[Bug target/105192] ICE: in gen_reg_rtx, at emit-rtl.cc:1167 (during RTL pass: ce3) with -O -fno-if-conversion -fharden-compares

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105192

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2022-11-16
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Something like this might fix the issue:
apinski@xeond:~/src/upstream-gcc/gcc/gcc$ git diff config/mips/mips.md
diff --git a/gcc/config/mips/mips.md b/gcc/config/mips/mips.md
index e0f0a582732..3419f4df307 100644
--- a/gcc/config/mips/mips.md
+++ b/gcc/config/mips/mips.md
@@ -1217,7 +1217,7 @@ (define_expand "ctrap<mode>4"
                            [(match_operand:GPR 1 "reg_or_0_operand")
                             (match_operand:GPR 2 "arith_operand")])
            (match_operand 3 "const_0_operand"))]
-  "ISA_HAS_COND_TRAPI || ISA_HAS_COND_TRAP"
+  "can_create_pseudo_p () && (ISA_HAS_COND_TRAPI || ISA_HAS_COND_TRAP")
 {
   mips_expand_conditional_trap (operands[0]);
   DONE;

I have no way to test it though.
But what is happening from the backtrace and where the assert is:
* the post-reload ifcvt is detecting a conditional trap and the mips backend
didn't expect any post-reload (really !can_create_pseudo_p ()) detection of a
condtional trap happening so it thought it could use force_reg (which might
create a new pseudo register).