https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96865
Jeevitha <jeevitha at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jeevitha at gcc dot gnu.org
--- Comment #1 from Jeevitha <jeevitha at gcc dot gnu.org> ---
The issue was in the following RTL,
(insn 6 2 21 2 (parallel [
(set (reg:SI 96 lr)
(unspec:SI [
(symbol_ref:SI ("_GLOBAL_OFFSET_TABLE_") [flags 0x42])
(label_ref 0)
] UNSPEC_TOCPTR))
(code_label 0 0 0 2 (nil) [2 uses])
]) "test.c":6:6 798 {load_toc_v4_PIC_1b_normal}
The above RTL will generate the following assembly:
bcl 20,31,$+8
There is no need for 'label_ref 0' and 'code_label' in the RTL instruction, as
the assembly does not require them. Is my understanding correct?
Currently, we are encountering an ICE due to the following reason: In the
'hash_rtx' function [cse.cc], handling for 'LABEL_REF' is present but not for
'code_label', causing it to not return a hash directly. Instead, it searches
for the format corresponding to '(code_label 0 0 0 4 (nil))', which is
"uuB00is". In the provided code snippet below, 'B' and 'u' do not have case
handling, leading to a 'gcc_unreachable'. This is the reason for the ICE
currently.
fmt = GET_RTX_FORMAT (code); //code -> code_label
for (; i >= 0; i--)
{
switch (fmt[i])
{
case 'e':
/* If we are about to do the last recursive call
needed at this level, change it into iteration.
This function is called enough to be worth it. */
if (i == 0)
{
x = XEXP (x, i);
goto repeat;
}
hash += hash_rtx (XEXP (x, i), VOIDmode, do_not_record_p,
hash_arg_in_memory_p,
have_reg_qty, cb);
break;
case 'E':
for (j = 0; j < XVECLEN (x, i); j++)
hash += hash_rtx (XVECEXP (x, i, j), VOIDmode, do_not_record_p,
hash_arg_in_memory_p,
have_reg_qty, cb);
break;
case 's':
hash += hash_rtx_string (XSTR (x, i));
break;
case 'i':
hash += (unsigned int) XINT (x, i);
break;
case 'p':
hash += constant_lower_bound (SUBREG_BYTE (x));
break;
case '0': case 't':
/* Unused. */
break;
default:
gcc_unreachable ();
}
}
To address this, I've removed 'operand1' and adjusted the respective
'match_dup' to 'operand0' in the below pattern.
(define_insn "load_toc_v4_PIC_1b_normal"
[(set (reg:SI LR_REGNO)
(unspec:SI [(match_operand:SI 0 "immediate_operand" "s")
(label_ref (match_operand 1 "" ""))]
UNSPEC_TOCPTR))
(match_dup 1)]
After implementing the mentioned change, there were no ICEs or regressions. Is
this approach correct?
