This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Stage1 compiler miscompiles Ada on ia64
- From: Andreas Schwab <schwab at suse dot de>
- To: gcc at gcc dot gnu dot org
- Date: 30 Dec 2001 23:10:53 +0100
- Subject: Stage1 compiler miscompiles Ada on ia64
The stage1 compiler miscompiles the ada compiler on ia64 with -O2. The
symptoms are:
a-colire.adb:75:23: warning: loop range is null, loop will not execute
a-colire.adb:82:26: warning: loop range is null, loop will not execute
a-colire.adb:106:24: warning: loop range is null, loop will not execute
a-colire.adb:114:26: warning: loop range is null, loop will not execute
One of the miscompiled functions is Checks.Determine_Range, the damage
happens during flow2. The epilogue is clobbering register r9 which is
also used to return Hi, so that Sem_Ch5.Analyze_Iteration_Scheme is
receiving garbage. Here are the relevant parts of checks.adb.23.flow2:
(insn 3619 3618 2549 (set (reg:DI 9 r9)
(mem:DI (reg:DI 18 r18) [67 S8 A64])) 6 {*movdi_internal} (insn_list 3617 (nil))
(expr_list:REG_DEAD (reg:DI 18 r18)
(nil)))
(insn 2549 3619 3701 (use (reg/i:TI 8 r8)) -1 (insn_list 3618 (nil))
(expr_list:REG_DEAD (reg:DI 9 r9)
(nil)))
(note 3701 2549 3704 NOTE_INSN_EPILOGUE_BEG)
[...]
(insn 3730 3728 3732 (set (reg:DI 9 r9)
(mem:DI (post_modify:DI (reg:DI 28 r28)
(plus:DI (reg:DI 28 r28)
(const_int 16 [0x10]))) [68 S8 A64])) -1 (insn_list 3726 (nil))
(expr_list:REG_INC (reg:DI 28 r28)
(nil)))
(insn 3732 3730 3734 (set (reg:DI 320 b0)
(reg:DI 9 r9)) -1 (insn_list 3730 (nil))
(nil))
The basic problem seems to be that insn 2549 marks register r9 as dead.
Andreas.
--
Andreas Schwab "And now for something
Andreas.Schwab@suse.de completely different."
SuSE Labs, SuSE GmbH, Schanzäckerstr. 10, D-90443 Nürnberg
Key fingerprint = 58CA 54C7 6D53 942B 1756 01D3 44D5 214B 8276 4ED5