diff options
Diffstat (limited to 'gcc/global.c')
| -rw-r--r-- | gcc/global.c | 25 |
1 files changed, 0 insertions, 25 deletions
diff --git a/gcc/global.c b/gcc/global.c index befdc9f..cddcac3 100644 --- a/gcc/global.c +++ b/gcc/global.c @@ -131,31 +131,6 @@ static int local_reg_live_length[FIRST_PSEUDO_REGISTER]; #define SET_REGBIT(TABLE, I, J) SET_HARD_REG_BIT (allocno[I].TABLE, J) -/* This is turned off because it doesn't work right for DImode. - (And it is only used for DImode, so the other cases are worthless.) - The problem is that it isn't true that there is NO possibility of conflict; - only that there is no conflict if the two pseudos get the exact same regs. - If they were allocated with a partial overlap, there would be a conflict. - We can't safely turn off the conflict unless we have another way to - prevent the partial overlap. - - Idea: change hard_reg_conflicts so that instead of recording which - hard regs the allocno may not overlap, it records where the allocno - may not start. Change both where it is used and where it is updated. - Then there is a way to record that (reg:DI 108) may start at 10 - but not at 9 or 11. There is still the question of how to record - this semi-conflict between two pseudos. */ -#if 0 -/* Reg pairs for which conflict after the current insn - is inhibited by a REG_NO_CONFLICT note. - If the table gets full, we ignore any other notes--that is conservative. */ -#define NUM_NO_CONFLICT_PAIRS 4 -/* Number of pairs in use in this insn. */ -int n_no_conflict_pairs; -static struct { int allocno1, allocno2;} - no_conflict_pairs[NUM_NO_CONFLICT_PAIRS]; -#endif /* 0 */ - /* Return true if *LOC contains an asm. */ static int |