https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117562
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|ASSIGNED |NEW
Assignee|rguenth at gcc dot gnu.org |unassigned at gcc dot
gnu.org
Keywords| |ra
--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
Removing the alternative fixes the slowdown, both LTO and non-LTO, the
multi-epilogue version is now faster.
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index 72acd5bde5e..0a3ee38f55a 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -11045,11 +11045,11 @@
})
(define_insn "sse_movhlps"
- [(set (match_operand:V4SF 0 "nonimmediate_operand" "=x,v,x,v,m")
+ [(set (match_operand:V4SF 0 "nonimmediate_operand" "=x,v,x,v")
(vec_select:V4SF
(vec_concat:V8SF
- (match_operand:V4SF 1 "nonimmediate_operand" " 0,v,0,v,0")
- (match_operand:V4SF 2 "nonimmediate_operand" " x,v,o,o,v"))
+ (match_operand:V4SF 1 "nonimmediate_operand" " 0,v,0,v")
+ (match_operand:V4SF 2 "nonimmediate_operand" " x,v,o,o"))
(parallel [(const_int 6)
(const_int 7)
(const_int 2)
@@ -11059,12 +11059,11 @@
movhlps\t{%2, %0|%0, %2}
vmovhlps\t{%2, %1, %0|%0, %1, %2}
movlps\t{%H2, %0|%0, %H2}
- vmovlps\t{%H2, %1, %0|%0, %1, %H2}
- %vmovhps\t{%2, %0|%q0, %2}"
- [(set_attr "isa" "noavx,avx,noavx,avx,*")
+ vmovlps\t{%H2, %1, %0|%0, %1, %H2}"
+ [(set_attr "isa" "noavx,avx,noavx,avx")
(set_attr "type" "ssemov2")
- (set_attr "prefix" "orig,maybe_evex,orig,maybe_evex,maybe_vex")
- (set_attr "mode" "V4SF,V4SF,V2SF,V2SF,V2SF")])
+ (set_attr "prefix" "orig,maybe_evex,orig,maybe_evex")
+ (set_attr "mode" "V4SF,V4SF,V2SF,V2SF")])
(define_expand "sse_movlhps_exp"
[(set (match_operand:V4SF 0 "nonimmediate_operand")
