https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117562
--- Comment #7 from Hongtao Liu <liuhongt at gcc dot gnu.org> ---
> Huh. It looks like this is from a V4SF -> 2xV2DF extension via
> vec_unpack_{hi,lo}_expr.
>
> Originally this is
>
> (insn 1161 1160 1162 58 (set (reg:V4SF 853)
> (vec_select:V4SF (vec_concat:V8SF (reg:V4SF 853)
> (reg:V4SF 566 [ vect__811.229 ]))
> (parallel [
> (const_int 6 [0x6])
> (const_int 7 [0x7])
> (const_int 2 [0x2])
> (const_int 3 [0x3])
> ]))) "cont_mgau.c":157:9 5181 {sse_movhlps}
> (expr_list:REG_DEAD (reg:V4SF 566 [ vect__811.229 ])
> (nil)))
>
vec_unpacks_hi_v4sf create an unintialized (reg:V4SF 853), I guess it may
confuse LRA to allocate a mem for it.
10413(define_expand "vec_unpacks_hi_v4sf"
10414 [(set (match_dup 2)
10415 (vec_select:V4SF
10416 (vec_concat:V8SF
10417 (match_dup 2) ------------ here, use without initialization.
10418 (match_operand:V4SF 1 "vector_operand"))
10419 (parallel [(const_int 6) (const_int 7)
10420 (const_int 2) (const_int 3)])))
10421 (set (match_operand:V2DF 0 "register_operand")
10422 (float_extend:V2DF
10423 (vec_select:V2SF
10424 (match_dup 2)
10425 (parallel [(const_int 0) (const_int 1)]))))]
10426 "TARGET_SSE2"
10427 "operands[2] = gen_reg_rtx (V4SFmode);")
Explicitly initialize the operands[2] with 0, the spill is gone, Could you try
below patch to see if it also fixed ths issue?
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index e50355f4839..e1ee688d5e4 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -10500,7 +10500,10 @@ (define_expand "vec_unpacks_hi_v4sf"
(match_dup 2)
(parallel [(const_int 0) (const_int 1)]))))]
"TARGET_SSE2"
- "operands[2] = gen_reg_rtx (V4SFmode);")
+{
+ operands[2] = gen_reg_rtx (V4SFmode);
+ emit_move_insn (operands[2], CONST0_RTX (V4SFmode));
+})
(define_expand "vec_unpacks_hi_v8sf"
[(set (match_dup 2)
