diff options
| author | Sunil K Pandey <skpgkp2@gmail.com> | 2021-12-29 08:23:33 -0800 |
|---|---|---|
| committer | Sunil K Pandey <skpgkp2@gmail.com> | 2021-12-29 11:36:46 -0800 |
| commit | 146310177aa9f2c7d990ef856ed6e8bb94407f06 (patch) | |
| tree | 1f81aa90d5cf28eab3442be8f3889aa5ff6a23d6 | |
| parent | 5d28a8962dcb6ec056b81d730e3c6fb57185a210 (diff) | |
| download | glibc-146310177aa9f2c7d990ef856ed6e8bb94407f06.zip glibc-146310177aa9f2c7d990ef856ed6e8bb94407f06.tar.gz glibc-146310177aa9f2c7d990ef856ed6e8bb94407f06.tar.bz2 | |
x86-64: Add vector atan/atanf implementation to libmvec
Implement vectorized atan/atanf containing SSE, AVX, AVX2 and
AVX512 versions for libmvec as per vector ABI. It also contains
accuracy and ABI tests for vector atan/atanf with regenerated ulps.
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
50 files changed, 1741 insertions, 1 deletions
diff --git a/bits/libm-simd-decl-stubs.h b/bits/libm-simd-decl-stubs.h index 2ccdd1f..b4647ca 100644 --- a/bits/libm-simd-decl-stubs.h +++ b/bits/libm-simd-decl-stubs.h @@ -109,4 +109,15 @@ #define __DECL_SIMD_acosf32x #define __DECL_SIMD_acosf64x #define __DECL_SIMD_acosf128x + +#define __DECL_SIMD_atan +#define __DECL_SIMD_atanf +#define __DECL_SIMD_atanl +#define __DECL_SIMD_atanf16 +#define __DECL_SIMD_atanf32 +#define __DECL_SIMD_atanf64 +#define __DECL_SIMD_atanf128 +#define __DECL_SIMD_atanf32x +#define __DECL_SIMD_atanf64x +#define __DECL_SIMD_atanf128x #endif diff --git a/math/bits/mathcalls.h b/math/bits/mathcalls.h index 2cc6654..3e27c21 100644 --- a/math/bits/mathcalls.h +++ b/math/bits/mathcalls.h @@ -54,7 +54,7 @@ __MATHCALL_VEC (acos,, (_Mdouble_ __x)); /* Arc sine of X. */ __MATHCALL (asin,, (_Mdouble_ __x)); /* Arc tangent of X. */ -__MATHCALL (atan,, (_Mdouble_ __x)); +__MATHCALL_VEC (atan,, (_Mdouble_ __x)); /* Arc tangent of Y/X. */ __MATHCALL (atan2,, (_Mdouble_ __y, _Mdouble_ __x)); diff --git a/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist b/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist index b37b557..a93258d 100644 --- a/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist +++ b/sysdeps/unix/sysv/linux/x86_64/libmvec.abilist @@ -47,10 +47,18 @@ GLIBC_2.22 _ZGVeN8v_sin F GLIBC_2.22 _ZGVeN8vv_pow F GLIBC_2.22 _ZGVeN8vvv_sincos F GLIBC_2.35 _ZGVbN2v_acos F +GLIBC_2.35 _ZGVbN2v_atan F GLIBC_2.35 _ZGVbN4v_acosf F +GLIBC_2.35 _ZGVbN4v_atanf F GLIBC_2.35 _ZGVcN4v_acos F +GLIBC_2.35 _ZGVcN4v_atan F GLIBC_2.35 _ZGVcN8v_acosf F +GLIBC_2.35 _ZGVcN8v_atanf F GLIBC_2.35 _ZGVdN4v_acos F +GLIBC_2.35 _ZGVdN4v_atan F GLIBC_2.35 _ZGVdN8v_acosf F +GLIBC_2.35 _ZGVdN8v_atanf F GLIBC_2.35 _ZGVeN16v_acosf F +GLIBC_2.35 _ZGVeN16v_atanf F GLIBC_2.35 _ZGVeN8v_acos F +GLIBC_2.35 _ZGVeN8v_atan F diff --git a/sysdeps/x86/fpu/bits/math-vector.h b/sysdeps/x86/fpu/bits/math-vector.h index dabb74c..1c0e5c5 100644 --- a/sysdeps/x86/fpu/bits/math-vector.h +++ b/sysdeps/x86/fpu/bits/math-vector.h @@ -62,6 +62,10 @@ # define __DECL_SIMD_acos __DECL_SIMD_x86_64 # undef __DECL_SIMD_acosf # define __DECL_SIMD_acosf __DECL_SIMD_x86_64 +# undef __DECL_SIMD_atan +# define __DECL_SIMD_atan __DECL_SIMD_x86_64 +# undef __DECL_SIMD_atanf +# define __DECL_SIMD_atanf __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 4bcbd1f..ddcccb1 100644 --- a/sysdeps/x86/fpu/finclude/math-vector-fortran.h +++ b/sysdeps/x86/fpu/finclude/math-vector-fortran.h @@ -30,6 +30,8 @@ !GCC$ builtin (powf) attributes simd (notinbranch) if('x86_64') !GCC$ builtin (acos) attributes simd (notinbranch) if('x86_64') !GCC$ builtin (acosf) attributes simd (notinbranch) if('x86_64') +!GCC$ builtin (atan) attributes simd (notinbranch) if('x86_64') +!GCC$ builtin (atanf) attributes simd (notinbranch) if('x86_64') !GCC$ builtin (cos) attributes simd (notinbranch) if('x32') !GCC$ builtin (cosf) attributes simd (notinbranch) if('x32') @@ -45,3 +47,5 @@ !GCC$ builtin (powf) attributes simd (notinbranch) if('x32') !GCC$ builtin (acos) attributes simd (notinbranch) if('x32') !GCC$ builtin (acosf) attributes simd (notinbranch) if('x32') +!GCC$ builtin (atan) attributes simd (notinbranch) if('x32') +!GCC$ builtin (atanf) attributes simd (notinbranch) if('x32') diff --git a/sysdeps/x86_64/fpu/Makeconfig b/sysdeps/x86_64/fpu/Makeconfig index 7acf1f3..dae0887 100644 --- a/sysdeps/x86_64/fpu/Makeconfig +++ b/sysdeps/x86_64/fpu/Makeconfig @@ -23,6 +23,7 @@ postclean-generated += libmvec.mk # Define for both math and mathvec directories. libmvec-funcs = \ acos \ + atan \ cos \ exp \ log \ diff --git a/sysdeps/x86_64/fpu/Versions b/sysdeps/x86_64/fpu/Versions index 2985fe7..424f6d5 100644 --- a/sysdeps/x86_64/fpu/Versions +++ b/sysdeps/x86_64/fpu/Versions @@ -15,6 +15,8 @@ libmvec { } GLIBC_2.35 { _ZGVbN2v_acos; _ZGVcN4v_acos; _ZGVdN4v_acos; _ZGVeN8v_acos; + _ZGVbN2v_atan; _ZGVcN4v_atan; _ZGVdN4v_atan; _ZGVeN8v_atan; _ZGVbN4v_acosf; _ZGVcN8v_acosf; _ZGVdN8v_acosf; _ZGVeN16v_acosf; + _ZGVbN4v_atanf; _ZGVcN8v_atanf; _ZGVdN8v_atanf; _ZGVeN16v_atanf; } } diff --git a/sysdeps/x86_64/fpu/libm-test-ulps b/sysdeps/x86_64/fpu/libm-test-ulps index 6c12976..2e64e59 100644 --- a/sysdeps/x86_64/fpu/libm-test-ulps +++ b/sysdeps/x86_64/fpu/libm-test-ulps @@ -164,6 +164,26 @@ float: 2 float128: 2 ldouble: 1 +Function: "atan_vlen16": +float: 1 + +Function: "atan_vlen2": +double: 1 + +Function: "atan_vlen4": +double: 1 +float: 1 + +Function: "atan_vlen4_avx2": +double: 1 + +Function: "atan_vlen8": +double: 1 +float: 1 + +Function: "atan_vlen8_avx2": +float: 1 + Function: "atanh": double: 2 float: 2 diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan2_core-sse2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_atan2_core-sse2.S new file mode 100644 index 0000000..115e522 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan2_core-sse2.S @@ -0,0 +1,20 @@ +/* SSE2 version of vectorized atan, 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 + <https://www.gnu.org/licenses/>. */ + +#define _ZGVbN2v_atan _ZGVbN2v_atan_sse2 +#include "../svml_d_atan2_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan2_core.c b/sysdeps/x86_64/fpu/multiarch/svml_d_atan2_core.c new file mode 100644 index 0000000..93f079f --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan2_core.c @@ -0,0 +1,27 @@ +/* Multiple versions of vectorized atan, 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 + <https://www.gnu.org/licenses/>. */ + +#define SYMBOL_NAME _ZGVbN2v_atan +#include "ifunc-mathvec-sse4_1.h" + +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); + +#ifdef SHARED +__hidden_ver1 (_ZGVbN2v_atan, __GI__ZGVbN2v_atan, __redirect__ZGVbN2v_atan) + __attribute__ ((visibility ("hidden"))); +#endif diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan2_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_d_atan2_core_sse4.S new file mode 100644 index 0000000..f0ad036 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan2_core_sse4.S @@ -0,0 +1,245 @@ +/* Function atan 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: + * + * For 0.0 <= x <= 7.0/16.0: atan(x) = atan(0.0) + atan(s), where s=(x-0.0)/(1.0+0.0*x) + * For 7.0/16.0 <= x <= 11.0/16.0: atan(x) = atan(0.5) + atan(s), where s=(x-0.5)/(1.0+0.5*x) + * For 11.0/16.0 <= x <= 19.0/16.0: atan(x) = atan(1.0) + atan(s), where s=(x-1.0)/(1.0+1.0*x) + * For 19.0/16.0 <= x <= 39.0/16.0: atan(x) = atan(1.5) + atan(s), where s=(x-1.5)/(1.0+1.5*x) + * For 39.0/16.0 <= x <= inf : atan(x) = atan(inf) + atan(s), where s=-1.0/x + * Where atan(s) ~= s+s^3*Poly11(s^2) on interval |s|<7.0/0.16. + * + */ + +/* Offsets for data table __svml_datan_data_internal_avx512 + */ +#define AbsMask 0 +#define Shifter 16 +#define MaxThreshold 32 +#define MOne 48 +#define One 64 +#define LargeX 80 +#define Zero 96 +#define Tbl_H 112 +#define Tbl_L 368 +#define dIndexMed 624 +#define Pi2 640 +#define Pi2_low 656 +#define coeff 672 + +#include <sysdep.h> + + .text + .section .text.sse4,"ax",@progbits +ENTRY(_ZGVbN2v_atan_sse4) + lea Tbl_H+128+__svml_datan_data_internal_avx512(%rip), %rcx + movups __svml_datan_data_internal_avx512(%rip), %xmm4 + movups Shifter+__svml_datan_data_internal_avx512(%rip), %xmm3 + andps %xmm0, %xmm4 + movaps %xmm3, %xmm12 + movaps %xmm4, %xmm5 + addpd %xmm4, %xmm12 + movaps %xmm12, %xmm7 + +/* + * table lookup sequence + * VPERMUTE not available + */ + movaps %xmm12, %xmm10 + subpd %xmm3, %xmm7 + subpd %xmm7, %xmm5 + mulpd %xmm4, %xmm7 + movups MaxThreshold+__svml_datan_data_internal_avx512(%rip), %xmm2 + psllq $3, %xmm10 + +/* saturate X range */ + movups LargeX+__svml_datan_data_internal_avx512(%rip), %xmm8 + pxor %xmm4, %xmm0 + cmplepd %xmm4, %xmm2 + addpd One+__svml_datan_data_internal_avx512(%rip), %xmm7 + minpd %xmm4, %xmm8 + movups MOne+__svml_datan_data_internal_avx512(%rip), %xmm6 + movaps %xmm2, %xmm1 + movaps %xmm2, %xmm9 + andnps %xmm5, %xmm1 + andps %xmm2, %xmm6 + andnps %xmm7, %xmm9 + andps %xmm2, %xmm8 + orps %xmm6, %xmm1 + orps %xmm8, %xmm9 + +/* R+Rl = DiffX/Y */ + divpd %xmm9, %xmm1 + pand .FLT_11(%rip), %xmm10 + +/* set table value to Pi/2 for large X */ + movups Pi2+__svml_datan_data_internal_avx512(%rip), %xmm4 + movd %xmm10, %eax + andps %xmm2, %xmm4 + pshufd $2, %xmm10, %xmm11 + movaps %xmm2, %xmm10 + +/* polynomial evaluation */ + movaps %xmm1, %xmm2 + mulpd %xmm1, %xmm2 + movd %xmm11, %edx + movups coeff+__svml_datan_data_internal_avx512(%rip), %xmm5 + movaps %xmm2, %xmm7 + movups coeff+32+__svml_datan_data_internal_avx512(%rip), %xmm6 + movaps %xmm2, %xmm9 + mulpd %xmm2, %xmm5 + mulpd %xmm2, %xmm7 + addpd coeff+16+__svml_datan_data_internal_avx512(%rip), %xmm5 + mulpd %xmm2, %xmm6 + mulpd %xmm7, %xmm5 + addpd coeff+48+__svml_datan_data_internal_avx512(%rip), %xmm6 + mulpd %xmm1, %xmm9 + addpd %xmm5, %xmm6 + movups coeff+64+__svml_datan_data_internal_avx512(%rip), %xmm8 + mulpd %xmm2, %xmm8 + mulpd %xmm6, %xmm7 + addpd coeff+80+__svml_datan_data_internal_avx512(%rip), %xmm8 + addpd %xmm7, %xmm8 + mulpd %xmm8, %xmm9 + movups dIndexMed+__svml_datan_data_internal_avx512(%rip), %xmm14 + cmplepd %xmm12, %xmm14 + addpd %xmm9, %xmm1 + movslq %eax, %rax + movaps %xmm14, %xmm3 + movslq %edx, %rdx + movsd -128(%rax,%rcx), %xmm13 + movsd (%rcx,%rax), %xmm15 + movhpd -128(%rdx,%rcx), %xmm13 + movhpd (%rcx,%rdx), %xmm15 + andnps %xmm13, %xmm3 + andps %xmm14, %xmm15 + orps %xmm15, %xmm3 + andnps %xmm3, %xmm10 + orps %xmm4, %xmm10 + addpd %xmm1, %xmm10 + pxor %xmm10, %xmm0 + ret + +END(_ZGVbN2v_atan_sse4) + + .section .rodata, "a" + .align 16 + +#ifdef __svml_datan_data_internal_avx512_typedef +typedef unsigned int VUINT32; +typedef struct { + __declspec(align(16)) VUINT32 AbsMask[2][2]; + __declspec(align(16)) VUINT32 Shifter[2][2]; + __declspec(align(16)) VUINT32 MaxThreshold[2][2]; + __declspec(align(16)) VUINT32 MOne[2][2]; + __declspec(align(16)) VUINT32 One[2][2]; + __declspec(align(16)) VUINT32 LargeX[2][2]; + __declspec(align(16)) VUINT32 Zero[2][2]; + __declspec(align(16)) VUINT32 Tbl_H[32][2]; + __declspec(align(16)) VUINT32 Tbl_L[32][2]; + __declspec(align(16)) VUINT32 dIndexMed[2][2]; + __declspec(align(16)) VUINT32 Pi2[2][2]; + __declspec(align(16)) VUINT32 Pi2_low[2][2]; + __declspec(align(16)) VUINT32 coeff[6][2][2]; + } __svml_datan_data_internal_avx512; +#endif +__svml_datan_data_internal_avx512: + /*== AbsMask ==*/ + .quad 0x7fffffffffffffff, 0x7fffffffffffffff + /*== Shifter ==*/ + .align 16 + .quad 0x4318000000000000, 0x4318000000000000 + /*== MaxThreshold ==*/ + .align 16 + .quad 0x401f800000000000, 0x401f800000000000 + /*== MOne ==*/ + .align 16 + .quad 0xbff0000000000000, 0xbff0000000000000 + /*== One ==*/ + .align 16 + .quad 0x3ff0000000000000, 0x3ff0000000000000 + /*== LargeX ==*/ + .align 16 + .quad 0x47f0000000000000, 0x47f0000000000000 + /*== Zero ==*/ + .align 16 + .quad 0x0000000000000000, 0x0000000000000000 + /*== Tbl_H ==*/ + .align 16 + .quad 0x0000000000000000, 0x3fcf5b75f92c80dd + .quad 0x3fddac670561bb4f, 0x3fe4978fa3269ee1 + .quad 0x3fe921fb54442d18, 0x3fecac7c57846f9e + .quad 0x3fef730bd281f69b, 0x3ff0d38f2c5ba09f + .quad 0x3ff1b6e192ebbe44, 0x3ff270ef55a53a25 + .quad 0x3ff30b6d796a4da8, 0x3ff38d6a6ce13353 + .quad 0x3ff3fc176b7a8560, 0x3ff45b54837351a0 + .quad 0x3ff4ae10fc6589a5, 0x3ff4f68dea672617 + .quad 0x3ff5368c951e9cfd, 0x3ff56f6f33a3e6a7 + .quad 0x3ff5a25052114e60, 0x3ff5d013c41adabd + .quad 0x3ff5f97315254857, 0x3ff61f06c6a92b89 + .quad 0x3ff6414d44094c7c, 0x3ff660b02c736a06 + .quad 0x3ff67d8863bc99bd, 0x3ff698213a9d5053 + .quad 0x3ff6b0bae830c070, 0x3ff6c78c7edeb195 + .quad 0x3ff6dcc57bb565fd, 0x3ff6f08f07435fec + .quad 0x3ff7030cf9403197, 0x3ff7145eac2088a4 + /*== Tbl_L ==*/ + .align 16 + .quad 0x0000000000000000, 0x3c68ab6e3cf7afbd + .quad 0x3c7a2b7f222f65e2, 0x3c72419a87f2a458 + .quad 0x3c81a62633145c07, 0x3c80dae13ad18a6b + .quad 0x3c7007887af0cbbd, 0xbc9bd0dc231bfd70 + .quad 0x3c9b1b466a88828e, 0xbc9a66b1af5f84fb + .quad 0x3c96254cb03bb199, 0xbc812c77e8a80f5c + .quad 0xbc4441a3bd3f1084, 0x3c79e4a72eedacc4 + .quad 0xbc93b03e8a27f555, 0x3c9934f9f2b0020e + .quad 0xbc996f47948a99f1, 0xbc7df6edd6f1ec3b + .quad 0x3c78c2d0c89de218, 0x3c9f82bba194dd5d + .quad 0xbc831151a43b51ca, 0xbc8487d50bceb1a5 + .quad 0xbc9c5f60a65c7397, 0xbc7acb6afb332a0f + .quad 0xbc99b7bd2e1e8c9c, 0xbc9b9839085189e3 + .quad 0xbc97d1ab82ffb70b, 0x3c99239ad620ffe2 + .quad 0xbc929c86447928e7, 0xbc8957a7170df016 + .quad 0xbc7cbe1896221608, 0xbc9fda5797b32a0b + /*== dIndexMed ==*/ + .align 16 + .quad 0x4318000000000010, 0x4318000000000010 + /*== Pi2 ==*/ + .align 16 + .quad 0x3ff921fb54442d18, 0x3ff921fb54442d18 + /*== Pi2_low ==*/ + .align 16 + .quad 0x3c91a62633145c07, 0x3c91a62633145c07 + /*== coeff6 ==*/ + .align 16 + .quad 0x3fb2e9b9f5c4fe97, 0x3fb2e9b9f5c4fe97 + .quad 0xbfb74257c46790cc, 0xbfb74257c46790cc + .quad 0x3fbc71bfeff916a0, 0x3fbc71bfeff916a0 + .quad 0xbfc249248eef04da, 0xbfc249248eef04da + .quad 0x3fc999999998741e, 0x3fc999999998741e + .quad 0xbfd555555555554d, 0xbfd555555555554d + .align 16 + .type __svml_datan_data_internal_avx512,@object + .size __svml_datan_data_internal_avx512,.-__svml_datan_data_internal_avx512 + .align 16 + +.FLT_11: + .long 0x00000078,0x00000000,0x00000078,0x00000000 + .type .FLT_11,@object + .size .FLT_11,16 diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan4_core-sse.S b/sysdeps/x86_64/fpu/multiarch/svml_d_atan4_core-sse.S new file mode 100644 index 0000000..79c48db --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan4_core-sse.S @@ -0,0 +1,20 @@ +/* SSE version of vectorized