From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-pj1-x1031.google.com (mail-pj1-x1031.google.com [IPv6:2607:f8b0:4864:20::1031]) by sourceware.org (Postfix) with ESMTPS id 2B1593858D39 for ; Wed, 29 Dec 2021 21:25:02 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 2B1593858D39 Received: by mail-pj1-x1031.google.com with SMTP id a11-20020a17090a854b00b001b11aae38d6so20962087pjw.2 for ; Wed, 29 Dec 2021 13:25:02 -0800 (PST) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:date:from:to:cc:subject:message-id:references :mime-version:content-disposition:in-reply-to; bh=Oz6dYh4B6HAVVCJSqn1WIFSS6CSdvhD+Bpq97Wlz7b8=; b=q5aO2Ziv6jzRLxc+Q6gS/VeLALQRX5oYUDpPD7NrdoQUipISNep7txCjSkW3sCe+mO GXm8JU3yUW7NangrUhDz3TCiOpscHIvjIHbv+Vt+OQY38kmhbpkylyo/45N4WH5tU/9y kxqKdDARklFLPil+aO32akx6qx8zW46bjlfI5LRHFBuvuHzVMufo6pt/qKEWRb+M68zK Lozme612k3U7ttyfFFzAsfBbqM/TTHoD05nM06K93AnHdUDCBNFrJmJm6oNMmixUeSYj 9WfyZN62qeabE48sKZAbyp1OcW7XddIg5l0gsePYOrJBixmbqJJUGX1SDJRBFhleO2Ht YZ7g== X-Gm-Message-State: AOAM533Zpz1SlHYQ7qH30KoaOk7esAUcCQlok1/T41DYPCTyK87D1FZD b6TjKSjtN4AE6mBZw3APhsU= X-Google-Smtp-Source: ABdhPJwa+0JIiyyMBmXr5wpSLduWt6l3xUQxLt6uKrAeFrxtMz3HRNbD/nSMNcedeAT0f+JC6uzxrA== X-Received: by 2002:a17:90b:3143:: with SMTP id ip3mr33977915pjb.58.1640813100147; Wed, 29 Dec 2021 13:25:00 -0800 (PST) Received: from gnu-tgl-3.localdomain ([172.58.35.133]) by smtp.gmail.com with ESMTPSA id u3sm26009394pfk.82.2021.12.29.13.24.59 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Wed, 29 Dec 2021 13:24:59 -0800 (PST) Received: by gnu-tgl-3.localdomain (Postfix, from userid 1000) id 569C1C08AD; Wed, 29 Dec 2021 13:24:58 -0800 (PST) Date: Wed, 29 Dec 2021 13:24:58 -0800 From: "H.J. Lu" To: Sunil K Pandey Cc: libc-alpha@sourceware.org, hjl.tools@gmail.com, andrey.kolesov@intel.com, marius.cornea@intel.com Subject: Re: [PATCH v5 03/18] x86-64: Add vector hypot/hypotf implementation to libmvec Message-ID: References: <20211229064000.1465621-1-skpgkp2@gmail.com> <20211229064000.1465621-4-skpgkp2@gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20211229064000.1465621-4-skpgkp2@gmail.com> X-Spam-Status: No, score=-3028.1 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, FREEMAIL_FROM, GIT_PATCH_0, KAM_SHORT, KAM_STOCKGEN, RCVD_IN_BARRACUDACENTRAL, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Wed, 29 Dec 2021 21:25:07 -0000 On Tue, Dec 28, 2021 at 10:39:45PM -0800, Sunil K Pandey wrote: > Implement vectorized hypot/hypotf containing SSE, AVX, AVX2 and > AVX512 versions for libmvec as per vector ABI. It also contains > accuracy and ABI tests for vector hypot/hypotf with regenerated ulps. > --- > bits/libm-simd-decl-stubs.h | 11 + > math/bits/mathcalls.h | 2 +- > .../unix/sysv/linux/x86_64/libmvec.abilist | 8 + > sysdeps/x86/fpu/bits/math-vector.h | 4 + > .../x86/fpu/finclude/math-vector-fortran.h | 4 + > sysdeps/x86_64/fpu/Makeconfig | 1 + > sysdeps/x86_64/fpu/Versions | 2 + > sysdeps/x86_64/fpu/libm-test-ulps | 20 ++ > .../fpu/multiarch/svml_d_hypot2_core-sse2.S | 20 ++ > .../x86_64/fpu/multiarch/svml_d_hypot2_core.c | 28 ++ > .../fpu/multiarch/svml_d_hypot2_core_sse4.S | 279 +++++++++++++++++ > .../fpu/multiarch/svml_d_hypot4_core-sse.S | 20 ++ > .../x86_64/fpu/multiarch/svml_d_hypot4_core.c | 28 ++ > .../fpu/multiarch/svml_d_hypot4_core_avx2.S | 289 ++++++++++++++++++ > .../fpu/multiarch/svml_d_hypot8_core-avx2.S | 20 ++ > .../x86_64/fpu/multiarch/svml_d_hypot8_core.c | 28 ++ > .../fpu/multiarch/svml_d_hypot8_core_avx512.S | 235 ++++++++++++++ > .../fpu/multiarch/svml_s_hypotf16_core-avx2.S | 20 ++ > .../fpu/multiarch/svml_s_hypotf16_core.c | 28 ++ > .../multiarch/svml_s_hypotf16_core_avx512.S | 239 +++++++++++++++ > .../fpu/multiarch/svml_s_hypotf4_core-sse2.S | 20 ++ > .../fpu/multiarch/svml_s_hypotf4_core.c | 28 ++ > .../fpu/multiarch/svml_s_hypotf4_core_sse4.S | 265 ++++++++++++++++ > .../fpu/multiarch/svml_s_hypotf8_core-sse.S | 20 ++ > .../fpu/multiarch/svml_s_hypotf8_core.c | 28 ++ > .../fpu/multiarch/svml_s_hypotf8_core_avx2.S | 269 ++++++++++++++++ > sysdeps/x86_64/fpu/svml_d_hypot2_core.S | 29 ++ > sysdeps/x86_64/fpu/svml_d_hypot4_core.S | 29 ++ > sysdeps/x86_64/fpu/svml_d_hypot4_core_avx.S | 25 ++ > sysdeps/x86_64/fpu/svml_d_hypot8_core.S | 25 ++ > sysdeps/x86_64/fpu/svml_s_hypotf16_core.S | 25 ++ > sysdeps/x86_64/fpu/svml_s_hypotf4_core.S | 29 ++ > sysdeps/x86_64/fpu/svml_s_hypotf8_core.S | 29 ++ > sysdeps/x86_64/fpu/svml_s_hypotf8_core_avx.S | 25 ++ > .../fpu/test-double-libmvec-hypot-avx.c | 1 + > .../fpu/test-double-libmvec-hypot-avx2.c | 1 + > .../fpu/test-double-libmvec-hypot-avx512f.c | 1 + > .../x86_64/fpu/test-double-libmvec-hypot.c | 3 + > .../x86_64/fpu/test-double-vlen2-wrappers.c | 1 + > .../fpu/test-double-vlen4-avx2-wrappers.c | 1 + > .../x86_64/fpu/test-double-vlen4-wrappers.c | 1 + > .../x86_64/fpu/test-double-vlen8-wrappers.c | 1 + > .../fpu/test-float-libmvec-hypotf-avx.c | 1 + > .../fpu/test-float-libmvec-hypotf-avx2.c | 1 + > .../fpu/test-float-libmvec-hypotf-avx512f.c | 1 + > .../x86_64/fpu/test-float-libmvec-hypotf.c | 3 + > .../x86_64/fpu/test-float-vlen16-wrappers.c | 1 + > .../x86_64/fpu/test-float-vlen4-wrappers.c | 1 + > .../fpu/test-float-vlen8-avx2-wrappers.c | 1 + > .../x86_64/fpu/test-float-vlen8-wrappers.c | 1 + > 50 files changed, 2151 insertions(+), 1 deletion(-) > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_hypot2_core-sse2.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_hypot2_core.c > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_hypot2_core_sse4.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_hypot4_core-sse.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_hypot4_core.c > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_hypot4_core_avx2.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_hypot8_core-avx2.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_hypot8_core.c > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_hypot8_core_avx512.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_hypotf16_core-avx2.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_hypotf16_core.c > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_hypotf16_core_avx512.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core-sse2.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core.c > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core_sse4.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_hypotf8_core-sse.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_hypotf8_core.c > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_hypotf8_core_avx2.S > create mode 100644 sysdeps/x86_64/fpu/svml_d_hypot2_core.S > create mode 100644 sysdeps/x86_64/fpu/svml_d_hypot4_core.S > create mode 100644 sysdeps/x86_64/fpu/svml_d_hypot4_core_avx.S > create mode 100644 sysdeps/x86_64/fpu/svml_d_hypot8_core.S > create mode 100644 sysdeps/x86_64/fpu/svml_s_hypotf16_core.S > create mode 100644 sysdeps/x86_64/fpu/svml_s_hypotf4_core.S > create mode 100644 sysdeps/x86_64/fpu/svml_s_hypotf8_core.S > create mode 100644 sysdeps/x86_64/fpu/svml_s_hypotf8_core_avx.S > create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-hypot-avx.c > create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-hypot-avx2.c > create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-hypot-avx512f.c > create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-hypot.c > create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-hypotf-avx.c > create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-hypotf-avx2.c > create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-hypotf-avx512f.c > create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-hypotf.c > > diff --git a/bits/libm-simd-decl-stubs.h b/bits/libm-simd-decl-stubs.h > index ae8ee882d0..adf65f6bc2 100644 > --- a/bits/libm-simd-decl-stubs.h > +++ b/bits/libm-simd-decl-stubs.h > @@ -131,4 +131,15 @@ > #define __DECL_SIMD_asinf32x > #define __DECL_SIMD_asinf64x > #define __DECL_SIMD_asinf128x > + > +#define __DECL_SIMD_hypot > +#define __DECL_SIMD_hypotf > +#define __DECL_SIMD_hypotl > +#define __DECL_SIMD_hypotf16 > +#define __DECL_SIMD_hypotf32 > +#define __DECL_SIMD_hypotf64 > +#define __DECL_SIMD_hypotf128 > +#define __DECL_SIMD_hypotf32x > +#define __DECL_SIMD_hypotf64x > +#define __DECL_SIMD_hypotf128x > #endif > diff --git a/math/bits/mathcalls.h b/math/bits/mathcalls.h > index bb53b7021e..2ed820a0dc 100644 > --- a/math/bits/mathcalls.h > +++ b/math/bits/mathcalls.h > @@ -144,7 +144,7 @@ __MATHCALL (sqrt,, (_Mdouble_ __x)); > > #if defined __USE_XOPEN || defined __USE_ISOC99 > /* Return `sqrt(X*X + Y*Y)'. */ > -__MATHCALL (hypot,, (_Mdouble_ __x, _Mdouble_ __y)); > +__MATHCALL_VEC (hypot,, (_Mdouble_ __x, _Mdouble_ __y)); > #endif > > #if defined __USE_XOPEN_EXTENDED || defined __USE_ISOC99 > diff --git a/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist b/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist > index ab03a07f92..12bb03245b 100644 > --- a/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist > +++ b/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist > @@ -49,24 +49,32 @@ GLIBC_2.22 _ZGVeN8vvv_sincos F > GLIBC_2.35 _ZGVbN2v_acos F > GLIBC_2.35 _ZGVbN2v_asin F > GLIBC_2.35 _ZGVbN2v_atan F > +GLIBC_2.35 _ZGVbN2vv_hypot F > GLIBC_2.35 _ZGVbN4v_acosf F > GLIBC_2.35 _ZGVbN4v_asinf F > GLIBC_2.35 _ZGVbN4v_atanf F > +GLIBC_2.35 _ZGVbN4vv_hypotf F > GLIBC_2.35 _ZGVcN4v_acos F > GLIBC_2.35 _ZGVcN4v_asin F > GLIBC_2.35 _ZGVcN4v_atan F > +GLIBC_2.35 _ZGVcN4vv_hypot F > GLIBC_2.35 _ZGVcN8v_acosf F > GLIBC_2.35 _ZGVcN8v_asinf F > GLIBC_2.35 _ZGVcN8v_atanf F > +GLIBC_2.35 _ZGVcN8vv_hypotf F > GLIBC_2.35 _ZGVdN4v_acos F > GLIBC_2.35 _ZGVdN4v_asin F > GLIBC_2.35 _ZGVdN4v_atan F > +GLIBC_2.35 _ZGVdN4vv_hypot F > GLIBC_2.35 _ZGVdN8v_acosf F > GLIBC_2.35 _ZGVdN8v_asinf F > GLIBC_2.35 _ZGVdN8v_atanf F > +GLIBC_2.35 _ZGVdN8vv_hypotf F > GLIBC_2.35 _ZGVeN16v_acosf F > GLIBC_2.35 _ZGVeN16v_asinf F > GLIBC_2.35 _ZGVeN16v_atanf F > +GLIBC_2.35 _ZGVeN16vv_hypotf F > GLIBC_2.35 _ZGVeN8v_acos F > GLIBC_2.35 _ZGVeN8v_asin F > GLIBC_2.35 _ZGVeN8v_atan F > +GLIBC_2.35 _ZGVeN8vv_hypot F > diff --git a/sysdeps/x86/fpu/bits/math-vector.h b/sysdeps/x86/fpu/bits/math-vector.h > index 73cb8849ff..437977c5fd 100644 > --- a/sysdeps/x86/fpu/bits/math-vector.h > +++ b/sysdeps/x86/fpu/bits/math-vector.h > @@ -70,6 +70,10 @@ > # define __DECL_SIMD_asin __DECL_SIMD_x86_64 > # undef __DECL_SIMD_asinf > # define __DECL_SIMD_asinf __DECL_SIMD_x86_64 > +# undef __DECL_SIMD_hypot > +# define __DECL_SIMD_hypot __DECL_SIMD_x86_64 > +# undef __DECL_SIMD_hypotf > +# define __DECL_SIMD_hypotf __DECL_SIMD_x86_64 > > # endif > #endif > diff --git a/sysdeps/x86/fpu/finclude/math-vector-fortran.h b/sysdeps/x86/fpu/finclude/math-vector-fortran.h > index 4552c2bdfa..cda31479a6 100644 > --- a/sysdeps/x86/fpu/finclude/math-vector-fortran.h > +++ b/sysdeps/x86/fpu/finclude/math-vector-fortran.h > @@ -34,6 +34,8 @@ > !GCC$ builtin (atanf) attributes simd (notinbranch) if('x86_64') > !GCC$ builtin (asin) attributes simd (notinbranch) if('x86_64') > !GCC$ builtin (asinf) attributes simd (notinbranch) if('x86_64') > +!GCC$ builtin (hypot) attributes simd (notinbranch) if('x86_64') > +!GCC$ builtin (hypotf) attributes simd (notinbranch) if('x86_64') > > !GCC$ builtin (cos) attributes simd (notinbranch) if('x32') > !GCC$ builtin (cosf) attributes simd (notinbranch) if('x32') > @@ -53,3 +55,5 @@ > !GCC$ builtin (atanf) attributes simd (notinbranch) if('x32') > !GCC$ builtin (asin) attributes simd (notinbranch) if('x32') > !GCC$ builtin (asinf) attributes simd (notinbranch) if('x32') > +!GCC$ builtin (hypot) attributes simd (notinbranch) if('x32') > +!GCC$ builtin (hypotf) attributes simd (notinbranch) if('x32') > diff --git a/sysdeps/x86_64/fpu/Makeconfig b/sysdeps/x86_64/fpu/Makeconfig > index e0eae0b196..7769a02731 100644 > --- a/sysdeps/x86_64/fpu/Makeconfig > +++ b/sysdeps/x86_64/fpu/Makeconfig > @@ -27,6 +27,7 @@ libmvec-funcs = \ > atan \ > cos \ > exp \ > + hypot \ > log \ > pow \ > sin \ > diff --git a/sysdeps/x86_64/fpu/Versions b/sysdeps/x86_64/fpu/Versions > index 10baf869a5..e359e5dc2c 100644 > --- a/sysdeps/x86_64/fpu/Versions > +++ b/sysdeps/x86_64/fpu/Versions > @@ -17,8 +17,10 @@ libmvec { > _ZGVbN2v_acos; _ZGVcN4v_acos; _ZGVdN4v_acos; _ZGVeN8v_acos; > _ZGVbN2v_asin; _ZGVcN4v_asin; _ZGVdN4v_asin; _ZGVeN8v_asin; > _ZGVbN2v_atan; _ZGVcN4v_atan; _ZGVdN4v_atan; _ZGVeN8v_atan; > + _ZGVbN2vv_hypot; _ZGVcN4vv_hypot; _ZGVdN4vv_hypot; _ZGVeN8vv_hypot; > _ZGVbN4v_acosf; _ZGVcN8v_acosf; _ZGVdN8v_acosf; _ZGVeN16v_acosf; > _ZGVbN4v_asinf; _ZGVcN8v_asinf; _ZGVdN8v_asinf; _ZGVeN16v_asinf; > _ZGVbN4v_atanf; _ZGVcN8v_atanf; _ZGVdN8v_atanf; _ZGVeN16v_atanf; > + _ZGVbN4vv_hypotf; _ZGVcN8vv_hypotf; _ZGVdN8vv_hypotf; _ZGVeN16vv_hypotf; > } > } > diff --git a/sysdeps/x86_64/fpu/libm-test-ulps b/sysdeps/x86_64/fpu/libm-test-ulps > index ea0f833381..a7513ec94e 100644 > --- a/sysdeps/x86_64/fpu/libm-test-ulps > +++ b/sysdeps/x86_64/fpu/libm-test-ulps > @@ -1375,6 +1375,26 @@ double: 1 > float128: 1 > ldouble: 1 > > +Function: "hypot_vlen16": > +float: 1 > + > +Function: "hypot_vlen2": > +double: 1 > + > +Function: "hypot_vlen4": > +double: 1 > +float: 1 > + > +Function: "hypot_vlen4_avx2": > +double: 1 > + > +Function: "hypot_vlen8": > +double: 1 > +float: 1 > + > +Function: "hypot_vlen8_avx2": > +float: 1 > + > Function: "j0": > double: 3 > float: 9 > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_hypot2_core-sse2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot2_core-sse2.S > new file mode 100644 > index 0000000000..237e38459e > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot2_core-sse2.S > @@ -0,0 +1,20 @@ > +/* SSE2 version of vectorized hypot. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#define _ZGVbN2vv_hypot _ZGVbN2vv_hypot_sse2 > +#include "../svml_d_hypot2_core.S" > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_hypot2_core.c b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot2_core.c > new file mode 100644 > index 0000000000..3f0865f05d > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot2_core.c > @@ -0,0 +1,28 @@ > +/* Multiple versions of vectorized hypot, vector length is 2. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#define SYMBOL_NAME _ZGVbN2vv_hypot > +#include "ifunc-mathvec-sse4_1.h" > + > +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); > + > +#ifdef SHARED > +__hidden_ver1 (_ZGVbN2vv_hypot, __GI__ZGVbN2vv_hypot, > + __redirect__ZGVbN2vv_hypot) > + __attribute__ ((visibility ("hidden"))); > +#endif > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_hypot2_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot2_core_sse4.S > new file mode 100644 > index 0000000000..931f34e5f2 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot2_core_sse4.S > @@ -0,0 +1,279 @@ > +/* Function hypot vectorized with SSE4. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + https://www.gnu.org/licenses/. */ > + > +/* > + * ALGORITHM DESCRIPTION: > + * > + * HIGH LEVEL OVERVIEW > + * > + * Calculate z = (x*x+y*y) > + * Calculate reciplicle sqrt (z) > + * Calculate error = z*(rsqrt(z)*rsqrt(z)) - 1 > + * Calculate fixing part p with polynom > + * Fix answer with sqrt(z) = z * rsqrt(z) + error * p * z > + * > + * ALGORITHM DETAILS > + * > + * Multiprecision branch for _HA_ only > + * Remove sigm from both arguments > + * Find maximum (_x) and minimum (_y) (by abs value) between arguments > + * Split _x int _a and _b for multiprecision > + * If _x >> _y we will we will not split _y for multiprecision > + * all _y will be put into lower part (_d) and higher part (_c = 0) > + * Fixing _hilo_mask for the case _x >> _y > + * Split _y into _c and _d for multiprecision with fixed mask > + * > + * compute Hi and Lo parts of _z = _x*_x + _y*_y > + * > + * _zHi = _a*_a + _c*_c > + * _zLo = (_x + _a)*_b + _d*_y + _d*_c > + * _z = _zHi + _zLo > + * > + * No multiprecision branch for _LA_ and _EP_ > + * _z = _VARG1 * _VARG1 + _VARG2 * _VARG2 > + * > + * Check _z exponent to be withing borders [3BC ; 441] else goto Callout > + * > + * _s ~ 1.0/sqrt(_z) > + * _s2 ~ 1.0/(sqrt(_z)*sqrt(_z)) ~ 1.0/_z = (1.0/_z + O) > + * _e[rror] = (1.0/_z + O) * _z - 1.0 > + * calculate fixing part _p > + * _p = (((_POLY_C5*_e + _POLY_C4)*_e +_POLY_C3)*_e +_POLY_C2)*_e + _POLY_C1 > + * some parts of polynom are skipped for lower flav > + * > + * result = _z * (1.0/sqrt(_z) + O) + _p * _e[rror] * _z > + * > + * > + */ > + > +/* Offsets for data table __svml_dhypot_data_internal > + */ > +#define _dHiLoMask 0 > +#define _dAbsMask 16 > +#define _dOne 32 > +#define _POLY_C5 48 > +#define _POLY_C4 64 > +#define _POLY_C3 80 > +#define _POLY_C2 96 > +#define _POLY_C1 112 > +#define _LowBoundary 128 > +#define _HighBoundary 144 > + > +#include > + > + .text > + .section .text.sse4,"ax",@progbits > +ENTRY(_ZGVbN2vv_hypot_sse4) > + subq $88, %rsp > + cfi_def_cfa_offset(96) > + > +/* > + * Defines > + * Implementation > + * Multiprecision branch for _HA_ only > + * _z = _VARG1 * _VARG1 + _VARG2 * _VARG2 > + */ > + movaps %xmm0, %xmm10 > + movaps %xmm1, %xmm2 > + mulpd %xmm0, %xmm10 > + mulpd %xmm1, %xmm2 > + addpd %xmm2, %xmm10 > + > +/* > + * _s ~ 1.0/sqrt(_z) > + * _s2 ~ 1.0/(sqrt(_z)*sqrt(_z)) ~ 1.0/_z > + */ > + cvtpd2ps %xmm10, %xmm7 > + movlhps %xmm7, %xmm7 > + rsqrtps %xmm7, %xmm8 > + cvtps2pd %xmm8, %xmm11 > + movaps %xmm11, %xmm2 > + mulpd %xmm11, %xmm2 > + > +/* _e[rror] ~ (1.0/_z + O) * _z - 1.0 */ > + mulpd %xmm10, %xmm2 > + subpd _dOne+__svml_dhypot_data_internal(%rip), %xmm2 > + > +/* > + * calculate fixing part _p > + * _p = (((_POLY_C5*_e + _POLY_C4)*_e +_POLY_C3)*_e +_POLY_C2)*_e + _POLY_C1 > + * some parts of polynom are skipped for lower flav > + */ > + movups _POLY_C4+__svml_dhypot_data_internal(%rip), %xmm9 > + mulpd %xmm2, %xmm9 > + addpd _POLY_C3+__svml_dhypot_data_internal(%rip), %xmm9 > + mulpd %xmm2, %xmm9 > + addpd _POLY_C2+__svml_dhypot_data_internal(%rip), %xmm9 > + mulpd %xmm2, %xmm9 > + addpd _POLY_C1+__svml_dhypot_data_internal(%rip), %xmm9 > + > +/* result = _z * (1.0/sqrt(_z) + O) + _p * _e[rror] * _z */ > + mulpd %xmm9, %xmm2 > + mulpd %xmm11, %xmm2 > + mulpd %xmm10, %xmm11 > + mulpd %xmm10, %xmm2 > + > +/* Check _z exponent to be withing borders [3BC ; 441] else goto Callout */ > + movq _LowBoundary+__svml_dhypot_data_internal(%rip), %xmm5 > + movq _HighBoundary+__svml_dhypot_data_internal(%rip), %xmm3 > + pshufd $221, %xmm10, %xmm4 > + pcmpgtd %xmm4, %xmm5 > + pcmpgtd %xmm3, %xmm4 > + por %xmm4, %xmm5 > + pshufd $80, %xmm5, %xmm6 > + movmskpd %xmm6, %edx > + addpd %xmm11, %xmm2 > + > +/* The end of implementation */ > + testl %edx, %edx > + > +/* Go to special inputs processing branch */ > + jne L(SPECIAL_VALUES_BRANCH) > + # LOE rbx rbp r12 r13 r14 r15 edx xmm0 xmm1 xmm2 > + > +/* Restore registers > + * and exit the function > + */ > + > +L(EXIT): > + movaps %xmm2, %xmm0 > + addq $88, %rsp > + cfi_def_cfa_offset(8) > + ret > + cfi_def_cfa_offset(96) > + > +/* Branch to process > + * special inputs > + */ > + > +L(SPECIAL_VALUES_BRANCH): > + movups %xmm0, 32(%rsp) > + movups %xmm1, 48(%rsp) > + movups %xmm2, 64(%rsp) > + # LOE rbx rbp r12 r13 r14 r15 edx > + > + xorl %eax, %eax > + movq %r12, 16(%rsp) > + cfi_offset(12, -80) > + movl %eax, %r12d > + movq %r13, 8(%rsp) > + cfi_offset(13, -88) > + movl %edx, %r13d > + movq %r14, (%rsp) > + cfi_offset(14, -96) > + # LOE rbx rbp r15 r12d r13d > + > +/* Range mask > + * bits