From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-pj1-x1036.google.com (mail-pj1-x1036.google.com [IPv6:2607:f8b0:4864:20::1036]) by sourceware.org (Postfix) with ESMTPS id 361113858402 for ; Wed, 29 Dec 2021 21:26:03 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 361113858402 Received: by mail-pj1-x1036.google.com with SMTP id mj19so19502210pjb.3 for ; Wed, 29 Dec 2021 13:26:03 -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=taqgx1aQLpp++PG2II22Ezqjw1gjiHlB4oVCE2g24v0=; b=22B2KoV96gc7xZiwwUzyAK6MiLkkMkk+BNjd7fAg7yWwR7r6nkQh/jo9f4TLZIFX50 ztGy1Dm/OllfQpa0qkL2ChIQXlEneDdnifAQTUejhhmuRhDEOqGdBcj2wlXXSabLqU4p eEu155ittSQLwXKN9wGF+O3/9fMT+s/vlR7xd1oMhdOy01jPfptJ63IhXJMaVJD2Nc9U KmnuMkKUv1OeSP9AyrFqhhPQCLIWGntYSSXiEKRIjffRd8kOcFvxQitD4Ba1eifsfjr3 IIN2ACeUARLZukoTHU1OmVqPD/q3bIr2FHe94H3cb/2+OlH8iDdvb4ud8ofNyEKTbjMd jRmg== X-Gm-Message-State: AOAM532s+HLbJMBdiP8f3w/1ryaK6QY+W+g09ovqGMoBySmJYoGbR+Tv xjE1SNbpuzUfWU1T4e3OdBE= X-Google-Smtp-Source: ABdhPJzDkmBSaz6Op2Yv9BkdMiflHXQmoKKjjtOauD9mq0bOWJqgMOR1HP2E+x5o/BaFzKPZZP6XXw== X-Received: by 2002:a17:902:7688:b0:148:be1c:18b8 with SMTP id m8-20020a170902768800b00148be1c18b8mr27832169pll.146.1640813161279; Wed, 29 Dec 2021 13:26:01 -0800 (PST) Received: from gnu-tgl-3.localdomain ([172.58.35.133]) by smtp.gmail.com with ESMTPSA id p10sm24575978pfw.69.2021.12.29.13.26.00 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Wed, 29 Dec 2021 13:26:00 -0800 (PST) Received: by gnu-tgl-3.localdomain (Postfix, from userid 1000) id B0720C08AD; Wed, 29 Dec 2021 13:25:59 -0800 (PST) Date: Wed, 29 Dec 2021 13:25:59 -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 05/18] x86-64: Add vector exp10/exp10f implementation to libmvec Message-ID: References: <20211229064000.1465621-1-skpgkp2@gmail.com> <20211229064000.1465621-6-skpgkp2@gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20211229064000.1465621-6-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 21:26:08 -0000 On Tue, Dec 28, 2021 at 10:39:47PM -0800, Sunil K Pandey wrote: > Implement vectorized exp10/exp10f containing SSE, AVX, AVX2 and > AVX512 versions for libmvec as per vector ABI. It also contains > accuracy and ABI tests for vector exp10/exp10f 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_exp102_core-sse2.S | 20 + > .../x86_64/fpu/multiarch/svml_d_exp102_core.c | 27 ++ > .../fpu/multiarch/svml_d_exp102_core_sse4.S | 418 +++++++++++++++++ > .../fpu/multiarch/svml_d_exp104_core-sse.S | 20 + > .../x86_64/fpu/multiarch/svml_d_exp104_core.c | 27 ++ > .../fpu/multiarch/svml_d_exp104_core_avx2.S | 429 ++++++++++++++++++ > .../fpu/multiarch/svml_d_exp108_core-avx2.S | 20 + > .../x86_64/fpu/multiarch/svml_d_exp108_core.c | 27 ++ > .../fpu/multiarch/svml_d_exp108_core_avx512.S | 287 ++++++++++++ > .../fpu/multiarch/svml_s_exp10f16_core-avx2.S | 20 + > .../fpu/multiarch/svml_s_exp10f16_core.c | 28 ++ > .../multiarch/svml_s_exp10f16_core_avx512.S | 269 +++++++++++ > .../fpu/multiarch/svml_s_exp10f4_core-sse2.S | 20 + > .../fpu/multiarch/svml_s_exp10f4_core.c | 28 ++ > .../fpu/multiarch/svml_s_exp10f4_core_sse4.S | 311 +++++++++++++ > .../fpu/multiarch/svml_s_exp10f8_core-sse.S | 20 + > .../fpu/multiarch/svml_s_exp10f8_core.c | 28 ++ > .../fpu/multiarch/svml_s_exp10f8_core_avx2.S | 331 ++++++++++++++ > sysdeps/x86_64/fpu/svml_d_exp102_core.S | 29 ++ > sysdeps/x86_64/fpu/svml_d_exp104_core.S | 29 ++ > sysdeps/x86_64/fpu/svml_d_exp104_core_avx.S | 25 + > sysdeps/x86_64/fpu/svml_d_exp108_core.S | 25 + > sysdeps/x86_64/fpu/svml_s_exp10f16_core.S | 25 + > sysdeps/x86_64/fpu/svml_s_exp10f4_core.S | 29 ++ > sysdeps/x86_64/fpu/svml_s_exp10f8_core.S | 29 ++ > sysdeps/x86_64/fpu/svml_s_exp10f8_core_avx.S | 25 + > .../fpu/test-double-libmvec-exp10-avx.c | 1 + > .../fpu/test-double-libmvec-exp10-avx2.c | 1 + > .../fpu/test-double-libmvec-exp10-avx512f.c | 1 + > .../x86_64/fpu/test-double-libmvec-exp10.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-exp10f-avx.c | 1 + > .../fpu/test-float-libmvec-exp10f-avx2.c | 1 + > .../fpu/test-float-libmvec-exp10f-avx512f.c | 1 + > .../x86_64/fpu/test-float-libmvec-exp10f.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, 2617 insertions(+), 1 deletion(-) > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_exp102_core-sse2.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_exp102_core.c > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_exp102_core_sse4.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_exp104_core-sse.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_exp104_core.c > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_exp104_core_avx2.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_exp108_core-avx2.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_exp108_core.c > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_exp108_core_avx512.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_exp10f16_core-avx2.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_exp10f16_core.c > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_exp10f16_core_avx512.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core-sse2.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core.c > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core_sse4.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_exp10f8_core-sse.S > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_exp10f8_core.c > create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_exp10f8_core_avx2.S > create mode 100644 sysdeps/x86_64/fpu/svml_d_exp102_core.S > create mode 100644 sysdeps/x86_64/fpu/svml_d_exp104_core.S > create mode 100644 sysdeps/x86_64/fpu/svml_d_exp104_core_avx.S > create mode 100644 sysdeps/x86_64/fpu/svml_d_exp108_core.S > create mode 100644 sysdeps/x86_64/fpu/svml_s_exp10f16_core.S > create mode 100644 sysdeps/x86_64/fpu/svml_s_exp10f4_core.S > create mode 100644 sysdeps/x86_64/fpu/svml_s_exp10f8_core.S > create mode 100644 sysdeps/x86_64/fpu/svml_s_exp10f8_core_avx.S > create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-exp10-avx.c > create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-exp10-avx2.c > create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-exp10-avx512f.c > create mode 100644 sysdeps/x86_64/fpu/test-double-libmvec-exp10.c > create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-exp10f-avx.c > create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-exp10f-avx2.c > create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-exp10f-avx512f.c > create mode 100644 sysdeps/x86_64/fpu/test-float-libmvec-exp10f.c > > diff --git a/bits/libm-simd-decl-stubs.h b/bits/libm-simd-decl-stubs.h > index 36d6643eb9..bc18621f17 100644 > --- a/bits/libm-simd-decl-stubs.h > +++ b/bits/libm-simd-decl-stubs.h > @@ -153,4 +153,15 @@ > #define __DECL_SIMD_exp2f32x > #define __DECL_SIMD_exp2f64x > #define __DECL_SIMD_exp2f128x > + > +#define __DECL_SIMD_exp10 > +#define __DECL_SIMD_exp10f > +#define __DECL_SIMD_exp10l > +#define __DECL_SIMD_exp10f16 > +#define __DECL_SIMD_exp10f32 > +#define __DECL_SIMD_exp10f64 > +#define __DECL_SIMD_exp10f128 > +#define __DECL_SIMD_exp10f32x > +#define __DECL_SIMD_exp10f64x > +#define __DECL_SIMD_exp10f128x > #endif > diff --git a/math/bits/mathcalls.h b/math/bits/mathcalls.h > index 645088cbf3..870778457f 100644 > --- a/math/bits/mathcalls.h > +++ b/math/bits/mathcalls.h > @@ -111,7 +111,7 @@ __MATHCALL (modf,, (_Mdouble_ __x, _Mdouble_ *__iptr)) __nonnull ((2)); > > #if __GLIBC_USE (IEC_60559_FUNCS_EXT_C2X) > /* Compute exponent to base ten. */ > -__MATHCALL (exp10,, (_Mdouble_ __x)); > +__MATHCALL_VEC (exp10,, (_Mdouble_ __x)); > #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 1717f2dee9..b3c1f59593 100644 > --- a/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist > +++ b/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist > @@ -49,40 +49,48 @@ 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 _ZGVbN2v_exp10 F > GLIBC_2.35 _ZGVbN2v_exp2 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 _ZGVbN4v_exp10f F > GLIBC_2.35 _ZGVbN4v_exp2f 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 _ZGVcN4v_exp10 F > GLIBC_2.35 _ZGVcN4v_exp2 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 _ZGVcN8v_exp10f F > GLIBC_2.35 _ZGVcN8v_exp2f 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 _ZGVdN4v_exp10 F > GLIBC_2.35 _ZGVdN4v_exp2 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 _ZGVdN8v_exp10f F > GLIBC_2.35 _ZGVdN8v_exp2f 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 _ZGVeN16v_exp10f F > GLIBC_2.35 _ZGVeN16v_exp2f 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 _ZGVeN8v_exp10 F > GLIBC_2.35 _ZGVeN8v_exp2 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 c7a972521b..f3f9c2e092 100644 > --- a/sysdeps/x86/fpu/bits/math-vector.h > +++ b/sysdeps/x86/fpu/bits/math-vector.h > @@ -78,6 +78,10 @@ > # define __DECL_SIMD_exp2 __DECL_SIMD_x86_64 > # undef __DECL_SIMD_exp2f > # define __DECL_SIMD_exp2f __DECL_SIMD_x86_64 > +# undef __DECL_SIMD_exp10 > +# define __DECL_SIMD_exp10 __DECL_SIMD_x86_64 > +# undef __DECL_SIMD_exp10f > +# define __DECL_SIMD_exp10f __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 0994e6dfac..c033abbedc 100644 > --- a/sysdeps/x86/fpu/finclude/math-vector-fortran.h > +++ b/sysdeps/x86/fpu/finclude/math-vector-fortran.h > @@ -38,6 +38,8 @@ > !GCC$ builtin (hypotf) attributes simd (notinbranch) if('x86_64') > !GCC$ builtin (exp2) attributes simd (notinbranch) if('x86_64') > !GCC$ builtin (exp2f) attributes simd (notinbranch) if('x86_64') > +!GCC$ builtin (exp10) attributes simd (notinbranch) if('x86_64') > +!GCC$ builtin (exp10f) attributes simd (notinbranch) if('x86_64') > > !GCC$ builtin (cos) attributes simd (notinbranch) if('x32') > !GCC$ builtin (cosf) attributes simd (notinbranch) if('x32') > @@ -61,3 +63,5 @@ > !GCC$ builtin (hypotf) attributes simd (notinbranch) if('x32') > !GCC$ builtin (exp2) attributes simd (notinbranch) if('x32') > !GCC$ builtin (exp2f) attributes simd (notinbranch) if('x32') > +!GCC$ builtin (exp10) attributes simd (notinbranch) if('x32') > +!GCC$ builtin (exp10f) attributes simd (notinbranch) if('x32') > diff --git a/sysdeps/x86_64/fpu/Makeconfig b/sysdeps/x86_64/fpu/Makeconfig > index 03b2364417..fd0a9da439 100644 > --- a/sysdeps/x86_64/fpu/Makeconfig > +++ b/sysdeps/x86_64/fpu/Makeconfig > @@ -27,6 +27,7 @@ libmvec-funcs = \ > atan \ > cos \ > exp \ > + exp10 \ > exp2 \ > hypot \ > log \ > diff --git a/sysdeps/x86_64/fpu/Versions b/sysdeps/x86_64/fpu/Versions > index 12b7ad1830..f29cfa4cbf 100644 > --- a/sysdeps/x86_64/fpu/Versions > +++ b/sysdeps/x86_64/fpu/Versions > @@ -17,11 +17,13 @@ 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; > + _ZGVbN2v_exp10; _ZGVcN4v_exp10; _ZGVdN4v_exp10; _ZGVeN8v_exp10; > _ZGVbN2v_exp2; _ZGVcN4v_exp2; _ZGVdN4v_exp2; _ZGVeN8v_exp2; > _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; > + _ZGVbN4v_exp10f; _ZGVcN8v_exp10f; _ZGVdN8v_exp10f; _ZGVeN16v_exp10f; > _ZGVbN4v_exp2f; _ZGVcN8v_exp2f; _ZGVdN8v_exp2f; _ZGVeN16v_exp2f; > _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 bc4479ad39..45f2e4bb53 100644 > --- a/sysdeps/x86_64/fpu/libm-test-ulps > +++ b/sysdeps/x86_64/fpu/libm-test-ulps > @@ -1252,6 +1252,26 @@ float: 1 > float128: 3 > ldouble: 