On 2/10/19, H.J. Lu <hjl.to...@gmail.com> wrote:
> Emulate MMX sse_cvtps2pi/sse_cvttps2pi with SSE.
>
> PR target/89021
> * config/i386/mmx.md (sse_cvtps2pi): Add SSE emulation.
> (sse_cvttps2pi): Likewise.
It looks to me that this description is wrong. We don't have V4SF
modes here, but V2SF, so we have to fake 64bit load in case of MMX.
The cvtps2dq will access memory in true 128bit width, so this is
wrong.
We have to fix the description to not fake wide mode.
Uros.
> ---
> gcc/config/i386/sse.md | 30 ++++++++++++++++++------------
> 1 file changed, 18 insertions(+), 12 deletions(-)
>
> diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
> index 7d2c0367911..ad31ac3d9e6 100644
> --- a/gcc/config/i386/sse.md
> +++ b/gcc/config/i386/sse.md
> @@ -4668,26 +4668,32 @@
> (set_attr "mode" "V4SF")])
>
> (define_insn "sse_cvtps2pi"
> - [(set (match_operand:V2SI 0 "register_operand" "=y")
> + [(set (match_operand:V2SI 0 "register_operand" "=y,Yv")
> (vec_select:V2SI
> - (unspec:V4SI [(match_operand:V4SF 1 "nonimmediate_operand" "xm")]
> + (unspec:V4SI [(match_operand:V4SF 1 "nonimmediate_operand" "xm,YvBm")]
> UNSPEC_FIX_NOTRUNC)
> (parallel [(const_int 0) (const_int 1)])))]
> - "TARGET_SSE"
> - "cvtps2pi\t{%1, %0|%0, %q1}"
> - [(set_attr "type" "ssecvt")
> - (set_attr "unit" "mmx")
> + "(TARGET_MMX || TARGET_MMX_WITH_SSE) && TARGET_SSE"
> + "@
> + cvtps2pi\t{%1, %0|%0, %q1}
> + %vcvtps2dq\t{%1, %0|%0, %1}"
> + [(set_attr "mmx_isa" "native,x64")
> + (set_attr "type" "ssecvt")
> + (set_attr "unit" "mmx,*")
> (set_attr "mode" "DI")])
>
> (define_insn "sse_cvttps2pi"
> - [(set (match_operand:V2SI 0 "register_operand" "=y")
> + [(set (match_operand:V2SI 0 "register_operand" "=y,Yv")
> (vec_select:V2SI
> - (fix:V4SI (match_operand:V4SF 1 "nonimmediate_operand" "xm"))
> + (fix:V4SI (match_operand:V4SF 1 "nonimmediate_operand" "xm,YvBm"))
> (parallel [(const_int 0) (const_int 1)])))]
> - "TARGET_SSE"
> - "cvttps2pi\t{%1, %0|%0, %q1}"
> - [(set_attr "type" "ssecvt")
> - (set_attr "unit" "mmx")
> + "(TARGET_MMX || TARGET_MMX_WITH_SSE) && TARGET_SSE"
> + "@
> + cvttps2pi\t{%1, %0|%0, %q1}
> + %vcvttps2dq\t{%1, %0|%0, %1}"
> + [(set_attr "mmx_isa" "native,x64")
> + (set_attr "type" "ssecvt")
> + (set_attr "unit" "mmx,*")
> (set_attr "prefix_rep" "0")
> (set_attr "mode" "SF")])
>
> --
> 2.20.1
>
>
