[PATCH][AArch64] Add neon_pairwise_add & neon_pairwise_add_q types
James Greenhalgh
james.greenhalgh@arm.com
Thu Mar 9 10:54:00 GMT 2017
On Wed, Mar 08, 2017 at 06:09:52PM +0000, James Greenhalgh wrote:
> On Mon, Mar 06, 2017 at 05:09:44AM +0000, Hurugalawadi, Naveen wrote:
> > Hi,
> >
> > Please find attached the patch that adds "neon_pairwise_add" &
> > "neon_pairwise_add_qcrypto_pmull" for AArch64.
> >
> > The patch doesn't change spec but improve other benchmarks.
> >
> > Bootstrapped and Regression tested on aarch64-thunder-linux.
> > Please review the patch and let us know if its okay for Stage-1?
>
> The whitespace in various places in this patch is inconsistent with the
> whitespace around the modified line. For example:
>
> --- a/gcc/config/arm/cortex-a9-neon.md
> +++ b/gcc/config/arm/cortex-a9-neon.md
> @@ -52,6 +52,7 @@
> (const_string "neon_int_2")
> (eq_attr "type" "neon_neg, neon_neg_q,\
> neon_reduc_add, neon_reduc_add_q,\
> + neon_pairwise_add, neon_pairwise_add_q,\
> neon_reduc_add_long,\
> neon_add_long, neon_sub_long")
> (const_string "neon_int_3")
>
> Where most lines use 8 spaces, and you've used tabs.
>
> While all these lines should really have been tabs to begin with, they
> didn't, and introducing inconsistency from line to line is worse than
> just following the existing style.
>
> With the whitespace fixed, this patch is OK for stage 1.
Having double checked after looking at the pmull patch, you've missed
the same cores here. So this patch isn't OK without fixes for the models
in cortex-a53.md and exynos-m1.md
Additionally, this will need an ARM port maintainer to take a look before it
can go in.
Thanks,
James
>
> Thanks,
> James
>
> > 2017-03-06 Julian Brown <julian@codesourcery.com>
> > Naveen H.S <Naveen.Hurugalawadi@cavium.com>
> >
> > * config/aarch64/aarch64-simd.md (aarch64_reduc_plus_internal<mode>)
> > (aarch64_reduc_plus_internalv2si, aarch64_addp<mode>, aarch64_addpdi):
> > Use neon_pairwise_add/neon_pairwise_add_q as appropriate.
> > * config/aarch64/iterators.md (reduc_pairwise): New mode attribute.
> > * config/aarch64/thunderx.md (thunderx_neon_add, thunderx_neon_add_q):
> > Tweak for neon_pairwise_add split.
> > * config/aarch64/thunderx2t99.md (thunderx2t99_asimd_int): Add
> > neon_pairwise_add/neon_pairwise_add_q types.
> > * config/arm/cortex-a15-neon.md (cortex_a15_neon_type): Likewise.
> > * config/arm/cortex-a17-neon.md (cortex_a17_neon_type): Likewise.
> > * config/arm/cortex-a57.md (cortex_a57_neon_type): Likewise.
> > * config/arm/cortex-a8-neon.md (cortex_a8_neon_type): Likewise.
> > * config/arm/cortex-a9-neon.md (cortex_a9_neon_type): Likewise.
> > * config/arm/xgene1.md (xgene1_neon_arith): Likewise.
> > * config/arm/types.md (neon_pairwise_add, neon_pairwise_add_q): Add.
>
> > diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md
> > index 338b9f8..878f86a 100644
> > --- a/gcc/config/aarch64/aarch64-simd.md
> > +++ b/gcc/config/aarch64/aarch64-simd.md
> > @@ -2101,7 +2101,7 @@
> > UNSPEC_ADDV))]
> > "TARGET_SIMD"
> > "add<VDQV:vp>\\t%<Vetype>0, %1.<Vtype>"
> > - [(set_attr "type" "neon_reduc_add<q>")]
> > + [(set_attr "type" "neon_<reduc_pairwise>_add<q>")]
> > )
> >
> > (define_insn "aarch64_reduc_plus_internalv2si"
> > @@ -2110,7 +2110,7 @@
> > UNSPEC_ADDV))]
> > "TARGET_SIMD"
> > "addp\\t%0.2s, %1.2s, %1.2s"
> > - [(set_attr "type" "neon_reduc_add")]
> > + [(set_attr "type" "neon_pairwise_add")]
> > )
> >
> > (define_insn "reduc_plus_scal_<mode>"