check > + */ > + > +L(RANGEMASK_CHECK): > + btl %r12d, %r13d > + > +/* Call scalar math function */ > + jc L(SCALAR_MATH_CALL) > + # LOE rbx rbp r15 r12d r13d > + > +/* Special inputs > + * processing loop > + */ > + > +L(SPECIAL_VALUES_LOOP): > + incl %r12d > + cmpl $2, %r12d > + > +/* Check bits in range mask */ > + jl L(RANGEMASK_CHECK) > + # LOE rbx rbp r15 r12d r13d > + > + movq 16(%rsp), %r12 > + cfi_restore(12) > + movq 8(%rsp), %r13 > + cfi_restore(13) > + movq (%rsp), %r14 > + cfi_restore(14) > + movups 64(%rsp), %xmm2 > + > +/* Go to exit */ > + jmp L(EXIT) > + cfi_offset(12, -80) > + cfi_offset(13, -88) > + cfi_offset(14, -96) > + # LOE rbx rbp r12 r13 r14 r15 xmm2 > + > +/* Scalar math fucntion call > + * to process special input > + */ > + > +L(SCALAR_MATH_CALL): > + movl %r12d, %r14d > + movsd 32(%rsp,%r14,8), %xmm0 > + movsd 48(%rsp,%r14,8), %xmm1 > + call hypot@PLT > + # LOE rbx rbp r14 r15 r12d r13d xmm0 > + > + movsd %xmm0, 64(%rsp,%r14,8) > + > +/* Process special inputs in loop */ > + jmp L(SPECIAL_VALUES_LOOP) > + # LOE rbx rbp r15 r12d r13d > +END(_ZGVbN2vv_hypot_sse4) > + > + .section .rodata, "a" > + .align 16 > + > +#ifdef __svml_dhypot_data_internal_typedef > +typedef unsigned int VUINT32; > +typedef struct > +{ > + __declspec(align(16)) VUINT32 _dHiLoMask[2][2]; > + __declspec(align(16)) VUINT32 _dAbsMask[2][2]; > + __declspec(align(16)) VUINT32 _dOne[2][2]; > + __declspec(align(16)) VUINT32 _POLY_C5[2][2]; > + __declspec(align(16)) VUINT32 _POLY_C4[2][2]; > + __declspec(align(16)) VUINT32 _POLY_C3[2][2]; > + __declspec(align(16)) VUINT32 _POLY_C2[2][2]; > + __declspec(align(16)) VUINT32 _POLY_C1[2][2]; > + __declspec(align(16)) VUINT32 _LowBoundary[4][1]; > + __declspec(align(16)) VUINT32 _HighBoundary[4][1]; > +} __svml_dhypot_data_internal; > +#endif > +__svml_dhypot_data_internal: > + /* legacy algorithm */ > + .quad 0xffffc00000000000, 0xffffc00000000000 /* _dHiLoMask */ > + .align 16 > + .quad 0x7fffffffffffffff, 0x7fffffffffffffff /* _dAbsMask */ > + .align 16 > + .quad 0x3FF0000000000000, 0x3FF0000000000000 /* _dOne */ > + .align 16 > + .quad 0xBFCF800000000000, 0xBFCF800000000000 /* _POLY_C5 */ > + .align 16 > + .quad 0x3FD1800000000000, 0x3FD1800000000000 /* _POLY_C4 */ > + .align 16 > + .quad 0xBFD4000000000000, 0xBFD4000000000000 /* _POLY_C3 */ > + .align 16 > + .quad 0x3FD8000000000000, 0x3FD8000000000000 /* _POLY_C2 */ > + .align 16 > + .quad 0xBFE0000000000000, 0xBFE0000000000000 /* _POLY_C1 */ > + .align 16 > + .long 0x3BC00000, 0x3BC00000, 0x3BC00000, 0x3BC00000 /* _LowBoundary */ > + .align 16 > + .long 0x44100000, 0x44100000, 0x44100000, 0x44100000 /* _HighBoundary */ > + .align 16 > + .type __svml_dhypot_data_internal,@object > + .size __svml_dhypot_data_internal,.-__svml_dhypot_data_internal > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_hypot4_core-sse.S b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot4_core-sse.S > new file mode 100644 > index 0000000000..5e7c75c44c > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot4_core-sse.S > @@ -0,0 +1,20 @@ > +/* SSE version of vectorized hypot. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#define _ZGVdN4vv_hypot _ZGVdN4vv_hypot_sse_wrapper > +#include "../svml_d_hypot4_core.S" > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_hypot4_core.c b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot4_core.c > new file mode 100644 > index 0000000000..06f34d35e1 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot4_core.c > @@ -0,0 +1,28 @@ > +/* Multiple versions of vectorized hypot, vector length is 4. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#define SYMBOL_NAME _ZGVdN4vv_hypot > +#include "ifunc-mathvec-avx2.h" > + > +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); > + > +#ifdef SHARED > +__hidden_ver1 (_ZGVdN4vv_hypot, __GI__ZGVdN4vv_hypot, > + __redirect__ZGVdN4vv_hypot) > + __attribute__ ((visibility ("hidden"))); > +#endif > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_hypot4_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot4_core_avx2.S > new file mode 100644 > index 0000000000..45028ab7e9 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot4_core_avx2.S > @@ -0,0 +1,289 @@ > +/* Function hypot vectorized with AVX2. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + https://www.gnu.org/licenses/. */ > + > +/* > + * ALGORITHM DESCRIPTION: > + * > + * HIGH LEVEL OVERVIEW > + * > + * Calculate z = (x*x+y*y) > + * Calculate reciplicle sqrt (z) > + * Calculate error = z*(rsqrt(z)*rsqrt(z)) - 1 > + * Calculate fixing part p with polynom > + * Fix answer with sqrt(z) = z * rsqrt(z) + error * p * z > + * > + * ALGORITHM DETAILS > + * > + * Multiprecision branch for _HA_ only > + * Remove sigm from both arguments > + * Find maximum (_x) and minimum (_y) (by abs value) between arguments > + * Split _x int _a and _b for multiprecision > + * If _x >> _y we will we will not split _y for multiprecision > + * all _y will be put into lower part (_d) and higher part (_c = 0) > + * Fixing _hilo_mask for the case _x >> _y > + * Split _y into _c and _d for multiprecision with fixed mask > + * > + * compute Hi and Lo parts of _z = _x*_x + _y*_y > + * > + * _zHi = _a*_a + _c*_c > + * _zLo = (_x + _a)*_b + _d*_y + _d*_c > + * _z = _zHi + _zLo > + * > + * No multiprecision branch for _LA_ and _EP_ > + * _z = _VARG1 * _VARG1 + _VARG2 * _VARG2 > + * > + * Check _z exponent to be withing borders [3BC ; 441] else goto Callout > + * > + * _s ~ 1.0/sqrt(_z) > + * _s2 ~ 1.0/(sqrt(_z)*sqrt(_z)) ~ 1.0/_z = (1.0/_z + O) > + * _e[rror] = (1.0/_z + O) * _z - 1.0 > + * calculate fixing part _p > + * _p = (((_POLY_C5*_e + _POLY_C4)*_e +_POLY_C3)*_e +_POLY_C2)*_e + _POLY_C1 > + * some parts of polynom are skipped for lower flav > + * > + * result = _z * (1.0/sqrt(_z) + O) + _p * _e[rror] * _z > + * > + * > + */ > + > +/* Offsets for data table __svml_dhypot_data_internal > + */ > +#define _dHiLoMask 0 > +#define _dAbsMask 32 > +#define _dOne 64 > +#define _POLY_C5 96 > +#define _POLY_C4 128 > +#define _POLY_C3 160 > +#define _POLY_C2 192 > +#define _POLY_C1 224 > +#define _LowBoundary 256 > +#define _HighBoundary 288 > + > +#include > + > + .text > + .section .text.avx2,"ax",@progbits > +ENTRY(_ZGVdN4vv_hypot_avx2) > + pushq %rbp > + cfi_def_cfa_offset(16) > + movq %rsp, %rbp > + cfi_def_cfa(6, 16) > + cfi_offset(6, -16) > + andq $-32, %rsp > + subq $128, %rsp > + vmovapd %ymm1, %ymm2 > + vmovapd %ymm0, %ymm1 > + > +/* > + * Defines > + * Implementation > + * Multiprecision branch for _HA_ only > + * _z = _VARG1 * _VARG1 + _VARG2 * _VARG2 > + */ > + vmulpd %ymm1, %ymm1, %ymm0 > + > +/* > + * calculate fixing part _p > + * _p = (((_POLY_C5*_e + _POLY_C4)*_e +_POLY_C3)*_e +_POLY_C2)*_e + _POLY_C1 > + * some parts of polynom are skipped for lower flav > + */ > + vmovupd _POLY_C4+__svml_dhypot_data_internal(%rip), %ymm15 > + vmovups _LowBoundary+__svml_dhypot_data_internal(%rip), %xmm4 > + vfmadd231pd %ymm2, %ymm2, %ymm0 > + > +/* > + * _s ~ 1.0/sqrt(_z) > + * _s2 ~ 1.0/(sqrt(_z)*sqrt(_z)) ~ 1.0/_z > + */ > + vcvtpd2ps %ymm0, %xmm12 > + > +/* Check _z exponent to be withing borders [3BC ; 441] else goto Callout */ > + vextractf128 $1, %ymm0, %xmm3 > + vrsqrtps %xmm12, %xmm13 > + vshufps $221, %xmm3, %xmm0, %xmm5 > + vcvtps2pd %xmm13, %ymm3 > + vpcmpgtd %xmm5, %xmm4, %xmm6 > + vpcmpgtd _HighBoundary+__svml_dhypot_data_internal(%rip), %xmm5, %xmm7 > + vpor %xmm7, %xmm6, %xmm9 > + vpshufd $80, %xmm9, %xmm8 > + vmulpd %ymm3, %ymm3, %ymm14 > + vpshufd $250, %xmm9, %xmm10 > + > +/* _e[rror] ~ (1.0/_z + O) * _z - 1.0 */ > + vfmsub213pd _dOne+__svml_dhypot_data_internal(%rip), %ymm0, %ymm14 > + vfmadd213pd _POLY_C3+__svml_dhypot_data_internal(%rip), %ymm14, %ymm15 > + vfmadd213pd _POLY_C2+__svml_dhypot_data_internal(%rip), %ymm14, %ymm15 > + vfmadd213pd _POLY_C1+__svml_dhypot_data_internal(%rip), %ymm14, %ymm15 > + > +/* result = _z * (1.0/sqrt(_z) + O) + _p * _e[rror] * _z */ > + vmulpd %ymm15, %ymm14, %ymm14 > + vmulpd %ymm14, %ymm3, %ymm15 > + vmulpd %ymm15, %ymm0, %ymm4 > + vfmadd213pd %ymm4, %ymm3, %ymm0 > + vinsertf128 $1, %xmm10, %ymm8, %ymm11 > + vmovmskpd %ymm11, %edx > + > +/* The end of implementation */ > + testl %edx, %edx > + > +/* Go to special inputs processing branch */ > + jne L(SPECIAL_VALUES_BRANCH) > + # LOE rbx r12 r13 r14 r15 edx ymm0 ymm1 ymm2 > + > +/* Restore registers > + * and exit the function > + */ > + > +L(EXIT): > + movq %rbp, %rsp > + popq %rbp > + cfi_def_cfa(7, 8) > + cfi_restore(6) > + ret > + cfi_def_cfa(6, 16) > + cfi_offset(6, -16) > + > +/* Branch to process > + * special inputs > + */ > + > +L(SPECIAL_VALUES_BRANCH): > + vmovupd %ymm1, 32(%rsp) > + vmovupd %ymm2, 64(%rsp) > + vmovupd %ymm0, 96(%rsp) > + # LOE rbx r12 r13 r14 r15 edx ymm0 > + > + xorl %eax, %eax > + # LOE rbx r12 r13 r14 r15 eax edx > + > + vzeroupper > + movq %r12, 16(%rsp) > + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -112; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x90, 0xff, 0xff, 0xff, 0x22 > + movl %eax, %r12d > + movq %r13, 8(%rsp) > + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -120; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x88, 0xff, 0xff, 0xff, 0x22 > + movl %edx, %r13d > + movq %r14, (%rsp) > + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -128; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x80, 0xff, 0xff, 0xff, 0x22 > + # LOE rbx r15 r12d r13d > + > +/* Range mask > + * bits check > + */ > + > +L(RANGEMASK_CHECK): > + btl %r12d, %r13d > + > +/* Call scalar math function */ > + jc L(SCALAR_MATH_CALL) > + # LOE rbx r15 r12d r13d > + > +/* Special inputs > + * processing loop > + */ > + > +L(SPECIAL_VALUES_LOOP): > + incl %r12d > + cmpl $4, %r12d > + > +/* Check bits in range mask */ > + jl L(RANGEMASK_CHECK) > + # LOE rbx r15 r12d r13d > + > + movq 16(%rsp), %r12 > + cfi_restore(12) > + movq 8(%rsp), %r13 > + cfi_restore(13) > + movq (%rsp), %r14 > + cfi_restore(14) > + vmovupd 96(%rsp), %ymm0 > + > +/* Go to exit */ > + jmp L(EXIT) > + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -112; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x90, 0xff, 0xff, 0xff, 0x22 > + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -120; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x88, 0xff, 0xff, 0xff, 0x22 > + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -128; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x80, 0xff, 0xff, 0xff, 0x22 > + # LOE rbx r12 r13 r14 r15 ymm0 > + > +/* Scalar math fucntion call > + * to process special input > + */ > + > +L(SCALAR_MATH_CALL): > + movl %r12d, %r14d > + movsd 32(%rsp,%r14,8), %xmm0 > + movsd 64(%rsp,%r14,8), %xmm1 > + call hypot@PLT > + # LOE rbx r14 r15 r12d r13d xmm0 > + > + movsd %xmm0, 96(%rsp,%r14,8) > + > +/* Process special inputs in loop */ > + jmp L(SPECIAL_VALUES_LOOP) > + # LOE rbx r15 r12d r13d > +END(_ZGVdN4vv_hypot_avx2) > + > + .section .rodata, "a" > + .align 32 > + > +#ifdef __svml_dhypot_data_internal_typedef > +typedef unsigned int VUINT32; > +typedef struct > +{ > + __declspec(align(32)) VUINT32 _dHiLoMask[4][2]; > + __declspec(align(32)) VUINT32 _dAbsMask[4][2]; > + __declspec(align(32)) VUINT32 _dOne[4][2]; > + __declspec(align(32)) VUINT32 _POLY_C5[4][2]; > + __declspec(align(32)) VUINT32 _POLY_C4[4][2]; > + __declspec(align(32)) VUINT32 _POLY_C3[4][2]; > + __declspec(align(32)) VUINT32 _POLY_C2[4][2]; > + __declspec(align(32)) VUINT32 _POLY_C1[4][2]; > + __declspec(align(32)) VUINT32 _LowBoundary[8][1]; > + __declspec(align(32)) VUINT32 _HighBoundary[8][1]; > +} __svml_dhypot_data_internal; > +#endif > +__svml_dhypot_data_internal: > + /* legacy algorithm */ > + .quad 0xffffc00000000000, 0xffffc00000000000, 0xffffc00000000000, 0xffffc00000000000 /* _dHiLoMask */ > + .align 32 > + .quad 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff /* _dAbsMask */ > + .align 32 > + .quad 0x3FF0000000000000, 0x3FF0000000000000, 0x3FF0000000000000, 0x3FF0000000000000 /* _dOne */ > + .align 32 > + .quad 0xBFCF800000000000, 0xBFCF800000000000, 0xBFCF800000000000, 0xBFCF800000000000 /* _POLY_C5 */ > + .align 32 > + .quad 0x3FD1800000000000, 0x3FD1800000000000, 0x3FD1800000000000, 0x3FD1800000000000 /* _POLY_C4 */ > + .align 32 > + .quad 0xBFD4000000000000, 0xBFD4000000000000, 0xBFD4000000000000, 0xBFD4000000000000 /* _POLY_C3 */ > + .align 32 > + .quad 0x3FD8000000000000, 0x3FD8000000000000, 0x3FD8000000000000, 0x3FD8000000000000 /* _POLY_C2 */ > + .align 32 > + .quad 0xBFE0000000000000, 0xBFE0000000000000, 0xBFE0000000000000, 0xBFE0000000000000 /* _POLY_C1 */ > + .align 32 > + .long 0x3BC00000, 0x3BC00000, 0x3BC00000, 0x3BC00000, 0x3BC00000, 0x3BC00000, 0x3BC00000, 0x3BC00000 /* _LowBoundary */ > + .align 32 > + .long 0x44100000, 0x44100000, 0x44100000, 0x44100000, 0x44100000, 0x44100000, 0x44100000, 0x44100000 /* _HighBoundary */ > + .align 32 > + .type __svml_dhypot_data_internal,@object > + .size __svml_dhypot_data_internal,.-__svml_dhypot_data_internal > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_hypot8_core-avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot8_core-avx2.S > new file mode 100644 > index 0000000000..a53e82cf9a > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot8_core-avx2.S > @@ -0,0 +1,20 @@ > +/* AVX2 version of vectorized hypot. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#define _ZGVeN8vv_hypot _ZGVeN8vv_hypot_avx2_wrapper > +#include "../svml_d_hypot8_core.S" > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_hypot8_core.c b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot8_core.c > new file mode 100644 > index 0000000000..6052c752c9 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot8_core.c > @@ -0,0 +1,28 @@ > +/* Multiple versions of vectorized hypot, vector length is 8. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#define SYMBOL_NAME _ZGVeN8vv_hypot > +#include "ifunc-mathvec-avx512-skx.h" > + > +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); > + > +#ifdef SHARED > +__hidden_ver1 (_ZGVeN8vv_hypot, __GI__ZGVeN8vv_hypot, > + __redirect__ZGVeN8vv_hypot) > + __attribute__ ((visibility ("hidden"))); > +#endif > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_hypot8_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot8_core_avx512.S > new file mode 100644 > index 0000000000..1e5e716a8d > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_hypot8_core_avx512.S > @@ -0,0 +1,235 @@ > +/* Function hypot vectorized with AVX-512. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + https://www.gnu.org/licenses/. */ > + > +/* > + * ALGORITHM DESCRIPTION: > + * > + * HIGH LEVEL OVERVIEW > + * > + * Calculate z = (x*x+y*y) > + * Calculate reciplicle sqrt (z) > + * Calculate error = z*(rsqrt(z)*rsqrt(z)) - 1 > + * Calculate fixing part p with polynom > + * Fix answer with sqrt(z) = z * rsqrt(z) + error * p * z > + * > + * ALGORITHM DETAILS > + * > + * Multiprecision branch for _HA_ only > + * Remove sigm from both arguments > + * Find maximum (_x) and minimum (_y) (by abs value) between arguments > + * Split _x int _a and _b for multiprecision > + * If _x >> _y we will we will not split _y for multiprecision > + * all _y will be put into lower part (_d) and higher part (_c = 0) > + * Fixing _hilo_mask for the case _x >> _y > + * Split _y into _c and _d for multiprecision with fixed mask > + * > + * compute Hi and Lo parts of _z = _x*_x + _y*_y > + * > + * _zHi = _a*_a + _c*_c > + * _zLo = (_x + _a)*_b + _d*_y + _d*_c > + * _z = _zHi + _zLo > + * > + * No multiprecision branch for _LA_ and _EP_ > + * _z = _VARG1 * _VARG1 + _VARG2 * _VARG2 > + * > + * Check _z exponent to be withing borders [3BC ; 441] else goto Callout > + * > + * _s ~ 1.0/sqrt(_z) > + * _s2 ~ 1.0/(sqrt(_z)*sqrt(_z)) ~ 1.0/_z = (1.0/_z + O) > + * _e[rror] = (1.0/_z + O) * _z - 1.0 > + * calculate fixing part _p > + * _p = (((_POLY_C5*_e + _POLY_C4)*_e +_POLY_C3)*_e +_POLY_C2)*_e + _POLY_C1 > + * some parts of polynom are skipped for lower flav > + * > + * result = _z * (1.0/sqrt(_z) + O) + _p * _e[rror] * _z > + * > + * > + */ > + > +/* Offsets for data table __svml_dhypot_data_internal > + */ > +#define _dAbsMask 0 > +#define _lExpBound_uisa 64 > +#define _lExpBound 128 > +#define _dHalf 192 > + > +#include > + > + .text > + .section .text.evex512,"ax",@progbits > +ENTRY(_ZGVeN8vv_hypot_skx) > + pushq %rbp > + cfi_def_cfa_offset(16) > + movq %rsp, %rbp > + cfi_def_cfa(6, 16) > + cfi_offset(6, -16) > + andq $-64, %rsp > + subq $256, %rsp > + vgetexppd {sae}, %zmm0, %zmm2 > + vgetexppd {sae}, %zmm1, %zmm3 > + vmovups _dHalf+__svml_dhypot_data_internal(%rip), %zmm9 > + vmaxpd {sae}, %zmm3, %zmm2, %zmm4 > + vmulpd {rn-sae}, %zmm0, %zmm0, %zmm2 > + vandpd _dAbsMask+__svml_dhypot_data_internal(%rip), %zmm4, %zmm5 > + vfmadd231pd {rn-sae}, %zmm1, %zmm1, %zmm2 > + > +/* Select exponent bound so that no scaling is needed */ > + vpcmpq $5, _lExpBound_uisa+__svml_dhypot_data_internal(%rip), %zmm5, %k0 > + vrsqrt14pd %zmm2, %zmm6 > + kmovw %k0, %edx > + vmulpd {rn-sae}, %zmm6, %zmm2, %zmm7 > + vmulpd {rn-sae}, %zmm6, %zmm9, %zmm8 > + vfnmadd231pd {rn-sae}, %zmm7, %zmm8, %zmm9 > + vfmadd231pd {rn-sae}, %zmm9, %zmm8, %zmm8 > + vfmadd213pd {rn-sae}, %zmm7, %zmm7, %zmm9 > + vfnmadd231pd {rn-sae}, %zmm9, %zmm9, %zmm2 > + vfmadd213pd {rn-sae}, %zmm9, %zmm8, %zmm2 > + > +/* The end of implementation */ > + testl %edx, %edx > + > +/* Go to special inputs processing branch */ > + jne L(SPECIAL_VALUES_BRANCH) > + # LOE rbx r12 r13 r14 r15 edx zmm0 zmm1 zmm2 > + > +/* Restore registers > + * and exit the function > + */ > + > +L(EXIT): > + vmovaps %zmm2, %zmm0 > + movq %rbp, %rsp > + popq %rbp > + cfi_def_cfa(7, 8) > + cfi_restore(6) > + ret > + cfi_def_cfa(6, 16) > + cfi_offset(6, -16) > + > +/* Branch to process > + * special inputs > + */ > + > +L(SPECIAL_VALUES_BRANCH): > + vmovups %zmm0, 64(%rsp) > + vmovups %zmm1, 128(%rsp) > + vmovups %zmm2, 192(%rsp) > + # LOE rbx r12 r13 r14 r15 edx zmm2 > + > + xorl %eax, %eax > + # LOE rbx r12 r13 r14 r15 eax edx > + > + vzeroupper > + movq %r12, 16(%rsp) > + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -240; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x10, 0xff, 0xff, 0xff, 0x22 > + movl %eax, %r12d > + movq %r13, 8(%rsp) > + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -248; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x08, 0xff, 0xff, 0xff, 0x22 > + movl %edx, %r13d > + movq %r14, (%rsp) > + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -256; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x00, 0xff, 0xff, 0xff, 0x22 > + # LOE rbx r15 r12d r13d > + > +/* Range mask > + * bits check > + */ > + > +L(RANGEMASK_CHECK): > + btl %r12d, %r13d > + > +/* Call scalar math function */ > + jc L(SCALAR_MATH_CALL) > + # LOE rbx r15 r12d r13d > + > +/* Special inputs > + * processing loop > + */ > + > +L(SPECIAL_VALUES_LOOP): > + incl %r12d > + cmpl $8, %r12d > + > +/* Check bits in range mask */ > + jl L(RANGEMASK_CHECK) > + # LOE rbx r15 r12d r13d > + > + movq 16(%rsp), %r12 > + cfi_restore(12) > + movq 8(%rsp), %r13 > + cfi_restore(13) > + movq (%rsp), %r14 > + cfi_restore(14) > + vmovups 192(%rsp), %zmm2 > + > +/* Go to exit */ > + jmp L(EXIT) > + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -240; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x10, 0xff, 0xff, 0xff, 0x22 > + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -248; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x08, 0xff, 