[Bug middle-end/53823] New: [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[Bug middle-end/53823] New: [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

             Bug #: 53823
           Summary: [4.8 Regression] FAIL:
                    gcc.c-torture/execute/930921-1.c execution at -O0 and
                    -O1
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: danglin@gcc.gnu.org
              Host: hppa*-*-* (32-bit)
            Target: hppa*-*-* (32-bit)
             Build: hppa*-*-* (32-bit)


Executing on host: /home/dave/gnu/gcc/objdir/gcc/xgcc
-B/home/dave/gnu/gcc/objdir/gcc/
/home/dave/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/930921-1.c 
-fno-diagnostics-show-caret  -w  -O0   -lm   -o
/home/dave/gnu/gcc/objdir/gcc/testsuite/gcc/930921-1.x0    (timeout = 300)
spawn /home/dave/gnu/gcc/objdir/gcc/xgcc -B/home/dave/gnu/gcc/objdir/gcc/
/home/dave/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/930921-1.c
-fno-diagnostics-show-caret -w -O0 -lm -o
/home/dave/gnu/gcc/objdir/gcc/testsuite/gcc/930921-1.x0
PASS: gcc.c-torture/execute/930921-1.c compilation,  -O0 
Setting LD_LIBRARY_PATH to
:/home/dave/gnu/gcc/objdir/gcc::/home/dave/gnu/gcc/objdir/gcc:/home/dave/gnu/gcc/objdir/hppa-linux-gnu/libstdc++-v3/src/.libs:/home/dave/gnu/gcc/objdir/hppa-linux-gnu/libmudflap/.libs:/home/dave/gnu/gcc/objdir/hppa-linux-gnu/libssp/.libs:/home/dave/gnu/gcc/objdir/hppa-linux-gnu/libgomp/.libs:/home/dave/gnu/gcc/objdir/hppa-linux-gnu/libatomic/.libs:/home/dave/gnu/gcc/objdir/./gcc:/home/dave/gnu/gcc/objdir/./prev-gcc
spawn [open ...]
FAIL: gcc.c-torture/execute/930921-1.c execution,  -O0 

dave@mx3210:~/gnu/gcc/objdir/gcc$ ./xgcc -B./ -v
Reading specs from ./specs
COLLECT_GCC=./xgcc
COLLECT_LTO_WRAPPER=./lto-wrapper
Target: hppa-linux-gnu
Configured with: ../gcc/configure --with-gnu-as --with-gnu-ld --enable-shared
--enable-multiarch --with-multiarch-defaults=hppa-linux-gnu
--enable-linker-build-id --build=hppa-linux-gnu --host=hppa-linux-gnu
--target=hppa-linux-gnu --prefix=/home/dave/opt/gnu/gcc/gcc-4.7.0
--with-local-prefix=/home/dave/opt/gnu --enable-threads=posix
--enable-__cxa_atexit --build=hppa-linux-gnu --enable-clocale=gnu
--enable-java-gc=boehm
--enable-languages=c,c++,objc,fortran,obj-c++,java,ada,lto
Thread model: posix
gcc version 4.8.0 20120630 (experimental) [trunk revision 189094] (GCC)

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.8.0

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

John David Anglin <danglin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rth at gcc dot gnu.org

--- Comment #1 from John David Anglin <danglin at gcc dot gnu.org> 2012-07-11 10:50:57 UTC ---
Introduced in revision 188786.

