[PATCH 08/40] i386: Emulate MMX ashr<mode>3/<shift_insn><mode>3 with SSE

Uros Bizjak ubizjak@gmail.com
Mon Feb 11 23:35:00 GMT 2019


On Tue, Feb 12, 2019 at 12:15 AM Uros Bizjak <ubizjak@gmail.com> wrote:
>
> On Mon, Feb 11, 2019 at 11:55 PM H.J. Lu <hjl.tools@gmail.com> wrote:
> >
> > Emulate MMX ashr<mode>3/<shift_insn><mode>3 with SSE.  Only SSE register
> > source operand is allowed.
> >
> >         PR target/89021
> >         * config/i386/mmx.md (mmx_ashr<mode>3): Changed to define_expand.
> >         Disallow TARGET_MMX_WITH_SSE.
> >         (mmx_<shift_insn><mode>3): Likewise.
> >         (ashr<mode>3): New.
> >         (*ashr<mode>3): Likewise.
> >         (<shift_insn><mode>3): Likewise.
> >         (*<shift_insn><mode>3): Likewise.
> > ---
> >  gcc/config/i386/mmx.md | 68 ++++++++++++++++++++++++++++++++----------
> >  1 file changed, 52 insertions(+), 16 deletions(-)
> >
> > diff --git a/gcc/config/i386/mmx.md b/gcc/config/i386/mmx.md
> > index 0e44b3ce9b8..1b4f67be902 100644
> > --- a/gcc/config/i386/mmx.md
> > +++ b/gcc/config/i386/mmx.md
> > @@ -983,33 +983,69 @@
> >    [(set_attr "type" "mmxadd")
> >     (set_attr "mode" "DI")])
> >
> > -(define_insn "mmx_ashr<mode>3"
> > -  [(set (match_operand:MMXMODE24 0 "register_operand" "=y")
> > +(define_expand "mmx_ashr<mode>3"
> > +  [(set (match_operand:MMXMODE24 0 "register_operand")
> >          (ashiftrt:MMXMODE24
> > -         (match_operand:MMXMODE24 1 "register_operand" "0")
> > -         (match_operand:DI 2 "nonmemory_operand" "yN")))]
> > -  "TARGET_MMX"
> > -  "psra<mmxvecsize>\t{%2, %0|%0, %2}"
> > -  [(set_attr "type" "mmxshft")
> > +         (match_operand:MMXMODE24 1 "register_operand")
> > +         (match_operand:DI 2 "nonmemory_operand")))]
> > +  "TARGET_MMX && !TARGET_MMX_WITH_SSE")
>
> Are you sure this is the correct condition? This pattern is called
> from a builtin, which should be enabled for TARGET_MMX *or*
> TARGET_MMX_WITH_SSE.

It looks to me that "TARGET_MMX || TARGET_MMX_WITH_SSE" should be used
with mmx_* patterns (and new SSE constraints should be added to these
patterns) and new named shift expanders should be added here.

Uros.

> > +(define_expand "ashr<mode>3"
> > +  [(set (match_operand:MMXMODE24 0 "register_operand")
> > +        (ashiftrt:MMXMODE24
> > +         (match_operand:MMXMODE24 1 "register_operand")
> > +         (match_operand:DI 2 "nonmemory_operand")))]
> > +  "TARGET_MMX_WITH_SSE")
> > +
> > +(define_insn "*ashr<mode>3"
> > +  [(set (match_operand:MMXMODE24 0 "register_operand" "=y,x,Yv")
> > +        (ashiftrt:MMXMODE24
> > +         (match_operand:MMXMODE24 1 "register_operand" "0,0,Yv")
> > +         (match_operand:DI 2 "nonmemory_operand" "yN,xN,YvN")))]
> > +  "TARGET_MMX || TARGET_MMX_WITH_SSE"
> > +  "@
> > +   psra<mmxvecsize>\t{%2, %0|%0, %2}
> > +   psra<mmxvecsize>\t{%2, %0|%0, %2}
> > +   vpsra<mmxvecsize>\t{%2, %1, %0|%0, %1, %2}"
> > +  [(set_attr "mmx_isa" "native,x64_noavx,x64_avx")
> > +   (set_attr "type" "mmxshft,sseishft,sseishft")
> >     (set (attr "length_immediate")
> >       (if_then_else (match_operand 2 "const_int_operand")
> >         (const_string "1")
> >         (const_string "0")))
> > -   (set_attr "mode" "DI")])
> > +   (set_attr "mode" "DI,TI,TI")])
> >
> > -(define_insn "mmx_<shift_insn><mode>3"
> > -  [(set (match_operand:MMXMODE248 0 "register_operand" "=y")
> > +(define_expand "mmx_<shift_insn><mode>3"
> > +  [(set (match_operand:MMXMODE248 0 "register_operand")
> >          (any_lshift:MMXMODE248
> > -         (match_operand:MMXMODE248 1 "register_operand" "0")
> > -         (match_operand:DI 2 "nonmemory_operand" "yN")))]
> > -  "TARGET_MMX"
> > -  "p<vshift><mmxvecsize>\t{%2, %0|%0, %2}"
> > -  [(set_attr "type" "mmxshft")
> > +         (match_operand:MMXMODE248 1 "register_operand")
> > +         (match_operand:DI 2 "nonmemory_operand")))]
> > +  "TARGET_MMX && !TARGET_MMX_WITH_SSE")
> > +
> > +(define_expand "<shift_insn><mode>3"
> > +  [(set (match_operand:MMXMODE248 0 "register_operand")
> > +        (any_lshift:MMXMODE248
> > +         (match_operand:MMXMODE248 1 "register_operand")
> > +         (match_operand:DI 2 "nonmemory_operand")))]
> > +  "TARGET_MMX_WITH_SSE")
> > +
> > +(define_insn "*<shift_insn><mode>3"
> > +  [(set (match_operand:MMXMODE248 0 "register_operand" "=y,x,Yv")
> > +        (any_lshift:MMXMODE248
> > +         (match_operand:MMXMODE248 1 "register_operand" "0,0,Yv")
> > +         (match_operand:DI 2 "nonmemory_operand" "yN,xN,YvN")))]
> > +  "TARGET_MMX || TARGET_MMX_WITH_SSE"
> > +  "@
> > +   p<vshift><mmxvecsize>\t{%2, %0|%0, %2}
> > +   p<vshift><mmxvecsize>\t{%2, %0|%0, %2}
> > +   vp<vshift><mmxvecsize>\t{%2, %1, %0|%0, %1, %2}"
> > +  [(set_attr "mmx_isa" "native,x64_noavx,x64_avx")
> > +   (set_attr "type" "mmxshft,sseishft,sseishft")
> >     (set (attr "length_immediate")
> >       (if_then_else (match_operand 2 "const_int_operand")
> >         (const_string "1")
> >         (const_string "0")))
> > -   (set_attr "mode" "DI")])
> > +   (set_attr "mode" "DI,TI,TI")])
> >
> >  ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
> >  ;;
> > --
> > 2.20.1
> >



More information about the Gcc-patches mailing list