[Bug target/59289] New: [ARM] regression on unsigned-extend-2.c

[Bug target/59289] New: [ARM] regression on unsigned-extend-2.c

[christophe.lyon at st dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59289

            Bug ID: 59289
           Summary: [ARM] regression on unsigned-extend-2.c
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: christophe.lyon at st dot com

Since commit 203160 (New rtx costs infrastructure for ARM), I have noticed that
gcc.target/arm/unsigned-extend-2.c scan-assembler ands
gcc.target/arm/unsigned-extend-2.c scan-assembler-not cmp
now FAIL (used to PASS).

I have configured GCC as:
target: arm-none-linux-gnueabihf
mode: thumb
cpu: cortex-a15
fpu: neon-vfpv4

With r203159:
    movs    r3, #8
.L3:
    lsrs    r0, r0, #1
    subs    r3, r3, #1
    ands    r3, r3, #255
    bne    .L3
    bx    lr

with r203160:
    movs    r3, #0
.L3:
    lsrs    r0, r0, #1
    adds    r3, r3, #1
    cmp    r3, #8
    bne    .L3
    bx    lr

It seems that code gen has slightly evolved in trunk since r203160, but the
test still fails.

The code sequence does not seem really worse than the original one, but the
testcase should at least be updated.

[Bug target/59289] [4.9 Regression][ARM] regression on unsigned-extend-2.c

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59289

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Version|unknown                     |4.9.0
   Target Milestone|---                         |4.9.0
            Summary|[ARM] regression on         |[4.9 Regression][ARM]
                   |unsigned-extend-2.c         |regression on
                   |                            |unsigned-extend-2.c

[Bug target/59289] [ARM] regression on unsigned-extend-2.c

[ktkachov at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59289

ktkachov at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|arm                         |arm-*-*
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2013-11-27
      Known to work|                            |4.8.2
            Version|4.9.0                       |unknown
           Assignee|unassigned at gcc dot gnu.org      |ktkachov at gcc dot gnu.org
            Summary|[4.9 Regression][ARM]       |[ARM] regression on
                   |regression on               |unsigned-extend-2.c
                   |unsigned-extend-2.c         |
     Ever confirmed|0                           |1
      Known to fail|                            |4.9.0

--- Comment #1 from ktkachov at gcc dot gnu.org ---
Looking a bit into it, it seems that I wrote the A15 costs without considering
that they're supposed to represent the latency cost minus 1 (since every insn
has an implicit cost of COSTS_N_INSNS (1) already). Those were the early days
of the new costs tables. If I adjust the costs for the A15 for this, it gives
the old sequence again.

Regardless of whether this new sequence is good or not, the A15 costs should be
fixed. I have a patch in testing

[Bug target/59289] [ARM] regression on unsigned-extend-2.c

[ktkachov at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59289

--- Comment #2 from ktkachov at gcc dot gnu.org ---
Author: ktkachov
Date: Fri Nov 29 15:19:34 2013
New Revision: 205529

URL: http://gcc.gnu.org/viewcvs?rev=205529&root=gcc&view=rev
Log:
2013-11-29  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>

    PR target/59289
    * config/arm/arm.c (cortexa15_extra_costs): Adjust costs.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/arm/arm.c

[Bug target/59289] [ARM] regression on unsigned-extend-2.c

[ktkachov at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59289

ktkachov at gcc dot gnu.org changed:

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

--- Comment #3 from ktkachov at gcc dot gnu.org ---
Fixed on trunk.