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #2 from John David Anglin <danglin at gcc dot gnu.org> 2012-07-12 01:43:11 UTC ---
This hunk of RTL was generated in .expand for function f:

;; D.1356_3 = D.1355_2 * 2863311531;
(insn 9 8 10 (set (reg:SI 104)
        (subreg:SI (reg:DI 96 [ D.1355 ]) 4))
/test/gnu/gcc/gcc/gcc/testsuite/gc
c.c-torture/execute/930921-1.c:4 -1
     (nil))

(insn 10 9 11 (set (subreg:SI (reg:DI 97 [ D.1356 ]) 0)
        (ashift:SI (reg:SI 104)
            (const_int 31 [0x1f])))
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/930921-1.c:4 -1
     (expr_list:REG_EQUAL (ashift:SI (subreg:SI (reg:DI 96 [ D.1355 ]) 4)
            (const_int 31 [0x1f]))
        (nil)))

(insn 11 10 0 (set (subreg:SI (reg:DI 97 [ D.1356 ]) 4)
        (const_int 0 [0]))
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/930921-1.c:4 -1
     (nil))

Looks wrong to me.  Fails when x=1.

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #3 from John David Anglin <danglin at gcc dot gnu.org> 2012-07-27 18:19:26 UTC ---
*** Bug 53974 has been marked as a duplicate of this bug. ***

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #4 from John David Anglin <danglin at gcc dot gnu.org> 2012-07-29 16:31:22 UTC ---
Created attachment 27888
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27888
Patch

Fixes 930921-1.c and four original acats fails, but not
additional acats fails introduced in 189441.

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[rth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

Richard Henderson <rth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-07-30
     Ever Confirmed|0                           |1

--- Comment #5 from Richard Henderson <rth at gcc dot gnu.org> 2012-07-30 13:31:15 UTC ---
That patch is obviously correct.

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[steven at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

Steven Bosscher <steven at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |uros at gcc dot gnu.org

--- Comment #6 from Steven Bosscher <steven at gcc dot gnu.org> 2012-07-30 14:13:04 UTC ---
And this is why "uninitialized" warnings shouldn't be silenced like this...

    * expmed.c (expand_mult): Initialize coeff and is_neg.

http://gcc.gnu.org/viewcvs?view=revision&revision=189281

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #7 from Uros Bizjak <ubizjak at gmail dot com> 2012-07-30 14:25:32 UTC ---
(In reply to comment #6)
> And this is why "uninitialized" warnings shouldn't be silenced like this...
> 
>     * expmed.c (expand_mult): Initialize coeff and is_neg.
> 
> http://gcc.gnu.org/viewcvs?view=revision&revision=189281

Sure, considering that following is way better:

# These files are to have specific diagnostics suppressed, or are not to
# be subject to -Werror:
# flex output may yield harmless "no previous prototype" warnings
build/gengtype-lex.o-warn = -Wno-error
gengtype-lex.o-warn = -Wno-error
expmed.o-warn = -Wno-error

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #8 from John David Anglin <danglin at gcc dot gnu.org> 2012-07-30 15:46:13 UTC ---
Author: danglin
Date: Mon Jul 30 15:46:08 2012
New Revision: 189980

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=189980
Log:
    PR middle-end/53823
    * expmed.c (expand_mult): Skip synth_mult for constant double op1 except
    for special cases.  Don't initialize coeff and is_neg.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/expmed.c

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #9 from John David Anglin <danglin at gcc dot gnu.org> 2012-07-31 18:51:26 UTC ---
Regarding the remaining acats fails, the following line in a-calend.adb
appears to be miscompiled:

            Date_Dur := Date_Dur - Time_Dur (Four_Year_Segs) *
                                   Secs_In_Four_Years;

The resulting value for Date_Dur is wrong.

The generated code appears to use synth_mult.