0xff, 0xff, 0x22 > + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -256; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x00, 0xff, 0xff, 0xff, 0x22 > + # LOE rbx r12 r13 r14 r15 zmm2 > + > +/* Scalar math fucntion call > + * to process special input > + */ > + > +L(SCALAR_MATH_CALL): > + movl %r12d, %r14d > + movsd 64(%rsp,%r14,8), %xmm0 > + movsd 128(%rsp,%r14,8), %xmm1 > + call hypot@PLT > + # LOE rbx r14 r15 r12d r13d xmm0 > + > + movsd %xmm0, 192(%rsp,%r14,8) > + > +/* Process special inputs in loop */ > + jmp L(SPECIAL_VALUES_LOOP) > + # LOE rbx r15 r12d r13d > +END(_ZGVeN8vv_hypot_skx) > + > + .section .rodata, "a" > + .align 64 > + > +#ifdef __svml_dhypot_data_internal_typedef > +typedef unsigned int VUINT32; > +typedef struct > +{ > + __declspec(align(64)) VUINT32 _dAbsMask[8][2]; > + __declspec(align(64)) VUINT32 _lExpBound_uisa[8][2]; > + __declspec(align(64)) VUINT32 _lExpBound[8][2]; > + __declspec(align(64)) VUINT32 _dHalf[8][2]; > +} __svml_dhypot_data_internal; > +#endif > +__svml_dhypot_data_internal: > + /* legacy algorithm */ > + .quad 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff /* _dAbsMask */ > + /* fma based algorithm*/ > + .align 64 > + .quad 0x407ff00000000000, 0x407ff00000000000, 0x407ff00000000000, 0x407ff00000000000, 0x407ff00000000000, 0x407ff00000000000, 0x407ff00000000000, 0x407ff00000000000 /* _lExpBound_uisa */ > + .align 64 > + .quad 0x404f800000000000, 0x404f800000000000, 0x404f800000000000, 0x404f800000000000, 0x404f800000000000, 0x404f800000000000, 0x404f800000000000, 0x404f800000000000 /* _lExpBound */ > + .align 64 > + .quad 0x3FE0000000000000, 0x3FE0000000000000, 0x3FE0000000000000, 0x3FE0000000000000, 0x3FE0000000000000, 0x3FE0000000000000, 0x3FE0000000000000, 0x3FE0000000000000 /* _dHalf */ > + .align 64 > + .type __svml_dhypot_data_internal,@object > + .size __svml_dhypot_data_internal,.-__svml_dhypot_data_internal > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf16_core-avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf16_core-avx2.S > new file mode 100644 > index 0000000000..a6ba40df4d > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf16_core-avx2.S > @@ -0,0 +1,20 @@ > +/* AVX2 version of vectorized hypotf. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#define _ZGVeN16vv_hypotf _ZGVeN16vv_hypotf_avx2_wrapper > +#include "../svml_s_hypotf16_core.S" > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf16_core.c b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf16_core.c > new file mode 100644 > index 0000000000..0c9eb6a364 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf16_core.c > @@ -0,0 +1,28 @@ > +/* Multiple versions of vectorized hypotf, vector length is 16. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#define SYMBOL_NAME _ZGVeN16vv_hypotf > +#include "ifunc-mathvec-avx512-skx.h" > + > +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); > + > +#ifdef SHARED > +__hidden_ver1 (_ZGVeN16vv_hypotf, __GI__ZGVeN16vv_hypotf, > + __redirect__ZGVeN16vv_hypotf) > + __attribute__ ((visibility ("hidden"))); > +#endif > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf16_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf16_core_avx512.S > new file mode 100644 > index 0000000000..46a156d136 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf16_core_avx512.S > @@ -0,0 +1,239 @@ > +/* Function hypotf vectorized with AVX-512. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + https://www.gnu.org/licenses/. */ > + > +/* > + * ALGORITHM DESCRIPTION: > + * > + * HIGH LEVEL OVERVIEW > + * > + * Calculate z = (x*x+y*y) > + * Calculate reciplicle sqrt (z) > + * Calculate make two NR iterations > + * > + * ALGORITHM DETAILS > + * > + * Multiprecision branch for _HA_ only > + * Remove sigm from both arguments > + * Find maximum (_x) and minimum (_y) (by abs value) between arguments > + * Split _x int _a and _b for multiprecision > + * If _x >> _y we will we will not split _y for multiprecision > + * all _y will be put into lower part (_d) and higher part (_c = 0) > + * Fixing _hilo_mask for the case _x >> _y > + * Split _y into _c and _d for multiprecision with fixed mask > + * > + * compute Hi and Lo parts of _z = _x*_x + _y*_y > + * > + * _zHi = _a*_a + _c*_c > + * _zLo = (_x + _a)*_b + _d*_y + _d*_c > + * _z = _zHi + _zLo > + * > + * No multiprecision branch for _LA_ and _EP_ > + * _z = _VARG1 * _VARG1 + _VARG2 * _VARG2 > + * > + * Check _z exponent to be withing borders [1E3 ; 60A] else goto Callout > + * > + * Compute resciplicle sqrt s0 ~ 1.0/sqrt(_z), > + * that multiplied by _z, is final result for _EP_ version. > + * > + * First iteration (or zero iteration): > + * s = z * s0 > + * h = .5 * s0 > + * d = s * h - .5 > + * > + * Second iteration: > + * h = d * h + h > + * s = s * d + s > + * d = s * s - z (in multiprecision for _HA_) > + * > + * result = s - h * d > + * > + * EP version of the function can be implemented as y[i]=sqrt(a[i]^2+b[i]^2) > + * with all intermediate operations done in target precision for i=1,..,n. > + * It can return result y[i]=0 in case a[i]^2 and b[i]^2 underflow in target > + * precision (for some i). It can return result y[i]=NAN in case > + * a[i]^2+b[i]^2 overflow in target precision, for some i. It can return > + * result y[i]=NAN in case a[i] or b[i] is infinite, for some i. > + * > + * > + */ > + > +/* Offsets for data table __svml_shypot_data_internal > + */ > +#define _sAbsMask 0 > +#define _sHalf 64 > +#define _iExpBound 128 > + > +#include > + > + .text > + .section .text.exex512,"ax",@progbits > +ENTRY(_ZGVeN16vv_hypotf_skx) > + pushq %rbp > + cfi_def_cfa_offset(16) > + movq %rsp, %rbp > + cfi_def_cfa(6, 16) > + cfi_offset(6, -16) > + andq $-64, %rsp > + subq $256, %rsp > + vgetexpps {sae}, %zmm0, %zmm2 > + vgetexpps {sae}, %zmm1, %zmm3 > + vmovups _sHalf+__svml_shypot_data_internal(%rip), %zmm6 > + vmaxps {sae}, %zmm3, %zmm2, %zmm4 > + vmulps {rn-sae}, %zmm0, %zmm0, %zmm2 > + vandps _sAbsMask+__svml_shypot_data_internal(%rip), %zmm4, %zmm5 > + vfmadd231ps {rn-sae}, %zmm1, %zmm1, %zmm2 > + vpcmpd $5, _iExpBound+__svml_shypot_data_internal(%rip), %zmm5, %k0 > + vrsqrt14ps %zmm2, %zmm7 > + kmovw %k0, %edx > + vmulps {rn-sae}, %zmm7, %zmm2, %zmm9 > + vmulps {rn-sae}, %zmm7, %zmm6, %zmm8 > + vfnmadd231ps {rn-sae}, %zmm9, %zmm9, %zmm2 > + vfmadd213ps {rn-sae}, %zmm9, %zmm8, %zmm2 > + > +/* > + * VSCALEF( S, _VRES1, _VRES1, sExp ); > + * The end of implementation > + */ > + testl %edx, %edx > + > +/* Go to special inputs processing branch */ > + jne L(SPECIAL_VALUES_BRANCH) > + # LOE rbx r12 r13 r14 r15 edx zmm0 zmm1 zmm2 > + > +/* Restore registers > + * and exit the function > + */ > + > +L(EXIT): > + vmovaps %zmm2, %zmm0 > + movq %rbp, %rsp > + popq %rbp > + cfi_def_cfa(7, 8) > + cfi_restore(6) > + ret > + cfi_def_cfa(6, 16) > + cfi_offset(6, -16) > + > +/* Branch to process > + * special inputs > + */ > + > +L(SPECIAL_VALUES_BRANCH): > + vmovups %zmm0, 64(%rsp) > + vmovups %zmm1, 128(%rsp) > + vmovups %zmm2, 192(%rsp) > + # LOE rbx r12 r13 r14 r15 edx zmm2 > + > + xorl %eax, %eax > + # LOE rbx r12 r13 r14 r15 eax edx > + > + vzeroupper > + movq %r12, 16(%rsp) > + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -240; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x10, 0xff, 0xff, 0xff, 0x22 > + movl %eax, %r12d > + movq %r13, 8(%rsp) > + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -248; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x08, 0xff, 0xff, 0xff, 0x22 > + movl %edx, %r13d > + movq %r14, (%rsp) > + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -256; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x00, 0xff, 0xff, 0xff, 0x22 > + # LOE rbx r15 r12d r13d > + > +/* Range mask > + * bits check > + */ > + > +L(RANGEMASK_CHECK): > + btl %r12d, %r13d > + > +/* Call scalar math function */ > + jc L(SCALAR_MATH_CALL) > + # LOE rbx r15 r12d r13d > + > +/* Special inputs > + * processing loop > + */ > + > +L(SPECIAL_VALUES_LOOP): > + incl %r12d > + cmpl $16, %r12d > + > +/* Check bits in range mask */ > + jl L(RANGEMASK_CHECK) > + # LOE rbx r15 r12d r13d > + > + movq 16(%rsp), %r12 > + cfi_restore(12) > + movq 8(%rsp), %r13 > + cfi_restore(13) > + movq (%rsp), %r14 > + cfi_restore(14) > + vmovups 192(%rsp), %zmm2 > + > +/* Go to exit */ > + jmp L(EXIT) > + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -240; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x10, 0xff, 0xff, 0xff, 0x22 > + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -248; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x08, 0xff, 0xff, 0xff, 0x22 > + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -256; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x00, 0xff, 0xff, 0xff, 0x22 > + # LOE rbx r12 r13 r14 r15 zmm2 > + > +/* Scalar math fucntion call > + * to process special input > + */ > + > +L(SCALAR_MATH_CALL): > + movl %r12d, %r14d > + movss 64(%rsp,%r14,4), %xmm0 > + movss 128(%rsp,%r14,4), %xmm1 > + call hypotf@PLT > + # LOE rbx r14 r15 r12d r13d xmm0 > + > + movss %xmm0, 192(%rsp,%r14,4) > + > +/* Process special inputs in loop */ > + jmp L(SPECIAL_VALUES_LOOP) > + # LOE rbx r15 r12d r13d > +END(_ZGVeN16vv_hypotf_skx) > + > + .section .rodata, "a" > + .align 64 > + > +#ifdef __svml_shypot_data_internal_typedef > +typedef unsigned int VUINT32; > +typedef struct > +{ > + __declspec(align(64)) VUINT32 _sAbsMask[16][1]; > + __declspec(align(64)) VUINT32 _sHalf[16][1]; > + __declspec(align(64)) VUINT32 _iExpBound[16][1]; > +} __svml_shypot_data_internal; > +#endif > +__svml_shypot_data_internal: > + .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff /* _sAbsMask */ > + .align 64 > + .