From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from foss.arm.com (foss.arm.com [217.140.110.172]) by sourceware.org (Postfix) with ESMTP id 1A6043858D28 for ; Tue, 27 Jun 2023 18:35:35 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 1A6043858D28 Authentication-Results: sourceware.org; dmarc=pass (p=none dis=none) header.from=arm.com Authentication-Results: sourceware.org; spf=pass smtp.mailfrom=arm.com Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.121.207.14]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id CF2082F4; Tue, 27 Jun 2023 11:36:18 -0700 (PDT) Received: from localhost (e121540-lin.manchester.arm.com [10.32.110.72]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPSA id 434AC3F64C; Tue, 27 Jun 2023 11:35:34 -0700 (PDT) From: Richard Sandiford To: Prathamesh Kulkarni Mail-Followup-To: Prathamesh Kulkarni ,gcc Patches , richard.sandiford@arm.com Cc: gcc Patches Subject: Re: [SVE] Fold svdupq to VEC_PERM_EXPR if elements are not constant References: Date: Tue, 27 Jun 2023 19:35:33 +0100 In-Reply-To: (Prathamesh Kulkarni's message of "Tue, 27 Jun 2023 17:31:18 +0530") Message-ID: User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/26.3 (gnu/linux) MIME-Version: 1.0 Content-Type: text/plain X-Spam-Status: No, score=-27.3 required=5.0 tests=BAYES_00,GIT_PATCH_0,KAM_DMARC_NONE,KAM_DMARC_STATUS,KAM_LAZY_DOMAIN_SECURITY,KAM_SHORT,SPF_HELO_NONE,SPF_NONE,TXREP,T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org List-Id: Prathamesh Kulkarni writes: > Hi Richard, > Sorry I forgot to commit this patch, which you had approved in: > https://gcc.gnu.org/pipermail/gcc-patches/2023-April/615308.html > > Just for context for the following test: > svint32_t f_s32(int32x4_t x) > { > return svdupq_s32 (x[0], x[1], x[2], x[3]); > } > > -O3 -mcpu=generic+sve generates following code after interleave+zip1 patch: > f_s32: > dup s31, v0.s[1] > mov v30.8b, v0.8b > ins v31.s[1], v0.s[3] > ins v30.s[1], v0.s[2] > zip1 v0.4s, v30.4s, v31.4s > dup z0.q, z0.q[0] > ret > > Code-gen with attached patch: > f_s32: > dup z0.q, z0.q[0] > ret > > Bootstrapped+tested on aarch64-linux-gnu. > OK to commit ? > > Thanks, > Prathamesh > > [SVE] Fold svdupq to VEC_PERM_EXPR if elements are not constant. > > gcc/ChangeLog: > * config/aarch64/aarch64-sve-builtins-base.cc > (svdupq_impl::fold_nonconst_dupq): New method. > (svdupq_impl::fold): Call fold_nonconst_dupq. > > gcc/testsuite/ChangeLog: > * gcc.target/aarch64/sve/acle/general/dupq_11.c: New test. OK, thanks. Richard > diff --git a/gcc/config/aarch64/aarch64-sve-builtins-base.cc b/gcc/config/aarch64/aarch64-sve-builtins-base.cc > index 95b4cb8a943..9010ecca6da 100644 > --- a/gcc/config/aarch64/aarch64-sve-builtins-base.cc > +++ b/gcc/config/aarch64/aarch64-sve-builtins-base.cc > @@ -817,6 +817,52 @@ public: > > class svdupq_impl : public quiet > { > +private: > + gimple * > + fold_nonconst_dupq (gimple_folder &f) const > + { > + /* Lower lhs = svdupq (arg0, arg1, ..., argN} into: > + tmp = {arg0, arg1, ..., arg} > + lhs = VEC_PERM_EXPR (tmp, tmp, {0, 1, 2, N-1, ...