAIL: gcc.dg/torture/builtin-nonneg-1.c (test for excess errors)
         FAIL: gcc.dg/torture/builtin-nonneg-1.c (test for excess errors)

20040313-1.c is the test case for PR 14470.  The two loop failures
happen only at -Os on x86, it does not seem to be a tree optimizer
failure because we emit the same trees.  I'm looking at those today. 
The builtin-nonneg failure is another case of tree-ssa not folding
builtins properly.

None of the failures seemed important enough to hold the merge which was
already getting old.


Diego.