[Bug rtl-optimization/65123] New: lra remat places insn which breaks data flow

[Bug rtl-optimization/65123] New: lra remat places insn which breaks data flow

[krebbel at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65123

            Bug ID: 65123
           Summary: lra remat places insn which breaks data flow
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: rtl-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: krebbel at gcc dot gnu.org

The vector-compare-1.c testcase fails on S/390 when compiled with -march=z196

Steps to reproduce:
make check-gcc RUNTESTFLAGS="--target_board='unix/-m31/-march=z196'
dg-torture.exp=vector-compare-1.c"

FAIL: c-c++-common/torture/vector-compare-1.c   -O3 -fomit-frame-pointer 
execution test
FAIL: c-c++-common/torture/vector-compare-1.c   -O3 -fomit-frame-pointer
-funroll-loops  execution test
FAIL: c-c++-common/torture/vector-compare-1.c   -O3 -fomit-frame-pointer
-funroll-all-loops -finline-functions  execution test
FAIL: c-c++-common/torture/vector-compare-1.c   -O3 -g  execution test

lra-remat places an insn (14287) which reads and clobbers the condition code
hard register 33 right after (insn 1330) each other making the second depend on
the clobbered cc.

IRA:

;; basic block 134, loop depth 0, count 0, freq 619, maybe hot
;;  prev block 133, next block 135, flags: (RTL)
;;  pred:       133 [100.0%]  (FALLTHRU)
;; bb 134 artificial_defs: { }
;; bb 134 artificial_uses: { u1044(11){ }u1045(15){ }u1046(32){ }u1047(34){ }}
;; lr  in       11 [%r11] 14 [%r14] 15 [%r15] 32 [%ap] 34 [%fp] 109 110 2132
2134 2268 2311 6202 6445 6446 6447 6448 6449 6450
;; lr  use      11 [%r11] 15 [%r15] 32 [%ap] 34 [%fp] 6202 6446 6449
;; lr  def      33 [%cc] 116 2317 2318
(note 1327 1326 1329 134 [bb 134] NOTE_INSN_BASIC_BLOCK)
(insn 1329 1327 1330 134 (parallel [
            (set (reg:CCL3 33 %cc)
                (compare:CCL3 (reg:SI 6446 [ D.2538 ])
                    (reg:SI 6449 [ D.2538 ])))
            (clobber (scratch:SI))
        ])
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 380
{*subsi3_cconly2}
     (nil))
(insn 1330 1329 1328 134 (parallel [
            (set (reg:SI 2317 [ D.2534+-3 ])
                (geu:SI (reg:CCL3 33 %cc)
                    (const_int 0 [0])))
            (clobber (reg:CC 33 %cc))
        ])
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 413
{*scondsi}
     (expr_list:REG_DEAD (reg:CCL3 33 %cc)
        (expr_list:REG_UNUSED (reg:CC 33 %cc)
            (nil))))
(insn 1328 1330 1331 134 (set (reg:SI 116 [ D.2533 ])
        (const_int -1 [0xffffffffffffffff]))
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 68
{*movsi_esa}
     (nil))
(insn 1331 1328 1332 134 (parallel [
            (set (reg:SI 2318 [ D.2533 ])
                (neg:SI (reg:SI 2317 [ D.2534+-3 ])))
            (clobber (reg:CC 33 %cc))
        ])
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 490
{*negsi2}
     (expr_list:REG_DEAD (reg:SI 2317 [ D.2534+-3 ])
        (expr_list:REG_UNUSED (reg:CC 33 %cc)
            (nil))))
(insn 1332 1331 1333 134 (set (reg:CCZ 33 %cc)
        (compare:CCZ (reg:SI 6202 [ D.2537+4 ])
            (reg:SI 2318 [ D.2533 ])))
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 26
{*cmpsi_cct}
     (expr_list:REG_DEAD (reg:SI 2318 [ D.2533 ])
        (expr_list:REG_EQUAL (compare:CCZ (const_int -1 [0xffffffffffffffff])
                (reg:SI 2318 [ D.2533 ]))
            (nil))))
(jump_insn 1333 1332 1334 134 (set (pc)
        (if_then_else (ne (reg:CCZ 33 %cc)
                (const_int 0 [0]))
            (label_ref:SI 9319)
            (pc)))
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 598
{*cjump_64}
     (expr_list:REG_DEAD (reg:CCZ 33 %cc)
        (int_list:REG_BR_PROB 4 (nil)))
 -> 9319)