}) */ > + > + if (f.type_suffix (0).bool_p > + || BYTES_BIG_ENDIAN) > + return NULL; > + > + tree lhs = gimple_call_lhs (f.call); > + tree lhs_type = TREE_TYPE (lhs); > + tree elt_type = TREE_TYPE (lhs_type); > + scalar_mode elt_mode = SCALAR_TYPE_MODE (elt_type); > + machine_mode vq_mode = aarch64_vq_mode (elt_mode).require (); > + tree vq_type = build_vector_type_for_mode (elt_type, vq_mode); > + > + unsigned nargs = gimple_call_num_args (f.call); > + vec *v; > + vec_alloc (v, nargs); > + for (unsigned i = 0; i < nargs; i++) > + CONSTRUCTOR_APPEND_ELT (v, NULL_TREE, gimple_call_arg (f.call, i)); > + tree vec = build_constructor (vq_type, v); > + tree tmp = make_ssa_name_fn (cfun, vq_type, 0); > + gimple *g = gimple_build_assign (tmp, vec); > + > + gimple_seq stmts = NULL; > + gimple_seq_add_stmt_without_update (&stmts, g); > + > + poly_uint64 lhs_len = TYPE_VECTOR_SUBPARTS (lhs_type); > + vec_perm_builder sel (lhs_len, nargs, 1); > + for (unsigned i = 0; i < nargs; i++) > + sel.quick_push (i); > + > + vec_perm_indices indices (sel, 1, nargs); > + tree mask_type = build_vector_type (ssizetype, lhs_len); > + tree mask = vec_perm_indices_to_tree (mask_type, indices); > + > + gimple *g2 = gimple_build_assign (lhs, VEC_PERM_EXPR, tmp, tmp, mask); > + gimple_seq_add_stmt_without_update (&stmts, g2); > + gsi_replace_with_seq (f.gsi, stmts, false); > + return g2; > + } > + > public: > gimple * > fold (gimple_folder &f) const override > @@ -832,7 +878,7 @@ public: > { > tree elt = gimple_call_arg (f.call, i); > if (!CONSTANT_CLASS_P (elt)) > - return NULL; > + return fold_nonconst_dupq (f); > builder.quick_push (elt); > for (unsigned int j = 1; j < factor; ++j) > builder.quick_push (build_zero_cst (TREE_TYPE (vec_type))); > diff --git a/gcc/testsuite/gcc.target/aarch64/sve/acle/general/dupq_11.c b/gcc/testsuite/gcc.target/aarch64/sve/acle/general/dupq_11.c > new file mode 100644 > index 00000000000..f19f8deb1e5 > --- /dev/null > +++ b/gcc/testsuite/gcc.target/aarch64/sve/acle/general/dupq_11.c > @@ -0,0 +1,31 @@ > +/* { dg-do compile } */ > +/* { dg-options "-O3 -fdump-tree-optimized" } */ > + > +#include > +#include > + > +svint8_t f_s8(int8x16_t x) > +{ > + return svdupq_s8 (x[0], x[1], x[2], x[3], x[4], x[5], x[6], x[7], > + x[8], x[9], x[10], x[11], x[12], x[13], x[14], x[15]); > +} > + > +svint16_t f_s16(int16x8_t x) > +{ > + return svdupq_s16 (x[0], x[1], x[2], x[3], x[4], x[5], x[6], x[7]); > +} > + > +svint32_t f_s32(int32x4_t x) > +{ > + return svdupq_s32 (x[0], x[1], x[2], x[3]); > +} > + > +svint64_t f_s64(int64x2_t x) > +{ > + return svdupq_s64 (x[0], x[1]); > +} > + > +/* { dg-final { scan-tree-dump "VEC_PERM_EXPR" "optimized" } } */ > +/* { dg-final { scan-tree-dump-not "svdupq" "optimized" } } */ > + > +/* { dg-final { scan-assembler-times {\tdup\tz[0-9]+\.q, z[0-9]+\.q\[0\]\n} 4 } } */