From 04f496d6025753058bdd071fd711e9f56df149a7 Mon Sep 17 00:00:00 2001 From: Andrew Senkevich Date: Tue, 9 Jun 2015 18:29:47 +0300 Subject: [PATCH] Vector cosf for x86_64. Here is implementation of vectorized cosf containing SSE, AVX, AVX2 and AVX512 versions according to Vector ABI . * sysdeps/x86_64/fpu/Makefile (libmvec-support): Added new files. * sysdeps/x86_64/fpu/Versions: New versions added. * sysdeps/x86_64/fpu/svml_s_cosf4_core.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_s_cosf4_core.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_s_cosf4_core_sse4.S: New file. * sysdeps/x86_64/fpu/svml_s_cosf8_core_avx.S: New file. * sysdeps/x86_64/fpu/svml_s_cosf8_core.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_s_cosf8_core.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_s_cosf8_core_avx2.S: New file. * sysdeps/x86_64/fpu/svml_s_cosf16_core.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core_avx512.S: New file. * sysdeps/x86_64/fpu/svml_s_wrapper_impl.h: New file. * sysdeps/x86_64/fpu/svml_s_cosf_data.S: New file. * sysdeps/x86_64/fpu/svml_s_cosf_data.h: New file. * sysdeps/x86_64/fpu/multiarch/Makefile (libmvec-sysdep_routines): Added build of SSE, AVX2 and AVX512 IFUNC versions. * sysdeps/unix/sysv/linux/x86_64/libmvec.abilist: New versions added. * sysdeps/x86/fpu/bits/math-vector.h: Added SIMD declaration for cosf. * NEWS: Mention addition of x86_64 vector cosf. --- ChangeLog | 21 + NEWS | 2 +- sysdeps/unix/sysv/linux/x86_64/libmvec.abilist | 4 + sysdeps/x86/fpu/bits/math-vector.h | 2 + sysdeps/x86_64/fpu/Makefile | 4 +- sysdeps/x86_64/fpu/Versions | 1 + sysdeps/x86_64/fpu/multiarch/Makefile | 3 +- sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core.S | 39 + .../fpu/multiarch/svml_s_cosf16_core_avx512.S | 460 ++++++++ sysdeps/x86_64/fpu/multiarch/svml_s_cosf4_core.S | 38 + .../x86_64/fpu/multiarch/svml_s_cosf4_core_sse4.S | 227 ++++ sysdeps/x86_64/fpu/multiarch/svml_s_cosf8_core.S | 38 + .../x86_64/fpu/multiarch/svml_s_cosf8_core_avx2.S | 215 ++++ sysdeps/x86_64/fpu/svml_s_cosf16_core.S | 25 + sysdeps/x86_64/fpu/svml_s_cosf4_core.S | 29 + sysdeps/x86_64/fpu/svml_s_cosf8_core.S | 29 + sysdeps/x86_64/fpu/svml_s_cosf8_core_avx.S | 25 + sysdeps/x86_64/fpu/svml_s_cosf_data.S | 1130 ++++++++++++++++++++ sysdeps/x86_64/fpu/svml_s_cosf_data.h | 58 + sysdeps/x86_64/fpu/svml_s_wrapper_impl.h | 111 ++ 20 files changed, 2458 insertions(+), 3 deletions(-) create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core_avx512.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_cosf4_core.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_cosf4_core_sse4.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_cosf8_core.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_s_cosf8_core_avx2.S create mode 100644 sysdeps/x86_64/fpu/svml_s_cosf16_core.S create mode 100644 sysdeps/x86_64/fpu/svml_s_cosf4_core.S create mode 100644 sysdeps/x86_64/fpu/svml_s_cosf8_core.S create mode 100644 sysdeps/x86_64/fpu/svml_s_cosf8_core_avx.S create mode 100644 sysdeps/x86_64/fpu/svml_s_cosf_data.S create mode 100644 sysdeps/x86_64/fpu/svml_s_cosf_data.h create mode 100644 sysdeps/x86_64/fpu/svml_s_wrapper_impl.h diff --git a/ChangeLog b/ChangeLog index 47318a7..c3e52b2 100644 --- a/ChangeLog +++ b/ChangeLog @@ -47,6 +47,27 @@ * sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c: New file. * sysdeps/x86_64/fpu/test-double-vlen8.c: New file. + * sysdeps/x86_64/fpu/Makefile (libmvec-support): Added new files. + * sysdeps/x86_64/fpu/Versions: New versions added. + * sysdeps/x86_64/fpu/svml_s_cosf4_core.S: New file. + * sysdeps/x86_64/fpu/multiarch/svml_s_cosf4_core.S: New file. + * sysdeps/x86_64/fpu/multiarch/svml_s_cosf4_core_sse4.S: New file. + * sysdeps/x86_64/fpu/svml_s_cosf8_core_avx.S: New file. + * sysdeps/x86_64/fpu/svml_s_cosf8_core.S: New file. + * sysdeps/x86_64/fpu/multiarch/svml_s_cosf8_core.S: New file. + * sysdeps/x86_64/fpu/multiarch/svml_s_cosf8_core_avx2.S: New file. + * sysdeps/x86_64/fpu/svml_s_cosf16_core.S: New file. + * sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core.S: New file. + * sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core_avx512.S: New file. + * sysdeps/x86_64/fpu/svml_s_wrapper_impl.h: New file. + * sysdeps/x86_64/fpu/svml_s_cosf_data.S: New file. + * sysdeps/x86_64/fpu/svml_s_cosf_data.h: New file. + * sysdeps/x86_64/fpu/multiarch/Makefile (libmvec-sysdep_routines): Added + build of SSE, AVX2 and AVX512 IFUNC versions. + * sysdeps/unix/sysv/linux/x86_64/libmvec.abilist: New versions added. + * sysdeps/x86/fpu/bits/math-vector.h: Added SIMD declaration for cosf. + * NEWS: Mention addition of x86_64 vector cosf. + 2015-06-09 Marko Myllynen * locale/C-ctype.c (PREDEFINED_CLASSES): Remove. diff --git a/NEWS b/NEWS index 5e223a1..53f244d 100644 --- a/NEWS +++ b/NEWS @@ -52,7 +52,7 @@ Version 2.22 condition in some applications. * Added vector math library named libmvec with the following vectorized x86_64 - implementations: cos. + implementations: cos, cosf. The library can be disabled with --disable-mathvec. Use of the functions is enabled with -fopenmp -ffast-math starting from -O1 for GCC version >= 4.9.0. The library is linked in as needed when using -lm (no need to specify -lmvec diff --git a/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist b/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist index be6eaed..acabb8a 100644 --- a/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist +++ b/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist @@ -1,6 +1,10 @@ GLIBC_2.22 GLIBC_2.22 A _ZGVbN2v_cos F + _ZGVbN4v_cosf F _ZGVcN4v_cos F + _ZGVcN8v_cosf F _ZGVdN4v_cos F + _ZGVdN8v_cosf F + _ZGVeN16v_cosf F _ZGVeN8v_cos F diff --git a/sysdeps/x86/fpu/bits/math-vector.h b/sysdeps/x86/fpu/bits/math-vector.h index 27294ce..b3ef833 100644 --- a/sysdeps/x86/fpu/bits/math-vector.h +++ b/sysdeps/x86/fpu/bits/math-vector.h @@ -30,5 +30,7 @@ # define __DECL_SIMD_x86_64 _Pragma ("omp declare simd notinbranch") # undef __DECL_SIMD_cos # define __DECL_SIMD_cos __DECL_SIMD_x86_64 +# undef __DECL_SIMD_cosf +# define __DECL_SIMD_cosf __DECL_SIMD_x86_64 # endif #endif diff --git a/sysdeps/x86_64/fpu/Makefile b/sysdeps/x86_64/fpu/Makefile index 2d5fcf8..af0d2af 100644 --- a/sysdeps/x86_64/fpu/Makefile +++ b/sysdeps/x86_64/fpu/Makefile @@ -1,7 +1,9 @@ ifeq ($(subdir),mathvec) libmvec-support += svml_d_cos2_core svml_d_cos4_core_avx \ svml_d_cos4_core svml_d_cos8_core \ - svml_d_cos_data init-arch + svml_d_cos_data svml_s_cosf4_core svml_s_cosf8_core_avx \ + svml_s_cosf8_core svml_s_cosf16_core svml_s_cosf_data \ + init-arch endif # Variables for libmvec tests. diff --git a/sysdeps/x86_64/fpu/Versions b/sysdeps/x86_64/fpu/Versions index 4290e73..f85c28b 100644 --- a/sysdeps/x86_64/fpu/Versions +++ b/sysdeps/x86_64/fpu/Versions @@ -1,5 +1,6 @@ libmvec { GLIBC_2.22 { _ZGVbN2v_cos; _ZGVcN4v_cos; _ZGVdN4v_cos; _ZGVeN8v_cos; + _ZGVbN4v_cosf; _ZGVcN8v_cosf; _ZGVdN8v_cosf; _ZGVeN16v_cosf; } } diff --git a/sysdeps/x86_64/fpu/multiarch/Makefile b/sysdeps/x86_64/fpu/multiarch/Makefile index b2f3266..6b50475 100644 --- a/sysdeps/x86_64/fpu/multiarch/Makefile +++ b/sysdeps/x86_64/fpu/multiarch/Makefile @@ -54,5 +54,6 @@ endif ifeq ($(subdir),mathvec) libmvec-sysdep_routines += svml_d_cos2_core_sse4 svml_d_cos4_core_avx2 \ - svml_d_cos8_core_avx512 + svml_d_cos8_core_avx512 svml_s_cosf4_core_sse4 \ + svml_s_cosf8_core_avx2 svml_s_cosf16_core_avx512 