2 > > +Function: "exp10_vlen16": > +float: 3 > + > +Function: "exp10_vlen2": > +double: 1 > + > +Function: "exp10_vlen4": > +double: 1 > +float: 1 > + > +Function: "exp10_vlen4_avx2": > +double: 1 > + > +Function: "exp10_vlen8": > +double: 1 > +float: 1 > + > +Function: "exp10_vlen8_avx2": > +float: 1 > + > Function: "exp2": > double: 1 > float: 1 > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp102_core-sse2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_exp102_core-sse2.S > new file mode 100644 > index 0000000000..ab615c0323 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp102_core-sse2.S > @@ -0,0 +1,20 @@ > +/* SSE2 version of vectorized exp10, 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 _ZGVbN2v_exp10 _ZGVbN2v_exp10_sse2 > +#include "../svml_d_exp102_core.S" > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp102_core.c b/sysdeps/x86_64/fpu/multiarch/svml_d_exp102_core.c > new file mode 100644 > index 0000000000..5c5625b278 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp102_core.c > @@ -0,0 +1,27 @@ > +/* Multiple versions of vectorized exp10, 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 _ZGVbN2v_exp10 > +#include "ifunc-mathvec-sse4_1.h" > + > +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); > + > +#ifdef SHARED > +__hidden_ver1 (_ZGVbN2v_exp10, __GI__ZGVbN2v_exp10, __redirect__ZGVbN2v_exp10) > + __attribute__ ((visibility ("hidden"))); > +#endif > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp102_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_d_exp102_core_sse4.S > new file mode 100644 > index 0000000000..7c6e5de3e0 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp102_core_sse4.S > @@ -0,0 +1,418 @@ > +/* Function exp10 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: > + * > + * exp10(x) = 2^x/log10(2) = 2^n * (1 + T[j]) * (1 + P(y)) > + * where > + * x = m*log10(2)/K + y, y in [-log10(2)/K..log10(2)/K] > + * m = n*K + j, m,n,j - signed integer, j in [-K/2..K/2] > + * > + * values of 2^j/K are tabulated > + * > + * P(y) is a minimax polynomial approximation of exp10(x)-1 > + * on small interval [-log10(2)/K..log10(2)/K] > + * > + * Special cases: > + * > + * exp10(NaN) = NaN > + * exp10(+INF) = +INF > + * exp10(-INF) = 0 > + * exp10(x) = 1 for subnormals > + * For IEEE double > + * if x > 3.39782712893383973096e+02 then exp10(x) overflow > + * if x < -3.45133219101941108420e+02 then exp10(x) underflow > + * > + */ > + > +/* Offsets for data table __svml_dexp10_data_internal > + */ > +#define _dbT 0 > +#define _dbLg2_10 1024 > +#define _dbShifter 1040 > +#define _dbInvLg2_10hi 1056 > +#define _dbInvLg2_10lo 1072 > +#define _dPC1 1088 > +#define _dPC2 1104 > +#define _dPC3 1120 > +#define _dPC4 1136 > +#define _dPC5 1152 > +#define _lExpMask 1168 > +#define _iIndexMask 1184 > +#define _iAbsMask 1200 > +#define _iDomainRange 1216 > + > +#include > + > + .text > + .section .text.sse4,"ax",@progbits > +ENTRY(_ZGVbN2v_exp10_sse4) > + subq $72, %rsp > + cfi_def_cfa_offset(80) > + > +/* R */ > + movaps %xmm0, %xmm12 > + > +/* Load arument */ > + movups _dbLg2_10+__svml_dexp10_data_internal(%rip), %xmm13 > + lea __svml_dexp10_data_internal(%rip), %rsi > + mulpd %xmm0, %xmm13 > + movups _dbShifter+__svml_dexp10_data_internal(%rip), %xmm1 > + addpd %xmm1, %xmm13 > + movaps %xmm13, %xmm9 > + subpd %xmm1, %xmm9 > + movups _dbInvLg2_10hi+__svml_dexp10_data_internal(%rip), %xmm8 > + mulpd %xmm9, %xmm8 > + movups _dbInvLg2_10lo+__svml_dexp10_data_internal(%rip), %xmm10 > + mulpd %xmm9, %xmm10 > + subpd %xmm8, %xmm12 > + subpd %xmm10, %xmm12 > + > +/* > + * Polynomial > + * poly(dN) = a1*dR+...+a5*dR^5 > + */ > + movups _dPC5+__svml_dexp10_data_internal(%rip), %xmm11 > + mulpd %xmm12, %xmm11 > + addpd _dPC4+__svml_dexp10_data_internal(%rip), %xmm11 > + mulpd %xmm12, %xmm11 > + addpd _dPC3+__svml_dexp10_data_internal(%rip), %xmm11 > + mulpd %xmm12, %xmm11 > + addpd _dPC2+__svml_dexp10_data_internal(%rip), %xmm11 > + > +/* a1+...+a5*dR^4 ! */ > + mulpd %xmm12, %xmm11 > + addpd _dPC1+__svml_dexp10_data_internal(%rip), %xmm11 > + movq _iIndexMask+__svml_dexp10_data_internal(%rip), %xmm5 > + > +/* Index and lookup */ > + pshufd $136, %xmm13, %xmm6 > + > +/* 2^N */ > + psllq $45, %xmm13 > + pand %xmm5, %xmm6 > + > +/* iIndex*=sizeof(D); */ > + pslld $3, %xmm6 > + movd %xmm6, %eax > + pshufd $1, %xmm6, %xmm7 > + movq _iAbsMask+__svml_dexp10_data_internal(%rip), %xmm2 > + > +/* a1*dR+...+a5*dR^5 */ > + mulpd %xmm11, %xmm12 > + movd %xmm7, %ecx > + > +/* Check for overflow\underflow */ > + pshufd $221, %xmm0, %xmm4 > + movq _iDomainRange+__svml_dexp10_data_internal(%rip), %xmm3 > + pand %xmm2, %xmm4 > + movslq %eax, %rax > + pcmpgtd %xmm3, %xmm4 > + movslq %ecx, %rcx > + movmskps %xmm4, %edx > + > +/* lM==EXP(2^N) */ > + pand _lExpMask+__svml_dexp10_data_internal(%rip), %xmm13 > + movsd (%rsi,%rax), %xmm1 > + movhpd (%rsi,%rcx), %xmm1 > + > +/* Tj*poly */ > + mulpd %xmm1, %xmm12 > + addpd %xmm12, %xmm1 > + > +/* quick 2^N */ > + paddq %xmm13, %xmm1 > + andl $3, %edx > + > +/* Finish */ > + > +/* Go to special inputs processing branch */ > + jne L(SPECIAL_VALUES_BRANCH) > + # LOE rbx rbp r12 r13 r14 r15 edx xmm0 xmm1 > + > +/* Restore registers > + * and exit the function > + */ > + > +L(EXIT): > + movaps %xmm1, %xmm0 > + addq $72, %rsp > + cfi_def_cfa_offset(8) > + ret > + cfi_def_cfa_offset(80) > + > +/* Branch to process > + * special inputs > + */ > + > +L(SPECIAL_VALUES_BRANCH): > + movups %xmm0, 32(%rsp) > + movups %xmm1, 48(%rsp) > + # LOE rbx rbp r12 r13 r14 r15 edx xmm1 > + > + xorl %eax, %eax > + movq %r12, 16(%rsp) > + cfi_offset(12, -64) > + movl %eax, %r12d > + movq %r13, 8(%rsp) > + cfi_offset(13, -72) > + movl %edx, %r13d > + movq %r14, (%rsp) > + cfi_offset(14, -80) > + # 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 48(%rsp), %xmm1 > + > +/* Go to exit */ > + jmp L(EXIT) > + cfi_offset(12, -64) > + cfi_offset(13, -72) > + cfi_offset(14, -80) > + # LOE rbx rbp r12 r13 r14 r15 xmm1 > + > +/* Scalar math fucntion call > + * to process special input > + */ > + > +L(SCALAR_MATH_CALL): > + movl %r12d, %r14d > + movsd 32(%rsp,%r14,8), %xmm0 > + call exp10@PLT > + # LOE rbx rbp r14 r15 r12d r13d xmm0 > + > + movsd %xmm0, 48(%rsp,%r14,8) > + > +/* Process special inputs in loop */ > + jmp L(SPECIAL_VALUES_LOOP) > + # LOE rbx rbp r15 r12d r13d > +END(_ZGVbN2v_exp10_sse4) > + > + .section .rodata, "a" > + .align 16 > + > +#ifdef __svml_dexp10_data_internal_typedef > +typedef unsigned int VUINT32; > +typedef struct > +{ > + __declspec(align(16)) VUINT32 _dbT[(1<<7)][2]; > + __declspec(align(16)) VUINT32 _dbLg2_10[2][2]; > + __declspec(align(16)) VUINT32 _dbShifter[2][2]; > + __declspec(align(16)) VUINT32 _dbInvLg2_10hi[2][2]; > + __declspec(align(16)) VUINT32 _dbInvLg2_10lo[2][2]; > + __declspec(align(16)) VUINT32 _dPC1[2][2]; > + __declspec(align(16)) VUINT32 _dPC2[2][2]; > + __declspec(align(16)) VUINT32 _dPC3[2][2]; > + __declspec(align(16)) VUINT32 _dPC4[2][2]; > + __declspec(align(16)) VUINT32 _dPC5[2][2]; > + __declspec(align(16)) VUINT32 _lExpMask[2][2]; > + __declspec(align(16)) VUINT32 _iIndexMask[4][1]; > + __declspec(align(16)) VUINT32 _iAbsMask[4][1]; > + __declspec(align(16)) VUINT32 _iDomainRange[4][1]; > +} __svml_dexp10_data_internal; > +#endif > +__svml_dexp10_data_internal: > + /*== _dbT ==*/ > + .quad 0x3ff0000000000000 /*2^( 0 /128)*/ > + .quad 0x3ff0163da9fb3335 /*2^( 1 /128)*/ > + .quad 0x3ff02c9a3e778061 /*2^( 2 /128)*/ > + .quad 0x3ff04315e86e7f85 /*2^( 3 /128)*/ > + .quad 0x3ff059b0d3158574 /*2^( 4 /128)*/ > + .quad 0x3ff0706b29ddf6de /*2^( 5 /128)*/ > + .quad 0x3ff0874518759bc8 /*2^( 6 /128)*/ > + .quad 0x3ff09e3ecac6f383 /*2^( 7 /128)*/ > + .quad 0x3ff0b5586cf9890f /*2^( 8 /128)*/ > + .quad 0x3ff0cc922b7247f7 /*2^( 9 /128)*/ > + .quad 0x3ff0e3ec32d3d1a2 /*2^( 10 /128)*/ > + .quad 0x3ff0fb66affed31b /*2^( 11 /128)*/ > + .quad 0x3ff11301d0125b51 /*2^( 12 /128)*/ > + .quad 0x3ff12abdc06c31cc /*2^( 13 /128)*/ > + .quad 0x3ff1429aaea92de0 /*2^( 14 /128)*/ > + .quad 0x3ff15a98c8a58e51 /*2^( 15 /128)*/ > + .quad 0x3ff172b83c7d517b /*2^( 16 /128)*/ > + .quad 0x3ff18af9388c8dea /*2^( 17 /128)*/ > + .quad 0x3ff1a35beb6fcb75 /*2^( 18 /128)*/ > + .quad 0x3ff1bbe084045cd4 /*2^( 19 /128)*/ > + .quad 0x3ff1d4873168b9aa /*2^( 20 /128)*/ > + .quad 0x3ff1ed5022fcd91d /*2^( 21 /128)*/ > + .quad 0x3ff2063b88628cd6 /*2^( 22 /128)*/ > + .quad 0x3ff21f49917ddc96 /*2^( 23 /128)*/ > + .quad 0x3ff2387a6e756238 /*2^( 24 /128)*/ > + .quad 0x3ff251ce4fb2a63f /*2^( 25 /128)*/ > + .quad 0x3ff26b4565e27cdd /*2^( 26 /128)*/ > + .quad 0x3ff284dfe1f56381 /*2^( 27 /128)*/ > + .quad 0x3ff29e9df51fdee1 /*2^( 28 /128)*/ > + .quad 0x3ff2b87fd0dad990 /*2^( 29 /128)*/ > + .quad 0x3ff2d285a6e4030b /*2^( 30 /128)*/ > + .quad 0x3ff2ecafa93e2f56 /*2^( 31 /128)*/ > + .quad 0x3ff306fe0a31b715 /*2^( 32 /128)*/ > + .quad 0x3ff32170fc4cd831 /*2^( 33 /128)*/ > + .quad 0x3ff33c08b26416ff /*2^( 34 /128)*/ > + .quad 0x3ff356c55f929ff1 /*2^( 35 /128)*/ > + .quad 0x3ff371a7373aa9cb /*2^( 36 /128)*/ > + .quad 0x3ff38cae6d05d866 /*2^( 37 /128)*/ > + .quad 0x3ff3a7db34e59ff7 /*2^( 38 /128)*/ > + .quad 0x3ff3c32dc313a8e5 /*2^( 39 /128)*/ > + .quad 0x3ff3dea64c123422 /*2^( 40 /128)*/ > + .quad 0x3ff3fa4504ac801c /*2^( 41 /128)*/ > + .quad 0x3ff4160a21f72e2a /*2^( 42 /128)*/ > + .quad 0x3ff431f5d950a897 /*2^( 43 /128)*/ > + .quad 0x3ff44e086061892d /*2^( 44 /128)*/ > + .quad 0x3ff46a41ed1d0057 /*2^( 45 /128)*/ > + .quad 0x3ff486a2b5c13cd0 /*2^( 46 /128)*/ > + .quad 0x3ff4a32af0d7d3de /*2^( 47 /128)*/ > + .quad 0x3ff4bfdad5362a27 /*2^( 48 /128)*/ > + .quad 0x3ff4dcb299fddd0d /*2^( 49 /128)*/ > + .quad 0x3ff4f9b2769d2ca7 /*2^( 50 /128)*/ > + .quad 0x3ff516daa2cf6642 /*2^( 51 /128)*/ > + .quad 0x3ff5342b569d4f82 /*2^( 52 /128)*/ > + .quad 0x3ff551a4ca5d920f /*2^( 53 /128)*/ > + .quad 0x3ff56f4736b527da /*2^( 54 /128)*/ > + .quad 0x3ff58d12d497c7fd /*2^( 55 /128)*/ > + .quad 0x3ff5ab07dd485429 /*2^( 56 /128)*/ > + .quad 0x3ff5c9268a5946b7 /*2^( 57 /128)*/ > + .quad 0x3ff5e76f15ad2148 /*2^( 58 /128)*/ > + .quad 0x3ff605e1b976dc09 /*2^( 59 /128)*/ > + .quad 0x3ff6247eb03a5585 /*2^( 60 /128)*/ > + .quad 0x3ff6434634ccc320 /*2^( 61 /128)*/ > + .quad 0x3ff6623882552225 /*2^( 62 /128)*/ > + .quad 0x3ff68155d44ca973 /*2^( 63 /128)*/ > + .quad 0x3ff6a09e667f3bcd /*2^( 64 /128)*/ > + .quad 0x3ff6c012750bdabf /*2^( 65 /128)*/ > + .quad 0x3ff6dfb23c651a2f /*2^( 66 /128)*/ > + .quad 0x3ff6ff7df9519484 /*2^( 67 /128)*/ > + .quad 0x3ff71f75e8ec5f74 /*2^( 68 /128)*/ > + .quad 0x3ff73f9a48a58174 /*2^( 69 /128)*/ > + .quad 0x3ff75feb564267c9 /*2^( 70 /128)*/ > + .quad 0x3ff780694fde5d3f /*2^( 71 /128)*/ > + .quad 0x3ff7a11473eb0187 /*2^( 72 /128)*/ > + .quad 0x3ff7c1ed0130c132 /*2^( 73 /128)*/ > + .quad 0x3ff7e2f336cf4e62 /*2^( 74 /128)*/ > + .quad 0x3ff80427543e1a12 /*2^( 75 /128)*/ > + .quad 0x3ff82589994cce13 /*2^( 76 /128)*/ > + .quad 0x3ff8471a4623c7ad /*2^( 77 /128)*/ > + .quad 0x3ff868d99b4492ed /*2^( 78 /128)*/ > + .quad 0x3ff88ac7d98a6699 /*2^( 79 /128)*/ > + .quad 0x3ff8ace5422aa0db /*2^( 80 /128)*/ > + .quad 0x3ff8cf3216b5448c /*2^( 81 /128)*/ > + .quad 0x3ff8f1ae99157736 /*2^( 82 /128)*/ > + .quad 0x3ff9145b0b91ffc6 /*2^( 83 /128)*/ > + .quad 0x3ff93737b0cdc5e5 /*2^( 84 /128)*/ > + .quad 0x3ff95a44cbc8520f /*2^( 85 /128)*/ > + .quad 0x3ff97d829fde4e50 /*2^( 86 /128)*/ > + .quad 0x3ff9a0f170ca07ba /*2^( 87 /128)*/ > + .quad 0x3ff9c49182a3f090 /*2^( 88 /128)*/ > + .quad 0x3ff9e86319e32323 /*2^( 89 /128)*/ > + .quad 0x3ffa0c667b5de565 /*2^( 90 /128)*/ > + .quad 0x3ffa309bec4a2d33 /*2^( 91 /128)*/ > + .quad 0x3ffa5503b23e255d /*2^( 92 /128)*/ > + .quad 0x3ffa799e1330b358 /*2^( 93 /128)*/ > + .quad 0x3ffa9e6b5579fdbf /*2^( 94 /128)*/ > + .quad 0x3ffac36bbfd3f37a /*2^( 95 /128)*/ > + .quad 0x3ffae89f995ad3ad /*2^( 96 /128)*/ > + .quad 0x3ffb0e07298db666 /*2^( 97 /128)*/ > + .quad 0x3ffb33a2b84f15fb /*2^( 98 /128)*/ > + .quad 0x3ffb59728de5593a /*2^( 99 /128)*/ > + .quad 0x3ffb7f76f2fb5e47 /*2^( 100 /128)*/ > + .quad 0x3ffba5b030a1064a /*2^( 101 /128)*/ > + .quad 0x3ffbcc1e904bc1d2 /*2^( 102 /128)*/ > + .quad 0x3ffbf2c25bd71e09 /*2^( 103 /128)*/ > + .quad 0x3ffc199bdd85529c /*2^( 104 /128)*/ > + .quad 0x3ffc40ab5fffd07a /*2^( 105 /128)*/ > + .quad 0x3ffc67f12e57d14b /*2^( 106 /128)*/ > + .quad 0x3ffc8f6d9406e7b5 /*2^( 107 /128)*/ > + .quad 0x3ffcb720dcef9069 /*2^( 108 /128)*/ > + .quad 0x3ffcdf0b555dc3fa /*2^( 109 /128)*/ > + .quad 0x3ffd072d4a07897c /*2^( 110 /128)*/ > + .quad 0x3ffd2f87080d89f2 /*2^( 111 /128)*/ > + .quad 0x3ffd5818dcfba487 /*2^( 112 /128)*/ > + .quad 0x3ffd80e316c98398 /*2^( 113 /128)*/ > + .quad 0x3ffda9e603db3285 /*2^( 114 /128)*/ > + .quad 0x3ffdd321f301b460 /*2^( 115 /128)*/ > + .quad 0x3ffdfc97337b9b5f /*2^( 116 /128)*/ > + .quad 0x3ffe264614f5a129 /*2^( 117 /128)*/ > + .quad 0x3ffe502ee78b3ff6 /*2^( 118 /128)*/ > + .quad 0x3ffe7a51fbc74c83 /*2^( 119 /128)*/ > + .quad 0x3ffea4afa2a490da /*2^( 120 /128)*/ > + .quad 0x3ffecf482d8e67f1 /*2^( 121 /128)*/ > + .quad 0x3ffefa1bee615a27 /*2^( 122 /128)*/ > + .quad 0x3fff252b376bba97 /*2^( 123 /128)*/ > + .quad 0x3fff50765b6e4540 /*2^( 124 /128)*/ > + .quad 0x3fff7bfdad9cbe14 /*2^( 125 /128)*/ > + .quad 0x3fffa7c1819e90d8 /*2^( 126 /128)*/ > + .quad 0x3fffd3c22b8f71f1 /*2^( 127 /128)*/ > + .align 16 > + .quad 0x407a934f0979a371, 0x407a934f0979a371 /* _dbLg2_10*2^K */ > + .align 16 > + .quad 0x4338800000000000, 0x4338800000000000 /* _dbShifter */ > + .align 16 > + .quad 0x3f63441350a00000, 0x3f63441350a00000 /* _dbInvLg2_10hi/2^K 53-11-K bits*/ > + .align 16 > + .quad 0xbd10c0219dc1da99, 0xbd10c0219dc1da99 /* _dbInvLg2_10lo/2^K */ > + //PC0 = 1.0 > + .align 16 > + .quad 0x40026bb1bbb55516, 0x40026bb1bbb55516 /* _dPC1 */ > + .align 16 > + .quad 0x40053524c73ce8e3, 0x40053524c73ce8e3 /* _dPC2 */ > + .align 16 > + .quad 0x4000470591ccea8b, 0x4000470591ccea8b /* _dPC3 */ > + .align 16 > + .quad 0x3ff2bd767584db59, 0x3ff2bd767584db59 /* _dPC4 */ > + .align 16 > + .quad 0x3fe144c03efafb54, 0x3fe144c03efafb54 /* _dPC5 */ > + .align 16 > + .quad 0xfff0000000000000, 0xfff0000000000000 /* _lExpMask */ > + .align 16 > + .long 0x0000007f, 0x0000007f, 0x0000007f, 0x0000007f /* _iIndexMask =(2^K-1)*/ > + //common > + .align 16 > + .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff /* _iAbsMask */ > + .align 16 > + .long 0x40733a70, 0x40733a70, 0x40733a70, 0x40733a70 /* _iDomainRange */ > + .align 16 > + .type __svml_dexp10_data_internal,@object > + .size __svml_dexp10_data_internal,.-__svml_dexp10_data_internal > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp104_core-sse.S b/sysdeps/x86_64/fpu/multiarch/svml_d_exp104_core-sse.S > new file mode 100644 > index 0000000000..260c052143 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp104_core-sse.S > @@ -0,0 +1,20 @@ > +/* SSE version of vectorized exp10, 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 _ZGVdN4v_exp10 _ZGVdN4v_exp10_sse_wrapper > +#include "../svml_d_exp104_core.S" > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp104_core.c b/sysdeps/x86_64/fpu/multiarch/svml_d_exp104_core.c > new file mode 100644 > index 0000000000..e3e302be72 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp104_core.c > @@ -0,0 +1,27 @@ > +/* Multiple versions of vectorized exp10, 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 _ZGVdN4v_exp10 > +#include "ifunc-mathvec-avx2.h" > + > +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); > + > +#ifdef SHARED > +__hidden_ver1 (_ZGVdN4v_exp10, __GI__ZGVdN4v_exp10, __redirect__ZGVdN4v_exp10) > + __attribute__ ((visibility ("hidden"))); > +#endif > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp104_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_exp104_core_avx2.S > new file mode 100644 > index 0000000000..1a53f43c9e > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp104_core_avx2.S > @@ -0,0 +1,429 @@ > +/* Function exp10 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: > + * > + * exp10(x) = 2^x/log10(2) = 2^n * (1 + T[j]) * (1 + P(y)) > + * where > + * x = m*log10(2)/K + y, y in [-log10(2)/K..log10(2)/K] > + * m = n*K + j, m,n,j - signed integer, j in [-K/2..K/2] > + * > + * values of 2^j/K are tabulated > + * > + * P(y) is a minimax polynomial approximation of exp10(x)-1 > + * on small interval [-log10(2)/K..log10(2)/K] > + * > + * Special cases: > + * > + * exp10(NaN) = NaN > + * exp10(+INF) = +INF > + * exp10(-INF) = 0 > + * exp10(x) = 1 for subnormals > + * For IEEE double > + * if x > 3.39782712893383973096e+02 then exp10(x) overflow > + * if x < -3.45133219101941108420e+02 then exp10(x) underflow > + * > + */ > + > +/* Offsets for data table __svml_dexp10_data_internal > + */ > +#define _dbT 0 > +#define _dbLg2_10 1024 > +#define _dbShifter 1056 > +#define _dbInvLg2_10hi 1088 > +#define _dbInvLg2_10lo 1120 > +#define _dPC1 1152 > +#define _dPC2 1184 > +#define _dPC3 1216 > +#define _dPC4 1248 > +#define _dPC5 1280 > +#define _lExpMask 1312 > +#define _iIndexMask 1344 > +#define _iAbsMask 1376 > +#define _iDomainRange 1408 > + > +#include > + > + .text > + .section .text.avx2,"ax",@progbits > +ENTRY(_ZGVdN4v_exp10_avx2) > + pushq %rbp > + cfi_def_cfa_offset(16) > + movq %rsp, %rbp > + cfi_def_cfa(6, 16) > + cfi_offset(6, -16) > + andq $-32, %rsp > + subq $96, %rsp > + lea __svml_dexp10_data_internal(%rip), %r8 > + vmovapd %ymm0, %ymm2 > + vmovupd _dbShifter+__svml_dexp10_data_internal(%rip), %ymm3 > + > +/* Load arument */ > + vmovupd _dbLg2_10+__svml_dexp10_data_internal(%rip), %ymm0 > + vfmadd213pd %ymm3, %ymm2, %ymm0 > + vsubpd %ymm3, %ymm0, %ymm1 > + > +/* R */ > + vmovupd _dbInvLg2_10hi+__svml_dexp10_data_internal(%rip), %ymm3 > + vfnmadd213pd %ymm2, %ymm1, %ymm3 > + > +/* Check for overflow\underflow */ > + vextractf128 $1, %ymm2, %xmm4 > + vfnmadd132pd _dbInvLg2_10lo+__svml_dexp10_data_internal(%rip), %ymm3, %ymm1 > + vshufps $221, %xmm4, %xmm2, %xmm5 > + vandps _iAbsMask+__svml_dexp10_data_internal(%rip), %xmm5, %xmm6 > + vpcmpgtd _iDomainRange+__svml_dexp10_data_internal(%rip), %xmm6, %xmm7 > + > +/* > + * Polynomial > + * poly(dN) = a1*dR+...+a5*dR^5 > + */ > + vmovupd _dPC5+__svml_dexp10_data_internal(%rip), %ymm4 > + vmovmskps %xmm7, %eax > + vfmadd213pd _dPC4+__svml_dexp10_data_internal(%rip), %ymm1, %ymm4 > + vfmadd213pd _dPC3+__svml_dexp10_data_internal(%rip), %ymm1, %ymm4 > + vfmadd213pd _dPC2+__svml_dexp10_data_internal(%rip), %ymm1, %ymm4 > + > +/* a1+...+a5*dR^4 ! */ > + vfmadd213pd _dPC1+__svml_dexp10_data_internal(%rip), %ymm1, %ymm4 > + > +/* a1*dR+...+a5*dR^5 */ > + vmulpd %ymm4, %ymm1, %ymm1 > + > +/* Index and lookup */ > + vextractf128 $1, %ymm0, %xmm8 > + vshufps $136, %xmm8, %xmm0, %xmm9 > + vandps _iIndexMask+__svml_dexp10_data_internal(%rip), %xmm9, %xmm10 > + > +/* iIndex*=sizeof(D); */ > + vpslld $3, %xmm10, %xmm13 > + vmovd %xmm13, %edx > + > +/* 2^N */ > + vpsllq $45, %ymm0, %ymm0 > + vpextrd $2, %xmm13, %esi > + movslq %edx, %rdx > + vpextrd $1, %xmm13, %ecx > + movslq %esi, %rsi > + vpextrd $3, %xmm13, %edi > + movslq %ecx, %rcx > + movslq %edi, %rdi > + vmovsd (%r8,%rdx), %xmm11 > + vmovsd (%r8,%rsi), %xmm14 > + vmovhpd (%r8,%rcx), %xmm11, %xmm12 > + vmovhpd (%r8,%rdi), %xmm14, %xmm15 > + > +/* lM==EXP(2^N) */ > + vpand _lExpMask+__svml_dexp10_data_internal(%rip), %ymm0, %ymm6 > + vinsertf128 $1, %xmm15, %ymm12, %ymm5 > + > +/* Tj*poly */ > + vfmadd213pd %ymm5, %ymm5, %ymm1 > + > +/* quick 2^N */ > + vpaddq %ymm6, %ymm1, %ymm0 > + > +/* Finish */ > + testl %eax, %eax > + > +/* Go to special inputs processing branch */ > + jne L(SPECIAL_VALUES_BRANCH) > + # LOE rbx r12 r13 r14 r15 eax ymm0 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 %ymm2, 32(%rsp) > + vmovupd %ymm0, 64(%rsp) > + # LOE rbx r12 r13 r14 r15 eax ymm0 > + > + xorl %edx, %edx > + # 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: -80; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22 > + movl %edx, %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: -88; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22 > + movl %eax, %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: -96; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 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 64(%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: -80; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 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: -88; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 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: -96; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 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 > + call exp10@PLT > + # LOE rbx r14 r15 r12d r13d xmm0 > + > + movsd %xmm0, 64(%rsp,%r14,8) > + > +/* Process special inputs in loop */ > + jmp L(SPECIAL_VALUES_LOOP) > + # LOE rbx r15 r12d r13d > +END(_ZGVdN4v_exp10_avx2) > + > + .section .rodata, "a" > + .align 32 > + > +#ifdef __svml_dexp10_data_internal_typedef > +typedef unsigned int VUINT32; > +typedef struct > +{ > + __declspec(align(32)) VUINT32 _dbT[(1<<7)][2]; > + __declspec(align(32)) VUINT32 _dbLg2_10[4][2]; > + __declspec(align(32)) VUINT32 _dbShifter[4][2]; > + __declspec(align(32)) VUINT32 _dbInvLg2_10hi[4][2]; > + __declspec(align(32)) VUINT32 _dbInvLg2_10lo[4][2]; > + __declspec(align(32)) VUINT32 _dPC1[4][2]; > + __declspec(align(32)) VUINT32 _dPC2[4][2]; > + __declspec(align(32)) VUINT32 _dPC3[4][2]; > + __declspec(align(32)) VUINT32 _dPC4[4][2]; > + __declspec(align(32)) VUINT32 _dPC5[4][2]; > + __declspec(align(32)) VUINT32 _lExpMask[4][2]; > + __declspec(align(32)) VUINT32 _iIndexMask[8][1]; > + __declspec(align(32)) VUINT32 _iAbsMask[8][1]; > + __declspec(align(32)) VUINT32 _iDomainRange[8][1]; > +} __svml_dexp10_data_internal; > +#endif > +__svml_dexp10_data_internal: > + /*== _dbT ==*/ > + .quad 0x3ff0000000000000 /*2^( 0 /128)*/ > + .quad 0x3ff0163da9fb3335 /*2^( 1 /128)*/ > + .quad 0x3ff02c9a3e778061 /*2^( 2 /128)*/ > + .quad 0x3ff04315e86e7f85 /*2^( 3 /128)*/ > + .quad 0x3ff059b0d3158574 /*2^( 4 /128)*/ > + .quad 0x3ff0706b29ddf6de /*2^( 5 /128)*/ > + .quad 0x3ff0874518759bc8 /*2^( 6 /128)*/ > + .quad 0x3ff09e3ecac6f383 /*2^( 7 /128)*/ > + .quad 0x3ff0b5586cf9890f /*2^( 8 /128)*/ > + .quad 0x3ff0cc922b7247f7 /*2^( 9 /128)*/ > + .quad 0x3ff0e3ec32d3d1a2 /*2^( 10 /128)*/ > + .quad 0x3ff0fb66affed31b /*2^( 11 /128)*/ > + .quad 0x3ff11301d0125b51 /*2^( 12 /128)*/ > + .quad 0x3ff12abdc06c31cc /*2^( 13 /128)*/ > + .quad 0x3ff1429aaea92de0 /*2^( 14 /128)*/ > + .quad 0x3ff15a98c8a58e51 /*2^( 15 /128)*/ > + .quad 0x3ff172b83c7d517b /*2^( 16 /128)*/ > + .quad 0x3ff18af9388c8dea /*2^( 17 /128)*/ > + .quad 0x3ff1a35beb6fcb75 /*2^( 18 /128)*/ > + .quad 0x3ff1bbe084045cd4 /*2^( 19 /128)*/ > + .quad 0x3ff1d4873168b9aa /*2^( 20 /128)*/ > + .quad 0x3ff1ed5022fcd91d /*2^( 21 /128)*/ > + .quad 0x3ff2063b88628cd6 /*2^( 22 /128)*/ > + .quad 0x3ff21f49917ddc96 /*2^( 23 /128)*/ > + .quad 0x3ff2387a6e756238 /*2^( 24 /128)*/ > + .quad 0x3ff251ce4fb2a63f /*2^( 25 /128)*/ > + .quad 0x3ff26b4565e27cdd /*2^( 26 /128)*/ > + .quad 0x3ff284dfe1f56381 /*2^( 27 /128)*/ > + .quad 0x3ff29e9df51fdee1 /*2^( 28 /128)*/ > + .quad 0x3ff2b87fd0dad990 /*2^( 29 /128)*/ > + .quad 0x3ff2d285a6e4030b /*2^( 30 /128)*/ > + .quad 0x3ff2ecafa93e2f56 /*2^( 31 /128)*/ > + .quad 0x3ff306fe0a31b715 /*2^( 32 /128)*/ > + .quad 0x3ff32170fc4cd831 /*2^( 33 /128)*/ > + .quad 0x3ff33c08b26416ff /*2^( 34 /128)*/ > + .quad 0x3ff356c55f929ff1 /*2^( 35 /128)*/ > + .quad 0x3ff371a7373aa9cb /*2^( 36 /128)*/ > + .quad 0x3ff38cae6d05d866 /*2^( 37 /128)*/ > + .quad 0x3ff3a7db34e59ff7 /*2^( 38 /128)*/ > + .quad 0x3ff3c32dc313a8e5 /*2^( 39 /128)*/ > + .quad 0x3ff3dea64c123422 /*2^( 40 /128)*/ > + .quad 0x3ff3fa4504ac801c /*2^( 41 /128)*/ > + .quad 0x3ff4160a21f72e2a /*2^( 42 /128)*/ > + .quad 0x3ff431f5d950a897 /*2^( 43 /128)*/ > + .quad 0x3ff44e086061892d /*2^( 44 /128)*/ > + .quad 0x3ff46a41ed1d0057 /*2^( 45 /128)*/ > + .quad 0x3ff486a2b5c13cd0 /*2^( 46 /128)*/ > + .quad 0x3ff4a32af0d7d3de /*2^( 47 /128)*/ > + .quad 0x3ff4bfdad5362a27 /*2^( 48 /128)*/ > + .quad 0x3ff4dcb299fddd0d /*2^( 49 /128)*/ > + .quad 0x3ff4f9b2769d2ca7 /*2^( 50 /128)*/ > + .quad 0x3ff516daa2cf6642 /*2^( 51 /128)*/ > + .quad 0x3ff5342b569d4f82 /*2^( 52 /128)*/ > + .quad 0x3ff551a4ca5d920f /*2^( 53 /128)*/ > + .quad 0x3ff56f4736b527da /*2^( 54 /128)*/ > + .quad 0x3ff58d12d497c7fd /*2^( 55 /128)*/ > + .quad 0x3ff5ab07dd485429 /*2^( 56 /128)*/ > + .quad 0x3ff5c9268a5946b7 /*2^( 57 /128)*/ > + .quad 0x3ff5e76f15ad2148 /*2^( 58 /128)*/ > + .quad 0x3ff605e1b976dc09 /*2^( 59 /128)*/ > + .quad 0x3ff6247eb03a5585 /*2^( 60 /128)*/ > + .quad 0x3ff6434634ccc320 /*2^( 61 /128)*/ > + .quad 0x3ff6623882552225 /*2^( 62 /128)*/ > + .quad 0x3ff68155d44ca973 /*2^( 63 /128)*/ > + .quad 0x3ff6a09e667f3bcd /*2^( 64 /128)*/ > + .quad 0x3ff6c012750bdabf /*2^( 65 /128)*/ > + .quad 0x3ff6dfb23c651a2f /*2^( 66 /128)*/ > + .quad 0x3ff6ff7df9519484 /*2^( 67 /128)*/ > + .quad 0x3ff71f75e8ec5f74 /*2^( 68 /128)*/ > + .quad 0x3ff73f9a48a58174 /*2^( 69 /128)*/ > + .quad 0x3ff75feb564267c9 /*2^( 70 /128)*/ > + .quad 0x3ff780694fde5d3f /*2^( 71 /128)*/ > + .quad 0x3ff7a11473eb0187 /*2^( 72 /128)*/ > + .quad 0x3ff7c1ed0130c132 /*2^( 73 /128)*/ > + .quad 0x3ff7e2f336cf4e62 /*2^( 74 /128)*/ > + .quad 0x3ff80427543e1a12 /*2^( 75 /128)*/ > + .quad 0x3ff82589994cce13 /*2^( 76 /128)*/ > + .quad 0x3ff8471a4623c7ad /*2^( 77 /128)*/ > + .quad 0x3ff868d99b4492ed /*2^( 78 /128)*/ > + .quad 0x3ff88ac7d98a6699 /*2^( 79 /128)*/ > + .quad 0x3ff8ace5422aa0db /*2^( 80 /128)*/ > + .quad 0x3ff8cf3216b5448c /*2^( 81 /128)*/ > + .quad 0x3ff8f1ae99157736 /*2^( 82 /128)*/ > + .quad 0x3ff9145b0b91ffc6 /*2^( 83 /128)*/ > + .quad 0x3ff93737b0cdc5e5 /*2^( 84 /128)*/ > + .quad 0x3ff95a44cbc8520f /*2^( 85 /128)*/ > + .quad 0x3ff97d829fde4e50 /*2^( 86 /128)*/ > + .quad 0x3ff9a0f170ca07ba /*2^( 87 /128)*/ > + .quad 0x3ff9c49182a3f090 /*2^( 88 /128)*/ > + .quad 0x3ff9e86319e32323 /*2^( 89 /128)*/ > + .quad 0x3ffa0c667b5de565 /*2^( 90 /128)*/ > + .quad 0x3ffa309bec4a2d33 /*2^( 91 /128)*/ > + .quad 0x3ffa5503b23e255d /*2^( 92 /128)*/ > + .quad 0x3ffa799e1330b358 /*2^( 93 /128)*/ > + .quad 0x3ffa9e6b5579fdbf /*2^( 94 /128)*/ > + .quad 0x3ffac36bbfd3f37a /*2^( 95 /128)*/ > + .quad 0x3ffae89f995ad3ad /*2^( 96 /128)*/ > + .quad 0x3ffb0e07298db666 /*2^( 97 /128)*/ > + .quad 0x3ffb33a2b84f15fb /*2^( 98 /128)*/ > + .quad 0x3ffb59728de5593a /*2^( 99 /128)*/ > + .quad 0x3ffb7f76f2fb5e47 /*2^( 100 /128)*/ > + .quad 0x3ffba5b030a1064a /*2^( 101 /128)*/ > + .quad 0x3ffbcc1e904bc1d2 /*2^( 102 /128)*/ > + .quad 0x3ffbf2c25bd71e09 /*2^( 103 /128)*/ > + .quad 0x3ffc199bdd85529c /*2^( 104 /128)*/ > + .quad 0x3ffc40ab5fffd07a /*2^( 105 /128)*/ > + .quad 0x3ffc67f12e57d14b /*2^( 106 /128)*/ > + .quad 0x3ffc8f6d9406e7b5 /*2^( 107 /128)*/ > + .quad 0x3ffcb720dcef9069 /*2^( 108 /128)*/ > + .quad 0x3ffcdf0b555dc3fa /*2^( 109 /128)*/ > + .quad 0x3ffd072d4a07897c /*2^( 110 /128)*/ > + .quad 0x3ffd2f87080d89f2 /*2^( 111 /128)*/ > + .quad 0x3ffd5818dcfba487 /*2^( 112 /128)*/ > + .quad 0x3ffd80e316c98398 /*2^( 113 /128)*/ > + .quad 0x3ffda9e603db3285 /*2^( 114 /128)*/ > + .quad 0x3ffdd321f301b460 /*2^( 115 /128)*/ > + .quad 0x3ffdfc97337b9b5f /*2^( 116 /128)*/ > + .quad 0x3ffe264614f5a129 /*2^( 117 /128)*/ > + .quad 0x3ffe502ee78b3ff6 /*2^( 118 /128)*/ > + .quad 0x3ffe7a51fbc74c83 /*2^( 119 /128)*/ > + .quad 0x3ffea4afa2a490da /*2^( 120 /128)*/ > + .quad 0x3ffecf482d8e67f1 /*2^( 121 /128)*/ > + .quad 0x3ffefa1bee615a27 /*2^( 122 /128)*/ > + .quad 0x3fff252b376bba97 /*2^( 123 /128)*/ > + .quad 0x3fff50765b6e4540 /*2^( 124 /128)*/ > + .quad 0x3fff7bfdad9cbe14 /*2^( 125 /128)*/ > + .quad 0x3fffa7c1819e90d8 /*2^( 126 /128)*/ > + .quad 0x3fffd3c22b8f71f1 /*2^( 127 /128)*/ > + .align 32 > + .quad 0x407a934f0979a371, 0x407a934f0979a371, 0x407a934f0979a371, 0x407a934f0979a371 /* _dbLg2_10*2^K */ > + .align 32 > + .quad 0x4338800000000000, 0x4338800000000000, 0x4338800000000000, 0x4338800000000000 /* _dbShifter */ > + .align 32 > + .quad 0x3f63441350a00000, 0x3f63441350a00000, 0x3f63441350a00000, 0x3f63441350a00000 /* _dbInvLg2_10hi/2^K 53-11-K bits*/ > + .align 32 > + .quad 0xbd10c0219dc1da99, 0xbd10c0219dc1da99, 0xbd10c0219dc1da99, 0xbd10c0219dc1da99 /* _dbInvLg2_10lo/2^K */ > + //PC0 = 1.0 > + .align 32 > + .quad 0x40026bb1bbb55516, 0x40026bb1bbb55516, 0x40026bb1bbb55516, 0x40026bb1bbb55516 /* _dPC1 */ > + .align 32 > + .quad 0x40053524c73ce8e3, 0x40053524c73ce8e3, 0x40053524c73ce8e3, 0x40053524c73ce8e3 /* _dPC2 */ > + .align 32 > + .quad 0x4000470591ccea8b, 0x4000470591ccea8b, 0x4000470591ccea8b, 0x4000470591ccea8b /* _dPC3 */ > + .align 32 > + .quad 0x3ff2bd767584db59, 0x3ff2bd767584db59, 0x3ff2bd767584db59, 0x3ff2bd767584db59 /* _dPC4 */ > + .align 32 > + .quad 0x3fe144c03efafb54, 0x3fe144c03efafb54, 0x3fe144c03efafb54, 0x3fe144c03efafb54 /* _dPC5 */ > + .align 32 > + .quad 0xfff0000000000000, 0xfff0000000000000, 0xfff0000000000000, 0xfff0000000000000 /* _lExpMask */ > + .align 32 > + .long 0x0000007f, 0x0000007f, 0x0000007f, 0x0000007f, 0x0000007f, 0x0000007f, 0x0000007f, 0x0000007f /* _iIndexMask =(2^K-1)*/ > + //common > + .align 32 > + .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff /* _iAbsMask */ > + .align 32 > + .long 0x40733a70, 0x40733a70, 0x40733a70, 0x40733a70, 0x40733a70, 0x40733a70, 0x40733a70, 0x40733a70 /* _iDomainRange */ > + .align 32 > + .type __svml_dexp10_data_internal,@object > + .size __svml_dexp10_data_internal,.-__svml_dexp10_data_internal > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp108_core-avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_exp108_core-avx2.S > new file mode 100644 > index 0000000000..3aff9446d3 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp108_core-avx2.S > @@ -0,0 +1,20 @@ > +/* AVX2 version of vectorized exp10, 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 _ZGVeN8v_exp10 _ZGVeN8v_exp10_avx2_wrapper > +#include "../svml_d_exp108_core.S" > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp108_core.c b/sysdeps/x86_64/fpu/multiarch/svml_d_exp108_core.c > new file mode 100644 > index 0000000000..d592663169 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp108_core.c > @@ -0,0 +1,27 @@ > +/* Multiple versions of vectorized exp10, 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 _ZGVeN8v_exp10 > +#include "ifunc-mathvec-avx512-skx.h" > + > +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); > + > +#ifdef SHARED > +__hidden_ver1 (_ZGVeN8v_exp10, __GI__ZGVeN8v_exp10, __redirect__ZGVeN8v_exp10) > + __attribute__ ((visibility ("hidden"))); > +#endif > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp108_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_d_exp108_core_avx512.S > new file mode 100644 > index 0000000000..953cb5bc1a > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp108_core_avx512.S > @@ -0,0 +1,287 @@ > +/* Function exp10 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: > + * Typical exp10() implementation, except that: > + * - tables are small (16 elements), allowing for fast gathers > + * - all arguments processed in the main path > + * - final VSCALEF assists branch-free design (correct overflow/underflow and special case responses) > + * - a VAND is used to ensure the reduced argument |R|<2, even for large inputs > + * - RZ mode used to avoid oveflow to +/-Inf for x*log2(e); helps with special case handling > + * - SAE used to avoid spurious flag settings > + * > + */ > + > +/* Offsets for data table __svml_dexp10_data_internal_avx512 > + */ > +#define Exp_tbl_H 0 > +#define L2E 128 > +#define Shifter 192 > +#define L2H 256 > +#define L2L 320 > +#define EMask 384 > +#define poly_coeff6 448 > +#define poly_coeff5 512 > +#define poly_coeff4 576 > +#define poly_coeff3 640 > +#define poly_coeff2 704 > +#define poly_coeff1 768 > +#define AbsMask 832 > +#define Threshold 896 > + > +#include > + > + .text > + .section .text.evex512,"ax",@progbits > +ENTRY(_ZGVeN8v_exp10_skx) > + pushq %rbp > + cfi_def_cfa_offset(16) > + movq %rsp, %rbp > + cfi_def_cfa(6, 16) > + cfi_offset(6, -16) > + andq $-64, %rsp > + subq $192, %rsp > + vmovups L2E+__svml_dexp10_data_internal_avx512(%rip), %zmm4 > + vmovups Shifter+__svml_dexp10_data_internal_avx512(%rip), %zmm2 > + vmovups L2H+__svml_dexp10_data_internal_avx512(%rip), %zmm5 > + vmovups L2L+__svml_dexp10_data_internal_avx512(%rip), %zmm3 > + > +/* polynomial */ > + vmovups poly_coeff6+__svml_dexp10_data_internal_avx512(%rip), %zmm6 > + vmovups poly_coeff4+__svml_dexp10_data_internal_avx512(%rip), %zmm7 > + vmovups poly_coeff3+__svml_dexp10_data_internal_avx512(%rip), %zmm9 > + vmovups poly_coeff2+__svml_dexp10_data_internal_avx512(%rip), %zmm8 > + vmovups poly_coeff1+__svml_dexp10_data_internal_avx512(%rip), %zmm11 > + vmovups Threshold+__svml_dexp10_data_internal_avx512(%rip), %zmm14 > + vmovaps %zmm0, %zmm1 > + > +/* 2^(52-4)*1.5 + x * log2(e) */ > + vfmadd213pd {rz-sae}, %zmm2, %zmm1, %zmm4 > + vandpd AbsMask+__svml_dexp10_data_internal_avx512(%rip), %zmm1, %zmm13 > + > +/* Z0 ~ x*log2(e), rounded down to 4 fractional bits */ > + vsubpd {rn-sae}, %zmm2, %zmm4, %zmm0 > + > +/* Table lookup: Th */ > + vmovups __svml_dexp10_data_internal_avx512(%rip), %zmm2 > + vcmppd $29, {sae}, %zmm14, %zmm13, %k0 > + > +/* R = x - Z0*log(2) */ > + vfnmadd213pd {rn-sae}, %zmm1, %zmm0, %zmm5 > + vpermt2pd Exp_tbl_H+64+__svml_dexp10_data_internal_avx512(%rip), %zmm4, %zmm2 > + kmovw %k0, %edx > + vfnmadd231pd {rn-sae}, %zmm0, %zmm3, %zmm5 > + vmovups poly_coeff5+__svml_dexp10_data_internal_avx512(%rip), %zmm3 > + > +/* ensure |R|<2 even for special cases */ > + vandpd EMask+__svml_dexp10_data_internal_avx512(%rip), %zmm5, %zmm12 > + vmulpd {rn-sae}, %zmm12, %zmm12, %zmm10 > + vmulpd {rn-sae}, %zmm12, %zmm2, %zmm15 > + vfmadd231pd {rn-sae}, %zmm12, %zmm6, %zmm3 > + vfmadd231pd {rn-sae}, %zmm12, %zmm7, %zmm9 > + vfmadd231pd {rn-sae}, %zmm12, %zmm8, %zmm11 > + vfmadd213pd {rn-sae}, %zmm9, %zmm10, %zmm3 > + vfmadd213pd {rn-sae}, %zmm11, %zmm10, %zmm3 > + vfmadd213pd {rn-sae}, %zmm2, %zmm15, %zmm3 > + vscalefpd {rn-sae}, %zmm0, %zmm3, %zmm0 > + testl %edx, %edx > + > +/* Go to special inputs processing branch */ > + jne L(SPECIAL_VALUES_BRANCH) > + # LOE rbx r12 r13 r14 r15 edx zmm0 zmm1 > + > +/* 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): > + vmovups %zmm1, 64(%rsp) > + vmovups %zmm0, 128(%rsp) > + # LOE rbx r12 r13 r14 r15 edx zmm0 > + > + 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: -176; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x50, 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: -184; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x48, 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: -192; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x40, 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 128(%rsp), %zmm0 > + > +/* 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: -176; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x50, 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: -184; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x48, 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: -192; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x40, 0xff, 0xff, 0xff, 0x22 > + # LOE rbx r12 r13 r14 r15 zmm0 > + > +/* Scalar math fucntion call > + * to process special input > + */ > + > +L(SCALAR_MATH_CALL): > + movl %r12d, %r14d > + movsd 64(%rsp,%r14,8), %xmm0 > + call exp10@PLT > + # LOE rbx r14 r15 r12d r13d xmm0 > + > + movsd %xmm0, 128(%rsp,%r14,8) > + > +/* Process special inputs in loop */ > + jmp L(SPECIAL_VALUES_LOOP) > + # LOE rbx r15 r12d r13d > +END(_ZGVeN8v_exp10_skx) > + > + .section .rodata, "a" > + .align 64 > + > +#ifdef __svml_dexp10_data_internal_avx512_typedef > +typedef unsigned int VUINT32; > +typedef struct { > + __declspec(align(64)) VUINT32 Exp_tbl_H[16][2]; > + __declspec(align(64)) VUINT32 L2E[8][2]; > + __declspec(align(64)) VUINT32 Shifter[8][2]; > + __declspec(align(64)) VUINT32 L2H[8][2]; > + __declspec(align(64)) VUINT32 L2L[8][2]; > + __declspec(align(64)) VUINT32 EMask[8][2]; > + __declspec(align(64)) VUINT32 poly_coeff6[8][2]; > + __declspec(align(64)) VUINT32 poly_coeff5[8][2]; > + __declspec(align(64)) VUINT32 poly_coeff4[8][2]; > + __declspec(align(64)) VUINT32 poly_coeff3[8][2]; > + __declspec(align(64)) VUINT32 poly_coeff2[8][2]; > + __declspec(align(64)) VUINT32 poly_coeff1[8][2]; > + __declspec(align(64)) VUINT32 AbsMask[8][2]; > + __declspec(align(64)) VUINT32 Threshold[8][2]; > + } __svml_dexp10_data_internal_avx512; > +#endif > +__svml_dexp10_data_internal_avx512: > + /*== Exp_tbl_H ==*/ > + .quad 0x3ff0000000000000 > + .quad 0x3ff0b5586cf9890f > + .quad 0x3ff172b83c7d517b > + .quad 0x3ff2387a6e756238 > + .quad 0x3ff306fe0a31b715 > + .quad 0x3ff3dea64c123422 > + .quad 0x3ff4bfdad5362a27 > + .quad 0x3ff5ab07dd485429 > + .quad 0x3ff6a09e667f3bcd > + .quad 0x3ff7a11473eb0187 > + .quad 0x3ff8ace5422aa0db > + .quad 0x3ff9c49182a3f090 > + .quad 0x3ffae89f995ad3ad > + .quad 0x3ffc199bdd85529c > + .quad 0x3ffd5818dcfba487 > + .quad 0x3ffea4afa2a490da > + /*== log2(e) ==*/ > + .align 64 > + .quad 0x400A934F0979A371, 0x400A934F0979A371, 0x400A934F0979A371, 0x400A934F0979A371, 0x400A934F0979A371, 0x400A934F0979A371, 0x400A934F0979A371, 0x400A934F0979A371 > + /*== Shifter=2^(52-4)*1.5 ==*/ > + .align 64 > + .quad 0x42f8000000003ff0, 0x42f8000000003ff0, 0x42f8000000003ff0, 0x42f8000000003ff0, 0x42f8000000003ff0, 0x42f8000000003ff0, 0x42f8000000003ff0, 0x42f8000000003ff0 > + /*== L2H = log(2)_high ==*/ > + .align 64 > + .quad 0x3fd34413509f79ff, 0x3fd34413509f79ff, 0x3fd34413509f79ff, 0x3fd34413509f79ff, 0x3fd34413509f79ff, 0x3fd34413509f79ff, 0x3fd34413509f79ff, 0x3fd34413509f79ff > + /*== L2L = log(2)_low ==*/ > + .align 64 > + .quad 0xbc49dc1da994fd21, 0xbc49dc1da994fd21, 0xbc49dc1da994fd21, 0xbc49dc1da994fd21, 0xbc49dc1da994fd21, 0xbc49dc1da994fd21, 0xbc49dc1da994fd21, 0xbc49dc1da994fd21 > + /*== EMask ==*/ > + .align 64 > + .quad 0xbfffffffffffffff, 0xbfffffffffffffff, 0xbfffffffffffffff, 0xbfffffffffffffff, 0xbfffffffffffffff, 0xbfffffffffffffff, 0xbfffffffffffffff, 0xbfffffffffffffff > + /*== poly_coeff6 ==*/ > + .align 64 > + .quad 0x3fcb137ed8ac2020, 0x3fcb137ed8ac2020, 0x3fcb137ed8ac2020, 0x3fcb137ed8ac2020, 0x3fcb137ed8ac2020, 0x3fcb137ed8ac2020, 0x3fcb137ed8ac2020, 0x3fcb137ed8ac2020 > + /*== poly_coeff5 ==*/ > + .align 64 > + .quad 0x3fe141a8e24f9424, 0x3fe141a8e24f9424, 0x3fe141a8e24f9424, 0x3fe141a8e24f9424, 0x3fe141a8e24f9424, 0x3fe141a8e24f9424, 0x3fe141a8e24f9424, 0x3fe141a8e24f9424 > + /*== poly_coeff4 ==*/ > + .align 64 > + .quad 0x3ff2bd77a0926c9d, 0x3ff2bd77a0926c9d, 0x3ff2bd77a0926c9d, 0x3ff2bd77a0926c9d, 0x3ff2bd77a0926c9d, 0x3ff2bd77a0926c9d, 0x3ff2bd77a0926c9d, 0x3ff2bd77a0926c9d > + /*== poly_coeff3 ==*/ > + .align 64 > + .quad 0x40004705908704c8, 0x40004705908704c8, 0x40004705908704c8, 0x40004705908704c8, 0x40004705908704c8, 0x40004705908704c8, 0x40004705908704c8, 0x40004705908704c8 > + /*== poly_coeff2 ==*/ > + .align 64 > + .quad 0x40053524c73dfe25, 0x40053524c73dfe25, 0x40053524c73dfe25, 0x40053524c73dfe25, 0x40053524c73dfe25, 0x40053524c73dfe25, 0x40053524c73dfe25, 0x40053524c73dfe25 > + /*== poly_coeff1 ==*/ > + .align 64 > + .quad 0x40026bb1bbb554c2, 0x40026bb1bbb554c2, 0x40026bb1bbb554c2, 0x40026bb1bbb554c2, 0x40026bb1bbb554c2, 0x40026bb1bbb554c2, 0x40026bb1bbb554c2, 0x40026bb1bbb554c2 > + /*== AbsMask ==*/ > + .align 64 > + .quad 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff > + /*== Threshold ==*/ > + .align 64 > + .quad 0x40733A7146F72A41, 0x40733A7146F72A41, 0x40733A7146F72A41, 0x40733A7146F72A41, 0x40733A7146F72A41, 0x40733A7146F72A41, 0x40733A7146F72A41, 0x40733A7146F72A41 > + .align 64 > + .type __svml_dexp10_data_internal_avx512,@object > + .size __svml_dexp10_data_internal_avx512,.-__svml_dexp10_data_internal_avx512 > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f16_core-avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f16_core-avx2.S > new file mode 100644 > index 0000000000..dda41c9c8f > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f16_core-avx2.S > @@ -0,0 +1,20 @@ > +/* AVX2 version of vectorized exp10f. > + 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 _ZGVeN16v_exp10f _ZGVeN16v_exp10f_avx2_wrapper > +#include "../svml_s_exp10f16_core.S" > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f16_core.c b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f16_core.c > new file mode 100644 > index 0000000000..8176a5912b > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f16_core.c > @@ -0,0 +1,28 @@ > +/* Multiple versions of vectorized exp10f, 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 _ZGVeN16v_exp10f > +#include "ifunc-mathvec-avx512-skx.h" > + > +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); > + > +#ifdef SHARED > +__hidden_ver1 (_ZGVeN16v_exp10f, __GI__ZGVeN16v_exp10f, > + __redirect__ZGVeN16v_exp10f) > + __attribute__ ((visibility ("hidden"))); > +#endif > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f16_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f16_core_avx512.S > new file mode 100644 > index 0000000000..fc9309c90f > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f16_core_avx512.S > @@ -0,0 +1,269 @@ > +/* Function exp10f 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: > + * Typical exp10() implementation, except that: > + * - tables are small (16 elements), allowing for fast gathers > + * - all arguments processed in the main path > + * - final VSCALEF assists branch-free design (correct overflow/underflow and special case responses) > + * - a VAND is used to ensure the reduced argument |R|<2, even for large inputs > + * - RZ mode used to avoid oveflow to +/-Inf for x*log2(e); helps with special case handling > + * - SAE used to avoid spurious flag settings > + * > + */ > + > +/* Offsets for data table __svml_sexp10_data_internal_avx512 > + */ > +#define Exp_tbl_L 0 > +#define Exp_tbl_H 128 > +#define L2E 256 > +#define Shifter 320 > +#define L2H 384 > +#define L2L 448 > +#define EMask 512 > +#define AbsMask 576 > +#define Threshold 640 > +#define poly_coeff2 704 > +#define poly_coeff1 768 > + > +#include > + > + .text > + .section .text.exex512,"ax",@progbits > +ENTRY(_ZGVeN16v_exp10f_skx) > + pushq %rbp > + cfi_def_cfa_offset(16) > + movq %rsp, %rbp > + cfi_def_cfa(6, 16) > + cfi_offset(6, -16) > + andq $-64, %rsp > + subq $192, %rsp > + vmovups L2E+__svml_sexp10_data_internal_avx512(%rip), %zmm2 > + vmovups Shifter+__svml_sexp10_data_internal_avx512(%rip), %zmm1 > + vmovups L2H+__svml_sexp10_data_internal_avx512(%rip), %zmm5 > + vmovups L2L+__svml_sexp10_data_internal_avx512(%rip), %zmm4 > + > +/* ensure |R|<2 even for special cases */ > + vmovups EMask+__svml_sexp10_data_internal_avx512(%rip), %zmm6 > + vmovups poly_coeff2+__svml_sexp10_data_internal_avx512(%rip), %zmm9 > + > +/* 2^(52-4)*1.5 + x * log2(e) */ > + vfmadd213ps {rz-sae}, %zmm1, %zmm0, %zmm2 > + vmovups poly_coeff1+__svml_sexp10_data_internal_avx512(%rip), %zmm10 > + vmovups __svml_sexp10_data_internal_avx512(%rip), %zmm8 > + vmovups Exp_tbl_H+__svml_sexp10_data_internal_avx512(%rip), %zmm15 > + vmovups Threshold+__svml_sexp10_data_internal_avx512(%rip), %zmm13 > + vpsrld $5, %zmm2, %zmm3 > + > +/* Z0 ~ x*log2(e), rounded down to 6 fractional bits */ > + vsubps {rn-sae}, %zmm1, %zmm2, %zmm1 > + vpermt2ps Exp_tbl_L+64+__svml_sexp10_data_internal_avx512(%rip), %zmm2, %zmm8 > + vpermt2ps Exp_tbl_H+64+__svml_sexp10_data_internal_avx512(%rip), %zmm3, %zmm15 > + vandps AbsMask+__svml_sexp10_data_internal_avx512(%rip), %zmm0, %zmm12 > + > +/* R = x - Z0*log(2) */ > + vfnmadd213ps {rn-sae}, %zmm0, %zmm1, %zmm5 > + vcmpps $29, {sae}, %zmm13, %zmm12, %k0 > + vfnmadd231ps {rn-sae}, %zmm1, %zmm4, %zmm5 > + kmovw %k0, %edx > + vrangeps $2, {sae}, %zmm6, %zmm5, %zmm11 > + vfmadd231ps {rn-sae}, %zmm11, %zmm9, %zmm10 > + vmulps {rn-sae}, %zmm11, %zmm10, %zmm14 > + > +/* x!