Bug 90913 - [10 Regression] ICE in maybe_gen_insn, at optabs.c:7341 since r272239
Summary: [10 Regression] ICE in maybe_gen_insn, at optabs.c:7341 since r272239
Alias: None
Product: gcc
Classification: Unclassified
Component: tree-optimization (show other bugs)
Version: 10.0
: P3 normal
Target Milestone: 10.0
Assignee: Richard Biener
Keywords: ice-on-valid-code
Depends on:
Blocks: spec
  Show dependency treegraph
Reported: 2019-06-18 14:49 UTC by Martin Liška
Modified: 2019-06-21 11:14 UTC (History)
2 users (show)

See Also:
Known to work: 9.1.0
Known to fail: 10.0
Last reconfirmed: 2019-06-18 00:00:00

test-case (457 bytes, text/plain)
2019-06-18 14:49 UTC, Martin Liška
optimized dump file (5.08 KB, text/plain)
2019-06-18 15:06 UTC, Martin Liška

Note You need to log in before you can comment on or make changes to this bug.
Description Martin Liška 2019-06-18 14:49:01 UTC
Created attachment 46499 [details]

Attached test-case (reduced from 521.wrf_r) fails:

$ gcc fice.f90 -c -march=native -Ofast

      |                              1
Warning: Deleted feature: End expression in DO loop at (1) must be integer
during RTL pass: expand

   18 |                        ccnfact(l,m,n)=cm3
internal compiler error: in maybe_gen_insn, at optabs.c:7341
0x68f8c8 maybe_gen_insn(insn_code, unsigned int, expand_operand*)
0xcf4888 maybe_expand_insn(insn_code, unsigned int, expand_operand*)
0xcf4888 expand_insn(insn_code, unsigned int, expand_operand*)
0xbd1c8b expand_mask_store_optab_fn
0x9a6b47 expand_call_stmt
0x9a6b47 expand_gimple_stmt_1
0x9a6b47 expand_gimple_stmt
0x9ac351 expand_gimple_basic_block
0x9ae0b7 execute
Comment 1 Martin Liška 2019-06-18 15:04:04 UTC
$ (gdb) bt
#0  fancy_abort (file=0x18cc0d0 "/home/marxin/Programming/gcc/gcc/optabs.c", line=7341, function=0x18cc738 "maybe_gen_insn") at /home/marxin/Programming/gcc/gcc/diagnostic.c:1614
#1  0x0000000000d16618 in maybe_gen_insn (icode=CODE_FOR_nothing, nops=3, ops=0x7fffffffd580) at /home/marxin/Programming/gcc/gcc/optabs.c:7341
#2  0x0000000000d16a68 in maybe_expand_insn (icode=CODE_FOR_nothing, nops=3, ops=0x7fffffffd580) at /home/marxin/Programming/gcc/gcc/optabs.c:7385
#3  0x0000000000d16afc in expand_insn (icode=CODE_FOR_nothing, nops=3, ops=0x7fffffffd580) at /home/marxin/Programming/gcc/gcc/optabs.c:7416
#4  0x0000000000be03e4 in expand_mask_store_optab_fn (stmt=0x7ffff756d7e0, optab=maskstore_optab) at /home/marxin/Programming/gcc/gcc/internal-fn.c:2536
#5  0x0000000000be3f18 in expand_MASK_STORE (fn=IFN_MASK_STORE, stmt=0x7ffff756d7e0) at /home/marxin/Programming/gcc/gcc/internal-fn.def:133
#6  0x0000000000be5406 in expand_internal_call (fn=IFN_MASK_STORE, stmt=0x7ffff756d7e0) at /home/marxin/Programming/gcc/gcc/internal-fn.c:3573
#7  0x0000000000be5431 in expand_internal_call (stmt=0x7ffff756d7e0) at /home/marxin/Programming/gcc/gcc/internal-fn.c:3581
#8  0x00000000009af7f8 in expand_call_stmt (stmt=0x7ffff756d7e0) at /home/marxin/Programming/gcc/gcc/cfgexpand.c:2638
#9  expand_gimple_stmt_1 (stmt=<gimple_call 0x7ffff756d7e0>) at /home/marxin/Programming/gcc/gcc/cfgexpand.c:3709
#10 expand_gimple_stmt (stmt=<gimple_call 0x7ffff756d7e0>) at /home/marxin/Programming/gcc/gcc/cfgexpand.c:3868
#11 0x00000000009b5002 in expand_gimple_basic_block (bb=<optimized out>, disable_tail_calls=<optimized out>) at /home/marxin/Programming/gcc/gcc/cfgexpand.c:5908
#12 0x00000000009b6d68 in (anonymous namespace)::pass_expand::execute (this=<optimized out>, fun=0x7ffff7903000) at /home/marxin/Programming/gcc/gcc/cfgexpand.c:6531
#13 0x0000000000d35cfa in execute_one_pass (pass=<opt_pass* 0x23cce30 "expand"(238)>) at /home/marxin/Programming/gcc/gcc/passes.c:2473
#14 0x0000000000d36450 in execute_pass_list_1 (pass=<opt_pass* 0x23cce30 "expand"(238)>) at /home/marxin/Programming/gcc/gcc/passes.c:2559
#15 0x0000000000d36489 in execute_pass_list (fn=0x7ffff7903000, pass=<optimized out>) at /home/marxin/Programming/gcc/gcc/passes.c:2570
#16 0x00000000009ec780 in cgraph_node::expand (this=<cgraph_node * const 0x7ffff77312d0 "prescribe_aerosol_mixactivate"/0>) at /home/marxin/Programming/gcc/gcc/context.h:48
#17 0x00000000009ed71c in expand_all_functions () at /home/marxin/Programming/gcc/gcc/cgraphunit.c:2332
#18 symbol_table::compile (this=0x7ffff7732100) at /home/marxin/Programming/gcc/gcc/cgraphunit.c:2683
#19 0x00000000009eff1d in symbol_table::compile (this=0x7ffff7732100) at /home/marxin/Programming/gcc/gcc/cgraphunit.c:2861
#20 symbol_table::finalize_compilation_unit (this=0x7ffff7732100) at /home/marxin/Programming/gcc/gcc/cgraphunit.c:2861
#21 0x0000000000e09e3d in compile_file () at /home/marxin/Programming/gcc/gcc/toplev.c:481
#22 0x00000000007f1938 in do_compile () at /home/marxin/Programming/gcc/gcc/toplev.c:2209
#23 toplev::main (this=this@entry=0x7fffffffdabe, argc=<optimized out>, argc@entry=14, argv=<optimized out>, argv@entry=0x7fffffffdbb8) at /home/marxin/Programming/gcc/gcc/toplev.c:2344
#24 0x00000000007f544f in main (argc=14, argv=0x7fffffffdbb8) at /home/marxin/Programming/gcc/gcc/main.c:39