endif diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core.S b/sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core.S new file mode 100644 index 0000000..91564de --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core.S @@ -0,0 +1,39 @@ +/* Multiple versions of vectorized cosf. + Copyright (C) 2014-2015 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 + + .text +ENTRY (_ZGVeN16v_cosf) + .type _ZGVeN16v_cosf, @gnu_indirect_function + cmpl $0, KIND_OFFSET+__cpu_features(%rip) + jne 1 + call __init_cpu_features +1: leaq _ZGVeN16v_cosf_skx(%rip), %rax + testl $bit_AVX512DQ_Usable, __cpu_features+FEATURE_OFFSET+index_AVX512DQ_Usable(%rip) + jnz 3 +2: leaq _ZGVeN16v_cosf_knl(%rip), %rax + testl $bit_AVX512F_Usable, __cpu_features+FEATURE_OFFSET+index_AVX512F_Usable(%rip) + jnz 3 + leaq _ZGVeN16v_cosf_avx2_wrapper(%rip), %rax +3: ret +END (_ZGVeN16v_cosf) + +#define _ZGVeN16v_cosf _ZGVeN16v_cosf_avx2_wrapper +#include "../svml_s_cosf16_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core_avx512.S new file mode 100644 index 0000000..a78ae2e --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core_avx512.S @@ -0,0 +1,460 @@ +/* Function cosf vectorized with AVX-512. KNL and SKX versions. + Copyright (C) 2014-2015 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_cosf_data.h" +#include "svml_s_wrapper_impl.h" + + .text +ENTRY (_ZGVeN16v_cosf_knl) +#ifndef HAVE_AVX512_ASM_SUPPORT +WRAPPER_IMPL_AVX512 _ZGVdN8v_cosf +#else +/* + ALGORITHM DESCRIPTION: + + 1) Range reduction to [-Pi/2; +Pi/2] interval + a) We remove sign using AND operation + b) Add Pi/2 value to argument X for Cos to Sin transformation + c) Getting octant Y by 1/Pi multiplication + d) Add "Right Shifter" value + e) Treat obtained value as integer for destination sign setting. + Shift first bit of this value to the last (sign) position + f) Subtract "Right Shifter" value + g) Subtract 0.5 from result for octant correction + h) Subtract Y*PI from X argument, where PI divided to 4 parts: + X = X - Y*PI1 - Y*PI2 - Y*PI3 - Y*PI4; + 2) Polynomial (minimax for sin within [-Pi/2; +Pi/2] interval) + a) Calculate X^2 = X * X + b) Calculate polynomial: + R = X + X * X^2 * (A3 + x^2 * (A5 + ..... + 3) Destination sign setting + a) Set shifted destination sign using XOR operation: + R = XOR( R, S ); + */ + pushq %rbp + cfi_adjust_cfa_offset (8) + cfi_rel_offset (%rbp, 0) + movq %rsp, %rbp + cfi_def_cfa_register (%rbp) + andq $-64, %rsp + subq $1280, %rsp + movq __svml_scos_data@GOTPCREL(%rip), %rdx + +/* + h) Subtract Y*PI from X argument, where PI divided to 4 parts: + X = X - Y*PI1 - Y*PI2 - Y*PI3 + */ + vmovaps %zmm0, %zmm6 + movl $-1, %eax + +/* b) Add Pi/2 value to argument X for Cos to Sin transformation */ + vaddps __sHalfPI(%rdx), %zmm0, %zmm2 + vmovups __sRShifter(%rdx), %zmm3 + +/* + 1) Range reduction to [-Pi/2; +Pi/2] interval + c) Getting octant Y by 1/Pi multiplication + d) Add "Right Shifter" (0x4B000000) value + */ + vfmadd132ps __sInvPI(%rdx), %zmm3, %zmm2 + vmovups __sPI1_FMA(%rdx), %zmm5 + +/* f) Subtract "Right Shifter" (0x4B000000) value */ + vsubps %zmm3, %zmm2, %zmm4 + vmovups __sA9_FMA(%rdx), %zmm9 + +/* Check for large and special arguments */ + vpandd __sAbsMask(%rdx), %zmm0, %zmm1 + +/* + e) Treat obtained value as integer for destination sign setting. + Shift first bit of this value to the last (sign) position (S << 31) + */ + vpslld $31, %zmm2, %zmm8 + vcmpps $22, __sRangeReductionVal(%rdx), %zmm1, %k1 + vpbroadcastd %eax, %zmm12{%k1}{z} + +/* g) Subtract 0.5 from result for octant correction */ + vsubps __sOneHalf(%rdx), %zmm4, %zmm7 + vptestmd %zmm12, %zmm12, %k0 + vfnmadd231ps %zmm7, %zmm5, %zmm6 + kmovw %k0, %ecx + vfnmadd231ps __sPI2_FMA(%rdx), %zmm7, %zmm6 + vfnmadd132ps __sPI3_FMA(%rdx), %zmm6, %zmm7 + +/* a) Calculate X^2 = X * X */ + vmulps %zmm7, %zmm7, %zmm10 + +/* + 3) Destination sign setting + a) Set shifted destination sign using XOR operation: + R = XOR( R, S ); + */ + vpxord %zmm8, %zmm7, %zmm11 + +/* + b) Calculate polynomial: + R = X + X * X^2 * (A3 + x^2 * (A5 + x^2 * (A7 + x^2 * (A9)))); + */ + vfmadd213ps __sA7_FMA(%rdx), %zmm10, %zmm9 + vfmadd213ps __sA5_FMA(%rdx), %zmm10, %zmm9 + vfmadd213ps __sA3(%rdx), %zmm10, %zmm9 + vmulps %zmm10, %zmm9, %zmm1 + vfmadd213ps %zmm11, %zmm11, %zmm1 + testl %ecx, %ecx + jne .LBL_1_3 + +.LBL_1_2: + cfi_remember_state + vmovaps %zmm1, %zmm0 + movq %rbp, %rsp + cfi_def_cfa_register (%rsp) + popq %rbp + cfi_adjust_cfa_offset (-8) + cfi_restore (%rbp) + ret + +.LBL_1_3: + cfi_restore_state + vmovups %zmm0, 1152(%rsp) + vmovups %zmm1, 1216(%rsp) + je .LBL_1_2 + + xorb %dl, %dl + kmovw %k4, 1048(%rsp) + xorl %eax, %eax + kmovw %k5, 1040(%rsp) + kmovw %k6, 1032(%rsp) + kmovw %k7, 1024(%rsp) + vmovups %zmm16, 960(%rsp) + vmovups %zmm17, 896(%rsp) + vmovups %zmm18, 832(%rsp) + vmovups %zmm19, 768(%rsp) + vmovups %zmm20, 704(%rsp) + vmovups %zmm21, 640(%rsp) + vmovups %zmm22, 576(%rsp) + vmovups %zmm23, 512(%rsp) + vmovups %zmm24, 448(%rsp) + vmovups %zmm25, 384(%rsp) + vmovups %zmm26, 320(%rsp) + vmovups %zmm27, 256(%rsp) + vmovups %zmm28, 192(%rsp) + vmovups %zmm29, 128(%rsp) + vmovups %zmm30, 64(%rsp) + vmovups %zmm31, (%rsp) + movq %rsi, 1064(%rsp) + movq %rdi, 1056(%rsp) + movq %r12, 1096(%rsp) + cfi_offset_rel_rsp (12, 1096) + movb %dl, %r12b + movq %r13, 1088(%rsp) + cfi_offset_rel_rsp (13, 1088) + movl %ecx, %r13d + movq %r14, 1080(%rsp) + cfi_offset_rel_rsp (14, 1080) + movl %eax, %r14d + movq %r15, 1072(%rsp) + cfi_offset_rel_rsp (15, 1072) + cfi_remember_state + +.LBL_1_6: + btl %r14d, %r13d + jc .LBL_1_12 + +.LBL_1_7: + lea 1(%r14), %esi + btl %esi, %r13d + jc .LBL_1_10 + +.LBL_1_8: + addb $1, %r12b + addl $2, %r14d + cmpb $16, %r12b + jb .LBL_1_6 + + kmovw 1048(%rsp), %k4 + movq 1064(%rsp), %rsi + kmovw 1040(%rsp), %k5 + movq 1056(%rsp), %rdi + kmovw 1032(%rsp), %k6 + movq 1096(%rsp), %r12 + cfi_restore (%r12) + movq 1088(%rsp), %r13 + cfi_restore (%r13) + kmovw 1024(%rsp), %k7 + vmovups 960(%rsp), %zmm16 + vmovups 896(%rsp), %zmm17 + vmovups 832(%rsp), %zmm18 + vmovups 768(%rsp), %zmm19 + vmovups 704(%rsp), %zmm20 + vmovups 640(%rsp), %zmm21 + vmovups 576(%rsp), %zmm22 + vmovups 512(%rsp), %zmm23 + vmovups 448(%rsp), %zmm24 + vmovups 384(%rsp), %zmm25 + vmovups 320(%rsp), %zmm26 + vmovups 256(%rsp), %zmm27 + vmovups 192(%rsp), %zmm28 + vmovups 128(%rsp), %zmm29 + vmovups 64(%rsp), %zmm30 + vmovups (%rsp), %zmm31 + movq 1080(%rsp), %r14 + cfi_restore (%r14) + movq 1072(%rsp), %r15 + cfi_restore (%r15) + vmovups 1216(%rsp), %zmm1 + jmp .LBL_1_2 + +.LBL_1_10: + cfi_restore_state + movzbl %r12b, %r15d + vmovss 1156(%rsp,%r15,8), %xmm0 + call cosf@PLT + vmovss %xmm0, 1220(%rsp,%r15,8) + jmp .LBL_1_8 + +.LBL_1_12: + movzbl %r12b, %r15d + vmovss 1152(%rsp,%r15,8), %xmm0 + call cosf@PLT + vmovss %xmm0, 1216(%rsp,%r15,8) + jmp .LBL_1_7 +#endif +END (_ZGVeN16v_cosf_knl) + +ENTRY (_ZGVeN16v_cosf_skx) +#ifndef HAVE_AVX512_ASM_SUPPORT +WRAPPER_IMPL_AVX512 _ZGVdN8v_cosf +#else +/* + ALGORITHM DESCRIPTION: + + 1) Range reduction to [-Pi/2; +Pi/2] interval + a) We remove sign using AND operation + b) Add Pi/2 value to argument X for Cos to Sin transformation + c) Getting octant Y by 1/Pi multiplication + d) Add "Right Shifter" value + e) Treat obtained value as integer for destination sign setting. + Shift first bit of this value to the last (sign) position + f) Subtract "Right Shifter" value + g) Subtract 0.5 from result for octant correction + h) Subtract Y*PI from X argument, where PI divided to 4 parts: + X = X - Y*PI1 - Y*PI2 - Y*PI3 - Y*PI4; + 2) Polynomial (minimax for sin within [-Pi/2; +Pi/2] interval) + a) Calculate X^2 = X * X + b) Calculate polynomial: + R = X + X * X^2 * (A3 + x^2 * (A5 + ..... + 3) Destination sign setting + a) Set shifted destination sign using XOR operation: + R = XOR( R, S ); + */ + pushq %rbp + cfi_adjust_cfa_offset (8) + cfi_rel_offset (%rbp, 0) + movq %rsp, %rbp + cfi_def_cfa_register (%rbp) + andq $-64, %rsp + subq $1280, %rsp + movq __svml_scos_data@GOTPCREL(%rip), %rax + +/* + h) Subtract Y*PI from X argument, where PI divided to 4 parts: + X = X - Y*PI1 - Y*PI2 - Y*PI3 + */ + vmovaps %zmm0, %zmm6 + vmovups .L_2il0floatpacket.13(%rip), %zmm12 + vmovups __sRShifter(%rax), %zmm3 + vmovups __sPI1_FMA(%rax), %zmm5 + vmovups __sA9_FMA(%rax), %zmm9 + +/* b) Add Pi/2 value to argument X for Cos to Sin transformation */ + vaddps __sHalfPI(%rax), %zmm0, %zmm2 + +/* Check for large and special arguments */ + vandps __sAbsMask(%rax), %zmm0, %zmm1 + +/* + 1) Range reduction to [-Pi/2; +Pi/2] interval + c) Getting octant Y by 1/Pi multiplication + d) Add "Right Shifter" (0x4B000000) value + */ + vfmadd132ps __sInvPI(%rax), %zmm3, %zmm2 + vcmpps $18, __sRangeReductionVal(%rax), %zmm1, %k1 + +/* + e) Treat obtained value as integer for destination sign setting. + Shift first bit of this value to the last (sign) position (S << 31) + */ + vpslld $31, %zmm2, %zmm8 + +/* f) Subtract "Right Shifter" (0x4B000000) value */ + vsubps %zmm3, %zmm2, %zmm4 + +/* g) Subtract 0.5 from result for octant correction */ + vsubps __sOneHalf(%rax), %zmm4, %zmm7 + vfnmadd231ps %zmm7, %zmm5, %zmm6 + vfnmadd231ps __sPI2_FMA(%rax), %zmm7, %zmm6 + vfnmadd132ps __sPI3_FMA(%rax), %zmm6, %zmm7 + +/* a) Calculate X^2 = X * X */ + vmulps %zmm7, %zmm7, %zmm10 + +/* + 3) Destination sign setting + a) Set shifted destination sign using XOR operation: + R = XOR( R, S ); + */ + vxorps %zmm8, %zmm7, %zmm11 + +/* + b) Calculate polynomial: + R = X + X * X^2 * (A3 + x^2 * (A5 + x^2 * (A7 + x^2 * (A9)))); + */ + vfmadd213ps __sA7_FMA(%rax), %zmm10, %zmm9 + vfmadd213ps __sA5_FMA(%rax), %zmm10, %zmm9 + vfmadd213ps __sA3(%rax), %zmm10, %zmm9 + vpandnd %zmm1, %zmm1, %zmm12{%k1} + vmulps %zmm10, %zmm9, %zmm1 + vptestmd %zmm12, %zmm12, %k0 + vfmadd213ps %zmm11, %zmm11, %zmm1 + kmovw %k0, %ecx + testl %ecx, %ecx + jne .LBL_2_3 +.LBL_2_2: + cfi_remember_state + vmovaps %zmm1, %zmm0 + movq %rbp, %rsp + cfi_def_cfa_register (%rsp) + popq %rbp + cfi_adjust_cfa_offset (-8) + cfi_restore (%rbp) + ret + +.LBL_2_3: + cfi_restore_state + vmovups %zmm0, 1152(%rsp) + vmovups %zmm1, 1216(%rsp) + je .LBL_2_2 + + xorb %dl, %dl + xorl %eax, %eax + kmovw %k4, 1048(%rsp) + kmovw %k5, 1040(%rsp) + kmovw %k6, 1032(%rsp) + kmovw %k7, 1024(%rsp) + vmovups %zmm16, 960(%rsp) + vmovups %zmm17, 896(%rsp) + vmovups %zmm18, 832(%rsp) + vmovups %zmm19, 768(%rsp) + vmovups %zmm20, 704(%rsp) + vmovups %zmm21, 640(%rsp) + vmovups %zmm22, 576(%rsp) + vmovups %zmm23, 512(%rsp) + vmovups %zmm24, 448(%rsp) + vmovups %zmm25, 384(%rsp) + vmovups %zmm26, 320(%rsp) + vmovups %zmm27, 256(%rsp) + vmovups %zmm28, 192(%rsp) + vmovups %zmm29, 128(%rsp) + vmovups %zmm30, 64(%rsp) + vmovups %zmm31, (%rsp) + movq %rsi, 1064(%rsp) + movq %rdi, 1056(%rsp) + movq %r12, 1096(%rsp) + cfi_offset_rel_rsp (12, 1096) + movb %dl, %r12b + movq %r13, 1088(%rsp) + cfi_offset_rel_rsp (13, 1088) + movl %ecx, %r13d + movq %r14, 1080(%rsp) + cfi_offset_rel_rsp (14, 1080) + movl %eax, %r14d + movq %r15, 1072(%rsp) + cfi_offset_rel_rsp (15, 1072) + cfi_remember_state + +.LBL_2_6: + btl %r14d, %r13d + jc .LBL_2_12 +.LBL_2_7: + lea 1(%r14), %esi + btl %esi, %r13d + jc .LBL_2_10 +.LBL_2_8: + incb %r12b + addl $2, %r14d + cmpb $16, %r12b + jb .LBL_2_6 + kmovw 1048(%rsp), %k4 + kmovw 1040(%rsp), %k5 + kmovw 1032(%rsp), %k6 + kmovw 1024(%rsp), %k7 + vmovups 960(%rsp), %zmm16 + vmovups 896(%rsp), %zmm17 + vmovups 832(%rsp), %zmm18 + vmovups 768(%rsp), %zmm19 + vmovups 704(%rsp), %zmm20 + vmovups 640(%rsp), %zmm21 + vmovups 576(%rsp), %zmm22 + vmovups 512(%rsp), %zmm23 + vmovups 448(%rsp), %zmm24 + vmovups 384(%rsp), %zmm25 + vmovups 320(%rsp), %zmm26 + vmovups 256(%rsp), %zmm27 + vmovups 192(%rsp), %zmm28 + vmovups 128(%rsp), %zmm29 + vmovups 64(%rsp), %zmm30 + vmovups (%rsp), %zmm31 + vmovups 1216(%rsp), %zmm1 + movq 1064(%rsp), %rsi + movq 1056(%rsp), %rdi + movq 1096(%rsp), %r12 + cfi_restore (%r12) + movq 1088(%rsp), %r13 + cfi_restore (%r13) + movq 1080(%rsp), %r14 + cfi_restore (%r14) + movq 1072(%rsp), %r15 + cfi_restore (%r15) + jmp .LBL_2_2 + +.LBL_2_10: + cfi_restore_state + movzbl %r12b, %r15d + vmovss 1156(%rsp,%r15,8), %xmm0 + vzeroupper + vmovss 1156(%rsp,%r15,8), %xmm0 + call cosf@PLT + vmovss %xmm0, 1220(%rsp,%r15,8) + jmp .LBL_2_8 +.LBL_2_12: + movzbl %r12b, %r15d + vmovss 1152(%rsp,%r15,8), %xmm0 + vzeroupper + vmovss 1152(%rsp,%r15,8), %xmm0 + call cosf@PLT + vmovss %xmm0, 1216(%rsp,%r15,8) + jmp .LBL_2_7 +#endif +END (_ZGVeN16v_cosf_skx) + + .section .rodata, "a" +.L_2il0floatpacket.13: + .long 0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff + .type .L_2il0floatpacket.13,@object diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_cosf4_core.S b/sysdeps/x86_64/fpu/multiarch/svml_s_cosf4_core.S new file mode 100644 index 0000000..fa2363b --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_cosf4_core.S @@ -0,0 +1,38 @@ +/* Multiple versions of vectorized cosf, vector length is 4. + Copyright (C) 2014-2015 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 + + .text +ENTRY (_ZGVbN4v_cosf) + .type _ZGVbN4v_cosf, @gnu_indirect_function + cmpl $0, KIND_OFFSET+__cpu_features(%rip) + jne 1f + call __init_cpu_features +1: leaq _ZGVbN4v_cosf_sse4(%rip), %rax + testl $bit_SSE4_1, __cpu_features+CPUID_OFFSET+index_SSE4_1(%rip) + jz 2f + ret +2: leaq _ZGVbN4v_cosf_sse2(%rip), %rax + ret +END (_ZGVbN4v_cosf) +libmvec_hidden_def (_ZGVbN4v_cosf) + +#define _ZGVbN4v_cosf _ZGVbN4v_cosf_sse2 +#include "../svml_s_cosf4_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_cosf4_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_s_cosf4_core_sse4.S new file mode 100644 index 0000000..f231ba2 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_cosf4_core_sse4.S @@ -0,0 +1,227 @@ +/* Function cosf vectorized with SSE4. + Copyright (C) 2014-2015 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_cosf_data.h" + + .text +ENTRY (_ZGVbN4v_cosf_sse4) +/* + ALGORITHM DESCRIPTION: + + 1) Range reduction to [-Pi/2; +Pi/2] interval + a) We remove sign using AND operation + b) Add Pi/2 value to argument X for Cos to Sin transformation + c) Getting octant Y by 1/Pi multiplication + d) Add "Right Shifter" value + e) Treat obtained value as integer for destination sign setting. + Shift first bit of this value to the last (sign) position + f) Subtract "Right Shifter" value + g) Subtract 0.5 from result for octant correction + h) Subtract Y*PI from X argument, where PI divided to 4 parts: + X = X - Y*PI1 - Y*PI2 - Y*PI3 - Y*PI4; + 2) Polynomial (minimax for sin within [-Pi/2; +Pi/2] interval) + a) Calculate X^2 = X * X + b) Calculate polynomial: + R = X + X * X^2 * (A3 + x^2 * (A5 + ..... + 3) Destination sign setting + a) Set shifted destination sign using XOR operation: + R = XOR( R, S ); + */ + pushq %rbp + cfi_adjust_cfa_offset (8) + cfi_rel_offset (%rbp, 0) + movq %rsp, %rbp + cfi_def_cfa_register (%rbp) + andq $-64, %rsp + subq $320, %rsp + movaps %xmm0, %xmm4 + movq __svml_scos_data@GOTPCREL(%rip), %rax + movups __sHalfPI(%rax), %xmm1 + movups __sRShifter(%rax), %xmm5 + +/* b) Add Pi/2 value to argument X for Cos to Sin transformation */ + addps %xmm4, %xmm1 + +/* + 1) Range reduction to [-Pi/2; +Pi/2] interval + c) Getting octant Y by 1/Pi multiplication + d) Add "Right Shifter" (0x4B000000) value + */ + mulps __sInvPI(%rax), %xmm1 + movups __sPI1(%rax), %xmm6 + addps %xmm5, %xmm1 + +/* + e) Treat obtained value as integer for destination sign setting. + Shift first bit of this value to the last (sign) position (S << 31) + */ + movaps %xmm1, %xmm2 + +/* f) Subtract "Right Shifter" (0x4B000000) value */ + subps %xmm5, %xmm1 + movups __sPI2(%rax), %xmm7 + pslld $31, %xmm2 + movups __sPI3(%rax), %xmm5 + movups __sAbsMask(%rax), %xmm3 + +/* Check for large and special arguments */ + andps %xmm4, %xmm3 + +/* g) Subtract 0.5 from result for octant correction */ + subps __sOneHalf(%rax), %xmm1 + cmpnleps __sRangeReductionVal(%rax), %xmm3 + +/* + h) Subtract Y*PI from X argument, where PI divided to 4 parts: + X = X - Y*PI1 - Y*PI2 - Y*PI3 - Y*PI4; + */ + mulps %xmm1, %xmm6 + mulps %xmm1, %xmm7 + mulps %xmm1, %xmm5 + subps %xmm6, %xmm0 + movmskps %xmm3, %ecx + movups __sPI4(%rax), %xmm6 + subps %xmm7, %xmm0 + mulps %xmm6, %xmm1 + subps %xmm5, %xmm0 + subps %xmm1, %xmm0 + +/* a) Calculate X^2 = X * X */ + movaps %xmm0, %xmm1 + mulps %xmm0, %xmm1 + +/* + 3) Destination sign setting + a) Set shifted destination sign using XOR operation: + R = XOR( R, S ); + */ + xorps %xmm2, %xmm0 + movups __sA9(%rax), %xmm2 + +/* + b) Calculate polynomial: + R = X + X * X^2 * (A3 + x^2 * (A5 + x^2 * (A7 + x^2 * (A9)))); + */ + mulps %xmm1, %xmm2 + addps __sA7(%rax), %xmm2 + mulps %xmm1, %xmm2 + addps __sA5(%rax), %xmm2 + mulps %xmm1, %xmm2 + addps __sA3(%rax), %xmm2 + mulps %xmm2, %xmm1 + mulps %xmm0, %xmm1 + addps %xmm1, %xmm0 + testl %ecx, %ecx + jne .LBL_1_3 + +.LBL_1_2: + cfi_remember_state + movq %rbp, %rsp + cfi_def_cfa_register (%rsp) + popq %rbp + cfi_adjust_cfa_offset (-8) + cfi_restore (%rbp) + ret + +.LBL_1_3: + cfi_restore_state + movups %xmm4, 192(%rsp) + movups %xmm0, 256(%rsp) + je .LBL_1_2 + + xorb %dl, %dl + xorl %eax, %eax + movups %xmm8, 112(%rsp) + movups %xmm9, 96(%rsp) + movups %xmm10, 80(%rsp) + movups %xmm11, 64(%rsp) + movups %xmm12, 48(%rsp) + movups %xmm13, 32(%rsp) + movups %xmm14, 16(%rsp) + movups %xmm15, (%rsp) + movq %rsi, 136(%rsp) + movq %rdi, 128(%rsp) + movq %r12, 168(%rsp) + cfi_offset_rel_rsp (12, 168) + movb %dl, %r12b + movq %r13, 160(%rsp) + cfi_offset_rel_rsp (13, 160) + movl %ecx, %r13d + movq %r14, 152(%rsp) + cfi_offset_rel_rsp (14, 152) + movl %eax, %r14d + movq %r15, 144(%rsp) + cfi_offset_rel_rsp (15, 144) + cfi_remember_state + +.LBL_1_6: + btl %r14d, %r13d + jc .LBL_1_12 + +.LBL_1_7: + lea 1(%r14), %esi + btl %esi, %r13d + jc .LBL_1_10 + +.LBL_1_8: + incb %r12b + addl $2, %r14d + cmpb $16, %r12b + jb .LBL_1_6 + + movups 112(%rsp), %xmm8 + movups 96(%rsp), %xmm9 + movups 80(%rsp), %xmm10 + movups 64(%rsp), %xmm11 + movups 48(%rsp), %xmm12 + movups 32(%rsp), %xmm13 + movups 16(%rsp), %xmm14 + movups (%rsp), %xmm15 + movq 136(%rsp), %rsi + movq 128(%rsp), %rdi + movq 168(%rsp), %r12 + cfi_restore (%r12) + movq 160(%rsp), %r13 + cfi_restore (%r13) + movq 152(%rsp), %r14 + cfi_restore (%r14) + movq 144(%rsp), %r15 + cfi_restore (%r15) + movups 256(%rsp), %xmm0 + jmp .LBL_1_2 + +.LBL_1_10: + cfi_restore_state + movzbl %r12b, %r15d + movss 196(%rsp,%r15,8), %xmm0 + + call cosf@PLT + + movss %xmm0, 260(%rsp,%r15,8) + jmp .LBL_1_8 + +.LBL_1_12: + movzbl %r12b, %r15d + movss 192(%rsp,%r15,8), %xmm0 + + call cosf@PLT + + movss %xmm0, 256(%rsp,%r15,8) + jmp .LBL_1_7 +END (_ZGVbN4v_cosf_sse4) diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_cosf8_core.S b/sysdeps/x86_64/fpu/multiarch/svml_s_cosf8_core.S new file mode 100644 index 0000000..e14bba4 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_cosf8_core.S @@ -0,0 +1,38 @@ +/* Multiple versions of vectorized cosf, vector length is 8. + Copyright (C) 2014-2015 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 + + .text +ENTRY (_ZGVdN8v_cosf) + .type _ZGVdN8v_cosf, @gnu_indirect_function + cmpl $0, KIND_OFFSET+__cpu_features(%rip) + jne 1f + call __init_cpu_features +1: leaq _ZGVdN8v_cosf_avx2(%rip), %rax + testl $bit_AVX2_Usable, __cpu_features+FEATURE_OFFSET+index_AVX2_Usable(%rip) + jz 2f + ret +2: leaq _ZGVdN8v_cosf_sse_wrapper(%rip), %rax + ret +END (_ZGVdN8v_cosf) +libmvec_hidden_def (_ZGVdN8v_cosf) + +#define _ZGVdN8v_cosf _ZGVdN8v_cosf_sse_wrapper +#include "../svml_s_cosf8_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_cosf8_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_cosf8_core_avx2.S new file mode 100644 index 0000000..6c25e14 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_cosf8_core_avx2.S @@ -0,0 +1,215 @@ +/* Function cosf vectorized with AVX2. + Copyright (C) 2014-2015 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_cosf_data.h" + + .text +ENTRY (_ZGVdN8v_cosf_avx2) +/* + ALGORITHM DESCRIPTION: + + 1) Range reduction to [-Pi/2; +Pi/2] interval + a) We remove sign using AND operation + b) Add Pi/2 value to argument X for Cos to Sin transformation + c) Getting octant Y by 1/Pi multiplication + d) Add "Right Shifter" value + e) Treat obtained value as integer for destination sign setting. + Shift first bit of this value to the last (sign) position + f) Subtract "Right Shifter" value + g) Subtract 0.5 from result for octant correction + h) Subtract Y*PI from X argument, where PI divided to 4 parts: + X = X - Y*PI1 - Y*PI2 - Y*PI3 - Y*PI4; + 2) Polynomial (minimax for sin within [-Pi/2; +Pi/2] interval) + a) Calculate X^2 = X * X + b) Calculate polynomial: + R = X + X * X^2 * (A3 + x^2 * (A5 + ..... + 3) Destination sign setting + a) Set shifted destination sign using XOR operation: + R = XOR( R, S ); + */ + pushq %rbp + cfi_adjust_cfa_offset (8) + cfi_rel_offset (%rbp, 0) + movq %rsp, %rbp + cfi_def_cfa_register (%rbp) + andq $-64, %rsp + subq $448, %rsp + movq __svml_scos_data@GOTPCREL(%rip), %rax + vmovaps %ymm0, %ymm2 + vmovups __sRShifter(%rax), %ymm5 + vmovups __sPI1_FMA(%rax), %ymm7 + +/* b) Add Pi/2 value to argument X for Cos to Sin transformation */ + vaddps __sHalfPI(%rax), %ymm2, %ymm4 + +/* + 1) Range reduction to [-Pi/2; +Pi/2] interval + c) Getting octant Y by 1/Pi multiplication + d) Add "Right Shifter" (0x4B000000) value + */ + vfmadd132ps __sInvPI(%rax), %ymm5, %ymm4 + +/* f) Subtract "Right Shifter" (0x4B000000) value */ + vsubps %ymm5, %ymm4, %ymm6 + +/* + e) Treat obtained value as integer for destination sign setting. + Shift first bit of this value to the last (sign) position (S << 31) + */ + vpslld $31, %ymm4, %ymm0 + +/* g) Subtract 0.5 from result for octant correction */ + vsubps __sOneHalf(%rax), %ymm6, %ymm4 + +/* Check for large and special arguments */ + vandps __sAbsMask(%rax), %ymm2, %ymm3 + vcmpnle_uqps __sRangeReductionVal(%rax), %ymm3, %ymm1 + +/* + h) Subtract Y*PI from X argument, where PI divided to 4 parts: + X = X - Y*PI1 - Y*PI2 - Y*PI3 + */ + vmovaps %ymm2, %ymm3 + vfnmadd231ps %ymm4, %ymm7, %ymm3 + vfnmadd231ps __sPI2_FMA(%rax), %ymm4, %ymm3 + vfnmadd132ps __sPI3_FMA(%rax), %ymm3, %ymm4 + +/* a) Calculate X^2 = X * X */ + vmulps %ymm4, %ymm4, %ymm5 + +/* + 3) Destination sign setting + a) Set shifted destination sign using XOR operation: + R = XOR( R, S ); + */ + vxorps %ymm0, %ymm4, %ymm6 + vmovups __sA9_FMA(%rax), %ymm0 + +/* + b) Calculate polynomial: + R = X + X * X^2 * (A3 + x^2 * (A5 + x^2 * (A7 + x^2 * (A9)))) + */ + vfmadd213ps __sA7_FMA(%rax), %ymm5, %ymm0 + vfmadd213ps __sA5_FMA(%rax), %ymm5, %ymm0 + vfmadd213ps __sA3(%rax), %ymm5, %ymm0 + vmulps %ymm5, %ymm0, %ymm0 + vmovmskps %ymm1, %ecx + vfmadd213ps %ymm6, %ymm6, %ymm0 + testl %ecx, %ecx + jne .LBL_1_3 + +.LBL_1_2: + cfi_remember_state + movq %rbp, %rsp + cfi_def_cfa_register (%rsp) + popq %rbp + cfi_adjust_cfa_offset (-8) + cfi_restore (%rbp) + ret + +.LBL_1_3: + cfi_restore_state + vmovups %ymm2, 320(%rsp) + vmovups %ymm0, 384(%rsp) + je .LBL_1_2 + + xorb %dl, %dl + xorl %eax, %eax + vmovups %ymm8, 224(%rsp) + vmovups %ymm9, 192(%rsp) + vmovups %ymm10, 160(%rsp) + vmovups %ymm11, 128(%rsp) + vmovups %ymm12, 96(%rsp) + vmovups %ymm13, 64(%rsp) + vmovups %ymm14, 32(%rsp) + vmovups %ymm15, (%rsp) + movq %rsi, 264(%rsp) + movq %rdi, 256(%rsp) + movq %r12, 296(%rsp) + cfi_offset_rel_rsp (12, 296) + movb %dl, %r12b + movq %r13, 288(%rsp) + cfi_offset_rel_rsp (13, 288) + movl %ecx, %r13d + movq %r14, 280(%rsp) + cfi_offset_rel_rsp (14, 280) + movl %eax, %r14d + movq %r15, 272(%rsp) + cfi_offset_rel_rsp (15, 272) + cfi_remember_state + +.LBL_1_6: + btl %r14d, %r13d + jc .LBL_1_12 + +.LBL_1_7: + lea 1(%r14), %esi + btl %esi, %r13d + jc .LBL_1_10 + +.LBL_1_8: + incb %r12b + addl $2, %r14d + cmpb $16, %r12b + jb .LBL_1_6 + + vmovups 224(%rsp), %ymm8 + vmovups 192(%rsp), %ymm9 + vmovups 160(%rsp), %ymm10 + vmovups 128(%rsp), %ymm11 + vmovups 96(%rsp), %ymm12 + vmovups 64(%rsp), %ymm13 + vmovups 32(%rsp), %ymm14 + vmovups (%rsp), %ymm15 + vmovups 384(%rsp), %ymm0 + movq 264(%rsp), %rsi + movq 256(%rsp), %rdi + movq 296(%rsp), %r12 + cfi_restore (%r12) + movq 288(%rsp), %r13 + cfi_restore (%r13) + movq 280(%rsp), %r14 + cfi_restore (%r14) + movq 272(%rsp), %r15 + cfi_restore (%r15) + jmp .LBL_1_2 + +.LBL_1_10: + cfi_restore_state + movzbl %r12b, %r15d + vmovss 324(%rsp,%r15,8), %xmm0 + vzeroupper + + call cosf@PLT + + vmovss %xmm0, 388(%rsp,%r15,8) + jmp .LBL_1_8 + +.LBL_1_12: + movzbl %r12b, %r15d + vmovss 320(%rsp,%r15,8), %xmm0 + vzeroupper + + call cosf@PLT + + vmovss %xmm0, 384(%rsp,%r15,8) + jmp .LBL_1_7 + +END (_ZGVdN8v_cosf_avx2) diff --git a/sysdeps/x86_64/fpu/svml_s_cosf16_core.S b/sysdeps/x86_64/fpu/svml_s_cosf16_core.S new file mode 100644 index 0000000..e623df5 --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_s_cosf16_core.S @@ -0,0 +1,25 @@ +/* Function cosf vectorized with AVX-512. Wrapper to AVX2 version. + Copyright (C) 2014-2015 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_cosf) +WRAPPER_IMPL_AVX512 _ZGVdN8v_cosf +END (_ZGVeN16v_cosf) diff --git a/sysdeps/x86_64/fpu/svml_s_cosf4_core.S b/sysdeps/x86_64/fpu/svml_s_cosf4_core.S new file mode 100644 index 0000000..9875cd7 --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_s_cosf4_core.S @@ -0,0 +1,29 @@ +/* Function cosf vectorized with SSE2, wrapper version. + Copyright (C) 2014-2015 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_cosf) +WRAPPER_IMPL_SSE2 cosf +END (_ZGVbN4v_cosf) + +#ifndef USE_MULTIARCH + libmvec_hidden_def (_ZGVbN4v_cosf) +#endif diff --git a/sysdeps/x86_64/fpu/svml_s_cosf8_core.S b/sysdeps/x86_64/fpu/svml_s_cosf8_core.S new file mode 100644 index 0000000..376ee35 --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_s_cosf8_core.S @@ -0,0 +1,29 @@ +/* Function cosf vectorized with AVX2, wrapper version. + Copyright (C) 2014-2015 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_cosf) +WRAPPER_IMPL_AVX _ZGVbN4v_cosf +END (_ZGVdN8v_cosf) + +#ifndef USE_MULTIARCH + libmvec_hidden_def (_ZGVdN8v_cosf) +#endif diff --git a/sysdeps/x86_64/fpu/svml_s_cosf8_core_avx.S b/sysdeps/x86_64/fpu/svml_s_cosf8_core_avx.S new file mode 100644 index 0000000..a443fd2 --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_s_cosf8_core_avx.S @@ -0,0 +1,25 @@ +/* Function cosf vectorized in AVX ISA as wrapper to SSE4 ISA version. + Copyright (C) 2014-2015 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_cosf) +WRAPPER_IMPL_AVX _ZGVbN4v_cosf +END (_ZGVcN8v_cosf) diff --git a/sysdeps/x86_64/fpu/svml_s_cosf_data.S b/sysdeps/x86_64/fpu/svml_s_cosf_data.S new file mode 100644 index 0000000..2f7303c --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_s_cosf_data.S @@ -0,0 +1,1130 @@ +/* Data for function cosf. + Copyright (C) 2014-2015 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 "svml_s_cosf_data.h" + + .section .rodata, "a" + .align 64 + +/* Data table for vector implementations of function cosf. + The table may contain polynomial, reduction, lookup coefficients + and other macro_names obtained through different methods + of research and experimental work. */ + + .globl __svml_scos_data +__svml_scos_data: + +/* Lookup table for high accuracy version (CHL,SHi,SLo,Sigma). */ +.if .-__svml_scos_data != __dT +.err +.endif + .long 0x00000000 + .long 0x3f800000 + .long 0x00000000 + .long 0x00000000 + .long 0x3bdbd541 + .long 0x3f7fec43 + .long 0x3084cd0d + .long 0xbd000000 + .long 0x3c5c1342 + .long 0x3f7fb10f + .long 0x31de5b5f + .long 0xbd800000 + .long 0xbc354825 + .long 0x3f7f4e6d + .long 0x32d01884 + .long 0xbd800000 + .long 0x3cdd0b28 + .long 0x3f7ec46d + .long 0x31f44949 + .long 0xbe000000 + .long 0x3b29b1a9 + .long 0x3f7e1324 + .long 0xb2f1e603 + .long 0xbe000000 + .long 0xbcb2041c + .long 0x3f7d3aac + .long 0xb0f75ae9 + .long 0xbe000000 + .long 0xbd3c4289 + .long 0x3f7c3b28 + .long 0xb231d68b + .long 0xbe000000 + .long 0x3d60e8f8 + .long 0x3f7b14be + .long 0x32ff75cb + .long 0xbe800000 + .long 0x3cfd1f65 + .long 0x3f79c79d + .long 0x32c64e59 + .long 0xbe800000 + .long 0x3be60685 + .long 0x3f7853f8 + .long 0xb20db9e5 + .long 0xbe800000 + .long 0xbc88e931 + .long 0x3f76ba07 + .long 0x326d092c + .long 0xbe800000 + .long 0xbd25018c + .long 0x3f74fa0b + .long 0xb2939d22 + .long 0xbe800000 + .long 0xbd826b93 + .long 0x3f731447 + .long 0x32c48e11 + .long 0xbe800000 + .long 0xbdb1f34f + .long 0x3f710908 + .long 0x321ed0dd + .long 0xbe800000 + .long 0x3e0f77ad + .long 0x3f6ed89e + .long 0xb29333dc + .long 0xbf000000 + .long 0x3df043ab + .long 0x3f6c835e + .long 0x32f328d4 + .long 0xbf000000 + .long 0x3dc210d8 + .long 0x3f6a09a7 + .long 0xb2eb236c + .long 0xbf000000 + .long 0x3d945dff + .long 0x3f676bd8 + .long 0xb2bc3389 + .long 0xbf000000 + .long 0x3d4e645a + .long 0x3f64aa59 + .long 0x311a08fa + .long 0xbf000000 + .long 0x3cea5164 + .long 0x3f61c598 + .long 0xb2e7f425 + .long 0xbf000000 + .long 0x3be8b648 + .long 0x3f5ebe05 + .long 0x32c6f953 + .long 0xbf000000 + .long 0xbc670f32 + .long 0x3f5b941a + .long 0x32232dc8 + .long 0xbf000000 + .long 0xbd0f59aa + .long 0x3f584853 + .long 0xb27d5fc0 + .long 0xbf000000 + .long 0xbd639d9d + .long 0x3f54db31 + .long 0x3290ea1a + .long 0xbf000000 + .long 0xbd9b4153 + .long 0x3f514d3d + .long 0x300c4f04 + .long 0xbf000000 + .long 0xbdc3fdff + .long 0x3f4d9f02 + .long 0x327e70e8 + .long 0xbf000000 + .long 0xbdebfe8a + .long 0x3f49d112 + .long 0x32992640 + .long 0xbf000000 + .long 0xbe099e65 + .long 0x3f45e403 + .long 0x32b15174 + .long 0xbf000000 + .long 0xbe1cd957 + .long 0x3f41d870 + .long 0x32bff977 + .long 0xbf000000 + .long 0xbe2fad27 + .long 0x3f3daef9 + .long 0x319aabec + .long 0xbf000000 + .long 0xbe4216eb + .long 0x3f396842 + .long 0xb2810007 + .long 0xbf000000 + .long 0x3e95f61a + .long 0x3f3504f3 + .long 0x324fe77a + .long 