=0? */ > + vpxord %zmm7, %zmm7, %zmm7 > + vcmpps $4, {sae}, %zmm7, %zmm0, %k1 > + > +/* Th*Tl */ > + vmulps {rn-sae}, %zmm8, %zmm15, %zmm15{%k1} > + vfmadd213ps {rn-sae}, %zmm15, %zmm14, %zmm15 > + vscalefps {rn-sae}, %zmm1, %zmm15, %zmm1 > + testl %edx, %edx > + > +/* Go to special inputs processing branch */ > + jne L(SPECIAL_VALUES_BRANCH) > + # LOE rbx r12 r13 r14 r15 edx zmm0 zmm1 > + > +/* Restore registers > + * and exit the function > + */ > + > +L(EXIT): > + vmovaps %zmm1, %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) > + # LOE rbx r12 r13 r14 r15 edx zmm1 > + > + 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: -176; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x50, 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: -184; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x48, 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: -192; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x40, 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 128(%rsp), %zmm1 > + > +/* 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: -176; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x50, 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: -184; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x48, 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: -192; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x40, 0xff, 0xff, 0xff, 0x22 > + # LOE rbx r12 r13 r14 r15 zmm1 > + > +/* Scalar math fucntion call > + * to process special input > + */ > + > +L(SCALAR_MATH_CALL): > + movl %r12d, %r14d > + movss 64(%rsp,%r14,4), %xmm0 > + call exp10f@PLT > + # LOE rbx r14 r15 r12d r13d xmm0 > + > + movss %xmm0, 128(%rsp,%r14,4) > + > +/* Process special inputs in loop */ > + jmp L(SPECIAL_VALUES_LOOP) > + # LOE rbx r15 r12d r13d > +END(_ZGVeN16v_exp10f_skx) > + > + .section .rodata, "a" > + .align 64 > + > +#ifdef __svml_sexp10_data_internal_avx512_typedef > +typedef unsigned int VUINT32; > +typedef struct { > + __declspec(align(64)) VUINT32 Exp_tbl_L[32][1]; > + __declspec(align(64)) VUINT32 Exp_tbl_H[32][1]; > + __declspec(align(64)) VUINT32 L2E[16][1]; > + __declspec(align(64)) VUINT32 Shifter[16][1]; > + __declspec(align(64)) VUINT32 L2H[16][1]; > + __declspec(align(64)) VUINT32 L2L[16][1]; > + __declspec(align(64)) VUINT32 EMask[16][1]; > + __declspec(align(64)) VUINT32 AbsMask[16][1]; > + __declspec(align(64)) VUINT32 Threshold[16][1]; > + __declspec(align(64)) VUINT32 poly_coeff2[16][1]; > + __declspec(align(64)) VUINT32 poly_coeff1[16][1]; > + } __svml_sexp10_data_internal_avx512; > +#endif > +__svml_sexp10_data_internal_avx512: > + /*== Exp_tbl_L ==*/ > + .long 0x3f800001, 0x3f801631, 0x3f802c65, 0x3f80429d > + .long 0x3f8058d9, 0x3f806f18, 0x3f80855c, 0x3f809ba3 > + .long 0x3f80b1ee, 0x3f80c83d, 0x3f80de90, 0x3f80f4e7 > + .long 0x3f810b42, 0x3f8121a0, 0x3f813803, 0x3f814e69 > + .long 0x3f8164d3, 0x3f817b41, 0x3f8191b3, 0x3f81a829 > + .long 0x3f81bea2, 0x3f81d520, 0x3f81eba2, 0x3f820227 > + .long 0x3f8218b0, 0x3f822f3d, 0x3f8245cf, 0x3f825c64 > + .long 0x3f8272fd, 0x3f828999, 0x3f82a03a, 0x3f82b6df > + /*== Exp_tbl_H ==*/ > + .align 64 > + .long 0x3f800000, 0x3f82cd87, 0x3f85aac3, 0x3f88980f > + .long 0x3f8b95c2, 0x3f8ea43a, 0x3f91c3d3, 0x3f94f4f0 > + .long 0x3f9837f0, 0x3f9b8d3a, 0x3f9ef532, 0x3fa27043 > + .long 0x3fa5fed7, 0x3fa9a15b, 0x3fad583f, 0x3fb123f6 > + .long 0x3fb504f3, 0x3fb8fbaf, 0x3fbd08a4, 0x3fc12c4d > + .long 0x3fc5672a, 0x3fc9b9be, 0x3fce248c, 0x3fd2a81e > + .long 0x3fd744fd, 0x3fdbfbb8, 0x3fe0ccdf, 0x3fe5b907 > + .long 0x3feac0c7, 0x3fefe4ba, 0x3ff5257d, 0x3ffa83b3 > + /*== log2(10) ==*/ > + .align 64 > + .long 0x40549A78, 0x40549A78, 0x40549A78, 0x40549A78, 0x40549A78, 0x40549A78, 0x40549A78, 0x40549A78, 0x40549A78, 0x40549A78, 0x40549A78, 0x40549A78, 0x40549A78, 0x40549A78, 0x40549A78, 0x40549A78 > + /*== Shifter=2^(23-10)*1.5 ==*/ > + .align 64 > + .long 0x46400000, 0x46400000, 0x46400000, 0x46400000, 0x46400000, 0x46400000, 0x46400000, 0x46400000, 0x46400000, 0x46400000, 0x46400000, 0x46400000, 0x46400000, 0x46400000, 0x46400000, 0x46400000 > + /*== L2H = log(2)_high ==*/ > + .align 64 > + .long 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b > + /*== L2L = log(2)_low ==*/ > + .align 64 > + .long 0xb2760860, 0xb2760860, 0xb2760860, 0xb2760860, 0xb2760860, 0xb2760860, 0xb2760860, 0xb2760860, 0xb2760860, 0xb2760860, 0xb2760860, 0xb2760860, 0xb2760860, 0xb2760860, 0xb2760860, 0xb2760860 > + /*== EMask ==*/ > + .align 64 > + .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000 > + /*== AbsMask ==*/ > + .align 64 > + .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff > + /*== Threshold ==*/ > + .align 64 > + .long 0x4217B818, 0x4217B818, 0x4217B818, 0x4217B818, 0x4217B818, 0x4217B818, 0x4217B818, 0x4217B818, 0x4217B818, 0x4217B818, 0x4217B818, 0x4217B818, 0x4217B818, 0x4217B818, 0x4217B818, 0x4217B818 > + /*== poly_coeff2 ==*/ > + .align 64 > + .long 0x4029B7DA, 0x4029B7DA, 0x4029B7DA, 0x4029B7DA, 0x4029B7DA, 0x4029B7DA, 0x4029B7DA, 0x4029B7DA, 0x4029B7DA, 0x4029B7DA, 0x4029B7DA, 0x4029B7DA, 0x4029B7DA, 0x4029B7DA, 0x4029B7DA, 0x4029B7DA > + /*== poly_coeff1 ==*/ > + .align 64 > + .long 0x40135D8D, 0x40135D8D, 0x40135D8D, 0x40135D8D, 0x40135D8D, 0x40135D8D, 0x40135D8D, 0x40135D8D, 0x40135D8D, 0x40135D8D, 0x40135D8D, 0x40135D8D, 0x40135D8D, 0x40135D8D, 0x40135D8D, 0x40135D8D > + .align 64 > + .type __svml_sexp10_data_internal_avx512,@object > + .size __svml_sexp10_data_internal_avx512,.-__svml_sexp10_data_internal_avx512 > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core-sse2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core-sse2.S > new file mode 100644 > index 0000000000..460d01357d > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core-sse2.S > @@ -0,0 +1,20 @@ > +/* SSE2 version of vectorized exp10f, 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 _ZGVbN4v_exp10f _ZGVbN4v_exp10f_sse2 > +#include "../svml_s_exp10f4_core.S" > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core.c b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core.c > new file mode 100644 > index 0000000000..7ce90a9bae > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core.c > @@ -0,0 +1,28 @@ > +/* Multiple versions of vectorized exp10f, 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 _ZGVbN4v_exp10f > +#include "ifunc-mathvec-sse4_1.h" > + > +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); > + > +#ifdef SHARED > +__hidden_ver1 (_ZGVbN4v_exp10f, __GI__ZGVbN4v_exp10f, > + __redirect__ZGVbN4v_exp10f) > + __attribute__ ((visibility ("hidden"))); > +#endif > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core_sse4.S > new file mode 100644 > index 0000000000..879592b789 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core_sse4.S > @@ -0,0 +1,311 @@ > +/* Function exp10f 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: > + * > + * exp10(x) = 2^x/log10(2) = 2^n * (1 + T[j]) * (1 + P(y)) > + * where > + * x = m*log10(2)/K + y, y in [-log10(2)/K..log10(2)/K] > + * m = n*K + j, m,n,j - signed integer, j in [-K/2..K/2] > + * > + * values of 2^j/K are tabulated > + * > + * P(y) is a minimax polynomial approximation of exp10(x)-1 > + * on small interval [-log10(2)/K..log10(2)/K] > + * > + * Special cases: > + * > + * exp10(NaN) = NaN > + * exp10(+INF) = +INF > + * exp10(-INF) = 0 > + * exp10(x) = 1 for subnormals > + * For IEEE float > + * if x > 38.5318412780761720 then exp10f(x) overflow > + * if x < -45.4555282592773440 then exp10f(x) underflow > + * > + */ > + > +/* Offsets for data table __svml_sexp10_data_internal > + */ > +#define _sT 0 > +#define _sLg2_10 128 > +#define _sShifter 144 > +#define _sInvLg2_10hi 160 > +#define _sInvLg2_10lo 176 > +#define _sPC0 192 > +#define _sPC1 208 > +#define _sPC2 224 > +#define _iIndexMask 240 > +#define _iAbsMask 256 > +#define _iDomainRange 272 > + > +#include > + > + .text > + .section .text.sse4,"ax",@progbits > +ENTRY(_ZGVbN4v_exp10f_sse4) > + subq $72, %rsp > + cfi_def_cfa_offset(80) > + movaps %xmm0, %xmm4 > + > +/* Load arument */ > + movups _sLg2_10+__svml_sexp10_data_internal(%rip), %xmm2 > + lea __svml_sexp10_data_internal(%rip), %r8 > + mulps %xmm4, %xmm2 > + movups _sShifter+__svml_sexp10_data_internal(%rip), %xmm5 > + > +/* R */ > + movups _sInvLg2_10hi+__svml_sexp10_data_internal(%rip), %xmm14 > + addps %xmm5, %xmm2 > + movaps %xmm2, %xmm1 > + movups _sInvLg2_10lo+__svml_sexp10_data_internal(%rip), %xmm15 > + subps %xmm5, %xmm1 > + mulps %xmm1, %xmm14 > + movaps %xmm4, %xmm5 > + mulps %xmm1, %xmm15 > + subps %xmm14, %xmm5 > + > +/* > + * Polynomial > + * exp10 = 2^N*(Tj+Tj*poly) > + * poly(sN) = {1+later} a0+a1*sR > + */ > + movups _sPC2+__svml_sexp10_data_internal(%rip), %xmm1 > + subps %xmm15, %xmm5 > + mulps %xmm5, %xmm1 > + movdqu _iIndexMask+__svml_sexp10_data_internal(%rip), %xmm3 > + > +/* Index and lookup */ > + movdqa %xmm3, %xmm10 > + > +/* remove index bits */ > + pandn %xmm2, %xmm3 > + pand %xmm2, %xmm10 > + > +/* 2^N */ > + pslld $18, %xmm3 > + > +/* iIndex *= sizeof(S); */ > + pslld $2, %xmm10 > + addps _sPC1+__svml_sexp10_data_internal(%rip), %xmm1 > + movd %xmm10, %edx > + pshufd $1, %xmm10, %xmm7 > + pshufd $2, %xmm10, %xmm9 > + pshufd $3, %xmm10, %xmm11 > + movd %xmm7, %ecx > + movd %xmm9, %esi > + movd %xmm11, %edi > + > +/* Check for overflow\underflow */ > + movdqu _iAbsMask+__svml_sexp10_data_internal(%rip), %xmm6 > + pand %xmm4, %xmm6 > + mulps %xmm1, %xmm5 > + movslq %edx, %rdx > + addps _sPC0+__svml_sexp10_data_internal(%rip), %xmm5 > + movslq %ecx, %rcx > + movslq %esi, %rsi > + movslq %edi, %rdi > + movd (%r8,%rdx), %xmm0 > + movd (%r8,%rcx), %xmm8 > + movd (%r8,%rsi), %xmm13 > + movd (%r8,%rdi), %xmm12 > + punpckldq %xmm8, %xmm0 > + punpckldq %xmm12, %xmm13 > + punpcklqdq %xmm13, %xmm0 > + > +/* Tj_l+Tj_h*poly */ > + mulps %xmm0, %xmm5 > + pcmpgtd _iDomainRange+__svml_sexp10_data_internal(%rip), %xmm6 > + addps %xmm5, %xmm0 > + movmskps %xmm6, %eax > + > +/* quick mul 2^N */ > + paddd %xmm3, %xmm0 > + > +/* Finish */ > + testl %eax, %eax > + > +/* Go to special inputs processing branch */ > + jne L(SPECIAL_VALUES_BRANCH) > + # LOE rbx rbp r12 r13 r14 r15 eax xmm0 xmm4 > + > +/* Restore registers > + * and exit the function > + */ > + > +L(EXIT): > + addq $72, %rsp > + cfi_def_cfa_offset(8) > + ret > + cfi_def_cfa_offset(80) > + > +/* Branch to process > + * special inputs > + */ > + > +L(SPECIAL_VALUES_BRANCH): > + movups %xmm4, 32(%rsp) > + movups %xmm0, 48(%rsp) > + # LOE rbx rbp r12 r13 r14 r15 eax > + > + xorl %edx, %edx > + movq %r12, 16(%rsp) > + cfi_offset(12, -64) > + movl %edx, %r12d > + movq %r13, 8(%rsp) > + cfi_offset(13, -72) > + movl %eax, %r13d > + movq %r14, (%rsp) > + cfi_offset(14, -80) > + # 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 48(%rsp), %xmm0 > + > +/* Go to exit */ > + jmp L(EXIT) > + cfi_offset(12, -64) > + cfi_offset(13, -72) > + cfi_offset(14, -80) > + # LOE rbx rbp r12 r13 r14 r15 xmm0 > + > +/* Scalar math fucntion call > + * to process special input > + */ > + > +L(SCALAR_MATH_CALL): > + movl %r12d, %r14d > + movss 32(%rsp,%r14,4), %xmm0 > + call exp10f@PLT > + # LOE rbx rbp r14 r15 r12d r13d xmm0 > + > + movss %xmm0, 48(%rsp,%r14,4) > + > +/* Process special inputs in loop */ > + jmp L(SPECIAL_VALUES_LOOP) > + # LOE rbx rbp r15 r12d r13d > +END(_ZGVbN4v_exp10f_sse4) > + > + .section .rodata, "a" > + .align 16 > + > +#ifdef __svml_sexp10_data_internal_typedef > +typedef unsigned int VUINT32; > +typedef struct > +{ > + __declspec(align(16)) VUINT32 _sT[(1<<5)][1]; > + __declspec(align(16)) VUINT32 _sLg2_10[4][1]; > + __declspec(align(16)) VUINT32 _sShifter[4][1]; > + __declspec(align(16)) VUINT32 _sInvLg2_10hi[4][1]; > + __declspec(align(16)) VUINT32 _sInvLg2_10lo[4][1]; > + __declspec(align(16)) VUINT32 _sPC0[4][1]; > + __declspec(align(16)) VUINT32 _sPC1[4][1]; > + __declspec(align(16)) VUINT32 _sPC2[4][1]; > + __declspec(align(16)) VUINT32 _iIndexMask[4][1]; > + __declspec(align(16)) VUINT32 _iAbsMask[4][1]; > + __declspec(align(16)) VUINT32 _iDomainRange[4][1]; > +} __svml_sexp10_data_internal; > +#endif > +__svml_sexp10_data_internal: > + /*== _sT ==*/ > + .long 0x3f800000 // 2^( 0 /32 ) > + .long 0x3f82cd87 // 2^( 1 /32 ) > + .long 0x3f85aac3 // 2^( 2 /32 ) > + .long 0x3f88980f // 2^( 3 /32 ) > + .long 0x3f8b95c2 // 2^( 4 /32 ) > + .long 0x3f8ea43a // 2^( 5 /32 ) > + .long 0x3f91c3d3 // 2^( 6 /32 ) > + .long 0x3f94f4f0 // 2^( 7 /32 ) > + .long 0x3f9837f0 // 2^( 8 /32 ) > + .long 0x3f9b8d3a // 2^( 9 /32 ) > + .long 0x3f9ef532 // 2^( 10/32 ) > + .long 0x3fa27043 // 2^( 11/32 ) > + .long 0x3fa5fed7 // 2^( 12/32 ) > + .long 0x3fa9a15b // 2^( 13/32 ) > + .long 0x3fad583f // 2^( 14/32 ) > + .long 0x3fb123f6 // 2^( 15/32 ) > + .long 0x3fb504f3 // 2^( 16/32 ) > + .long 0x3fb8fbaf // 2^( 17/32 ) > + .long 0x3fbd08a4 // 2^( 18/32 ) > + .long 0x3fc12c4d // 2^( 19/32 ) > + .long 0x3fc5672a // 2^( 20/32 ) > + .long 0x3fc9b9be // 2^( 21/32 ) > + .long 0x3fce248c // 2^( 22/32 ) > + .long 0x3fd2a81e // 2^( 23/32 ) > + .long 0x3fd744fd // 2^( 24/32 ) > + .long 0x3fdbfbb8 // 2^( 25/32 ) > + .long 0x3fe0ccdf // 2^( 26/32 ) > + .long 0x3fe5b907 // 2^( 27/32 ) > + .long 0x3feac0c7 // 2^( 28/32 ) > + .long 0x3fefe4ba // 2^( 29/32 ) > + .long 0x3ff5257d // 2^( 30/32 ) > + .long 0x3ffa83b3 // 2^( 31/32 ) > + .align 16 > + .long 0x42d49a78, 0x42d49a78, 0x42d49a78, 0x42d49a78 /* _sLg2_10*2^K */ > + .align 16 > + .long 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000 /* _sShifter) */ > + .align 16 > + .long 0x3c1a2000, 0x3c1a2000, 0x3c1a2000, 0x3c1a2000 /* _sInvLg2_10hi/2^K hi (24-K-7) bits*/ > + .align 16 > + .long 0x341a84fc, 0x341a84fc, 0x341a84fc, 0x341a84fc /* _sInvLg2_10lo/2^K lo bits */ > + // otherwise exp10(0) won't produce exact 1.0 > + .align 16 > + .long 0x2fecc868, 0x2fecc868, 0x2fecc868, 0x2fecc868 /* _sPC0 */ > + .align 16 > + .long 0x40135e1b, 0x40135e1b, 0x40135e1b, 0x40135e1b /* _sPC1 */ > + .align 16 > + .long 0x4029a8d2, 0x4029a8d2, 0x4029a8d2, 0x4029a8d2 /* _sPC2 */ > + .align 16 > + .long 0x0000001f, 0x0000001f, 0x0000001f, 0x0000001f /* _iIndexMask =(2^K-1)*/ > + //common > + .align 16 > + .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff /* _iAbsMask */ > + .align 16 > + .long 0x4217b818, 0x4217b818, 0x4217b818, 0x4217b818 /* _iDomainRange=-log10(max_denormal=0x007fffff) RZ */ > + .align 16 > + .type __svml_sexp10_data_internal,@object > + .size __svml_sexp10_data_internal,.-__svml_sexp10_data_internal > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f8_core-sse.S b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f8_core-sse.S > new file mode 100644 > index 0000000000..3f3fe252da > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f8_core-sse.S > @@ -0,0 +1,20 @@ > +/* SSE version of vectorized exp10f, 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 _ZGVdN8v_exp10f _ZGVdN8v_exp10f_sse_wrapper > +#include "../svml_s_exp10f8_core.S" > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f8_core.c b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f8_core.c > new file mode 100644 > index 0000000000..1f5ed5a59d > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f8_core.c > @@ -0,0 +1,28 @@ > +/* Multiple versions of vectorized exp10f, 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 _ZGVdN8v_exp10f > +#include "ifunc-mathvec-avx2.h" > + > +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); > + > +#ifdef SHARED > +__hidden_ver1 (_ZGVdN8v_exp10f, __GI__ZGVdN8v_exp10f, > + __redirect__ZGVdN8v_exp10f) > + __attribute__ ((visibility ("hidden"))); > +#endif > diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f8_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f8_core_avx2.S > new file mode 100644 > index 0000000000..b576412cf1 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f8_core_avx2.S > @@ -0,0 +1,331 @@ > +/* Function exp10f 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: > + * > + * exp10(x) = 2^x/log10(2) = 2^n * (1 + T[j]) * (1 + P(y)) > + * where > + * x = m*log10(2)/K + y, y in [-log10(2)/K..log10(2)/K] > + * m = n*K + j, m,n,j - signed integer, j in [-K/2..K/2] > + * > + * values of 2^j/K are tabulated > + * > + * P(y) is a minimax polynomial approximation of exp10(x)-1 > + * on small interval [-log10(2)/K..log10(2)/K] > + * > + * Special cases: > + * > + * exp10(NaN) = NaN > + * exp10(+INF) = +INF > + * exp10(-INF) = 0 > + * exp10(x) = 1 for subnormals > + * For IEEE float > + * if x > 38.5318412780761720 then exp10f(x) overflow > + * if x < -45.4555282592773440 then exp10f(x) underflow > + * > + */ > + > +/* Offsets for data table __svml_sexp10_data_internal > + */ > +#define _sT 0 > +#define _sLg2_10 128 > +#define _sShifter 160 > +#define _sInvLg2_10hi 192 > +#define _sInvLg2_10lo 224 > +#define _sPC0 256 > +#define _sPC1 288 > +#define _sPC2 320 > +#define _iIndexMask 352 > +#define _iAbsMask 384 > +#define _iDomainRange 416 > + > +#include > + > + .text > + .section .text.avx2,"ax",@progbits > +ENTRY(_ZGVdN8v_exp10f_avx2) > + pushq %rbp > + cfi_def_cfa_offset(16) > + movq %rsp, %rbp > + cfi_def_cfa(6, 16) > + cfi_offset(6, -16) > + andq $-32, %rsp > + subq $96, %rsp > + lea __svml_sexp10_data_internal(%rip), %rax > + vmovups _sShifter+__svml_sexp10_data_internal(%rip), %ymm4 > + > +/* Load arument */ > + vmovups _sLg2_10+__svml_sexp10_data_internal(%rip), %ymm1 > + vmovups _iIndexMask+__svml_sexp10_data_internal(%rip), %ymm2 > + vmovaps %ymm0, %ymm3 > + vfmadd213ps %ymm4, %ymm3, %ymm1 > + > +/* Index and lookup */ > + vandps %ymm2, %ymm1, %ymm7 > + > +/* iIndex *= sizeof(S); */ > + vpslld $2, %ymm7, %ymm10 > + vsubps %ymm4, %ymm1, %ymm0 > + > +/* Check for overflow\underflow */ > + vandps _iAbsMask+__svml_sexp10_data_internal(%rip), %ymm3, %ymm5 > + vpcmpgtd _iDomainRange+__svml_sexp10_data_internal(%rip), %ymm5, %ymm6 > + vmovmskps %ymm6, %edx > + vmovd %xmm10, %ecx > + vextractf128 $1, %ymm10, %xmm6 > + vpextrd $1, %xmm10, %esi > + vpextrd $2, %xmm10, %edi > + vpextrd $3, %xmm10, %r8d > + movslq %ecx, %rcx > + movslq %esi, %rsi > + movslq %edi, %rdi > + movslq %r8d, %r8 > + vmovd (%rax,%rcx), %xmm8 > + vmovd (%rax,%rsi), %xmm9 > + vmovd (%rax,%rdi), %xmm11 > + vmovd (%rax,%r8), %xmm12 > + vpunpckldq %xmm9, %xmm8, %xmm13 > + vpunpckldq %xmm12, %xmm11, %xmm14 > + vpunpcklqdq %xmm14, %xmm13, %xmm15 > + > +/* R */ > + vmovups _sInvLg2_10hi+__svml_sexp10_data_internal(%rip), %ymm13 > + vmovd %xmm6, %r9d > + vfnmadd213ps %ymm3, %ymm0, %ymm13 > + vpextrd $1, %xmm6, %r10d > + movslq %r9d, %r9 > + movslq %r10d, %r10 > + vfnmadd132ps _sInvLg2_10lo+__svml_sexp10_data_internal(%rip), %ymm13, %ymm0 > + vmovd (%rax,%r9), %xmm4 > + vmovd (%rax,%r10), %xmm5 > + vpunpckldq %xmm5, %xmm4, %xmm9 > + > +/* > + * Polynomial > + * exp10 = 2^N*(Tj+Tj*poly) > + * poly(sN) = {1+later} a0+a1*sR > + */ > + vmovups _sPC2+__svml_sexp10_data_internal(%rip), %ymm4 > + vfmadd213ps _sPC1+__svml_sexp10_data_internal(%rip), %ymm0, %ymm4 > + vpextrd $2, %xmm6, %r11d > + vpextrd $3, %xmm6, %ecx > + movslq %r11d, %r11 > + movslq %ecx, %rcx > + vfmadd213ps _sPC0+__svml_sexp10_data_internal(%rip), %ymm0, %ymm4 > + vmovd (%rax,%r11), %xmm7 > + vmovd (%rax,%rcx), %xmm8 > + vpunpckldq %xmm8, %xmm7, %xmm11 > + > +/* remove index bits */ > + vpandn %ymm1, %ymm2, %ymm0 > + vpunpcklqdq %xmm11, %xmm9, %xmm12 > + > +/* 2^N */ > + vpslld $18, %ymm0, %ymm1 > + vinsertf128 $1, %xmm12, %ymm15, %ymm14 > + > +/* Tj_l+Tj_h*poly */ > + vfmadd213ps %ymm14, %ymm14, %ymm4 > + > +/* quick mul 2^N */ > + vpaddd %ymm1, %ymm4, %ymm0 > + > +/* Finish */ > + testl %edx, %edx > + > +/* Go to special inputs processing branch */ > + jne L(SPECIAL_VALUES_BRANCH) > + # LOE rbx r12 r13 r14 r15 edx ymm0 ymm3 > + > +/* 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): > + vmovups %ymm3, 32(%rsp) > + vmovups %ymm0, 64(%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: -80; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 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: -88; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 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: -96; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 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 64(%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: -80; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 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: -88; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 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: -96; DW_OP_plus) */ > + .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 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 > + movss 32(%rsp,%r14,4), %xmm0 > + call exp10f@PLT > + # LOE rbx r14 r15 r12d r13d xmm0 > + > + movss %xmm0, 64(%rsp,%r14,4) > + > +/* Process special inputs in loop */ > + jmp L(SPECIAL_VALUES_LOOP) > + # LOE rbx r15 r12d r13d > +END(_ZGVdN8v_exp10f_avx2) > + > + .section .rodata, "a" > + .align 32 > + > +#ifdef __svml_sexp10_data_internal_typedef > +typedef unsigned int VUINT32; > +typedef struct > +{ > + __declspec(align(32)) VUINT32 _sT[(1<<5)][1]; > + __declspec(align(32)) VUINT32 _sLg2_10[8][1]; > + __declspec(align(32)) VUINT32 _sShifter[8][1]; > + __declspec(align(32)) VUINT32 _sInvLg2_10hi[8][1]; > + __declspec(align(32)) VUINT32 _sInvLg2_10lo[8][1]; > + __declspec(align(32)) VUINT32 _sPC0[8][1]; > + __declspec(align(32)) VUINT32 _sPC1[8][1]; > + __declspec(align(32)) VUINT32 _sPC2[8][1]; > + __declspec(align(32)) VUINT32 _iIndexMask[8][1]; > + __declspec(align(32)) VUINT32 _iAbsMask[8][1]; > + __declspec(align(32)) VUINT32 _iDomainRange[8][1]; > +} __svml_sexp10_data_internal; > +#endif > +__svml_sexp10_data_internal: > + /*== _sT ==*/ > + .long 0x3f800000 // 2^( 0 /32 ) > + .long 0x3f82cd87 // 2^( 1 /32 ) > + .long 0x3f85aac3 // 2^( 2 /32 ) > + .long 0x3f88980f // 2^( 3 /32 ) > + .long 0x3f8b95c2 // 2^( 4 /32 ) > + .long 0x3f8ea43a // 2^( 5 /32 ) > + .long 0x3f91c3d3 // 2^( 6 /32 ) > + .long 0x3f94f4f0 // 2^( 7 /32 ) > + .long 0x3f9837f0 // 2^( 8 /32 ) > + .long 0x3f9b8d3a // 2^( 9 /32 ) > + .long 0x3f9ef532 // 2^( 10/32 ) > + .long 0x3fa27043 // 2^( 11/32 ) > + .long 0x3fa5fed7 // 2^( 12/32 ) > + .long 0x3fa9a15b // 2^( 13/32 ) > + .long 0x3fad583f // 2^( 14/32 ) > + .long 0x3fb123f6 // 2^( 15/32 ) > + .long 0x3fb504f3 // 2^( 16/32 ) > + .long 0x3fb8fbaf // 2^( 17/32 ) > + .long 0x3fbd08a4 // 2^( 18/32 ) > + .long 0x3fc12c4d // 2^( 19/32 ) > + .long 0x3fc5672a // 2^( 20/32 ) > + .long 0x3fc9b9be // 2^( 21/32 ) > + .long 0x3fce248c // 2^( 22/32 ) > + .long 0x3fd2a81e // 2^( 23/32 ) > + .long 0x3fd744fd // 2^( 24/32 ) > + .long 0x3fdbfbb8 // 2^( 25/32 ) > + .long 0x3fe0ccdf // 2^( 26/32 ) > + .long 0x3fe5b907 // 2^( 27/32 ) > + .long 0x3feac0c7 // 2^( 28/32 ) > + .long 0x3fefe4ba // 2^( 29/32 ) > + .long 0x3ff5257d // 2^( 30/32 ) > + .long 0x3ffa83b3 // 2^( 31/32 ) > + .align 32 > + .long 0x42d49a78, 0x42d49a78, 0x42d49a78, 0x42d49a78, 0x42d49a78, 0x42d49a78, 0x42d49a78, 0x42d49a78 /* _sLg2_10*2^K */ > + .align 32 > + .long 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000 /* _sShifter) */ > + .align 32 > + .long 0x3c1a2000, 0x3c1a2000, 0x3c1a2000, 0x3c1a2000, 0x3c1a2000, 0x3c1a2000, 0x3c1a2000, 0x3c1a2000 /* _sInvLg2_10hi/2^K hi (24-K-7) bits*/ > + .align 32 > + .long 0x341a84fc, 0x341a84fc, 0x341a84fc, 0x341a84fc, 0x341a84fc, 0x341a84fc, 0x341a84fc, 0x341a84fc /* _sInvLg2_10lo/2^K lo bits */ > + // otherwise exp10(0) won't produce exact 1.0 > + .align 32 > + .long 0x2fecc868, 0x2fecc868, 0x2fecc868, 0x2fecc868, 0x2fecc868, 0x2fecc868, 0x2fecc868, 0x2fecc868 /* _sPC0 */ > + .align 32 > + .long 0x40135e1b, 0x40135e1b, 0x40135e1b, 0x40135e1b, 0x40135e1b, 0x40135e1b, 0x40135e1b, 0x40135e1b /* _sPC1 */ > + .align 32 > + .long 0x4029a8d2, 0x4029a8d2, 0x4029a8d2, 0x4029a8d2, 0x4029a8d2, 0x4029a8d2, 0x4029a8d2, 0x4029a8d2 /* _sPC2 */ > + .align 32 > + .long 0x0000001f, 0x0000001f, 0x0000001f, 0x0000001f, 0x0000001f, 0x0000001f, 0x0000001f, 0x0000001f /* _iIndexMask =(2^K-1)*/ > + //common > + .align 32 > + .