$ (gdb) frame 4
#4  0x0000000000be03e4 in expand_mask_store_optab_fn (stmt=0x7ffff756d7e0, optab=maskstore_optab) at /home/marxin/Programming/gcc/gcc/internal-fn.c:2536
2536	  expand_insn (icode, 3, ops);
(gdb) p stmt
$3 = (gcall *) 0x7ffff756d7e0
(gdb) pgg
warning: Expression is not an assignment (and might have no effect)
# .MEM_188 = VDEF <_173>
.MASK_STORE (_485, 32B, _157, cm3_125(D));

$ (gdb) p mem
$4 = (rtx) 0x7ffff76e38e8
(gdb) pr
warning: Expression is not an assignment (and might have no effect)
(mem:SF (reg/f:DI 316 [ _484 ]) [1 MEM[base: _484, offset: 0B]+0 S4 A32])
(gdb) p mask
$5 = (rtx) 0x7ffff76e3af8
(gdb) pr
warning: Expression is not an assignment (and might have no effect)
(subreg:QI (reg:HI 489) 0)
(gdb) p reg
$6 = (rtx) 0x7ffff76c9c18
(gdb) pr
warning: Expression is not an assignment (and might have no effect)
(reg/v:SF 151 [ cm3 ])
(gdb) p type
$7 = <real_type 0x7ffff774b2a0 real(kind=4)>
(gdb) p maskt
$8 = <ssa_name 0x7ffff7927798>
Comment 2 Martin Liška 2019-06-18 15:06:23 UTC
Created attachment 46500 [details]
optimized dump file
Comment 3 Richard Biener 2019-06-18 15:09:56 UTC
Reproduced with -Ofast -march=znver1 -mveclibabi=svml
Comment 4 Richard Biener 2019-06-18 15:14:44 UTC
Somehow we end up with