> > @@ -4405,7 +4405,7 @@
> > UNSPEC_ADDP))]
> > "TARGET_SIMD"
> > "addp\t%<v>0<Vmtype>, %<v>1<Vmtype>, %<v>2<Vmtype>"
> > - [(set_attr "type" "neon_reduc_add<q>")]
> > + [(set_attr "type" "neon_pairwise_add<q>")]
> > )
> >
> > (define_insn "aarch64_addpdi"
> > @@ -4415,7 +4415,7 @@
> > UNSPEC_ADDP))]
> > "TARGET_SIMD"
> > "addp\t%d0, %1.2d"
> > - [(set_attr "type" "neon_reduc_add")]
> > + [(set_attr "type" "neon_pairwise_add")]
> > )
> >
> > ;; sqrt
> > diff --git a/gcc/config/aarch64/iterators.md b/gcc/config/aarch64/iterators.md
> > index c59d31e..c829cb5 100644
> > --- a/gcc/config/aarch64/iterators.md
> > +++ b/gcc/config/aarch64/iterators.md
> > @@ -790,6 +790,12 @@
> > (V2SF "p") (V4SF "v")
> > (V4HF "v") (V8HF "v")])
> >
> > +(define_mode_attr reduc_pairwise [(V8QI "reduc") (V16QI "reduc")
> > + (V4HI "reduc") (V8HI "reduc")
> > + (V2SI "pairwise") (V4SI "reduc")
> > + (V2DI "pairwise") (V2DF "pairwise")
> > + (V2SF "pairwise") (V4SF "reduc")])
> > +
> > (define_mode_attr vsi2qi [(V2SI "v8qi") (V4SI "v16qi")])
> > (define_mode_attr VSI2QI [(V2SI "V8QI") (V4SI "V16QI")])
> >
> > diff --git a/gcc/config/aarch64/thunderx.md b/gcc/config/aarch64/thunderx.md
> > index b67671d..95bfad4 100644
> > --- a/gcc/config/aarch64/thunderx.md
> > +++ b/gcc/config/aarch64/thunderx.md
> > @@ -266,7 +266,8 @@
> >
> > (define_insn_reservation "thunderx_neon_add" 4
> > (and (eq_attr "tune" "thunderx")
> > - (eq_attr "type" "neon_reduc_add, neon_reduc_minmax, neon_fp_reduc_add_s, \
> > + (eq_attr "type" "neon_reduc_add, neon_pairwise_add, neon_reduc_minmax,\
> > + neon_fp_reduc_add_s, \
> > neon_fp_reduc_add_d, neon_fp_to_int_s, neon_fp_to_int_d, \
> > neon_add_halve, neon_sub_halve, neon_qadd, neon_compare, \
> > neon_compare_zero, neon_minmax, neon_abd, neon_add, neon_sub, \
> > @@ -280,7 +281,8 @@
> >
> > (define_insn_reservation "thunderx_neon_add_q" 5
> > (and (eq_attr "tune" "thunderx")
> > - (eq_attr "type" "neon_reduc_add_q, neon_reduc_minmax_q, neon_fp_reduc_add_s_q, \
> > + (eq_attr "type" "neon_reduc_add_q, neon_pairwise_add_q,\
> > + neon_reduc_minmax_q, neon_fp_reduc_add_s_q, \
> > neon_fp_reduc_add_d_q, neon_fp_to_int_s_q, neon_fp_to_int_d_q, \
> > neon_add_halve_q, neon_sub_halve_q, neon_qadd_q, neon_compare_q, \
> > neon_compare_zero_q, neon_minmax_q, neon_abd_q, neon_add_q, neon_sub_q, \
> > diff --git a/gcc/config/aarch64/thunderx2t99.md b/gcc/config/aarch64/thunderx2t99.md
> > index 67011ac..f807547 100644
> > --- a/gcc/config/aarch64/thunderx2t99.md
> > +++ b/gcc/config/aarch64/thunderx2t99.md
> > @@ -231,6 +231,7 @@
> > neon_abs,neon_abs_q,\
> > neon_add,neon_add_q,\
> > neon_neg,neon_neg_q,\
> > + neon_pairwise_add,neon_pairwise_add_q,\
> > neon_add_long,neon_add_widen,\
> > neon_add_halve,neon_add_halve_q,\
> > neon_sub_long,neon_sub_widen,\
> > diff --git a/gcc/config/arm/cortex-a15-neon.md b/gcc/config/arm/cortex-a15-neon.md
> > index 73ee84c..1a02fa2 100644
> > --- a/gcc/config/arm/cortex-a15-neon.md
> > +++ b/gcc/config/arm/cortex-a15-neon.md
> > @@ -48,6 +48,7 @@
> > (eq_attr "type" "neon_add, neon_add_q, neon_add_long,\
> > neon_add_widen, neon_neg, neon_neg_q,\
> > neon_reduc_add, neon_reduc_add_q,\
> > + neon_pairwise_add, neon_pairwise_add_q,\
> > neon_reduc_add_long, neon_sub, neon_sub_q,\
> > neon_sub_long, neon_sub_widen, neon_logic,\
> > neon_logic_q, neon_tst, neon_tst_q")
> > diff --git a/gcc/config/arm/cortex-a17-neon.md b/gcc/config/arm/cortex-a17-neon.md