=> 0x00019c28 <__gnat_split+480>:    cmpib,>= 0,ret1,0x19cd0 <__gnat_split+648>
   0x00019c2c <__gnat_split+484>:    copy ret1,r20
   0x00019c30 <__gnat_split+488>:    ldi 0,r21
   0x00019c34 <__gnat_split+492>:    ldi -1,r22
   0x00019c38 <__gnat_split+496>:    and r19,r21,ret0
   0x00019c3c <__gnat_split+500>:    and r20,r22,ret1
   0x00019c40 <__gnat_split+504>:    extrw,u ret1,4,5,r19
   0x00019c44 <__gnat_split+508>:    depw,z ret0,26,27,r7
   0x00019c48 <__gnat_split+512>:    depw,z ret1,26,27,r8
   0x00019c4c <__gnat_split+516>:    or r19,r7,r7
   0x00019c50 <__gnat_split+520>:    sub r8,ret1,r12
   0x00019c54 <__gnat_split+524>:    sub,b r7,ret0,r11
   0x00019c58 <__gnat_split+528>:    extrw,u r12,8,9,r31
   0x00019c5c <__gnat_split+532>:    depw,z r12,22,23,r26
   0x00019c60 <__gnat_split+536>:    depw,z r11,22,23,r25
   0x00019c64 <__gnat_split+540>:    or r31,r25,r25
   0x00019c68 <__gnat_split+544>:    add r26,r12,r8
   0x00019c6c <__gnat_split+548>:    add,c r25,r11,r7
   0x00019c70 <__gnat_split+552>:    extrw,u r8,0,1,r31
   0x00019c74 <__gnat_split+556>:    depw,z r8,30,31,r24
   0x00019c78 <__gnat_split+560>:    depw,z r7,30,31,r23
   0x00019c7c <__gnat_split+564>:    or r31,r23,r23
   0x00019c80 <__gnat_split+568>:    sub r24,ret1,r26
   0x00019c84 <__gnat_split+572>:    sub,b r23,ret0,r25
   0x00019c88 <__gnat_split+576>:    extrw,u r26,3,4,r31
   0x00019c8c <__gnat_split+580>:    depw,z r25,27,28,r21
   0x00019c90 <__gnat_split+584>:    or r31,r21,r21
   0x00019c94 <__gnat_split+588>:    depw,z r26,27,28,r22
   0x00019c98 <__gnat_split+592>:    sub r22,ret1,r24
   0x00019c9c <__gnat_split+596>:    sub,b r21,ret0,r23
   0x00019ca0 <__gnat_split+600>:    extrw,u r24,5,6,r21
   0x00019ca4 <__gnat_split+604>:    depw,z r23,25,26,r19
   0x00019ca8 <__gnat_split+608>:    depw,z r24,25,26,r20
   0x00019cac <__gnat_split+612>:    or r21,r19,r19
   0x00019cb0 <__gnat_split+616>:    add ret1,r20,r22
   0x00019cb4 <__gnat_split+620>:    add,c ret0,r19,r21
   0x00019cb8 <__gnat_split+624>:    shrpw r21,r22,25,r19
   0x00019cbc <__gnat_split+628>:    depw,z r22,24,25,ret0
   0x00019cc0 <__gnat_split+632>:    copy r19,r21
   0x00019cc4 <__gnat_split+636>:    copy ret0,r22
   0x00019cc8 <__gnat_split+640>:    add r22,r4,r8
   0x00019ccc <__gnat_split+644>:    add,c r21,r3,r7
   0x00019cd0 <__gnat_split+648>:    copy r7,r25
   0x00019cd4 <__gnat_split+652>:    copy r8,r26
   0x00019cd8 <__gnat_split+656>:    ldi 0,r23
   0x00019cdc <__gnat_split+660>:    ldil L%1e13000,r24
   0x00019ce0 <__gnat_split+664>:    ldo 380(r24),r24
   0x00019ce4 <__gnat_split+668>:    b,l 0x34288 <__divdi3>,rp
   0x00019ce8 <__gnat_split+672>:    nop

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #10 from John David Anglin <danglin at gcc dot gnu.org> 2012-07-31 19:41:20 UTC ---
Created attachment 27913
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27913
Testcase

Testcase produces wrong output when compiled at -O0 and -O1.

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[rth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

Richard Henderson <rth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
         AssignedTo|unassigned at gcc dot       |rth at gcc dot gnu.org
                   |gnu.org                     |

--- Comment #11 from Richard Henderson <rth at gcc dot gnu.org> 2012-07-31 19:58:49 UTC ---
Mine.

