Hi!
This patch allows to vectorize negations using 32-byte vectors.
2011-10-14 Jakub Jelinek <ja...@redhat.com>
* config/i386/sse.md (neg<mode>2): Use VI_AVX2 iterator instead
of VI_128.
--- gcc/config/i386/sse.md.jj 2011-10-14 13:05:58.000000000 +0200
+++ gcc/config/i386/sse.md 2011-10-14 13:56:55.000000000 +0200
@@ -4860,10 +4860,10 @@ (define_insn "*vec_concatv2df"
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
(define_expand "neg<mode>2"
- [(set (match_operand:VI_128 0 "register_operand" "")
- (minus:VI_128
+ [(set (match_operand:VI_AVX2 0 "register_operand" "")
+ (minus:VI_AVX2
(match_dup 2)
- (match_operand:VI_128 1 "nonimmediate_operand" "")))]
+ (match_operand:VI_AVX2 1 "nonimmediate_operand" "")))]
"TARGET_SSE2"
"operands[2] = force_reg (<MODE>mode, CONST0_RTX (<MODE>mode));")
Jakub