0xbf800000 + .long 0x3e8d2f7d + .long 0x3f3085bb + .long 0xb2ae2d32 + .long 0xbf800000 + .long 0x3e84a20e + .long 0x3f2beb4a + .long 0xb2b73136 + .long 0xbf800000 + .long 0x3e789e3f + .long 0x3f273656 + .long 0xb2038343 + .long 0xbf800000 + .long 0x3e686ff3 + .long 0x3f226799 + .long 0x322123bb + .long 0xbf800000 + .long 0x3e58bbb7 + .long 0x3f1d7fd1 + .long 0x3292050c + .long 0xbf800000 + .long 0x3e4983f7 + .long 0x3f187fc0 + .long 0xb1c7a3f3 + .long 0xbf800000 + .long 0x3e3acb0c + .long 0x3f13682a + .long 0x32cdd12e + .long 0xbf800000 + .long 0x3e2c933b + .long 0x3f0e39da + .long 0xb24a32e7 + .long 0xbf800000 + .long 0x3e1edeb5 + .long 0x3f08f59b + .long 0xb2be4b4e + .long 0xbf800000 + .long 0x3e11af97 + .long 0x3f039c3d + .long 0xb25ba002 + .long 0xbf800000 + .long 0x3e0507ea + .long 0x3efc5d27 + .long 0xb180eca9 + .long 0xbf800000 + .long 0x3df1d344 + .long 0x3ef15aea + .long 0xb1ff2139 + .long 0xbf800000 + .long 0x3ddaad38 + .long 0x3ee63375 + .long 0xb1d9c774 + .long 0xbf800000 + .long 0x3dc4a143 + .long 0x3edae880 + .long 0x321e15cc + .long 0xbf800000 + .long 0x3dafb2cc + .long 0x3ecf7bca + .long 0x316a3b63 + .long 0xbf800000 + .long 0x3d9be50c + .long 0x3ec3ef15 + .long 0x31d5d52c + .long 0xbf800000 + .long 0x3d893b12 + .long 0x3eb8442a + .long 0xb2705ba6 + .long 0xbf800000 + .long 0x3d6f6f7e + .long 0x3eac7cd4 + .long 0xb2254e02 + .long 0xbf800000 + .long 0x3d4ebb8a + .long 0x3ea09ae5 + .long 0xb23e89a0 + .long 0xbf800000 + .long 0x3d305f55 + .long 0x3e94a031 + .long 0x326d59f0 + .long 0xbf800000 + .long 0x3d145f8c + .long 0x3e888e93 + .long 0x312c7d9e + .long 0xbf800000 + .long 0x3cf58104 + .long 0x3e78cfcc + .long 0xb11bd41d + .long 0xbf800000 + .long 0x3cc70c54 + .long 0x3e605c13 + .long 0x31a7e4f6 + .long 0xbf800000 + .long 0x3c9d6830 + .long 0x3e47c5c2 + .long 0xb0e5967d + .long 0xbf800000 + .long 0x3c71360b + .long 0x3e2f10a2 + .long 0x311167f9 + .long 0xbf800000 + .long 0x3c315502 + .long 0x3e164083 + .long 0x31e8e614 + .long 0xbf800000 + .long 0x3bf66e3c + .long 0x3dfab273 + .long 0xb11568cf + .long 0xbf800000 + .long 0x3b9dc971 + .long 0x3dc8bd36 + .long 0xb07592f5 + .long 0xbf800000 + .long 0x3b319298 + .long 0x3d96a905 + .long 0xb1531e61 + .long 0xbf800000 + .long 0x3a9de1c8 + .long 0x3d48fb30 + .long 0xb0ef227f + .long 0xbf800000 + .long 0x399de7df + .long 0x3cc90ab0 + .long 0xb005c998 + .long 0xbf800000 + .long 0x00000000 + .long 0x00000000 + .long 0x00000000 + .long 0xbf800000 + .long 0x399de7df + .long 0xbcc90ab0 + .long 0x3005c998 + .long 0xbf800000 + .long 0x3a9de1c8 + .long 0xbd48fb30 + .long 0x30ef227f + .long 0xbf800000 + .long 0x3b319298 + .long 0xbd96a905 + .long 0x31531e61 + .long 0xbf800000 + .long 0x3b9dc971 + .long 0xbdc8bd36 + .long 0x307592f5 + .long 0xbf800000 + .long 0x3bf66e3c + .long 0xbdfab273 + .long 0x311568cf + .long 0xbf800000 + .long 0x3c315502 + .long 0xbe164083 + .long 0xb1e8e614 + .long 0xbf800000 + .long 0x3c71360b + .long 0xbe2f10a2 + .long 0xb11167f9 + .long 0xbf800000 + .long 0x3c9d6830 + .long 0xbe47c5c2 + .long 0x30e5967d + .long 0xbf800000 + .long 0x3cc70c54 + .long 0xbe605c13 + .long 0xb1a7e4f6 + .long 0xbf800000 + .long 0x3cf58104 + .long 0xbe78cfcc + .long 0x311bd41d + .long 0xbf800000 + .long 0x3d145f8c + .long 0xbe888e93 + .long 0xb12c7d9e + .long 0xbf800000 + .long 0x3d305f55 + .long 0xbe94a031 + .long 0xb26d59f0 + .long 0xbf800000 + .long 0x3d4ebb8a + .long 0xbea09ae5 + .long 0x323e89a0 + .long 0xbf800000 + .long 0x3d6f6f7e + .long 0xbeac7cd4 + .long 0x32254e02 + .long 0xbf800000 + .long 0x3d893b12 + .long 0xbeb8442a + .long 0x32705ba6 + .long 0xbf800000 + .long 0x3d9be50c + .long 0xbec3ef15 + .long 0xb1d5d52c + .long 0xbf800000 + .long 0x3dafb2cc + .long 0xbecf7bca + .long 0xb16a3b63 + .long 0xbf800000 + .long 0x3dc4a143 + .long 0xbedae880 + .long 0xb21e15cc + .long 0xbf800000 + .long 0x3ddaad38 + .long 0xbee63375 + .long 0x31d9c774 + .long 0xbf800000 + .long 0x3df1d344 + .long 0xbef15aea + .long 0x31ff2139 + .long 0xbf800000 + .long 0x3e0507ea + .long 0xbefc5d27 + .long 0x3180eca9 + .long 0xbf800000 + .long 0x3e11af97 + .long 0xbf039c3d + .long 0x325ba002 + .long 0xbf800000 + .long 0x3e1edeb5 + .long 0xbf08f59b + .long 0x32be4b4e + .long 0xbf800000 + .long 0x3e2c933b + .long 0xbf0e39da + .long 0x324a32e7 + .long 0xbf800000 + .long 0x3e3acb0c + .long 0xbf13682a + .long 0xb2cdd12e + .long 0xbf800000 + .long 0x3e4983f7 + .long 0xbf187fc0 + .long 0x31c7a3f3 + .long 0xbf800000 + .long 0x3e58bbb7 + .long 0xbf1d7fd1 + .long 0xb292050c + .long 0xbf800000 + .long 0x3e686ff3 + .long 0xbf226799 + .long 0xb22123bb + .long 0xbf800000 + .long 0x3e789e3f + .long 0xbf273656 + .long 0x32038343 + .long 0xbf800000 + .long 0x3e84a20e + .long 0xbf2beb4a + .long 0x32b73136 + .long 0xbf800000 + .long 0x3e8d2f7d + .long 0xbf3085bb + .long 0x32ae2d32 + .long 0xbf800000 + .long 0x3e95f61a + .long 0xbf3504f3 + .long 0xb24fe77a + .long 0xbf800000 + .long 0xbe4216eb + .long 0xbf396842 + .long 0x32810007 + .long 0xbf000000 + .long 0xbe2fad27 + .long 0xbf3daef9 + .long 0xb19aabec + .long 0xbf000000 + .long 0xbe1cd957 + .long 0xbf41d870 + .long 0xb2bff977 + .long 0xbf000000 + .long 0xbe099e65 + .long 0xbf45e403 + .long 0xb2b15174 + .long 0xbf000000 + .long 0xbdebfe8a + .long 0xbf49d112 + .long 0xb2992640 + .long 0xbf000000 + .long 0xbdc3fdff + .long 0xbf4d9f02 + .long 0xb27e70e8 + .long 0xbf000000 + .long 0xbd9b4153 + .long 0xbf514d3d + .long 0xb00c4f04 + .long 0xbf000000 + .long 0xbd639d9d + .long 0xbf54db31 + .long 0xb290ea1a + .long 0xbf000000 + .long 0xbd0f59aa + .long 0xbf584853 + .long 0x327d5fc0 + .long 0xbf000000 + .long 0xbc670f32 + .long 0xbf5b941a + .long 0xb2232dc8 + .long 0xbf000000 + .long 0x3be8b648 + .long 0xbf5ebe05 + .long 0xb2c6f953 + .long 0xbf000000 + .long 0x3cea5164 + .long 0xbf61c598 + .long 0x32e7f425 + .long 0xbf000000 + .long 0x3d4e645a + .long 0xbf64aa59 + .long 0xb11a08fa + .long 0xbf000000 + .long 0x3d945dff + .long 0xbf676bd8 + .long 0x32bc3389 + .long 0xbf000000 + .long 0x3dc210d8 + .long 0xbf6a09a7 + .long 0x32eb236c + .long 0xbf000000 + .long 0x3df043ab + .long 0xbf6c835e + .long 0xb2f328d4 + .long 0xbf000000 + .long 0x3e0f77ad + .long 0xbf6ed89e + .long 0x329333dc + .long 0xbf000000 + .long 0xbdb1f34f + .long 0xbf710908 + .long 0xb21ed0dd + .long 0xbe800000 + .long 0xbd826b93 + .long 0xbf731447 + .long 0xb2c48e11 + .long 0xbe800000 + .long 0xbd25018c + .long 0xbf74fa0b + .long 0x32939d22 + .long 0xbe800000 + .long 0xbc88e931 + .long 0xbf76ba07 + .long 0xb26d092c + .long 0xbe800000 + .long 0x3be60685 + .long 0xbf7853f8 + .long 0x320db9e5 + .long 0xbe800000 + .long 0x3cfd1f65 + .long 0xbf79c79d + .long 0xb2c64e59 + .long 0xbe800000 + .long 0x3d60e8f8 + .long 0xbf7b14be + .long 0xb2ff75cb + .long 0xbe800000 + .long 0xbd3c4289 + .long 0xbf7c3b28 + .long 0x3231d68b + .long 0xbe000000 + .long 0xbcb2041c + .long 0xbf7d3aac + .long 0x30f75ae9 + .long 0xbe000000 + .long 0x3b29b1a9 + .long 0xbf7e1324 + .long 0x32f1e603 + .long 0xbe000000 + .long 0x3cdd0b28 + .long 0xbf7ec46d + .long 0xb1f44949 + .long 0xbe000000 + .long 0xbc354825 + .long 0xbf7f4e6d + .long 0xb2d01884 + .long 0xbd800000 + .long 0x3c5c1342 + .long 0xbf7fb10f + .long 0xb1de5b5f + .long 0xbd800000 + .long 0x3bdbd541 + .long 0xbf7fec43 + .long 0xb084cd0d + .long 0xbd000000 + .long 0x00000000 + .long 0xbf800000 + .long 0x00000000 + .long 0x00000000 + .long 0xbbdbd541 + .long 0xbf7fec43 + .long 0xb084cd0d + .long 0x3d000000 + .long 0xbc5c1342 + .long 0xbf7fb10f + .long 0xb1de5b5f + .long 0x3d800000 + .long 0x3c354825 + .long 0xbf7f4e6d + .long 0xb2d01884 + .long 0x3d800000 + .long 0xbcdd0b28 + .long 0xbf7ec46d + .long 0xb1f44949 + .long 0x3e000000 + .long 0xbb29b1a9 + .long 0xbf7e1324 + .long 0x32f1e603 + .long 0x3e000000 + .long 0x3cb2041c + .long 0xbf7d3aac + .long 0x30f75ae9 + .long 0x3e000000 + .long 0x3d3c4289 + .long 0xbf7c3b28 + .long 0x3231d68b + .long 0x3e000000 + .long 0xbd60e8f8 + .long 0xbf7b14be + .long 0xb2ff75cb + .long 0x3e800000 + .long 0xbcfd1f65 + .long 0xbf79c79d + .long 0xb2c64e59 + .long 0x3e800000 + .long 0xbbe60685 + .long 0xbf7853f8 + .long 0x320db9e5 + .long 0x3e800000 + .long 0x3c88e931 + .long 0xbf76ba07 + .long 0xb26d092c + .long 0x3e800000 + .long 0x3d25018c + .long 0xbf74fa0b + .long 0x32939d22 + .long 0x3e800000 + .long 0x3d826b93 + .long 0xbf731447 + .long 0xb2c48e11 + .long 0x3e800000 + .long 0x3db1f34f + .long 0xbf710908 + .long 0xb21ed0dd + .long 0x3e800000 + .long 0xbe0f77ad + .long 0xbf6ed89e + .long 0x329333dc + .long 0x3f000000 + .long 0xbdf043ab + .long 0xbf6c835e + .long 0xb2f328d4 + .long 0x3f000000 + .long 0xbdc210d8 + .long 0xbf6a09a7 + .long 0x32eb236c + .long 0x3f000000 + .long 0xbd945dff + .long 0xbf676bd8 + .long 0x32bc3389 + .long 0x3f000000 + .long 0xbd4e645a + .long 0xbf64aa59 + .long 0xb11a08fa + .long 0x3f000000 + .long 0xbcea5164 + .long 0xbf61c598 + .long 0x32e7f425 + .long 0x3f000000 + .long 0xbbe8b648 + .long 0xbf5ebe05 + .long 0xb2c6f953 + .long 0x3f000000 + .long 0x3c670f32 + .long 0xbf5b941a + .long 0xb2232dc8 + .long 0x3f000000 + .long 0x3d0f59aa + .long 0xbf584853 + .long 0x327d5fc0 + .long 0x3f000000 + .long 0x3d639d9d + .long 0xbf54db31 + .long 0xb290ea1a + .long 0x3f000000 + .long 0x3d9b4153 + .long 0xbf514d3d + .long 0xb00c4f04 + .long 0x3f000000 + .long 0x3dc3fdff + .long 0xbf4d9f02 + .long 0xb27e70e8 + .long 0x3f000000 + .long 0x3debfe8a + .long 0xbf49d112 + .long 0xb2992640 + .long 0x3f000000 + .long 0x3e099e65 + .long 0xbf45e403 + .long 0xb2b15174 + .long 0x3f000000 + .long 0x3e1cd957 + .long 0xbf41d870 + .long 0xb2bff977 + .long 0x3f000000 + .long 0x3e2fad27 + .long 0xbf3daef9 + .long 0xb19aabec + .long 0x3f000000 + .long 0x3e4216eb + .long 0xbf396842 + .long 0x32810007 + .long 0x3f000000 + .long 0xbe95f61a + .long 0xbf3504f3 + .long 0xb24fe77a + .long 0x3f800000 + .long 0xbe8d2f7d + .long 0xbf3085bb + .long 0x32ae2d32 + .long 0x3f800000 + .long 0xbe84a20e + .long 0xbf2beb4a + .long 0x32b73136 + .long 0x3f800000 + .long 0xbe789e3f + .long 0xbf273656 + .long 0x32038343 + .long 0x3f800000 + .long 0xbe686ff3 + .long 0xbf226799 + .long 0xb22123bb + .long 0x3f800000 + .long 0xbe58bbb7 + .long 0xbf1d7fd1 + .long 0xb292050c + .long 0x3f800000 + .long 0xbe4983f7 + .long 0xbf187fc0 + .long 0x31c7a3f3 + .long 0x3f800000 + .long 0xbe3acb0c + .long 0xbf13682a + .long 0xb2cdd12e + .long 0x3f800000 + .long 0xbe2c933b + .long 0xbf0e39da + .long 0x324a32e7 + .long 0x3f800000 + .long 0xbe1edeb5 + .long 0xbf08f59b + .long 0x32be4b4e + .long 0x3f800000 + .long 0xbe11af97 + .long 0xbf039c3d + .long 0x325ba002 + .long 0x3f800000 + .long 0xbe0507ea + .long 0xbefc5d27 + .long 0x3180eca9 + .long 0x3f800000 + .long 0xbdf1d344 + .long 0xbef15aea + .long 0x31ff2139 + .long 0x3f800000 + .long 0xbddaad38 + .long 0xbee63375 + .long 0x31d9c774 + .long 0x3f800000 + .long 0xbdc4a143 + .long 0xbedae880 + .long 0xb21e15cc + .long 0x3f800000 + .long 0xbdafb2cc + .long 0xbecf7bca + .long 0xb16a3b63 + .long 0x3f800000 + .long 0xbd9be50c + .long 0xbec3ef15 + .long 0xb1d5d52c + .long 0x3f800000 + .long 0xbd893b12 + .long 0xbeb8442a + .long 0x32705ba6 + .long 0x3f800000 + .long 0xbd6f6f7e + .long 0xbeac7cd4 + .long 0x32254e02 + .long 0x3f800000 + .long 0xbd4ebb8a + .long 0xbea09ae5 + .long 0x323e89a0 + .long 0x3f800000 + .long 0xbd305f55 + .long 0xbe94a031 + .long 0xb26d59f0 + .long 0x3f800000 + .long 0xbd145f8c + .long 0xbe888e93 + .long 0xb12c7d9e + .long 0x3f800000 + .long 0xbcf58104 + .long 0xbe78cfcc + .long 0x311bd41d + .long 0x3f800000 + .long 0xbcc70c54 + .long 0xbe605c13 + .long 0xb1a7e4f6 + .long 0x3f800000 + .long 0xbc9d6830 + .long 0xbe47c5c2 + .long 0x30e5967d + .long 0x3f800000 + .long 0xbc71360b + .long 0xbe2f10a2 + .long 0xb11167f9 + .long 0x3f800000 + .long 0xbc315502 + .long 0xbe164083 + .long 0xb1e8e614 + .long 0x3f800000 + .long 0xbbf66e3c + .long 0xbdfab273 + .long 0x311568cf + .long 0x3f800000 + .long 0xbb9dc971 + .long 0xbdc8bd36 + .long 0x307592f5 + .long 0x3f800000 + .long 0xbb319298 + .long 0xbd96a905 + .long 0x31531e61 + .long 0x3f800000 + .long 0xba9de1c8 + .long 0xbd48fb30 + .long 0x30ef227f + .long 0x3f800000 + .long 0xb99de7df + .long 0xbcc90ab0 + .long 0x3005c998 + .long 0x3f800000 + .long 0x00000000 + .long 0x00000000 + .long 0x00000000 + .long 0x3f800000 + .long 0xb99de7df + .long 0x3cc90ab0 + .long 0xb005c998 + .long 0x3f800000 + .long 0xba9de1c8 + .long 0x3d48fb30 + .long 0xb0ef227f + .long 0x3f800000 + .long 0xbb319298 + .long 0x3d96a905 + .long 0xb1531e61 + .long 0x3f800000 + .long 0xbb9dc971 + .long 0x3dc8bd36 + .long 0xb07592f5 + .long 0x3f800000 + .long 0xbbf66e3c + .long 0x3dfab273 + .long 0xb11568cf + .long 0x3f800000 + .long 0xbc315502 + .long 0x3e164083 + .long 0x31e8e614 + .long 0x3f800000 + .long 0xbc71360b + .long 0x3e2f10a2 + .long 0x311167f9 + .long 0x3f800000 + .long 0xbc9d6830 + .long 0x3e47c5c2 + .long 0xb0e5967d + .long 0x3f800000 + .long 0xbcc70c54 + .long 0x3e605c13 + .long 0x31a7e4f6 + .long 0x3f800000 + .long 0xbcf58104 + .long 0x3e78cfcc + .long 0xb11bd41d + .long 0x3f800000 + .long 0xbd145f8c + .long 0x3e888e93 + .long 0x312c7d9e + .long 0x3f800000 + .long 0xbd305f55 + .long 0x3e94a031 + .long 0x326d59f0 + .long 0x3f800000 + .long 0xbd4ebb8a + .long 0x3ea09ae5 + .long 0xb23e89a0 + .long 0x3f800000 + .long 0xbd6f6f7e + .long 0x3eac7cd4 + .long 0xb2254e02 + .long 0x3f800000 + .long 0xbd893b12 + .long 0x3eb8442a + .long 0xb2705ba6 + .long 0x3f800000 + .long 0xbd9be50c + .long 0x3ec3ef15 + .long 0x31d5d52c + .long 0x3f800000 + .long 0xbdafb2cc + .long 0x3ecf7bca + .long 0x316a3b63 + .long 0x3f800000 + .long 0xbdc4a143 + .long 0x3edae880 + .long 0x321e15cc + .long 0x3f800000 + .long 0xbddaad38 + .long 0x3ee63375 + .long 0xb1d9c774 + .long 0x3f800000 + .long 0xbdf1d344 + .long 0x3ef15aea + .long 0xb1ff2139 + .long 0x3f800000 + .long 0xbe0507ea + .long 0x3efc5d27 + .long 0xb180eca9 + .long 0x3f800000 + .long 0xbe11af97 + .long 0x3f039c3d + .long 0xb25ba002 + .long 0x3f800000 + .long 0xbe1edeb5 + .long 0x3f08f59b + .long 0xb2be4b4e + .long 0x3f800000 + .long 0xbe2c933b + .long 0x3f0e39da + .long 0xb24a32e7 + .long 0x3f800000 + .long 0xbe3acb0c + .long 0x3f13682a + .long 0x32cdd12e + .long 0x3f800000 + .long 0xbe4983f7 + .long 0x3f187fc0 + .long 0xb1c7a3f3 + .long 0x3f800000 + .long 0xbe58bbb7 + .long 0x3f1d7fd1 + .long 0x3292050c + .long 0x3f800000 + .long 0xbe686ff3 + .long 0x3f226799 + .long 0x322123bb + .long 0x3f800000 + .long 0xbe789e3f + .long 0x3f273656 + .long 0xb2038343 + .long 0x3f800000 + .long 0xbe84a20e + .long 0x3f2beb4a + .long 0xb2b73136 + .long 0x3f800000 + .long 0xbe8d2f7d + .long 0x3f3085bb + .long 0xb2ae2d32 + .long 0x3f800000 + .long 0xbe95f61a + .long 0x3f3504f3 + .long 0x324fe77a + .long 0x3f800000 + .long 0x3e4216eb + .long 0x3f396842 + .long 0xb2810007 + .long 0x3f000000 + .long 0x3e2fad27 + .long 0x3f3daef9 + .long 0x319aabec + .long 0x3f000000 + .long 0x3e1cd957 + .long 0x3f41d870 + .long 0x32bff977 + .long 0x3f000000 + .long 0x3e099e65 + .long 0x3f45e403 + .long 