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[dave.anglin at bell dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #12 from dave.anglin at bell dot net 2012-07-31 20:10:55 UTC ---
> Mine.


Expected result for testcase is 112975202 (0x6bbdd62).  Miscompiled  
result is
116077092194 (0x1b06bbdd62).

--
John David Anglin    dave.anglin@bell.net

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[rth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #13 from Richard Henderson <rth at gcc dot gnu.org> 2012-07-31 21:29:52 UTC ---
Created attachment 27914
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27914
Cross-compile output

Hmm.  I can't seem to reproduce this via a cross-compiler.

I tried CC='gcc -m32' of hppa-linux, so as to get the same
behavior wrt HOST_BITS_PER_WIDE_INT as your native bootstrap.
I compared the output on the testcase with mainline and 4.7
and they come out identical.

How does this -O1 output compare with native?

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[dave.anglin at bell dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #14 from dave.anglin at bell dot net 2012-07-31 22:54:59 UTC ---
On 31-Jul-12, at 5:29 PM, rth at gcc dot gnu.org wrote:

> How does this -O1 output compare with native?


There seems to an off by one error in various shifts, etc.   For  
example,

;(insn 8 4 10 (set (reg:SI 19 %r19 [104])
;        (lshiftrt:SI (reg:SI 24 %r24 [orig:100 y+4 ] [100])
;            (const_int 27 [0x1b]))) xxx.c:5 179 {lshrsi3}
;     (nil))
         extru %r24,4,5,%r19     ; 8     lshrsi3/2       [length = 4]

With 4.6, I see something similar to what you see:

;(insn 8 62 10 (set (reg:SI 20 %r20 [104])
;        (lshiftrt:SI (reg:SI 24 %r24 [orig:136+4 ] [136])
;            (const_int 28 [0x1c]))) xxx.c:5 178 {lshrsi3}
;     (nil))
         extru %r24,3,4,%r20     ; 8     lshrsi3/2       [length = 4]

--
John David Anglin    dave.anglin@bell.net

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[dave.anglin at bell dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #15 from dave.anglin at bell dot net 2012-07-31 22:55:54 UTC ---
Created attachment 27916
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27916
zz.s.txt

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

John David Anglin <danglin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |NEW

--- Comment #16 from John David Anglin <danglin at gcc dot gnu.org> 2012-08-01 00:30:40 UTC ---
The first shift amount seems to come from this statement
in expand_mult_const:

     int log = alg->log[opno];

The algorithm choice seems cost related, so it's starting
to make a bit of sense that this broke in 189441.

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #17 from John David Anglin <danglin at gcc dot gnu.org> 2012-08-01 00:40:24 UTC ---
(gdb) p *(const struct algorithm *)$r17
$21 = {cost = {cost = 24, latency = 24}, ops = 7, op = {alg_m, alg_sub_t2_m, 
    alg_add_factor, alg_sub_t2_m, alg_sub_t2_m, alg_add_t2_m, alg_shift, 
    2622983, alg_unknown, 28, alg_unknown, alg_unknown, 4226821568, 9510444, 
    786444, alg_unknown, 4226821456, alg_add_t_m2, alg_m, alg_add_t_m2, 
    4226822048, 10689160, alg_add_t_m2, 2748287, 10691300, 4226821664, 32, 
    alg_sub_factor, 1079796408, alg_unknown, 24, alg_unknown, 1079821632, 
    alg_unknown, alg_unknown, alg_unknown, alg_unknown, 1080629840, 
    1080642416, alg_unknown, 4226822232, alg_unknown, 1080629840, 1080642416, 
    alg_unknown, 1080672448, 1080128768, 1080642600, 378, 9959428, 
    alg_unknown, alg_zero, 1079744120, alg_zero, 3709487, 2873783, 27157688, 
    alg_unknown, 9704808, alg_unknown, 1079796408, alg_unknown, alg_unknown, 
    alg_unknown}, 
  log =
"\000\005\t\001\004\006\a0\000\000\000\b\000\227\370\004\373\360\060X\000\000\000\000@i\032P@iKp\000\000\000\000\000\247\332\370@\\\305@\000\252>\270@iL\030@iL\030@au\000@iLH"}

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[rth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #18 from Richard Henderson <rth at gcc dot gnu.org> 2012-08-01 02:25:03 UTC ---
But this is what you get when cross-compiling from i686.