(gdb) p debug_tree (maskt)
 <ssa_name 0x7ffff6e35798
    type <boolean_type 0x7ffff6c52b28 logical(kind=1) public unsigned QI
        size <integer_cst 0x7ffff6c3dc60 constant 8>
        unit-size <integer_cst 0x7ffff6c3dc78 constant 1>
        align:8 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type 0x7ffff6c52b28 precision:1 min <integer_cst 0x7ffff6c3deb8 0> max <integer_cst 0x7ffff6c3dee8 1>>
    def_stmt _157 = _155 & _21;

this must be a latent issue.
Comment 5 Jakub Jelinek 2019-06-20 15:56:00 UTC
Somewhat reduced:
! PR tree-optimization/90913
! { dg-do compile }
! { dg-options "-O3 -ffast-math" }
! { dg-additional-options "-mavx -mveclibabi=svml" { target i?86-*-* x86_64-*-* } }
subroutine foo (a, b, c, d, e, f, g, h, k, l)
  implicit none
  integer :: d, e, f, g, i, j
  real :: a, b(5,6), c(6), h(6,10,5), k(5,10,2), l(10,5), m, n, o
  do i=1,5
    do j=1,6
      if (m<2) then
        if (m<-2) then
Comment 6 Jakub Jelinek 2019-06-20 16:15:58 UTC
The ICE is because .MASK_LOAD/.MASK_STORE calls are supported just for vectors.
The way it is done is that during ifcvt those internal calls are added to the .LOOP_VECTORIZED guarded loops, and either we successfully vectorize those, then the .MASK_LOAD/.MASK_STORE ifn calls are replaced with their vectorized versions and all the checking is performed, or we should DCE that loop.
The r272239 change which I haven't really understood breaks this on this testcase, we end up with
  _116 = .LOOP_VECTORIZED (1, 4);
  if (_116 != 0)
    goto <bb 25>; [100.00%]
    goto <bb 26>; [100.00%]
that guards these scalar .MASK_STORE calls being replaced with
  _116 = 0;
  if (_190 != 0)
    goto <bb 25>; [100.00%]
    goto <bb 26>; [100.00%]
that is just wrong, we can't and shouldn't reuse the for vectorization only loop if we don't vectorize it.  Both because of these .MASK* calls, but also because e.g. there are COND_EXPRs all around in that which often isn't beneficial in scalar loops, etc.
Comment 7 Richard Biener 2019-06-20 16:57:22 UTC
Aww.  OK, so this is when we vectorize the inner loop of an if-converted outer-loop, then the outer loop LOOP_VECTORIZED condition replaced with the versioning condition dispatches to the variant supposed to be _scalar_ and the fallback
is the if-converted nested copy designed for vectorization.

So re-using the if-conversion copy is only possible when loop_to_version is
the inner loop or we are vectorizing the outer loop (I think we never do
versioning on outer loops currently though).

Testing patch.
Comment 8 Martin Liška 2019-06-21 06:44:56 UTC
One more test-case:

$ gcc /home/marxin/Programming/gcc/gcc/testsuite/gcc.dg/pr70177.c -fwrapv -mavx5124fmaps -O3
during RTL pass: expand
/home/marxin/Programming/gcc/gcc/testsuite/gcc.dg/pr70177.c: In function ‘foo’:
/home/marxin/Programming/gcc/gcc/testsuite/gcc.dg/pr70177.c:8:1: internal compiler error: in maybe_gen_insn, at optabs.c:7379
    8 | foo (int i, int j)
      | ^~~
0x693c2c maybe_gen_insn(insn_code, unsigned int, expand_operand*)
0xc80838 maybe_expand_insn(insn_code, unsigned int, expand_operand*)
0xc80838 expand_insn(insn_code, unsigned int, expand_operand*)
0xb5d98b expand_mask_store_optab_fn
0x9345e7 expand_call_stmt
0x9345e7 expand_gimple_stmt_1
0x9345e7 expand_gimple_stmt
0x939df1 expand_gimple_basic_block
0x93bb57 execute
Comment 9 Richard Biener 2019-06-21 11:11:11 UTC
Author: rguenth
Date: Fri Jun 21 11:10:39 2019
New Revision: 272545

URL: https://gcc.gnu.org/viewcvs?rev=272545&root=gcc&view=rev
2019-06-21  Richard Biener  <rguenther@suse.de>

	PR tree-optimization/90913
	* tree-vect-loop-manip.c (vect_loop_versioning): Do not re-use
	the scalar variant of if-conversion versioning.

	* gfortran.dg/vect/pr90913.f90: New testcase.

Comment 10 Richard Biener 2019-06-21 11:14:01 UTC
Should be fixed now.