0x32b15174 + .long 0x3f000000 + .long 0x3debfe8a + .long 0x3f49d112 + .long 0x32992640 + .long 0x3f000000 + .long 0x3dc3fdff + .long 0x3f4d9f02 + .long 0x327e70e8 + .long 0x3f000000 + .long 0x3d9b4153 + .long 0x3f514d3d + .long 0x300c4f04 + .long 0x3f000000 + .long 0x3d639d9d + .long 0x3f54db31 + .long 0x3290ea1a + .long 0x3f000000 + .long 0x3d0f59aa + .long 0x3f584853 + .long 0xb27d5fc0 + .long 0x3f000000 + .long 0x3c670f32 + .long 0x3f5b941a + .long 0x32232dc8 + .long 0x3f000000 + .long 0xbbe8b648 + .long 0x3f5ebe05 + .long 0x32c6f953 + .long 0x3f000000 + .long 0xbcea5164 + .long 0x3f61c598 + .long 0xb2e7f425 + .long 0x3f000000 + .long 0xbd4e645a + .long 0x3f64aa59 + .long 0x311a08fa + .long 0x3f000000 + .long 0xbd945dff + .long 0x3f676bd8 + .long 0xb2bc3389 + .long 0x3f000000 + .long 0xbdc210d8 + .long 0x3f6a09a7 + .long 0xb2eb236c + .long 0x3f000000 + .long 0xbdf043ab + .long 0x3f6c835e + .long 0x32f328d4 + .long 0x3f000000 + .long 0xbe0f77ad + .long 0x3f6ed89e + .long 0xb29333dc + .long 0x3f000000 + .long 0x3db1f34f + .long 0x3f710908 + .long 0x321ed0dd + .long 0x3e800000 + .long 0x3d826b93 + .long 0x3f731447 + .long 0x32c48e11 + .long 0x3e800000 + .long 0x3d25018c + .long 0x3f74fa0b + .long 0xb2939d22 + .long 0x3e800000 + .long 0x3c88e931 + .long 0x3f76ba07 + .long 0x326d092c + .long 0x3e800000 + .long 0xbbe60685 + .long 0x3f7853f8 + .long 0xb20db9e5 + .long 0x3e800000 + .long 0xbcfd1f65 + .long 0x3f79c79d + .long 0x32c64e59 + .long 0x3e800000 + .long 0xbd60e8f8 + .long 0x3f7b14be + .long 0x32ff75cb + .long 0x3e800000 + .long 0x3d3c4289 + .long 0x3f7c3b28 + .long 0xb231d68b + .long 0x3e000000 + .long 0x3cb2041c + .long 0x3f7d3aac + .long 0xb0f75ae9 + .long 0x3e000000 + .long 0xbb29b1a9 + .long 0x3f7e1324 + .long 0xb2f1e603 + .long 0x3e000000 + .long 0xbcdd0b28 + .long 0x3f7ec46d + .long 0x31f44949 + .long 0x3e000000 + .long 0x3c354825 + .long 0x3f7f4e6d + .long 0x32d01884 + .long 0x3d800000 + .long 0xbc5c1342 + .long 0x3f7fb10f + .long 0x31de5b5f + .long 0x3d800000 + .long 0xbbdbd541 + .long 0x3f7fec43 + .long 0x3084cd0d + .long 0x3d000000 + +/* General purpose constants: + absolute value mask */ +float_vector __sAbsMask 0x7fffffff + +/* threshold for out-of-range values */ +float_vector __sRangeReductionVal 0x461c4000 + +/* +INF */ +float_vector __sRangeVal 0x7f800000 + +/* High Accuracy version polynomial coefficients: + S1 = -1.66666666664728165763e-01 */ +float_vector __sS1 0xbe2aaaab + +/* S2 = 8.33329173045453069014e-03 */ +float_vector __sS2 0x3c08885c + +/* C1 = -5.00000000000000000000e-01 */ +float_vector __sC1 0xbf000000 + +/* C2 = 4.16638942914469202550e-02 */ +float_vector __sC2 0x3d2aaa7c + +/* Range reduction PI-based constants: + PI high part */ +float_vector __sPI1 0x40490000 + +/* PI mid part 1 */ +float_vector __sPI2 0x3a7da000 + +/* PI mid part 2 */ +float_vector __sPI3 0x34222000 + +/* PI low part */ +float_vector __sPI4 0x2cb4611a + +/* PI1, PI2, and PI3 when FMA is available + PI high part (when FMA available) */ +float_vector __sPI1_FMA 0x40490fdb + +/* PI mid part (when FMA available) */ +float_vector __sPI2_FMA 0xb3bbbd2e + +/* PI low part (when FMA available) */ +float_vector __sPI3_FMA 0xa7772ced + +/* Polynomial constants for work w/o FMA, relative error ~ 2^(-26.625) */ +float_vector __sA3 0xbe2aaaa6 +float_vector __sA5 0x3c08876a +float_vector __sA7 0xb94fb7ff +float_vector __sA9 0x362edef8 + +/* Polynomial constants, work with FMA, relative error ~ 2^(-26.417) */ +float_vector __sA5_FMA 0x3c088768 +float_vector __sA7_FMA 0xb94fb6cf +float_vector __sA9_FMA 0x362ec335 + +/* 1/PI */ +float_vector __sInvPI 0x3ea2f983 + +/* right-shifter constant */ +float_vector __sRShifter 0x4b400000 + +/* PI/2 */ +float_vector __sHalfPI 0x3fc90fdb + +/* 1/2 */ +float_vector __sOneHalf 0x3f000000 + .type __svml_scos_data,@object + .size __svml_scos_data,.-__svml_scos_data diff --git a/sysdeps/x86_64/fpu/svml_s_cosf_data.h b/sysdeps/x86_64/fpu/svml_s_cosf_data.h new file mode 100644 index 0000000..1e25c5a --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_s_cosf_data.h @@ -0,0 +1,58 @@ +/* Offsets for data table for vectorized cosf. + Copyright (C) 2014-2015 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 + . */ + +#ifndef S_COSF_DATA_H +#define S_COSF_DATA_H + +.macro float_vector offset value +.if .-__svml_scos_data != \offset +.err +.endif +.rept 16 +.long \value +.endr +.endm + +#define __dT 0 +#define __sAbsMask 4096 +#define __sRangeReductionVal 4160 +#define __sRangeVal 4224 +#define __sS1 4288 +#define __sS2 4352 +#define __sC1 4416 +#define __sC2 4480 +#define __sPI1 4544 +#define __sPI2 4608 +#define __sPI3 4672 +#define __sPI4 4736 +#define __sPI1_FMA 4800 +#define __sPI2_FMA 4864 +#define __sPI3_FMA 4928 +#define __sA3 4992 +#define __sA5 5056 +#define __sA7 5120 +#define __sA9 5184 +#define __sA5_FMA 5248 +#define __sA7_FMA 5312 +#define __sA9_FMA 5376 +#define __sInvPI 5440 +#define __sRShifter 5504 +#define __sHalfPI 5568 +#define __sOneHalf 5632 + +#endif diff --git a/sysdeps/x86_64/fpu/svml_s_wrapper_impl.h b/sysdeps/x86_64/fpu/svml_s_wrapper_impl.h new file mode 100644 index 0000000..d5b62ee --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_s_wrapper_impl.h @@ -0,0 +1,111 @@ +/* Wrapper implementations of vector math functions. + Copyright (C) 2014-2015 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 + . */ + +/* SSE2 ISA version as wrapper to scalar. */ +.macro WRAPPER_IMPL_SSE2 callee + subq $40, %rsp + cfi_adjust_cfa_offset(40) + movaps %xmm0, (%rsp) + call \callee@PLT + movss %xmm0, 16(%rsp) + movss 4(%rsp), %xmm0 + call \callee@PLT + movss %xmm0, 20(%rsp) + movss 8(%rsp), %xmm0 + call \callee@PLT + movss %xmm0, 24(%rsp) + movss 12(%rsp), %xmm0 + call \callee@PLT + movss 16(%rsp), %xmm3 + movss 20(%rsp), %xmm2 + movss 24(%rsp), %xmm1 + movss %xmm0, 28(%rsp) + unpcklps %xmm1, %xmm3 + unpcklps %xmm0, %xmm2 + unpcklps %xmm2, %xmm3 + movaps %xmm3, %xmm0 + addq $40, %rsp + cfi_adjust_cfa_offset(-40) + ret +.endm + +/* AVX/AVX2 ISA version as wrapper to SSE ISA version. */ +.macro WRAPPER_IMPL_AVX callee + pushq %rbp + cfi_adjust_cfa_offset (8) + cfi_rel_offset (%rbp, 0) + movq %rsp, %rbp + cfi_def_cfa_register (%rbp) + andq $-32, %rsp + subq $32, %rsp + vextractf128 $1, %ymm0, (%rsp) + vzeroupper + call HIDDEN_JUMPTARGET(\callee) + vmovaps %xmm0, 16(%rsp) + vmovaps (%rsp), %xmm0 + call HIDDEN_JUMPTARGET(\callee) + vmovaps %xmm0, %xmm1 + vmovaps 16(%rsp), %xmm0 + vinsertf128 $1, %xmm1, %ymm0, %ymm0 + movq %rbp, %rsp + cfi_def_cfa_register (%rsp) + popq %rbp + cfi_adjust_cfa_offset (-8) + cfi_restore (%rbp) + ret +.endm + +/* AVX512 ISA version as wrapper to AVX2 ISA version. */ +.macro WRAPPER_IMPL_AVX512 callee + pushq %rbp + cfi_adjust_cfa_offset (8) + cfi_rel_offset (%rbp, 0) + movq %rsp, %rbp + cfi_def_cfa_register (%rbp) + andq $-64, %rsp + subq $64, %rsp +/* Below is encoding for vmovaps %zmm0, (%rsp). */ + .byte 0x62 + .byte 0xf1 + .byte 0x7c + .byte 0x48 + .byte 0x29 + .byte 0x04 + .byte 0x24 +/* Below is encoding for vmovaps (%rsp), %ymm0. */ + .byte 0xc5 + .byte 0xfc + .byte 0x28 + .byte 0x04 + .byte 0x24 + call HIDDEN_JUMPTARGET(\callee) +/* Below is encoding for vmovaps 32(%rsp), %ymm0. */ + .byte 0xc5 + .byte 0xfc + .byte 0x28 + .byte 0x44 + .byte 0x24 + .byte 0x20 + call HIDDEN_JUMPTARGET(\callee) + movq %rbp, %rsp + cfi_def_cfa_register (%rsp) + popq %rbp + cfi_adjust_cfa_offset (-8) + cfi_restore (%rbp) + ret +.endm -- 2.7.4