https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121205
--- Comment #2 from Stefan Schulze Frielinghaus <stefansf at gcc dot gnu.org>
---
(In reply to Sam James from comment #1)
> With checking, I also see:
>
> +FAIL: gcc.target/i386/asm-hard-reg-1.c (internal compiler error: RTL check:
> expected elt 3 type 'e' or 'u', have '0' (rtx note) in PATTERN, at
> rtl.h:1523)
> +FAIL: gcc.target/i386/asm-hard-reg-2.c (internal compiler error: RTL check:
> expected elt 3 type 'e' or 'u', have '0' (rtx note) in PATTERN, at
> rtl.h:1523)
The checking error occurs for
void test (void)
{
int x, y;
__asm__ __volatile__ ("" : "={rdx}" (x), "=d" (y)); // expect error
__asm__ __volatile__ ("" :: "{rdx}" (x), "d" (y)); // expect error
__asm__ __volatile__ ("" : "=S" (x), "={rsi}" (y)); // expect error
}
were for each line we expect an error during LRA coming from
void
lra_asm_insn_error (rtx_insn *insn)
{
lra_asm_error_p = true;
error_for_asm (insn,
"%<asm%> operand has impossible constraints"
" or there are not enough registers");
/* Avoid further trouble with this insn. */
if (JUMP_P (insn))
{
ira_nullify_asm_goto (insn);
lra_invalidate_insn_data (insn);
}
else
{
PATTERN (insn) = gen_rtx_USE (VOIDmode, const0_rtx);
lra_set_insn_deleted (insn);
}
}
were we also set the pattern to zero in the else case. In
curr_insn_transform(), if an error occurs we still call in the end
lra_process_new_insns() with a insn deleted note. Instead we probably should
just return after reporting the error:
diff --git a/gcc/lra-constraints.cc b/gcc/lra-constraints.cc
index 83f8fda3b52..dc3c224a097 100644
--- a/gcc/lra-constraints.cc
+++ b/gcc/lra-constraints.cc
@@ -4929,7 +4929,10 @@ curr_insn_transform (bool check_only_p)
if (asm_noperands (PATTERN (curr_insn)) >= 0
&& ++curr_id->asm_reloads_num >= FIRST_PSEUDO_REGISTER)
/* Most probably there are no enough registers to satisfy asm insn: */
- lra_asm_insn_error (curr_insn);
+ {
+ lra_asm_insn_error (curr_insn);
+ return change_p;
+ }
}
if (goal_alt_out_sp_reload_p)
{