RELOAD:
;; basic block 134, loop depth 0, count 0, freq 619, maybe hot
;;  prev block 133, next block 135, flags: (REACHABLE, RTL, MODIFIED)
;;  pred:       133 [100.0%]  (FALLTHRU)
;; bb 134 artificial_defs: { }
;; bb 134 artificial_uses: { u-1(11){ }u-1(15){ }}
;; lr  in       0 [%r0] 1 [%r1] 3 [%r3] 4 [%r4] 5 [%r5] 6 [%r6] 7 [%r7] 8 [%r8]
9 [%r9] 10 [%r10] 11 [%r11] 12 [%r12] 14 [%r14] 15 [%r15] 34 [%fp]
;; lr  use      6 [%r6] 9 [%r9] 10 [%r10] 11 [%r11] 15 [%r15]
;; lr  def      2 [%r2] 33 [%cc]
;; live  in       0 [%r0] 1 [%r1] 3 [%r3] 4 [%r4] 5 [%r5] 6 [%r6] 7 [%r7] 8
[%r8] 9 [%r9] 10 [%r10] 11 [%r11] 12 [%r12] 14 [%r14] 15 [%r15] 34 [%fp]
;; live  gen      2 [%r2] 33 [%cc]
;; live  kill     2 [%r2] 33 [%cc]
(note 1327 1326 12665 134 [bb 134] NOTE_INSN_BASIC_BLOCK)
(insn 12665 1327 1329 134 (set (reg:SI 2 %r2 [6924])
        (reg:SI 9 %r9 [orig:6446 D.2538 ] [6446]))
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 68
{*movsi_esa}
     (nil))
(insn 1329 12665 1330 134 (parallel [
            (set (reg:CCL3 33 %cc)
                (compare:CCL3 (reg:SI 2 %r2 [6924])
                    (reg:SI 10 %r10 [orig:6449 D.2538 ] [6449])))
            (clobber (reg:SI 2 %r2 [6924]))
        ])
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 380
{*subsi3_cconly2}
     (nil))
(insn 1330 1329 14273 134 (parallel [
            (set (reg:SI 2 %r2 [orig:2317 D.2534+-3 ] [2317])
                (geu:SI (reg:CCL3 33 %cc)
                    (const_int 0 [0])))
            (clobber (reg:CC 33 %cc))
        ])
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 413
{*scondsi}
     (nil))
(note 14273 1330 14280 134 NOTE_INSN_DELETED)
(note 14280 14273 14117 134 NOTE_INSN_DELETED)
(note 14117 14280 1328 134 NOTE_INSN_DELETED)
(insn 1328 14117 14286 134 (set (reg:SI 2 %r2 [orig:116 D.2533 ] [116])
        (const_int -1 [0xffffffffffffffff]))
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 68
{*movsi_esa}
     (nil))
(insn 14286 1328 14287 134 (set (mem/c:SI (plus:SI (reg/f:SI 11 %r11)
                (const_int 464 [0x1d0])) [6 %sfp+-40 S4 A32])
        (reg:SI 2 %r2 [orig:116 D.2533 ] [116]))
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 68
{*movsi_esa}
     (nil))
(insn 14287 14286 14272 134 (parallel [
            (set (reg:SI 2 %r2 [orig:2317 D.2534+-3 ] [2317])
                (geu:SI (reg:CCL3 33 %cc)
                    (const_int 0 [0])))
            (clobber (reg:CC 33 %cc))
        ])
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 413
{*scondsi}
     (nil))
(note 14272 14287 1331 134 NOTE_INSN_DELETED)
(insn 1331 14272 14271 134 (parallel [
            (set (reg:SI 2 %r2 [orig:2318 D.2533 ] [2318])
                (neg:SI (reg:SI 2 %r2 [orig:2317 D.2534+-3 ] [2317])))
            (clobber (reg:CC 33 %cc))
        ])
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 490
{*negsi2}
     (nil))
(insn 14271 1331 14279 134 (set (reg:SI 2 %r2 [orig:2318 D.2533 ] [2318])
        (reg:SI 2 %r2 [orig:2318 D.2533 ] [2318]))
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 68
{*movsi_esa}
     (nil))
(note 14279 14271 14116 134 NOTE_INSN_DELETED)
(note 14116 14279 1332 134 NOTE_INSN_DELETED)
(insn 1332 14116 1333 134 (set (reg:CCZ 33 %cc)
        (compare:CCZ (reg:SI 6 %r6 [orig:6202 D.2537+4 ] [6202])
            (reg:SI 2 %r2 [orig:2318 D.2533 ] [2318])))
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 26
{*cmpsi_cct}
     (expr_list:REG_EQUAL (compare:CCZ (const_int -1 [0xffffffffffffffff])
            (reg:SI 2 %r2 [orig:2318 D.2533 ] [2318]))
        (nil)))
(jump_insn 1333 1332 1334 134 (set (pc)
        (if_then_else (ne (reg:CCZ 33 %cc)
                (const_int 0 [0]))
            (label_ref:SI 9319)
            (pc)))
/home/andreas/gcc/gcc/testsuite/c-c++-common/torture/vector-compare-1.c:54 598
{*cjump_64}
     (int_list:REG_BR_PROB 4 (nil))
 -> 9319)

[Bug rtl-optimization/65123] [5 regression] lra remat places insn which breaks data flow

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65123

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ra, wrong-code
           Priority|P3                          |P1
   Target Milestone|---                         |5.0
            Summary|[5.0 regression] lra remat  |[5 regression] lra remat
                   |places insn which breaks    |places insn which breaks
                   |data flow                   |data flow

[Bug rtl-optimization/65123] [5 regression] lra remat places insn which breaks data flow

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65123

Vladimir Makarov <vmakarov at gcc dot gnu.org> changed:

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

--- Comment #1 from Vladimir Makarov <vmakarov at gcc dot gnu.org> ---
Andreas, thanks for the analysis.  The rematerialized insn should be not a
candidate.  A check exists for this in la-remat.c but still it happens.  I'll
investigate this more on Monday and definitely fix it on the next week as this
is a serious problem.

[Bug rtl-optimization/65123] [5 regression] lra remat places insn which breaks data flow

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65123

--- Comment #2 from Vladimir Makarov <vmakarov at gcc dot gnu.org> ---
Author: vmakarov
Date: Tue Feb 24 18:11:38 2015
New Revision: 220946

URL: https://gcc.gnu.org/viewcvs?rev=220946&root=gcc&view=rev
Log:
2015-02-24  Vladimir Makarov  <vmakarov@redhat.com>

    PR rtl-optimization/65123
    * lra-remat.c (operand_to_remat): Check hard regs in insn
    definition too.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/lra-remat.c

[Bug rtl-optimization/65123] [5 regression] lra remat places insn which breaks data flow

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65123

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
                 CC|                            |jakub at gcc dot gnu.org
         Resolution|---                         |FIXED

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed, thanks.