3267     rtx temp = expand_mult_const (mode, op0, -coeff, NULL_RTX,
(gdb) p algorithm 
$1 = {cost = {cost = 24, latency = 24}, ops = 7, op = {alg_m, alg_add_t2_m, 
    alg_add_t2_m, alg_sub_t2_m, alg_sub_factor, alg_sub_factor, alg_shift, 
    4155963284, alg_unknown, 24, 4155924580, 144092293, 24, alg_unknown, 
    4294942376, 140383036, 4156006400, 167772184, 4294942392, 144092338, 
    148964452, 24, 4294942408, 140385161, 4156006400, 1180628613, 4155438304, 
    4155963284, 4155963284, 4155438304, 4155896864, 4155896864, alg_unknown, 
    alg_unknown, alg_unknown, alg_unknown, alg_unknown, 4155967104, 
    4155543416, 140385398, alg_zero, 167772160, 4294942616, 149524616, 
    alg_unknown, alg_unknown, 4294942552, 137729793, 4155963284, alg_unknown, 
    alg_unknown, alg_unknown, alg_unknown, 4155963284, 4155896864, 4155896864, 
    4155896864, 4155896864, 4155438304, 4155963284, 4155963284, 4155963284, 
    4155963284, 4155543416}, 
  log =
"\000\004\003\005\004\004\a\367\210\237\377\377\273\227\063\b\224\367\266\367",
'\000' <repeats 16 times>,
"\001\000\000\000\001\000\000\000\000\000\000\000\210\220\351\b\000\000\000\000ȟ\377\377\342y5\b"}

(We really ought to zero the structure first just to help debug.
But log[0] and op/log past [ops] are unused and uninitialized.)

The cross-compile *ought* not to affect costs, which means that
we ought to be making the same algorithm choices.  Which suggests
that -- if this is a fully bootstrapped pa compiler -- you're
looking at some part of expand_mult itself being mis-compiled.

I'm going to start a build on gcc61, but even checking out the
tree there has taken all day long...

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[dave.anglin at bell dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #19 from dave.anglin at bell dot net 2012-08-01 14:20:49 UTC ---
On 31-Jul-12, at 10:25 PM, rth at gcc dot gnu.org wrote:

> The cross-compile *ought* not to affect costs, which means that
> we ought to be making the same algorithm choices.  Which suggests
> that -- if this is a fully bootstrapped pa compiler -- you're
> looking at some part of expand_mult itself being mis-compiled.


I was afraid this might be the case.  However, I built a non bootstrap
compiler this morning with -g -O1.  The testcase still fails at -O0 and
O1, and works at -O2.

The compiler was configured as follows:

dave@mx3210:~/gnu/gcc/objdir/gcc$ ./xgcc -B./ -v
Reading specs from ./specs
COLLECT_GCC=./xgcc
COLLECT_LTO_WRAPPER=./lto-wrapper
Target: hppa-linux-gnu
Configured with: ../gcc/configure --with-gnu-as --with-gnu-ld --enable- 
shared --enable-multiarch --with-multiarch-defaults=hppa-linux-gnu -- 
enable-linker-build-id --build=hppa-linux-gnu --host=hppa-linux-gnu -- 
target=hppa-linux-gnu --prefix=/home/dave/opt/gnu/gcc/gcc-4.7.0 --with- 
local-prefix=/home/dave/opt/gnu --enable-threads=posix --enable- 
__cxa_atexit --build=hppa-linux-gnu --enable-clocale=gnu --enable-java- 
gc=boehm --enable-languages=c --disable-bootstrap
Thread model: posix
gcc version 4.8.0 20120801 (experimental) [trunk revision 190037] (GCC)

You probably would need to add --with-arch=1.1 to duplicate the  
default native settings with a cross.

