https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125953
Bug ID: 125953
Summary: [13/14/15/16/17 Regression] vect_recog_divmod_pattern
uses ranger on pattern stmts
Product: gcc
Version: 15.3.0
Status: UNCONFIRMED
Keywords: ice-on-valid-code
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: rguenth at gcc dot gnu.org
CC: amacleod at redhat dot com
Target Milestone: ---
Target: aarch64
Building firefox on aarch64 with -march=armv9-a results in
during GIMPLE pass: vect
In file included from Unified_cpp_image_decoders0.cpp:65:
../../../firefox-152.0.1/image/decoders/nsJXLDecoder.cpp: In member function
‘bool mozilla::image::nsJXLDecoder::WritePixelRowsToPipe()’:
../../../firefox-152.0.1/image/decoders/nsJXLDecoder.cpp:578:6: internal
compiler error: Segmentation fault
0x3403d7e internal_error(char const*, ...)
/space/rguenther/src/gcc-15-branch/gcc/diagnostic-global-context.cc:517
0x1a2a777 crash_signal
/space/rguenther/src/gcc-15-branch/gcc/toplev.cc:322
0x14476fd get_immediate_dominator(cdi_direction, basic_block_def*)
/space/rguenther/src/gcc-15-branch/gcc/dominance.cc:859
0x2ff4dfc ranger_cache::range_from_dom(vrange&, tree_node*, basic_block_def*,
ranger_cache::rfd_mode)
/space/rguenther/src/gcc-15-branch/gcc/gimple-range-cache.cc:1730
0x2ff3faa ranger_cache::fill_block_cache(tree_node*, basic_block_def*,
basic_block_def*)
/space/rguenther/src/gcc-15-branch/gcc/gimple-range-cache.cc:1486
...
0x2ff95fe fold_using_range::fold_stmt(vrange&, gimple*, fur_source&,
tree_node*)
/space/rguenther/src/gcc-15-branch/gcc/gimple-range-fold.cc:648
0x2fed54b gimple_ranger::fold_range_internal(vrange&, gimple*, tree_node*)
/space/rguenther/src/gcc-15-branch/gcc/gimple-range.cc:277
0x2fed871 gimple_ranger::range_of_stmt(vrange&, gimple*, tree_node*)
/space/rguenther/src/gcc-15-branch/gcc/gimple-range.cc:338
because we do
#32 0x00000000031abb4f in vect_recog_divmod_pattern (vinfo=0x69a0e50,
stmt_vinfo=0x52f73f0, type_out=0x7fffffffc1e8)
at /space/rguenther/src/gcc-15-branch/gcc/tree-vect-patterns.cc:4993
4993 if (ranger.range_of_expr (r, oprnd0, stmt) && !r.undefined_p
())
(gdb) p debug_tree (oprnd0)
<ssa_name 0x7fffb1bd8900
type <integer_type 0x7fffdd806000 public unsigned HI
size <integer_cst 0x7ffff681b3f0 constant 16>
unit-size <integer_cst 0x7ffff681b408 constant 2>
align:16 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
0x7fffdd806000 precision:16 min <integer_cst 0x7fffddbf1900 0> max <integer_cst
0x7fffddb84c90 65535>
pointer_to_this <pointer_type 0x7fffbd356e70>>
def_stmt patt_356 = patt_353 * patt_355;
version:356>
The above happens with the testcase I have only on the 15 branch. I'm
currently reducing it. I didn't try the 13 or 14 branches but we do the
same there. I'm not sure if we put any mitigation into ranger for stmts
not in the IL, did we?
