https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124696
Uroš Bizjak <ubizjak at gmail dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
Component|target |rtl-optimization
--- Comment #2 from Uroš Bizjak <ubizjak at gmail dot com> ---
RA is trying to allocate V4QI register as output reload for (insn 215):
(insn 215 31 32 4 (set (reg:V2QI 104 [ _22 ])
(reg:V2QI 274 [orig:104 _22 ] [104])) "pr124696.c":13:5 2089
{*movv2qi_internal}
(expr_list:REG_DEAD (reg:V2QI 274 [orig:104 _22 ] [104])
(nil)))
Creating newreg=281, assigning class NO_REGS to secondary r281
215: r104:V2QI=r274:V2QI
REG_DEAD r274:V2QI
Inserting the sec. move after:
224: r104:V2QI#0=r281:V4QI
215: r104:V2QI=r274:V2QI
REG_DEAD r274:V2QI
Changing on before:
225: r281:V4QI=r274:V2QI#0
deleting insn with uid = 215.
Creating newreg=282 from oldreg=274, assigning class ALL_REGS to
paradoxical subreg r282
225: r281:V4QI=r282:V4QI#0
Inserting paradoxical subreg reload before:
226: r282:V4QI#0=r274:V2QI
and after:
Creating newreg=283, assigning class NO_REGS to secondary r283
226: r282:V4QI#0=r274:V2QI
Inserting the sec. move after:
227: r282:V4QI#0=r283:V4QI
226: r282:V4QI#0=r274:V2QI
Changing on before:
228: r283:V4QI=r274:V2QI#0
deleting insn with uid = 226.
Creating newreg=284 from oldreg=274, assigning class ALL_REGS to
paradoxical subreg r284
228: r283:V4QI=r284:V4QI#0
Inserting paradoxical subreg reload before:
229: r284:V4QI#0=r274:V2QI
that continues in a reload loop.
I don't see anything around (insn 225, "*movv2qi_internal") that would require
V4QI mode. Looks like RA problem to me.
