On Thu, Aug 14, 2014 at 2:12 PM, Kirill Yukhin <kirill.yuk...@gmail.com> wrote:
> Hello,
> This patch introduces vcvt[t]pd2 patterns.
>
> Bootstrapped.
> New tests on top of patch-set all pass
> under simulator.
>
> Is it ok for trunk?
>
> gcc/
> * config/i386/i386.md
> (define_code_attr ufix_bool): New.
> * config/i386/sse.md
> (define_mode_iterator VF2_AVX512VL): New.
> (define_mode_attr sseintvecmode2): New.
> (define_insn "ufix_truncv2dfv2si2<mask_name>"): Add masking.
> (define_insn "<fixsuffix>fix_truncv4dfv4si2<mask_name>"): New.
> (define_insn
>
> "<fixsuffix>fix_trunc<mode><sseintvecmodelower>2<mask_name><round_saeonly_name>"):
> New.
> (define_insn
> "fix_notrunc<mode><sseintvecmodelower>2<mask_name><round_name>"):
> New.
> (define_insn
> "ufix_notrunc<mode><sseintvecmodelower>2<mask_name><round_name>"):
> New.
>
> --
> Thanks, K
>
> diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
> index 39fb2307..f6351cf 100644
> --- a/gcc/config/i386/i386.md
> +++ b/gcc/config/i386/i386.md
> @@ -881,6 +881,7 @@
> ;; Used in signed and unsigned fix.
> (define_code_iterator any_fix [fix unsigned_fix])
> (define_code_attr fixsuffix [(fix "") (unsigned_fix "u")])
> +(define_code_attr ufix_bool [(fix "false") (unsigned_fix "true")])
>
> ;; All integer modes.
> (define_mode_iterator SWI1248x [QI HI SI DI])
> diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
> index 3d3d1a0..eaf97ab 100644
> --- a/gcc/config/i386/sse.md
> +++ b/gcc/config/i386/sse.md
> @@ -228,6 +228,9 @@
> (define_mode_iterator VF_512
> [V16SF V8DF])
>
> +(define_mode_iterator VF2_AVX512VL
> + [V8DF (V4DF "TARGET_AVX512VL") (V2DF "TARGET_AVX512VL")])
> +
> ;; All vector integer modes
> (define_mode_iterator VI
> [(V16SI "TARGET_AVX512F") (V8DI "TARGET_AVX512F")
> @@ -523,6 +526,10 @@
> (V32HI "V32HI") (V64QI "V64QI")
> (V32QI "V32QI") (V16QI "V16QI")])
>
> +(define_mode_attr sseintvecmode2
> + [(V8DF "XI") (V4DF "OI") (V2DF "TI")
> + (V8SF "OI") (V4SF "TI")])
> +
> (define_mode_attr sseintvecmodelower
> [(V16SF "v16si") (V8DF "v8di")
> (V8SF "v8si") (V4DF "v4di")
> @@ -4236,15 +4243,58 @@
> (set_attr "prefix" "evex")
> (set_attr "mode" "OI")])
>
> -(define_insn "fix_truncv4dfv4si2"
> - [(set (match_operand:V4SI 0 "register_operand" "=x")
> - (fix:V4SI (match_operand:V4DF 1 "nonimmediate_operand" "xm")))]
> - "TARGET_AVX"
> - "vcvttpd2dq{y}\t{%1, %0|%0, %1}"
> +(define_insn "ufix_truncv2dfv2si2<mask_name>"
> + [(set (match_operand:V4SI 0 "register_operand" "=v")
> + (vec_concat:V4SI
> + (unsigned_fix:V2SI (match_operand:V2DF 1 "nonimmediate_operand"
> "vm"))
> + (const_vector:V2SI [(const_int 0) (const_int 0)])))]
> + "TARGET_AVX512VL"
> + "vcvttpd2udq{x}\t{%1, %0<mask_operand2>|%0<mask_operand2>, %1}"
> [(set_attr "type" "ssecvt")
> - (set_attr "prefix" "vex")
> + (set_attr "prefix" "evex")
> + (set_attr "mode" "TI")])
> +
> +(define_insn "<fixsuffix>fix_truncv4dfv4si2<mask_name>"
> + [(set (match_operand:V4SI 0 "register_operand" "=v")
> + (any_fix:V4SI (match_operand:V4DF 1 "nonimmediate_operand" "vm")))]
> + "(TARGET_AVX && !<ufix_bool>) || (TARGET_AVX512VL && TARGET_AVX512F)"
> + "vcvttpd2<fixsuffix>dq{y}\t{%1, %0<mask_operand2>|%0<mask_operand2>, %1}"
> + [(set_attr "type" "ssecvt")
> + (set_attr "prefix" "maybe_evex")
> (set_attr "mode" "OI")])
Please split this pattern to fix and unsigned_fix. The macroisation
here is confusing. You should then remove ufix_bool attribute.
OK with this change.
Thanks,
Uros.