> > index 29d08de..d4f9e03 100644
> > --- a/gcc/config/arm/cortex-a17-neon.md
> > +++ b/gcc/config/arm/cortex-a17-neon.md
> > @@ -47,6 +47,7 @@
> > (eq_attr "type" "neon_add, neon_add_q, neon_add_long,\
> > neon_add_widen, neon_neg, neon_neg_q,\
> > neon_reduc_add, neon_reduc_add_q,\
> > + neon_pairwise_add, neon_pairwise_add_q,\
> > neon_reduc_add_long, neon_sub, neon_sub_q,\
> > neon_sub_long, neon_sub_widen, neon_logic,\
> > neon_logic_q, neon_tst, neon_tst_q")
> > diff --git a/gcc/config/arm/cortex-a57.md b/gcc/config/arm/cortex-a57.md
> > index ebf4a49..cb5e181 100644
> > --- a/gcc/config/arm/cortex-a57.md
> > +++ b/gcc/config/arm/cortex-a57.md
> > @@ -50,6 +50,7 @@
> > (eq_attr "type" "neon_add, neon_add_q, neon_add_long,\
> > neon_add_widen, neon_neg, neon_neg_q,\
> > neon_reduc_add, neon_reduc_add_q,\
> > + neon_pairwise_add, neon_pairwise_add_q,\
> > neon_reduc_add_long, neon_sub, neon_sub_q,\
> > neon_sub_long, neon_sub_widen, neon_logic,\
> > neon_logic_q, neon_tst, neon_tst_q")
> > diff --git a/gcc/config/arm/cortex-a8-neon.md b/gcc/config/arm/cortex-a8-neon.md
> > index baa9180..652ba4e 100644
> > --- a/gcc/config/arm/cortex-a8-neon.md
> > +++ b/gcc/config/arm/cortex-a8-neon.md
> > @@ -51,6 +51,7 @@
> > (const_string "neon_int_2")
> > (eq_attr "type" "neon_neg, neon_neg_q,\
> > neon_reduc_add, neon_reduc_add_q,\
> > + neon_pairwise_add, neon_pairwise_add_q,\
> > neon_reduc_add_long,\
> > neon_add_long, neon_sub_long")
> > (const_string "neon_int_3")
> > diff --git a/gcc/config/arm/cortex-a9-neon.md b/gcc/config/arm/cortex-a9-neon.md
> > index 9e9827f..bb1e300 100644
> > --- a/gcc/config/arm/cortex-a9-neon.md
> > +++ b/gcc/config/arm/cortex-a9-neon.md
> > @@ -52,6 +52,7 @@
> > (const_string "neon_int_2")
> > (eq_attr "type" "neon_neg, neon_neg_q,\
> > neon_reduc_add, neon_reduc_add_q,\
> > + neon_pairwise_add, neon_pairwise_add_q,\
> > neon_reduc_add_long,\
> > neon_add_long, neon_sub_long")
> > (const_string "neon_int_3")
> > diff --git a/gcc/config/arm/types.md b/gcc/config/arm/types.md
> > index 253f496..ce1b6bc 100644
> > --- a/gcc/config/arm/types.md
> > +++ b/gcc/config/arm/types.md
> > @@ -274,6 +274,8 @@
> > ; neon_reduc_add_long
> > ; neon_reduc_add_acc
> > ; neon_reduc_add_acc_q
> > +; neon_pairwise_add
> > +; neon_pairwise_add_q
> > ; neon_reduc_minmax
> > ; neon_reduc_minmax_q
> > ; neon_logic
> > @@ -788,6 +790,8 @@
> > neon_reduc_add_long,\
> > neon_reduc_add_acc,\
> > neon_reduc_add_acc_q,\
> > + neon_pairwise_add,\
> > + neon_pairwise_add_q,\
> > neon_reduc_minmax,\
> > neon_reduc_minmax_q,\
> > neon_logic,\
> > @@ -1116,6 +1120,7 @@
> > neon_compare_q, neon_compare_zero, neon_compare_zero_q,\
> > neon_arith_acc, neon_arith_acc_q, neon_reduc_add,\
> > neon_reduc_add_q, neon_reduc_add_long, neon_reduc_add_acc,\
> > + neon_pairwise_add, neon_pairwise_add_q,\
> > neon_reduc_add_acc_q, neon_reduc_minmax, neon_reduc_minmax_q,\
> > neon_logic, neon_logic_q, neon_tst, neon_tst_q,\
> > neon_shift_imm, neon_shift_imm_q, neon_shift_imm_narrow_q,\
> > diff --git a/gcc/config/arm/xgene1.md b/gcc/config/arm/xgene1.md
> > index 34a13f4..1f68305 100644
> > --- a/gcc/config/arm/xgene1.md
> > +++ b/gcc/config/arm/xgene1.md
> > @@ -288,6 +288,8 @@
> > neon_arith_acc_q,\
> > neon_reduc_add,\
> > neon_reduc_add_q,\
> > + neon_pairwise_add,\
> > + neon_pairwise_add_q,\
> > neon_add_halve,\
> > neon_add_halve_q,\
> > neon_sub_halve,\
>
More information about the Gcc-patches
mailing list