https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118184
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2024-12-23
CC| |rsandifo at gcc dot gnu.org
Ever confirmed|0 |1
--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Expand we have:
;; u.d = Nan;
(insn 85 84 86 (set (reg:DI 134)
(high:DI (symbol_ref/u:DI ("*.LC0") [flags 0x82])))
"./strtod_nan_main.c":58:43 -1
(nil))
(insn 86 85 87 (set (reg/f:DI 133)
(lo_sum:DI (reg:DI 134)
(symbol_ref/u:DI ("*.LC0") [flags 0x82])))
"./strtod_nan_main.c":58:43 -1
(expr_list:REG_EQUAL (symbol_ref/u:DI ("*.LC0") [flags 0x82])
(nil)))
(insn 87 86 0 (set (reg/v:TI 110 [ u ])
(mem/u/c:TI (reg/f:DI 133) [0 S16 A128])) "./strtod_nan_main.c":58:43
-1
(expr_list:REG_EQUAL (const_wide_int 0x7fff8000000000000000000000000000)
(nil)))
;; MEM <unsigned long> [(union ieee854_long_double *)&u] = mant_21;
(insn 88 87 0 (set (subreg:DI (reg/v:TI 110 [ u ]) 0)
(reg/v:DI 109 [ mant ])) "./strtod_nan_main.c":58:165 -1
(nil))
;; retval_26 = u.d;
(insn 89 88 0 (set (reg/v:TF 111 [ <retval> ])
(subreg:TF (reg/v:TI 110 [ u ]) 0)) "./strtod_nan_main.c":58:270
discrim 1 -1
(nil))
;; u ={v} {CLOBBER(eos)};
(insn 90 89 0 (clobber (reg/v:TI 110 [ u ])) -1
(nil))
Which looks fine.
after Combine (and subreg3) we have:
(note 84 83 87 11 [bb 11] NOTE_INSN_BASIC_BLOCK)
(insn 87 84 88 11 (set (reg/v:TI 110 [ u ])
(const_wide_int 0x7fff8000000000000000000000000000))
"./strtod_nan_main.c":58:43 75 {*movti_aarch64}
(nil))
(insn 88 87 89 11 (set (subreg:DI (reg/v:TI 110 [ u ]) 0)
(reg/v:DI 109 [ mant ])) "./strtod_nan_main.c":58:165 70
{*movdi_aarch64}
(expr_list:REG_DEAD (reg/v:DI 109 [ mant ])
(nil)))
(insn 89 88 91 11 (set (reg/v:TF 111 [ <retval> ])
(subreg:TF (reg/v:TI 110 [ u ]) 0)) "./strtod_nan_main.c":58:270
discrim 1 83 {*movtf_aarch64}
(expr_list:REG_DEAD (reg/v:TI 110 [ u ])
(nil)))
Which also looks fine.
Early_ra messes it up.
Block 13:
46: 89: r111:TF=r110:TI#0
Creating allocnos [1:1] for r110
Not allocating region: r110's mode TI is not tieable to mode TF
48: 88: r110:TI#0=r140:DI
Block 13:
50: 87: r110:TI=<0x7fff800000000000,0>
Creating allocnos [0:0] for r110
-- dead
So yes this is 100% a target issue as early_ra is a target pass.