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000 /* _sHalf */ > + /* fma based algorithm*/ > + .align 64 > + .long 0x427C0000, 0x427C0000, 0x427C0000, 0x427C0000, 0x427C0000, 0x427C0000, 0x427C0000, 0x427C0000, 0x427C0000, 0x427C0000, 0x427C0000, 0x427C0000, 0x427C0000, 0x427C0000, 0x427C0000, 0x427C0000 /* _iExpBound */ > + .align 64 > + .type __svml_shypot_data_internal,@object > + .size __svml_shypot_data_internal,.-__svml_shypot_data_internal > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core-sse2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core-sse2.S > new file mode 100644 > index 0000000000..5e9dd22d94 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core-sse2.S > @@ -0,0 +1,20 @@ > +/* SSE2 version of vectorized hypotf. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#define _ZGVbN4vv_hypotf _ZGVbN4vv_hypotf_sse2 > +#include "../svml_s_hypotf4_core.S" > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core.c b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core.c > new file mode 100644 > index 0000000000..91c9f5ca3f > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core.c > @@ -0,0 +1,28 @@ > +/* Multiple versions of vectorized hypotf, vector length is 4. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#define SYMBOL_NAME _ZGVbN4vv_hypotf > +#include "ifunc-mathvec-sse4_1.h" > + > +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); > + > +#ifdef SHARED > +__hidden_ver1 (_ZGVbN4vv_hypotf, __GI__ZGVbN4vv_hypotf, > + __redirect__ZGVbN4vv_hypotf) > + __attribute__ ((visibility ("hidden"))); > +#endif > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core_sse4.S > new file mode 100644 > index 0000000000..a3f6d21ce1 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core_sse4.S > @@ -0,0 +1,265 @@ > +/* Function hypotf vectorized with SSE4. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + https://www.gnu.org/licenses/. */ > + > +/* > + * ALGORITHM DESCRIPTION: > + * > + * HIGH LEVEL OVERVIEW > + * > + * Calculate z = (x*x+y*y) > + * Calculate reciplicle sqrt (z) > + * Calculate make two NR iterations > + * > + * ALGORITHM DETAILS > + * > + * Multiprecision branch for _HA_ only > + * Remove sigm from both arguments > + * Find maximum (_x) and minimum (_y) (by abs value) between arguments > + * Split _x int _a and _b for multiprecision > + * If _x >> _y we will we will not split _y for multiprecision > + * all _y will be put into lower part (_d) and higher part (_c = 0) > + * Fixing _hilo_mask for the case _x >> _y > + * Split _y into _c and _d for multiprecision with fixed mask > + * > + * compute Hi and Lo parts of _z = _x*_x + _y*_y > + * > + * _zHi = _a*_a + _c*_c > + * _zLo = (_x + _a)*_b + _d*_y + _d*_c > + * _z = _zHi + _zLo > + * > + * No multiprecision branch for _LA_ and _EP_ > + * _z = _VARG1 * _VARG1 + _VARG2 * _VARG2 > + * > + * Check _z exponent to be withing borders [1E3 ; 60A] else goto Callout > + * > + * Compute resciplicle sqrt s0 ~ 1.0/sqrt(_z), > + * that multiplied by _z, is final result for _EP_ version. > + * > + * First iteration (or zero iteration): > + * s = z * s0 > + * h = .5 * s0 > + * d = s * h - .5 > + * > + * Second iteration: > + * h = d * h + h > + * s = s * d + s > + * d = s * s - z (in multiprecision for _HA_) > + * > + * result = s - h * d > + * > + * EP version of the function can be implemented as y[i]=sqrt(a[i]^2+b[i]^2) > + * with all intermediate operations done in target precision for i=1,..,n. > + * It can return result y[i]=0 in case a[i]^2 and b[i]^2 underflow in target > + * precision (for some i). It can return result y[i]=NAN in case > + * a[i]^2+b[i]^2 overflow in target precision, for some i. It can return > + * result y[i]=NAN in case a[i] or b[i] is infinite, for some i. > + * > + * > + */ > + > +/* Offsets for data table __svml_shypot_data_internal > + */ > +#define _sHiLoMask 0 > +#define _sAbsMask 16 > +#define _sHalf 32 > +#define _LowBoundary 48 > +#define _HighBoundary 64 > + > +#include > + > + .text > + .section .text.sse4,"ax",@progbits > +ENTRY(_ZGVbN4vv_hypotf_sse4) > + subq $88, %rsp > + cfi_def_cfa_offset(96) > + > +/* > + * Implementation > + * Multiprecision branch for _HA_ only > + * No multiprecision branch for _LA_ > + * _z = _VARG1 * _VARG1 + _VARG2 * _VARG2 > + */ > + movaps %xmm0, %xmm8 > + movaps %xmm1, %xmm2 > + mulps %xmm0, %xmm8 > + mulps %xmm1, %xmm2 > + > +/* > + * Variables > + * Defines > + * Constants loading > + */ > + movups _sHalf+__svml_shypot_data_internal(%rip), %xmm5 > + addps %xmm2, %xmm8 > + > +/* _s0 ~ 1.0/sqrt(_z) */ > + rsqrtps %xmm8, %xmm10 > + > +/* First iteration */ > + movaps %xmm10, %xmm2 > + movaps %xmm8, %xmm3 > + mulps %xmm8, %xmm2 > + mulps %xmm5, %xmm10 > + movaps %xmm2, %xmm6 > + mulps %xmm10, %xmm6 > + > +/* Check _z exponent to be withing borders [1E3 ; 60A] else goto Callout */ > + movdqu _LowBoundary+__svml_shypot_data_internal(%rip), %xmm4 > + subps %xmm6, %xmm5 > + > +/* Second iteration */ > + movaps %xmm5, %xmm7 > + pcmpgtd %xmm8, %xmm4 > + mulps %xmm2, %xmm5 > + mulps %xmm10, %xmm7 > + addps %xmm5, %xmm2 > + addps %xmm7, %xmm10 > + > +/* Finish second iteration in native precision for _LA_ */ > + movaps %xmm2, %xmm9 > + mulps %xmm2, %xmm9 > + pcmpgtd _HighBoundary+__svml_shypot_data_internal(%rip), %xmm3 > + subps %xmm8, %xmm9 > + mulps %xmm9, %xmm10 > + por %xmm3, %xmm4 > + movmskps %xmm4, %edx > + subps %xmm10, %xmm2 > + > +/* The end of implementation */ > + testl %edx, %edx > + > +/* Go to special inputs processing branch */ > + jne L(SPECIAL_VALUES_BRANCH) > + # LOE rbx rbp r12 r13 r14 r15 edx xmm0 xmm1 xmm2 > + > +/* Restore registers > + * and exit the function > + */ > + > +L(EXIT): > + movaps %xmm2, %xmm0 > + addq $88, %rsp > + cfi_def_cfa_offset(8) > + ret > + cfi_def_cfa_offset(96) > + > +/* Branch to process > + * special inputs > + */ > + > +L(SPECIAL_VALUES_BRANCH): > + movups %xmm0, 32(%rsp) > + movups %xmm1, 48(%rsp) > + movups %xmm2, 64(%rsp) > + # LOE rbx rbp r12 r13 r14 r15 edx > + > + xorl %eax, %eax > + movq %r12, 16(%rsp) > + cfi_offset(12, -80) > + movl %eax, %r12d > + movq %r13, 8(%rsp) > + cfi_offset(13, -88) > + movl %edx, %r13d > + movq %r14, (%rsp) > + cfi_offset(14, -96) > + # LOE rbx rbp r15 r12d r13d > + > +/* Range mask > + * bits check > + */ > + > +L(RANGEMASK_CHECK): > + btl %r12d, %r13d > + > +/* Call scalar math function */ > + jc L(SCALAR_MATH_CALL) > + # LOE rbx rbp r15 r12d r13d > + > +/* Special inputs > + * processing loop > + */ > + > +L(SPECIAL_VALUES_LOOP): > + incl %r12d > + cmpl $4, %r12d > + > +/* Check bits in range mask */ > + jl L(RANGEMASK_CHECK) > + # LOE rbx rbp r15 r12d r13d > + > + movq 16(%rsp), %r12 > + cfi_restore(12) > + movq 8(%rsp), %r13 > + cfi_restore(13) > + movq (%rsp), %r14 > + cfi_restore(14) > + movups 64(%rsp), %xmm2 > + > +/* Go to exit */ > + jmp L(EXIT) > + cfi_offset(12, -80) > + cfi_offset(13, -88) > + cfi_offset(14, -96) > + # LOE rbx rbp r12 r13 r14 r15 xmm2 > + > +/* Scalar math fucntion call > + * to process special input > + */ > + > +L(SCALAR_MATH_CALL): > + movl %r12d, %r14d > + movss 32(%rsp,%r14,4), %xmm0 > + movss 48(%rsp,%r14,4), %xmm1 > + call hypotf@PLT > + # LOE rbx rbp r14 r15 r12d r13d xmm0 > + > + movss %xmm0, 64(%rsp,%r14,4) > + > +/* Process special inputs in loop */ > + jmp L(SPECIAL_VALUES_LOOP) > + # LOE rbx rbp r15 r12d r13d > +END(_ZGVbN4vv_hypotf_sse4) > + > + .section .rodata, "a" > + .align 16 > + > +#ifdef __svml_shypot_data_internal_typedef > +typedef unsigned int VUINT32; > +typedef struct > +{ > + __declspec(align(16)) VUINT32 _sHiLoMask[4][1]; > + __declspec(align(16)) VUINT32 _sAbsMask[4][1]; > + __declspec(align(16)) VUINT32 _sHalf[4][1]; > + __declspec(align(16)) VUINT32 _LowBoundary[4][1]; > + __declspec(align(16)) VUINT32 _HighBoundary[4][1]; > +} __svml_shypot_data_internal; > +#endif > +__svml_shypot_data_internal: > + /* legacy algorithm */ > + .long 0xFFF80000, 0xFFF80000, 0xFFF80000, 0xFFF80000 /* _sHiLoMask */ > + .align 16 > + .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff /* _sAbsMask */ > + .align 16 > + .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000 /* _sHalf */ > + .align 16 > + .long 0x1E300000, 0x1E300000, 0x1E300000, 0x1E300000 /* _LowBoundary */ > + .align 16 > + .long 0x60A00000, 0x60A00000, 0x60A00000, 0x60A00000 /* _HighBoundary */ > + .align 16 > + .type __svml_shypot_data_internal,@object > + .size __svml_shypot_data_internal,.-__svml_shypot_data_internal > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf8_core-sse.S b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf8_core-sse.S > new file mode 100644 > index 0000000000..d37556e331 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf8_core-sse.S > @@ -0,0 +1,20 @@ > +/* SSE version of vectorized hypotf. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#define _ZGVdN8vv_hypotf _ZGVdN8vv_hypotf_sse_wrapper > +#include "../svml_s_hypotf8_core.S" > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf8_core.c b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf8_core.c > new file mode 100644 > index 0000000000..6cc497e73d > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf8_core.c > @@ -0,0 +1,28 @@ > +/* Multiple versions of vectorized sinf, vector length is 8. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#define SYMBOL_NAME _ZGVdN8vv_hypotf > +#include "ifunc-mathvec-avx2.h" > + > +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); > + > +#ifdef SHARED > +__hidden_ver1 (_ZGVdN8vv_hypotf, __GI__ZGVdN8vv_hypotf, > + __redirect__ZGVdN8vv_hypotf) > + __attribute__ ((visibility ("hidden"))); > +#endif > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf8_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf8_core_avx2.S > new file mode 100644 > index 0000000000..733022ff01 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf8_core_avx2.S > @@ -0,0 +1,269 @@ > +/* Function hypotf vectorized with AVX2. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + https://www.gnu.org/licenses/. */ > + > +/* > + * ALGORITHM DESCRIPTION: > + * > + * HIGH LEVEL OVERVIEW > + * > + * Calculate z = (x*x+y*y) > + * Calculate reciplicle sqrt (z) > + * Calculate make two NR iterations > + * > + * ALGORITHM DETAILS > + * > + * Multiprecision branch for _HA_ only > + * Remove sigm from both arguments > + * Find maximum (_x) and minimum (_y) (by abs value) between arguments > + * Split _x int _a and _b for multiprecision > + * If _x >> _y we will we will not split _y for multiprecision > + * all _y will be put into lower part (_d) and higher part (_c = 0) > + * Fixing _hilo_mask for the case _x >> _y > + * Split _y into _c and _d for multiprecision with fixed mask > + * > + * compute Hi and Lo parts of _z = _x*_x + _y*_y > + * > + * _zHi = _a*_a + _c*_c > + * _zLo = (_x + _a)*_b + _d*_y + _d*_c > + * _z = _zHi + _zLo > + * > + * No multiprecision branch for _LA_ and _EP_ > + * _z = _VARG1 * _VARG1 + _VARG2 * _VARG2 > + * > + * Check _z exponent to be withing borders [1E3 ; 60A] else goto Callout > + * > + * Compute resciplicle sqrt s0 ~ 1.0/sqrt(_z), > + * that multiplied by _z, is final result for _EP_ version. > + * > + * First iteration (or zero iteration): > + * s = z * s0 > + * h = .5 * s0 > + * d = s * h - .5 > + * > + * Second iteration: > + * h = d * h + h > + * s = s * d + s > + * d = s * s - z (in multiprecision for _HA_) > + * > + * result = s - h * d > + * > + * EP version of the function can be implemented as y[i]=sqrt(a[i]^2+b[i]^2) > + * with all intermediate operations done in target precision for i=1,..,n. > + * It can return result y[i]=0 in case a[i]^2 and b[i]^2 underflow in target > + * precision (for some i). It can return result y[i]=NAN in case > + * a[i]^2+b[i]^2 overflow in target precision, for some i. It can return > + * result y[i]=NAN in case a[i] or b[i] is infinite, for some i. > + * > + * > + */ > + > +/* Offsets for data table __svml_shypot_data_internal > + */ > +#define _sHiLoMask 0 > +#define _sAbsMask 32 > +#define _sHalf 64 > +#define _LowBoundary 96 > +#define _HighBoundary 128 > + > +#include > + > + .text > + .section .text.avx2,"ax",@progbits > +ENTRY(_ZGVdN8vv_hypotf_avx2) > + pushq %rbp > + cfi_def_cfa_offset(16) > + movq %rsp, %rbp > + cfi_def_cfa(6, 16) > + cfi_offset(6, -16) > + andq $-32, %rsp > + subq $128, %rsp > + > +/* > + * Implementation > + * Multiprecision branch for _HA_ only > + * No multiprecision branch for _LA_ > + * _z = _VARG1 * _VARG1 + _VARG2 * _VARG2 > + */ > + vmulps %ymm0, %ymm0, %ymm8 > + > +/* > + * Variables > + * Defines > + * Constants loading > + */ > + vmovups _sHalf+__svml_shypot_data_internal(%rip), %ymm7 > + > +/* Check _z exponent to be withing borders [1E3 ; 60A] else goto Callout */ > + vmovups _LowBoundary+__svml_shypot_data_internal(%rip), %ymm2 > + vfmadd231ps %ymm1, %ymm1, %ymm8 > + > +/* _s0 ~ 1.0/sqrt(_z) */ > + vrsqrtps %ymm8, %ymm6 > + vpcmpgtd %ymm8, %ymm2, %ymm3 > + > +/* First iteration */ > + vmulps %ymm8, %ymm6, %ymm9 > + vmulps %ymm7, %ymm6, %ymm2 > + vfnmadd231ps %ymm9, %ymm2, %ymm7 > + vfmadd213ps %ymm9, %ymm7, %ymm9 > + > +/* Second iteration */ > + vfmadd132ps %ymm7, %ymm2, %ymm2 > + vpcmpgtd _HighBoundary+__svml_shypot_data_internal(%rip), %ymm8, %ymm4 > + vpor %ymm4, %ymm3, %ymm5 > + > +/* Finish second iteration in native precision for _LA_ */ > + vfmsub231ps %ymm9, %ymm9, %ymm8 > + vmovmskps %ymm5, %edx > + vfnmadd213ps %ymm9, %ymm8, %ymm2 > + > +/* The end of implementation */ > + testl %edx, %edx > + > +/* Go to special inputs processing branch */ > + jne L(SPECIAL_VALUES_BRANCH) > + # LOE rbx r12 r13 r14 r15 edx ymm0 ymm1 ymm2 > + > +/* Restore registers > + * and exit the function > + */ > + > +L(EXIT): > + vmovaps %ymm2, %ymm0 > + movq %rbp, %rsp > + popq %rbp > + cfi_def_cfa(7, 8) > + cfi_restore(6) > + ret > + cfi_def_cfa(6, 16) > + cfi_offset(6, -16) > + > +/* Branch to process > + * special inputs > + */ > + > +L(SPECIAL_VALUES_BRANCH): > + vmovups %ymm0, 32(%rsp) > + vmovups %ymm1, 64(%rsp) > + vmovups %ymm2, 96(%rsp) > + # LOE rbx r12 r13 r14 r15 edx ymm2 > + > + xorl %eax, %eax > + # LOE rbx r12 r13 r14 r15 eax edx > + > + vzeroupper > + movq %r12, 16(%rsp) > + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -112; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x90, 0xff, 0xff, 0xff, 0x22 > + movl %eax, %r12d > + movq %r13, 8(%rsp) > + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -120; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x88, 0xff, 0xff, 0xff, 0x22 > + movl %edx, %r13d > + movq %r14, (%rsp) > + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -128; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x80, 0xff, 0xff, 0xff, 0x22 > + # LOE rbx r15 r12d r13d > + > +/* Range mask > + * bits check > + */ > + > +L(RANGEMASK_CHECK): > + btl %r12d, %r13d > + > +/* Call scalar math function */ > + jc L(SCALAR_MATH_CALL) > + # LOE rbx r15 r12d r13d > + > +/* Special inputs > + * processing loop > + */ > + > +L(SPECIAL_VALUES_LOOP): > + incl %r12d > + cmpl $8, %r12d > + > +/* Check bits in range mask */ > + jl L(RANGEMASK_CHECK) > + # LOE rbx r15 r12d r13d > + > + movq 16(%rsp), %r12 > + cfi_restore(12) > + movq 8(%rsp), %r13 > + cfi_restore(13) > + movq (%rsp), %r14 > + cfi_restore(14) > + vmovups 96(%rsp), %ymm2 > + > +/* Go to exit */ > + jmp L(EXIT) > + /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -112; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x90, 0xff, 0xff, 0xff, 0x22 > + /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -120; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x88, 0xff, 0xff, 0xff, 0x22 > + /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -128; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x80, 0xff, 0xff, 0xff, 0x22 > + # LOE rbx r12 r13 r14 r15 ymm2 > + > +/* Scalar math fucntion call > + * to process special input > + */ > + > +L(SCALAR_MATH_CALL): > + movl %r12d, %r14d > + movss 32(%rsp,%r14,4), %xmm0 > + movss 64(%rsp,%r14,4), %xmm1 > + call hypotf@PLT > + # LOE rbx r14 r15 r12d r13d xmm0 > + > + movss %xmm0, 96(%rsp,%r14,4) > + > +/* Process special inputs in loop */ > + jmp L(SPECIAL_VALUES_LOOP) > + # LOE rbx r15 r12d r13d > +END(_ZGVdN8vv_hypotf_avx2) > + > + .section .rodata, "a" > + .align 32 > + > +#ifdef __svml_shypot_data_internal_typedef > +typedef unsigned int VUINT32; > +typedef struct > +{ > + __declspec(align(32)) VUINT32 _sHiLoMask[8][1]; > + __declspec(align(32)) VUINT32 _sAbsMask[8][1]; > + __declspec(align(32)) VUINT32 _sHalf[8][1]; > + __declspec(align(32)) VUINT32 _LowBoundary[8][1]; > + __declspec(align(32)) VUINT32 _HighBoundary[8][1]; > +} __svml_shypot_data_internal; > +#endif > +__svml_shypot_data_internal: > + /* legacy algorithm */ > + .long 0xFFF80000, 0xFFF80000, 0xFFF80000, 0xFFF80000, 0xFFF80000, 0xFFF80000, 0xFFF80000, 0xFFF80000 /* _sHiLoMask */ > + .align 32 > + .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff /* _sAbsMask */ > + .align 32 > + .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000 /* _sHalf */ > + .align 32 > + .long 0x1E300000, 0x1E300000, 0x1E300000, 0x1E300000, 0x1E300000, 0x1E300000, 0x1E300000, 0x1E300000 /* _LowBoundary */ > + .align 32 > + .long 0x60A00000, 0x60A00000, 0x60A00000, 0x60A00000, 0x60A00000, 0x60A00000, 0x60A00000, 0x60A00000 /* _HighBoundary */ > + .align 32 > + .type __svml_shypot_data_internal,@object > + .size __svml_shypot_data_internal,.-__svml_shypot_data_internal > diff --git a/sysdeps/x86_64/fpu/svml_d_hypot2_core.S b/sysdeps/x86_64/fpu/svml_d_hypot2_core.S > new file mode 100644 > index 0000000000..ea98f36324 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_d_hypot2_core.S > @@ -0,0 +1,29 @@ > +/* Function hypot vectorized with SSE2. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#include > +#include "svml_d_wrapper_impl.h" > + > + .text > +ENTRY (_ZGVbN2vv_hypot) > +WRAPPER_IMPL_SSE2_ff hypot > +END (_ZGVbN2vv_hypot) > + > +#ifndef USE_MULTIARCH > + libmvec_hidden_def (_ZGVbN2vv_hypot) > +#endif > diff --git a/sysdeps/x86_64/fpu/svml_d_hypot4_core.S b/sysdeps/x86_64/fpu/svml_d_hypot4_core.S > new file mode 100644 > index 0000000000..cedbbff2b6 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_d_hypot4_core.S > @@ -0,0 +1,29 @@ > +/* Function hypot vectorized with AVX2, wrapper version. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#include > +#include "svml_d_wrapper_impl.h" > + > + .text > +ENTRY (_ZGVdN4vv_hypot) > +WRAPPER_IMPL_AVX_ff _ZGVbN2vv_hypot > +END (_ZGVdN4vv_hypot) > + > +#ifndef USE_MULTIARCH > + libmvec_hidden_def (_ZGVdN4vv_hypot) > +#endif > diff --git a/sysdeps/x86_64/fpu/svml_d_hypot4_core_avx.S b/sysdeps/x86_64/fpu/svml_d_hypot4_core_avx.S > new file mode 100644 > index 0000000000..e0fef5203d > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_d_hypot4_core_avx.S > @@ -0,0 +1,25 @@ > +/* Function hypot vectorized in AVX ISA as wrapper to SSE4 ISA version. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#include > +#include "svml_d_wrapper_impl.h" > + > + .text > +ENTRY (_ZGVcN4vv_hypot) > +WRAPPER_IMPL_AVX_ff _ZGVbN2vv_hypot > +END (_ZGVcN4vv_hypot) > diff --git a/sysdeps/x86_64/fpu/svml_d_hypot8_core.S b/sysdeps/x86_64/fpu/svml_d_hypot8_core.S > new file mode 100644 > index 0000000000..7588e4407b > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_d_hypot8_core.S > @@ -0,0 +1,25 @@ > +/* Function hypot vectorized with AVX-512. Wrapper to AVX2 version. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#include > +#include "svml_d_wrapper_impl.h" > + > + .text > +ENTRY (_ZGVeN8vv_hypot) > +WRAPPER_IMPL_AVX512_ff _ZGVdN4vv_hypot > +END (_ZGVeN8vv_hypot) > diff --git a/sysdeps/x86_64/fpu/svml_s_hypotf16_core.S b/sysdeps/x86_64/fpu/svml_s_hypotf16_core.S > new file mode 100644 > index 0000000000..06d421a926 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_s_hypotf16_core.S > @@ -0,0 +1,25 @@ > +/* Function hypotf vectorized with AVX-512. Wrapper to AVX2 version. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#include > +#include "svml_s_wrapper_impl.h" > + > + .text > +ENTRY (_ZGVeN16vv_hypotf) > +WRAPPER_IMPL_AVX512_ff _ZGVdN8vv_hypotf > +END (_ZGVeN16vv_hypotf) > diff --git a/sysdeps/x86_64/fpu/svml_s_hypotf4_core.S b/sysdeps/x86_64/fpu/svml_s_hypotf4_core.S > new file mode 100644 > index 0000000000..7e8553cae4 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_s_hypotf4_core.S > @@ -0,0 +1,29 @@ > +/* Function hypotf vectorized with SSE2. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#include > +#include "svml_s_wrapper_impl.h" > + > + .text > +ENTRY (_ZGVbN4vv_hypotf) > +WRAPPER_IMPL_SSE2_ff hypotf > +END (_ZGVbN4vv_hypotf) > + > +#ifndef USE_MULTIARCH > + libmvec_hidden_def (_ZGVbN4vv_hypotf) > +#endif > diff --git a/sysdeps/x86_64/fpu/svml_s_hypotf8_core.S b/sysdeps/x86_64/fpu/svml_s_hypotf8_core.S > new file mode 100644 > index 0000000000..a9bf27370b > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_s_hypotf8_core.S > @@ -0,0 +1,29 @@ > +/* Function hypotf vectorized with AVX2, wrapper version. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#include > +#include "svml_s_wrapper_impl.h" > + > + .text > +ENTRY (_ZGVdN8vv_hypotf) > +WRAPPER_IMPL_AVX_ff _ZGVbN4vv_hypotf > +END (_ZGVdN8vv_hypotf) > + > +#ifndef USE_MULTIARCH > + libmvec_hidden_def (_ZGVdN8vv_hypotf) > +#endif > diff --git a/sysdeps/x86_64/fpu/svml_s_hypotf8_core_avx.S b/sysdeps/x86_64/fpu/svml_s_hypotf8_core_avx.S > new file mode 100644 > index 0000000000..8b8008a7e9 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_s_hypotf8_core_avx.S > @@ -0,0 +1,25 @@ > +/* Function hypotf vectorized in AVX ISA as wrapper to SSE4 ISA version. > + Copyright (C) 2021 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#include > +#include "svml_s_wrapper_impl.h" > + > + .text > +ENTRY(_ZGVcN8vv_hypotf) > +WRAPPER_IMPL_AVX_ff _ZGVbN4vv_hypotf > +END(_ZGVcN8vv_hypotf) > diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-hypot-avx.c b/sysdeps/x86_64/fpu/test-double-libmvec-hypot-avx.c > new file mode 100644 > index 0000000000..c6a26a63e4 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-double-libmvec-hypot-avx.c > @@ -0,0 +1 @@ > +#include "test-double-libmvec-hypot.c" > diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-hypot-avx2.c b/sysdeps/x86_64/fpu/test-double-libmvec-hypot-avx2.c > new file mode 100644 > index 0000000000..c6a26a63e4 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-double-libmvec-hypot-avx2.c > @@ -0,0 +1 @@ > +#include "test-double-libmvec-hypot.c" > diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-hypot-avx512f.c b/sysdeps/x86_64/fpu/test-double-libmvec-hypot-avx512f.c > new file mode 100644 > index 0000000000..c6a26a63e4 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-double-libmvec-hypot-avx512f.c > @@ -0,0 +1 @@ > +#include "test-double-libmvec-hypot.c" > diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-hypot.c b/sysdeps/x86_64/fpu/test-double-libmvec-hypot.c > new file mode 100644 > index 0000000000..c0f600a443 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-double-libmvec-hypot.c > @@ -0,0 +1,3 @@ > +#define LIBMVEC_TYPE double > +#define LIBMVEC_FUNC hypot > +#include "test-vector-abi-arg2.h" > diff --git a/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c b/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c > index 5746bb5be3..9bc9d1dafa 100644 > --- a/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c > @@ -30,6 +30,7 @@ VECTOR_WRAPPER_ff (WRAPPER_NAME (pow), _ZGVbN2vv_pow) > VECTOR_WRAPPER (WRAPPER_NAME (acos), _ZGVbN2v_acos) > VECTOR_WRAPPER (WRAPPER_NAME (atan), _ZGVbN2v_atan) > VECTOR_WRAPPER (WRAPPER_NAME (asin), _ZGVbN2v_asin) > +VECTOR_WRAPPER_ff (WRAPPER_NAME (hypot), _ZGVbN2vv_hypot) > > #define VEC_INT_TYPE __m128i > > diff --git a/sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c b/sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c > index 8d3d5493ed..c41994d90a 100644 > --- a/sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c > @@ -33,6 +33,7 @@ VECTOR_WRAPPER_ff (WRAPPER_NAME (pow), _ZGVdN4vv_pow) > VECTOR_WRAPPER (WRAPPER_NAME (acos), _ZGVdN4v_acos) > VECTOR_WRAPPER (WRAPPER_NAME (atan), _ZGVdN4v_atan) > VECTOR_WRAPPER (WRAPPER_NAME (asin), _ZGVdN4v_asin) > +VECTOR_WRAPPER_ff (WRAPPER_NAME (hypot), _ZGVdN4vv_hypot) > > #ifndef __ILP32__ > # define VEC_INT_TYPE __m256i > diff --git a/sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c b/sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c > index f43328f2ff..881f6c801a 100644 > --- a/sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c > @@ -30,6 +30,7 @@ VECTOR_WRAPPER_ff (WRAPPER_NAME (pow), _ZGVcN4vv_pow) > VECTOR_WRAPPER (WRAPPER_NAME (acos), _ZGVcN4v_acos) > VECTOR_WRAPPER (WRAPPER_NAME (atan), _ZGVcN4v_atan) > VECTOR_WRAPPER (WRAPPER_NAME (asin), _ZGVcN4v_asin) > +VECTOR_WRAPPER_ff (WRAPPER_NAME (hypot), _ZGVcN4vv_hypot) > > #define VEC_INT_TYPE __m128i > > diff --git a/sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c b/sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c > index 8b566c199a..6fd106fe68 100644 > --- a/sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c > @@ -30,6 +30,7 @@ VECTOR_WRAPPER_ff (WRAPPER_NAME (pow), _ZGVeN8vv_pow) > VECTOR_WRAPPER (WRAPPER_NAME (acos), _ZGVeN8v_acos) > VECTOR_WRAPPER (WRAPPER_NAME (atan), _ZGVeN8v_atan) > VECTOR_WRAPPER (WRAPPER_NAME (asin), _ZGVeN8v_asin) > +VECTOR_WRAPPER_ff (WRAPPER_NAME (hypot), _ZGVeN8vv_hypot) > > #ifndef __ILP32__ > # define VEC_INT_TYPE __m512i > diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-hypotf-avx.c b/sysdeps/x86_64/fpu/test-float-libmvec-hypotf-avx.c > new file mode 100644 > index 0000000000..97d11ad1d3 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-float-libmvec-hypotf-avx.c > @@ -0,0 +1 @@ > +#include "test-float-libmvec-hypotf.c" > diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-hypotf-avx2.c b/sysdeps/x86_64/fpu/test-float-libmvec-hypotf-avx2.c > new file mode 100644 > index 0000000000..97d11ad1d3 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-float-libmvec-hypotf-avx2.c > @@ -0,0 +1 @@ > +#include "test-float-libmvec-hypotf.c" > diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-hypotf-avx512f.c b/sysdeps/x86_64/fpu/test-float-libmvec-hypotf-avx512f.c > new file mode 100644 > index 0000000000..97d11ad1d3 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-float-libmvec-hypotf-avx512f.c > @@ -0,0 +1 @@ > +#include "test-float-libmvec-hypotf.c" > diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-hypotf.c b/sysdeps/x86_64/fpu/test-float-libmvec-hypotf.c > new file mode 100644 > index 0000000000..38776fa724 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-float-libmvec-hypotf.c > @@ -0,0 +1,3 @@ > +#define LIBMVEC_TYPE float > +#define LIBMVEC_FUNC hypotf > +#include "test-vector-abi-arg2.h" > diff --git a/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c b/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c > index 3d3218a310..4c2ea6ddfe 100644 > --- a/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c > @@ -30,6 +30,7 @@ VECTOR_WRAPPER_ff (WRAPPER_NAME (powf), _ZGVeN16vv_powf) > VECTOR_WRAPPER (WRAPPER_NAME (acosf), _ZGVeN16v_acosf) > VECTOR_WRAPPER (WRAPPER_NAME (atanf), _ZGVeN16v_atanf) > VECTOR_WRAPPER (WRAPPER_NAME (asinf), _ZGVeN16v_asinf) > +VECTOR_WRAPPER_ff (WRAPPER_NAME (hypotf), _ZGVeN16vv_hypotf) > > #define VEC_INT_TYPE __m512i > > diff --git a/sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c b/sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c > index 7d75b9f60f..1d5d952d07 100644 > --- a/sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c > @@ -30,6 +30,7 @@ VECTOR_WRAPPER_ff (WRAPPER_NAME (powf), _ZGVbN4vv_powf) > VECTOR_WRAPPER (WRAPPER_NAME (acosf), _ZGVbN4v_acosf) > VECTOR_WRAPPER (WRAPPER_NAME (atanf), _ZGVbN4v_atanf) > VECTOR_WRAPPER (WRAPPER_NAME (asinf), _ZGVbN4v_asinf) > +VECTOR_WRAPPER_ff (WRAPPER_NAME (hypotf), _ZGVbN4vv_hypotf) > > #define VEC_INT_TYPE __m128i > > diff --git a/sysdeps/x86_64/fpu/test-float-vlen8-avx2-wrappers.c b/sysdeps/x86_64/fpu/test-float-vlen8-avx2-wrappers.c > index 405dde49bc..7a750f3781 100644 > --- a/sysdeps/x86_64/fpu/test-float-vlen8-avx2-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-float-vlen8-avx2-wrappers.c > @@ -33,6 +33,7 @@ VECTOR_WRAPPER_ff (WRAPPER_NAME (powf), _ZGVdN8vv_powf) > VECTOR_WRAPPER (WRAPPER_NAME (acosf), _ZGVdN8v_acosf) > VECTOR_WRAPPER (WRAPPER_NAME (atanf), _ZGVdN8v_atanf) > VECTOR_WRAPPER (WRAPPER_NAME (asinf), _ZGVdN8v_asinf) > +VECTOR_WRAPPER_ff (WRAPPER_NAME (hypotf), _ZGVdN8vv_hypotf) > > /* Redefinition of wrapper to be compatible with _ZGVdN8vvv_sincosf. */ > #undef VECTOR_WRAPPER_fFF > diff --git a/sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c b/sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c > index 7558443f2e..af816a7789 100644 > --- a/sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c > @@ -30,6 +30,7 @@ VECTOR_WRAPPER_ff (WRAPPER_NAME (powf), _ZGVcN8vv_powf) > VECTOR_WRAPPER (WRAPPER_NAME (acosf), _ZGVcN8v_acosf) > VECTOR_WRAPPER (WRAPPER_NAME (atanf), _ZGVcN8v_atanf) > VECTOR_WRAPPER (WRAPPER_NAME (asinf), _ZGVcN8v_asinf) > +VECTOR_WRAPPER_ff (WRAPPER_NAME (hypotf), _ZGVcN8vv_hypotf) > > #define VEC_INT_TYPE __m128i > > -- > 2.31.1 > LGTM. Reviewed-by: H.J. Lu Thanks. H.J.