Bug 38208 - [4.4 Regression] gcc.c-torture/compile/20080806-1.c
Summary: [4.4 Regression] gcc.c-torture/compile/20080806-1.c
Status: RESOLVED FIXED
Alias: None
Product: gcc
Classification: Unclassified
Component: target (show other bugs)
Version: 4.4.0
: P3 normal
Target Milestone: 4.4.0
Assignee: Jakub Jelinek
URL: http://gcc.gnu.org/ml/gcc-patches/200...
Keywords:
Depends on:
Blocks:
 
Reported: 2008-11-21 05:18 UTC by H.J. Lu
Modified: 2008-11-23 18:57 UTC (History)
4 users (show)

See Also:
Host:
Target:
Build:
Known to work:
Known to fail:
Last reconfirmed:

Attachments
gcc44-pr38208.patch (518 bytes, patch)
2008-11-21 08:43 UTC, Jakub Jelinek 		Details | Diff
View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description H.J. Lu 2008-11-21 05:18:42 UTC 
On Linux/x86, revision 142072 gave:

Executing on host: /export/gnu/import/svn/gcc-test/bld/gcc/xgcc -B/export/gnu/import/svn/gcc-test/bld/gcc/   -O3 -fomit-frame-pointer -funroll-loops  -w -c  -m32 -o 20080806-1.o /export/gnu/import/svn/gcc-test/src-trunk/gcc/testsuite/gcc.c-torture/compile/20080806-1.c    (timeout = 300)
/tmp/ccuL0Iug.s: Assembler messages:^M
/tmp/ccuL0Iug.s:9: Warning: 65544 shortened to 8^M

+FAIL: gcc.c-torture/compile/20080806-1.c  -O3 -fomit-frame-pointer  (test for excess errors)
+FAIL: gcc.c-torture/compile/20080806-1.c  -O3 -fomit-frame-pointer -funroll-all-loops -finline-functions  (test for excess errors)
+FAIL: gcc.c-torture/compile/20080806-1.c  -O3 -fomit-frame-pointer -funroll-loops  (test for excess errors)

Revision 142058 is OK.
Comment 1 H.J. Lu 2008-11-21 05:22:20 UTC 
Revision 142061 is bad.
Comment 2 H.J. Lu 2008-11-21 05:39:01 UTC 
Revision 142061:

http://gcc.gnu.org/ml/gcc-patches/2008-11/msg01051.html

is the cause.
Comment 3 Jakub Jelinek 2008-11-21 08:09:45 UTC 
Just the cause that the optimized dump looks differently, but is not wrong.
The bug happens during reload's register elimination.

(insn:HI 6 5 7 2 20080806-1.c:16 (parallel [
            (set (reg:HI 58 [ pretmp.42 ])
                (plus:HI (subreg:HI (reg/f:SI 20 frame) 0)
                    (const_int 136 [0x88])))
            (clobber (reg:CC 17 flags))
        ]) 293 {*addhi_1_lea} (expr_list:REG_UNUSED (reg:CC 17 flags)
        (nil)))

is reloaded into:

(insn 47 5 6 2 20080806-1.c:16 (set (reg:HI 1 dx [orig:58 pretmp.42 ] [58])
        (reg:HI 7 sp)) 53 {*movhi_1} (nil))

(insn:HI 6 47 7 2 20080806-1.c:16 (parallel [
            (set (reg:HI 1 dx [orig:58 pretmp.42 ] [58])
                (plus:HI (reg:HI 1 dx [orig:58 pretmp.42 ] [58])
                    (const_int 65544 [0x10008])))
            (clobber (reg:CC 17 flags))
        ]) 293 {*addhi_1_lea} (expr_list:REG_EQUIV (plus:HI (subreg:HI (plus:SI (reg/f:SI 7 sp)
                    (const_int 65408 [0xff80])) 0)
            (const_int 136 [0x88]))
        (nil)))

eliminate_regs_in_insn calls plus_constant with
to_rtx (subreg:HI (reg/f:SI 7 sp) 0) and offset 0x10008.  I guess it is
eliminate_regs_in_insn and not plus_constant responsibility to trunc_int_for_mode this.
Comment 4 Jakub Jelinek 2008-11-21 08:43:45 UTC 
Created attachment 16731 [details]
gcc44-pr38208.patch

I'll test the attached patch.  Related to PR29599.
Comment 5 Jakub Jelinek 2008-11-21 13:48:30 UTC 
Subject: Bug 38208

Author: jakub
Date: Fri Nov 21 13:47:05 2008
New Revision: 142088

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=142088
Log:
	PR target/38208
	* reload1.c (eliminate_regs_in_insn): For trunc_int_for_mode use
	mode of PLUS, not mode of the eliminated register.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/reload1.c
Comment 6 H.J. Lu 2008-11-21 17:04:19 UTC 
Fixed.