The difference in extracts and deposits may not be the problem.  The - 
O2 code appears to have the same extracts as the
-O1 code.  I'll see if I can find where the real difference arises.

--
John David Anglin    dave.anglin@bell.net

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[dave.anglin at bell dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #20 from dave.anglin at bell dot net 2012-08-01 14:27:30 UTC ---
On 1-Aug-12, at 10:20 AM, dave.anglin at bell dot net wrote:

> The difference in extracts and deposits may not be the problem.  The -
> O2 code appears to have the same extracts as the
> -O1 code.  I'll see if I can find where the real difference arises.


Doh, the correct result is inlined at -O2.  foo is still mis-compiled.

--
John David Anglin    dave.anglin@bell.net

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #21 from John David Anglin <danglin at gcc dot gnu.org> 2012-08-01 18:44:04 UTC ---
The issue is with the handling of negative constants.

In this bit of code,

          max_cost = (set_src_cost (gen_rtx_MULT (mode, fake_reg, op1), speed)
                      - neg_cost(speed, mode));
          if (max_cost > 0
              && choose_mult_variant (mode, -coeff, &algorithm,
                                      &variant, max_cost))

max_cost is computed to be 24 and choose_mult_variant returns 0.  This
causes the following hunk to be executed:

      max_cost = set_src_cost (gen_rtx_MULT (mode, fake_reg, op1), speed);
      if (choose_mult_variant (mode, coeff, &algorithm, &variant, max_cost))
        return expand_mult_const (mode, op0, coeff, target,
                                  &algorithm, variant);

max_cost is now 32 and choose_mult_variant succeeds.  However,
expand_mult_const
generates incorrect code when coeff is negative.

I hacked the max_cost in the negative case to be 32, and this produces
the correct result and code similar to your cross-compiled output.

If I remember correctly, the xmpyu instruction was introduced in PA 1.1
and this results in different costs for PA 1.0 and 1.1.

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #22 from John David Anglin <danglin at gcc dot gnu.org> 2012-08-12 14:30:12 UTC ---
Created attachment 27994
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27994
Patch

      /* synth_mult does an `unsigned int' multiply.  As long as the mode is
         less than or equal in size to `unsigned int' this doesn't matter.
         If the mode is larger than `unsigned int', then synth_mult works
         only if the constant value exactly fits in an `unsigned int' without
         any truncation.  This means that multiplying by negative values does
         not work; results are off by 2^32 on a 32 bit machine.  */

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[rth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #23 from Richard Henderson <rth at gcc dot gnu.org> 2012-08-13 15:51:37 UTC ---
On 08/12/2012 07:30 AM, danglin at gcc dot gnu.org wrote:
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823
> 
> --- Comment #22 from John David Anglin <danglin at gcc dot gnu.org> 2012-08-12 14:30:12 UTC ---
> Created attachment 27994
>   --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27994

Ok.


r~

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

--- Comment #24 from John David Anglin <danglin at gcc dot gnu.org> 2012-08-18 23:16:57 UTC ---
Author: danglin
Date: Sat Aug 18 23:16:53 2012
New Revision: 190505

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=190505
Log:
    PR middle-end/53823
    * expmed.c (expand_mult): Skip synth_mult for negative coefficients
    if the mode is larger than a wide int and it is too costly to multiply
    by a positive multiplier and negate the result.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/expmed.c

[Bug middle-end/53823] [4.8 Regression] FAIL: gcc.c-torture/execute/930921-1.c execution at -O0 and -O1

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53823

John David Anglin <danglin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED

--- Comment #25 from John David Anglin <danglin at gcc dot gnu.org> 2012-08-18 23:32:37 UTC ---
Although I've got another multiplication issue to resolve, it
was not introduced by the rearrangment of expand_mult.  So, I'm going
to start a new PR.