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff /* _iAbsMask */ > + .align 32 > + .long 0x4217b818, 0x4217b818, 0x4217b818, 0x4217b818, 0x4217b818, 0x4217b818, 0x4217b818, 0x4217b818 /* _iDomainRange=-log10(max_denormal=0x007fffff) RZ */ > + .align 32 > + .type __svml_sexp10_data_internal,@object > + .size __svml_sexp10_data_internal,.-__svml_sexp10_data_internal > diff --git a/sysdeps/x86_64/fpu/svml_d_exp102_core.S b/sysdeps/x86_64/fpu/svml_d_exp102_core.S > new file mode 100644 > index 0000000000..157fb3b7c0 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_d_exp102_core.S > @@ -0,0 +1,29 @@ > +/* Function exp10 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 (_ZGVbN2v_exp10) > +WRAPPER_IMPL_SSE2 exp10 > +END (_ZGVbN2v_exp10) > + > +#ifndef USE_MULTIARCH > + libmvec_hidden_def (_ZGVbN2v_exp10) > +#endif > diff --git a/sysdeps/x86_64/fpu/svml_d_exp104_core.S b/sysdeps/x86_64/fpu/svml_d_exp104_core.S > new file mode 100644 > index 0000000000..9b9d0a5d4b > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_d_exp104_core.S > @@ -0,0 +1,29 @@ > +/* Function exp10 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 (_ZGVdN4v_exp10) > +WRAPPER_IMPL_AVX _ZGVbN2v_exp10 > +END (_ZGVdN4v_exp10) > + > +#ifndef USE_MULTIARCH > + libmvec_hidden_def (_ZGVdN4v_exp10) > +#endif > diff --git a/sysdeps/x86_64/fpu/svml_d_exp104_core_avx.S b/sysdeps/x86_64/fpu/svml_d_exp104_core_avx.S > new file mode 100644 > index 0000000000..1ba1a819ed > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_d_exp104_core_avx.S > @@ -0,0 +1,25 @@ > +/* Function exp10 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 (_ZGVcN4v_exp10) > +WRAPPER_IMPL_AVX _ZGVbN2v_exp10 > +END (_ZGVcN4v_exp10) > diff --git a/sysdeps/x86_64/fpu/svml_d_exp108_core.S b/sysdeps/x86_64/fpu/svml_d_exp108_core.S > new file mode 100644 > index 0000000000..a530dc12de > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_d_exp108_core.S > @@ -0,0 +1,25 @@ > +/* Function exp10 vectorized with AVX-512, wrapper to 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 > + . */ > + > +#include > +#include "svml_d_wrapper_impl.h" > + > + .text > +ENTRY (_ZGVeN8v_exp10) > +WRAPPER_IMPL_AVX512 _ZGVdN4v_exp10 > +END (_ZGVeN8v_exp10) > diff --git a/sysdeps/x86_64/fpu/svml_s_exp10f16_core.S b/sysdeps/x86_64/fpu/svml_s_exp10f16_core.S > new file mode 100644 > index 0000000000..e5043bc875 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_s_exp10f16_core.S > @@ -0,0 +1,25 @@ > +/* Function exp10f 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 (_ZGVeN16v_exp10f) > +WRAPPER_IMPL_AVX512 _ZGVdN8v_exp10f > +END (_ZGVeN16v_exp10f) > diff --git a/sysdeps/x86_64/fpu/svml_s_exp10f4_core.S b/sysdeps/x86_64/fpu/svml_s_exp10f4_core.S > new file mode 100644 > index 0000000000..75e6637a82 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_s_exp10f4_core.S > @@ -0,0 +1,29 @@ > +/* Function exp10f vectorized with SSE2, 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 (_ZGVbN4v_exp10f) > +WRAPPER_IMPL_SSE2 exp10f > +END (_ZGVbN4v_exp10f) > + > +#ifndef USE_MULTIARCH > + libmvec_hidden_def (_ZGVbN4v_exp10f) > +#endif > diff --git a/sysdeps/x86_64/fpu/svml_s_exp10f8_core.S b/sysdeps/x86_64/fpu/svml_s_exp10f8_core.S > new file mode 100644 > index 0000000000..d481d2dee9 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_s_exp10f8_core.S > @@ -0,0 +1,29 @@ > +/* Function exp10f 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 (_ZGVdN8v_exp10f) > +WRAPPER_IMPL_AVX _ZGVbN4v_exp10f > +END (_ZGVdN8v_exp10f) > + > +#ifndef USE_MULTIARCH > + libmvec_hidden_def (_ZGVdN8v_exp10f) > +#endif > diff --git a/sysdeps/x86_64/fpu/svml_s_exp10f8_core_avx.S b/sysdeps/x86_64/fpu/svml_s_exp10f8_core_avx.S > new file mode 100644 > index 0000000000..65944bd4d2 > --- /dev/null > +++ b/sysdeps/x86_64/fpu/svml_s_exp10f8_core_avx.S > @@ -0,0 +1,25 @@ > +/* Function exp10f 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 (_ZGVcN8v_exp10f) > +WRAPPER_IMPL_AVX _ZGVbN4v_exp10f > +END (_ZGVcN8v_exp10f) > diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-exp10-avx.c b/sysdeps/x86_64/fpu/test-double-libmvec-exp10-avx.c > new file mode 100644 > index 0000000000..7cdda9895b > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-double-libmvec-exp10-avx.c > @@ -0,0 +1 @@ > +#include "test-double-libmvec-exp10.c" > diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-exp10-avx2.c b/sysdeps/x86_64/fpu/test-double-libmvec-exp10-avx2.c > new file mode 100644 > index 0000000000..7cdda9895b > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-double-libmvec-exp10-avx2.c > @@ -0,0 +1 @@ > +#include "test-double-libmvec-exp10.c" > diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-exp10-avx512f.c b/sysdeps/x86_64/fpu/test-double-libmvec-exp10-avx512f.c > new file mode 100644 > index 0000000000..7cdda9895b > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-double-libmvec-exp10-avx512f.c > @@ -0,0 +1 @@ > +#include "test-double-libmvec-exp10.c" > diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-exp10.c b/sysdeps/x86_64/fpu/test-double-libmvec-exp10.c > new file mode 100644 > index 0000000000..b1461ed85e > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-double-libmvec-exp10.c > @@ -0,0 +1,3 @@ > +#define LIBMVEC_TYPE double > +#define LIBMVEC_FUNC exp10 > +#include "test-vector-abi-arg1.h" > diff --git a/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c b/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c > index 2f7172bd7b..256e8f07c9 100644 > --- a/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c > @@ -32,6 +32,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (atan), _ZGVbN2v_atan) > VECTOR_WRAPPER (WRAPPER_NAME (asin), _ZGVbN2v_asin) > VECTOR_WRAPPER_ff (WRAPPER_NAME (hypot), _ZGVbN2vv_hypot) > VECTOR_WRAPPER (WRAPPER_NAME (exp2), _ZGVbN2v_exp2) > +VECTOR_WRAPPER (WRAPPER_NAME (exp10), _ZGVbN2v_exp10) > > #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 e2d519faac..9de1dab2c2 100644 > --- a/sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c > @@ -35,6 +35,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (atan), _ZGVdN4v_atan) > VECTOR_WRAPPER (WRAPPER_NAME (asin), _ZGVdN4v_asin) > VECTOR_WRAPPER_ff (WRAPPER_NAME (hypot), _ZGVdN4vv_hypot) > VECTOR_WRAPPER (WRAPPER_NAME (exp2), _ZGVdN4v_exp2) > +VECTOR_WRAPPER (WRAPPER_NAME (exp10), _ZGVdN4v_exp10) > > #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 1ce4d8b413..43865ab099 100644 > --- a/sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c > @@ -32,6 +32,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (atan), _ZGVcN4v_atan) > VECTOR_WRAPPER (WRAPPER_NAME (asin), _ZGVcN4v_asin) > VECTOR_WRAPPER_ff (WRAPPER_NAME (hypot), _ZGVcN4vv_hypot) > VECTOR_WRAPPER (WRAPPER_NAME (exp2), _ZGVcN4v_exp2) > +VECTOR_WRAPPER (WRAPPER_NAME (exp10), _ZGVcN4v_exp10) > > #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 6c87cec648..5dbdacf617 100644 > --- a/sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c > @@ -32,6 +32,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (atan), _ZGVeN8v_atan) > VECTOR_WRAPPER (WRAPPER_NAME (asin), _ZGVeN8v_asin) > VECTOR_WRAPPER_ff (WRAPPER_NAME (hypot), _ZGVeN8vv_hypot) > VECTOR_WRAPPER (WRAPPER_NAME (exp2), _ZGVeN8v_exp2) > +VECTOR_WRAPPER (WRAPPER_NAME (exp10), _ZGVeN8v_exp10) > > #ifndef __ILP32__ > # define VEC_INT_TYPE __m512i > diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-exp10f-avx.c b/sysdeps/x86_64/fpu/test-float-libmvec-exp10f-avx.c > new file mode 100644 > index 0000000000..be3cdaa80d > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-float-libmvec-exp10f-avx.c > @@ -0,0 +1 @@ > +#include "test-float-libmvec-exp10f.c" > diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-exp10f-avx2.c b/sysdeps/x86_64/fpu/test-float-libmvec-exp10f-avx2.c > new file mode 100644 > index 0000000000..be3cdaa80d > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-float-libmvec-exp10f-avx2.c > @@ -0,0 +1 @@ > +#include "test-float-libmvec-exp10f.c" > diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-exp10f-avx512f.c b/sysdeps/x86_64/fpu/test-float-libmvec-exp10f-avx512f.c > new file mode 100644 > index 0000000000..be3cdaa80d > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-float-libmvec-exp10f-avx512f.c > @@ -0,0 +1 @@ > +#include "test-float-libmvec-exp10f.c" > diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-exp10f.c b/sysdeps/x86_64/fpu/test-float-libmvec-exp10f.c > new file mode 100644 > index 0000000000..06f447eb8d > --- /dev/null > +++ b/sysdeps/x86_64/fpu/test-float-libmvec-exp10f.c > @@ -0,0 +1,3 @@ > +#define LIBMVEC_TYPE float > +#define LIBMVEC_FUNC exp10f > +#include "test-vector-abi-arg1.h" > diff --git a/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c b/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c > index 597d7d7598..c159c8f583 100644 > --- a/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c > @@ -32,6 +32,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (atanf), _ZGVeN16v_atanf) > VECTOR_WRAPPER (WRAPPER_NAME (asinf), _ZGVeN16v_asinf) > VECTOR_WRAPPER_ff (WRAPPER_NAME (hypotf), _ZGVeN16vv_hypotf) > VECTOR_WRAPPER (WRAPPER_NAME (exp2f), _ZGVeN16v_exp2f) > +VECTOR_WRAPPER (WRAPPER_NAME (exp10f), _ZGVeN16v_exp10f) > > #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 3500eec810..c745ef744a 100644 > --- a/sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c > @@ -32,6 +32,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (atanf), _ZGVbN4v_atanf) > VECTOR_WRAPPER (WRAPPER_NAME (asinf), _ZGVbN4v_asinf) > VECTOR_WRAPPER_ff (WRAPPER_NAME (hypotf), _ZGVbN4vv_hypotf) > VECTOR_WRAPPER (WRAPPER_NAME (exp2f), _ZGVbN4v_exp2f) > +VECTOR_WRAPPER (WRAPPER_NAME (exp10f), _ZGVbN4v_exp10f) > > #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 921b9c65d6..c9226cf4dc 100644 > --- a/sysdeps/x86_64/fpu/test-float-vlen8-avx2-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-float-vlen8-avx2-wrappers.c > @@ -35,6 +35,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (atanf), _ZGVdN8v_atanf) > VECTOR_WRAPPER (WRAPPER_NAME (asinf), _ZGVdN8v_asinf) > VECTOR_WRAPPER_ff (WRAPPER_NAME (hypotf), _ZGVdN8vv_hypotf) > VECTOR_WRAPPER (WRAPPER_NAME (exp2f), _ZGVdN8v_exp2f) > +VECTOR_WRAPPER (WRAPPER_NAME (exp10f), _ZGVdN8v_exp10f) > > /* 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 6cbcb57521..92970c5ace 100644 > --- a/sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c > +++ b/sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c > @@ -32,6 +32,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (atanf), _ZGVcN8v_atanf) > VECTOR_WRAPPER (WRAPPER_NAME (asinf), _ZGVcN8v_asinf) > VECTOR_WRAPPER_ff (WRAPPER_NAME (hypotf), _ZGVcN8vv_hypotf) > VECTOR_WRAPPER (WRAPPER_NAME (exp2f), _ZGVcN8v_exp2f) > +VECTOR_WRAPPER (WRAPPER_NAME (exp10f), _ZGVcN8v_exp10f) > > #define VEC_INT_TYPE __m128i > > -- > 2.31.1 > LGTM. Reviewed-by: H.J. Lu Thanks. H.J.