24315 – [3.4 Regression] amd64 fails -fpeephole2 (original) (raw)
Description Doug Coleman 2005-10-11 18:57:55 UTC
native/s48_bignum.c:456: internal compiler error: in final_scan_insn, at final.c:2514 Please submit a full bug report,
I'm still looking for where to attach my .i file. Hopefully I can attach it after submitting this page. uname -a: Linux frogger 2.6.12-9-amd64-generic #1 Tue Oct 4 19:23:45 BST 2005 x86_64 GNU/Linux
gcc --version: gcc (GCC) 4.0.2 20050808 (prerelease) (Ubuntu 4.0.1-4ubuntu9)
Comment 2 Drea Pinski 2005-10-11 19:00:03 UTC
Oh, can you show the command line invokation of gcc?
Comment 3 Doug Coleman 2005-10-11 19:03:00 UTC
gcc -fpeephole2 s48_bignum.i
Thanks!
Comment 4 Drea Pinski 2005-10-11 19:05:25 UTC
(In reply to comment #2)
Oh, can you show the command line invokation of gcc? Never mind, -O0 -fpeephole2 makes it fail.
Comment 5 Drea Pinski 2005-10-11 19:28:46 UTC
Reduced testcase for 4.0.x and above: void s48_double_to_bignum(int exponent){ long length = ((((exponent) + ((((sizeof (long)) * 8) - 2) - 1)) / (((sizeof (long)) * 8) - 2))); }
This really should be filed in a different PR as that one is a regression.
I will try to get a reduced testcase for 3.4.x and 3.3.x later today.
Comment 6 Drea Pinski 2005-10-12 13:30:41 UTC
Here is the one for 3.3.3 and above: typedef long bignum_digit_type; void bignum_divide_unsigned_normalized( ){ bignum_digit_type guess; guess = ((((unsigned long) 1) << (((sizeof (bignum_digit_type)) * 8) - 2)) - 1); }
I will file the other one seperate now.
Comment 8 Drea Pinski 2005-10-12 13:42:21 UTC
Confirmed that this is a real regression.
Comment 9 Drea Pinski 2005-10-12 13:45:19 UTC
The instruction we are should have split earlier: (insn/v 8 21 26 (set (mem/c/i:DI (plus:DI (reg/f:DI 6 bp) (const_int -8 [0xfffffffffffffff8])) [0 guess+0 S8 A64]) (const_int 4611686018427387903 [0x3fffffffffffffff])) 81 {*movdi_1_rex64} (nil) (nil))
Comment 10 Drea Pinski 2005-10-12 13:46:49 UTC
*** Bug 24330 has been marked as a duplicate of this bug. ***
Comment 11 Drea Pinski 2005-10-12 13:48:42 UTC
It turned out the one in comment #5 is the same issue as the orginal bug too.
Comment 13 Uroš Bizjak 2005-10-17 07:21:18 UTC
Fixed on mainline.
The bugfix is really trivial to fix, but I don't have older branches installed here. Could somebody help me with a backports?
Subject: Bug 24315
Author: uros Date: Tue Nov 8 07:58:51 2005 New Revision: 106633
URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=106633 Log: PR target/19340 * reg-stack.c (reg_to_stack): Update register liveness also for flag_sched2_use_traces.
[PR target/24315](show%5Fbug.cgi?id=24315 "RESOLVED FIXED - [3.4 Regression] amd64 fails -fpeephole2")
* config/i386/i386.md (*pushdi2_rex64 splitter)
(*movdi_1_rex64 splitter, *ashldi3_1 splitter)
(*ashrdi3_1 splitter, *lshrdi3_1 splitter): Delay splitting after
flow2 pass only when (optimize > 0 && flag_peephole2).testsuite/
[PR target/19340](show%5Fbug.cgi?id=19340 "RESOLVED FIXED - Compilation SEGFAULTs with -O1 -fschedule-insns2 -fsched2-use-traces on an x86 architecture.")
* gcc.dg/[pr19340](show%5Fbug.cgi?id=19340 "RESOLVED FIXED - Compilation SEGFAULTs with -O1 -fschedule-insns2 -fsched2-use-traces on an x86 architecture.").c: New test.
[PR target/24315](show%5Fbug.cgi?id=24315 "RESOLVED FIXED - [3.4 Regression] amd64 fails -fpeephole2")
* gcc.target/i386/[pr24315](show%5Fbug.cgi?id=24315 "RESOLVED FIXED - [3.4 Regression] amd64 fails -fpeephole2").c: New test.Added: branches/gcc-4_0-branch/gcc/testsuite/gcc.dg/pr19340.c branches/gcc-4_0-branch/gcc/testsuite/gcc.target/i386/pr24315.c Modified: branches/gcc-4_0-branch/gcc/ChangeLog branches/gcc-4_0-branch/gcc/config/i386/i386.md branches/gcc-4_0-branch/gcc/reg-stack.c branches/gcc-4_0-branch/gcc/testsuite/ChangeLog
Comment 15 Paolo Bonzini 2005-11-08 08:19:45 UTC
now fixed on 4.0 branch too
Subject: Bug 24315
Author: uros Date: Thu Nov 10 07:27:47 2005 New Revision: 106728
URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=106728 Log: PR target/24315 * config/i386/i386.md (*pushdi2_rex64 splitter) (*movdi_1_rex64 splitter): Delay splitting after flow2 pass only when (optimize > 0 && flag_peephole2).
[PR target/19340](show%5Fbug.cgi?id=19340 "RESOLVED FIXED - Compilation SEGFAULTs with -O1 -fschedule-insns2 -fsched2-use-traces on an x86 architecture.")
* reg-stack.c (reg_to_stack): Update register liveness also
for flag_sched2_use_traces.testsuite/
[PR target/24315](show%5Fbug.cgi?id=24315 "RESOLVED FIXED - [3.4 Regression] amd64 fails -fpeephole2")
* gcc.dg/[pr24315](show%5Fbug.cgi?id=24315 "RESOLVED FIXED - [3.4 Regression] amd64 fails -fpeephole2").c: New test.
[PR target/19340](show%5Fbug.cgi?id=19340 "RESOLVED FIXED - Compilation SEGFAULTs with -O1 -fschedule-insns2 -fsched2-use-traces on an x86 architecture.")
* gcc.dg/[pr19340](show%5Fbug.cgi?id=19340 "RESOLVED FIXED - Compilation SEGFAULTs with -O1 -fschedule-insns2 -fsched2-use-traces on an x86 architecture.").c: New test.Added: branches/gcc-3_4-branch/gcc/testsuite/gcc.dg/pr19340.c - copied unchanged from r106632, trunk/gcc/testsuite/gcc.dg/pr19340.c branches/gcc-3_4-branch/gcc/testsuite/gcc.dg/pr24315.c Modified: branches/gcc-3_4-branch/gcc/ChangeLog branches/gcc-3_4-branch/gcc/config/i386/i386.md branches/gcc-3_4-branch/gcc/reg-stack.c branches/gcc-3_4-branch/gcc/testsuite/ChangeLog
Comment 17 Uroš Bizjak 2005-11-10 07:31:55 UTC
Fixed on 3.4 branch.