atan, 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 + <https://www.gnu.org/licenses/>. */ + +#define _ZGVdN4v_atan _ZGVdN4v_atan_sse_wrapper +#include "../svml_d_atan4_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan4_core.c b/sysdeps/x86_64/fpu/multiarch/svml_d_atan4_core.c new file mode 100644 index 0000000..64ce66b --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan4_core.c @@ -0,0 +1,27 @@ +/* Multiple versions of vectorized atan, 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 + <https://www.gnu.org/licenses/>. */ + +#define SYMBOL_NAME _ZGVdN4v_atan +#include "ifunc-mathvec-avx2.h" + +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); + +#ifdef SHARED +__hidden_ver1 (_ZGVdN4v_atan, __GI__ZGVdN4v_atan, __redirect__ZGVdN4v_atan) + __attribute__ ((visibility ("hidden"))); +#endif diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan4_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_atan4_core_avx2.S new file mode 100644 index 0000000..5033651 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan4_core_avx2.S @@ -0,0 +1,225 @@ +/* Function atan 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: + * + * For 0.0 <= x <= 7.0/16.0: atan(x) = atan(0.0) + atan(s), where s=(x-0.0)/(1.0+0.0*x) + * For 7.0/16.0 <= x <= 11.0/16.0: atan(x) = atan(0.5) + atan(s), where s=(x-0.5)/(1.0+0.5*x) + * For 11.0/16.0 <= x <= 19.0/16.0: atan(x) = atan(1.0) + atan(s), where s=(x-1.0)/(1.0+1.0*x) + * For 19.0/16.0 <= x <= 39.0/16.0: atan(x) = atan(1.5) + atan(s), where s=(x-1.5)/(1.0+1.5*x) + * For 39.0/16.0 <= x <= inf : atan(x) = atan(inf) + atan(s), where s=-1.0/x + * Where atan(s) ~= s+s^3*Poly11(s^2) on interval |s|<7.0/0.16. + * + */ + +/* Offsets for data table __svml_datan_data_internal_avx512 + */ +#define AbsMask 0 +#define Shifter 32 +#define MaxThreshold 64 +#define MOne 96 +#define One 128 +#define LargeX 160 +#define Zero 192 +#define Tbl_H 224 +#define Tbl_L 480 +#define dIndexMed 736 +#define Pi2 768 +#define Pi2_low 800 +#define coeff 832 + +#include <sysdep.h> + + .text + .section .text.avx2,"ax",@progbits +ENTRY(_ZGVdN4v_atan_avx2) + lea Tbl_H+128+__svml_datan_data_internal_avx512(%rip), %rdi + vmovupd Shifter+__svml_datan_data_internal_avx512(%rip), %ymm4 + vmovupd One+__svml_datan_data_internal_avx512(%rip), %ymm9 + +/* saturate X range */ + vmovupd LargeX+__svml_datan_data_internal_avx512(%rip), %ymm6 + vandpd __svml_datan_data_internal_avx512(%rip), %ymm0, %ymm7 + vaddpd %ymm4, %ymm7, %ymm2 + vcmpge_oqpd MaxThreshold+__svml_datan_data_internal_avx512(%rip), %ymm7, %ymm3 + vminpd %ymm7, %ymm6, %ymm10 + vsubpd %ymm4, %ymm2, %ymm5 + +/* + * table lookup sequence + * VPERMUTE not available + */ + vpsllq $3, %ymm2, %ymm13 + vsubpd %ymm5, %ymm7, %ymm8 + vcmpge_oqpd dIndexMed+__svml_datan_data_internal_avx512(%rip), %ymm2, %ymm2 + vfmadd231pd %ymm7, %ymm5, %ymm9 + vpand .FLT_11(%rip), %ymm13, %ymm14 + vblendvpd %ymm3, MOne+__svml_datan_data_internal_avx512(%rip), %ymm8, %ymm11 + vblendvpd %ymm3, %ymm10, %ymm9, %ymm12 + vxorpd %ymm0, %ymm7, %ymm1 + +/* R+Rl = DiffX/Y */ + vdivpd %ymm12, %ymm11, %ymm0 + vextractf128 $1, %ymm14, %xmm4 + vmovd %xmm14, %eax + vmovd %xmm4, %ecx + movslq %eax, %rax + vpextrd $2, %xmm14, %edx + movslq %ecx, %rcx + vpextrd $2, %xmm4, %esi + movslq %edx, %rdx + movslq %esi, %rsi + vmovsd -128(%rax,%rdi), %xmm15 + vmovsd (%rdi,%rax), %xmm7 + vmovsd -128(%rcx,%rdi), %xmm5 + vmovsd (%rdi,%rcx), %xmm9 + vmovhpd -128(%rdx,%rdi), %xmm15, %xmm15 + vmovhpd (%rdi,%rdx), %xmm7, %xmm8 + vmovhpd -128(%rsi,%rdi), %xmm5, %xmm6 + vmovhpd (%rdi,%rsi), %xmm9, %xmm10 + +/* polynomial evaluation */ + vmulpd %ymm0, %ymm0, %ymm5 + vmulpd %ymm5, %ymm5, %ymm4 + vinsertf128 $1, %xmm6, %ymm15, %ymm11 + vinsertf128 $1, %xmm10, %ymm8, %ymm12 + vblendvpd %ymm2, %ymm12, %ymm11, %ymm13 + vmovupd coeff+__svml_datan_data_internal_avx512(%rip), %ymm8 + vmovupd coeff+64+__svml_datan_data_internal_avx512(%rip), %ymm2 + vmulpd %ymm5, %ymm0, %ymm6 + vfmadd213pd coeff+32+__svml_datan_data_internal_avx512(%rip), %ymm5, %ymm8 + vfmadd213pd coeff+96+__svml_datan_data_internal_avx512(%rip), %ymm5, %ymm2 + +/* set table value to Pi/2 for large X */ + vblendvpd %ymm3, Pi2+__svml_datan_data_internal_avx512(%rip), %ymm13, %ymm7 + vmovupd coeff+128+__svml_datan_data_internal_avx512(%rip), %ymm3 + vfmadd213pd %ymm2, %ymm4, %ymm8 + vfmadd213pd coeff+160+__svml_datan_data_internal_avx512(%rip), %ymm3, %ymm5 + vfmadd213pd %ymm5, %ymm4, %ymm8 + vfmadd213pd %ymm0, %ymm6, %ymm8 + vaddpd %ymm8, %ymm7, %ymm0 + vxorpd %ymm1, %ymm0, %ymm0 + ret + +END(_ZGVdN4v_atan_avx2) + + .section .rodata, "a" + .align 32 + +.FLT_11: + .long 0x00000078,0x00000000,0x00000078,0x00000000,0x00000078,0x00000000,0x00000078,0x00000000 + .type .FLT_11,@object + .size .FLT_11,32 + .align 32 + +#ifdef __svml_datan_data_internal_avx512_typedef +typedef unsigned int VUINT32; +typedef struct { + __declspec(align(32)) VUINT32 AbsMask[4][2]; + __declspec(align(32)) VUINT32 Shifter[4][2]; + __declspec(align(32)) VUINT32 MaxThreshold[4][2]; + __declspec(align(32)) VUINT32 MOne[4][2]; + __declspec(align(32)) VUINT32 One[4][2]; + __declspec(align(32)) VUINT32 LargeX[4][2]; + __declspec(align(32)) VUINT32 Zero[4][2]; + __declspec(align(32)) VUINT32 Tbl_H[32][2]; + __declspec(align(32)) VUINT32 Tbl_L[32][2]; + __declspec(align(32)) VUINT32 dIndexMed[4][2]; + __declspec(align(32)) VUINT32 Pi2[4][2]; + __declspec(align(32)) VUINT32 Pi2_low[4][2]; + __declspec(align(32)) VUINT32 coeff[6][4][2]; + } __svml_datan_data_internal_avx512; +#endif +__svml_datan_data_internal_avx512: + /*== AbsMask ==*/ + .quad 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff + /*== Shifter ==*/ + .align 32 + .quad 0x4318000000000000, 0x4318000000000000, 0x4318000000000000, 0x4318000000000000 + /*== MaxThreshold ==*/ + .align 32 + .quad 0x401f800000000000, 0x401f800000000000, 0x401f800000000000, 0x401f800000000000 + /*== MOne ==*/ + .align 32 + .quad 0xbff0000000000000, 0xbff0000000000000, 0xbff0000000000000, 0xbff0000000000000 + /*== One ==*/ + .align 32 + .quad 0x3ff0000000000000, 0x3ff0000000000000, 0x3ff0000000000000, 0x3ff0000000000000 + /*== LargeX ==*/ + .align 32 + .quad 0x47f0000000000000, 0x47f0000000000000, 0x47f0000000000000, 0x47f0000000000000 + /*== Zero ==*/ + .align 32 + .quad 0x0000000000000000, 0x0000000000000000, 0x0000000000000000, 0x0000000000000000 + /*== Tbl_H ==*/ + .align 32 + .quad 0x0000000000000000, 0x3fcf5b75f92c80dd + .quad 0x3fddac670561bb4f, 0x3fe4978fa3269ee1 + .quad 0x3fe921fb54442d18, 0x3fecac7c57846f9e + .quad 0x3fef730bd281f69b, 0x3ff0d38f2c5ba09f + .quad 0x3ff1b6e192ebbe44, 0x3ff270ef55a53a25 + .quad 0x3ff30b6d796a4da8, 0x3ff38d6a6ce13353 + .quad 0x3ff3fc176b7a8560, 0x3ff45b54837351a0 + .quad 0x3ff4ae10fc6589a5, 0x3ff4f68dea672617 + .quad 0x3ff5368c951e9cfd, 0x3ff56f6f33a3e6a7 + .quad 0x3ff5a25052114e60, 0x3ff5d013c41adabd + .quad 0x3ff5f97315254857, 0x3ff61f06c6a92b89 + .quad 0x3ff6414d44094c7c, 0x3ff660b02c736a06 + .quad 0x3ff67d8863bc99bd, 0x3ff698213a9d5053 + .quad 0x3ff6b0bae830c070, 0x3ff6c78c7edeb195 + .quad 0x3ff6dcc57bb565fd, 0x3ff6f08f07435fec + .quad 0x3ff7030cf9403197, 0x3ff7145eac2088a4 + /*== Tbl_L ==*/ + .align 32 + .quad 0x0000000000000000, 0x3c68ab6e3cf7afbd + .quad 0x3c7a2b7f222f65e2, 0x3c72419a87f2a458 + .quad 0x3c81a62633145c07, 0x3c80dae13ad18a6b + .quad 0x3c7007887af0cbbd, 0xbc9bd0dc231bfd70 + .quad 0x3c9b1b466a88828e, 0xbc9a66b1af5f84fb + .quad 0x3c96254cb03bb199, 0xbc812c77e8a80f5c + .quad 0xbc4441a3bd3f1084, 0x3c79e4a72eedacc4 + .quad 0xbc93b03e8a27f555, 0x3c9934f9f2b0020e + .quad 0xbc996f47948a99f1, 0xbc7df6edd6f1ec3b + .quad 0x3c78c2d0c89de218, 0x3c9f82bba194dd5d + .quad 0xbc831151a43b51ca, 0xbc8487d50bceb1a5 + .quad 0xbc9c5f60a65c7397, 0xbc7acb6afb332a0f + .quad 0xbc99b7bd2e1e8c9c, 0xbc9b9839085189e3 + .quad 0xbc97d1ab82ffb70b, 0x3c99239ad620ffe2 + .quad 0xbc929c86447928e7, 0xbc8957a7170df016 + .quad 0xbc7cbe1896221608, 0xbc9fda5797b32a0b + /*== dIndexMed ==*/ + .align 32 + .quad 0x4318000000000010, 0x4318000000000010, 0x4318000000000010, 0x4318000000000010 + /*== Pi2 ==*/ + .align 32 + .quad 0x3ff921fb54442d18, 0x3ff921fb54442d18, 0x3ff921fb54442d18, 0x3ff921fb54442d18 + /*== Pi2_low ==*/ + .align 32 + .quad 0x3c91a62633145c07, 0x3c91a62633145c07, 0x3c91a62633145c07, 0x3c91a62633145c07 + /*== coeff6 ==*/ + .align 32 + .quad 0x3fb2e9b9f5c4fe97, 0x3fb2e9b9f5c4fe97, 0x3fb2e9b9f5c4fe97, 0x3fb2e9b9f5c4fe97 + .quad 0xbfb74257c46790cc, 0xbfb74257c46790cc, 0xbfb74257c46790cc, 0xbfb74257c46790cc + .quad 0x3fbc71bfeff916a0, 0x3fbc71bfeff916a0, 0x3fbc71bfeff916a0, 0x3fbc71bfeff916a0 + .quad 0xbfc249248eef04da, 0xbfc249248eef04da, 0xbfc249248eef04da, 0xbfc249248eef04da + .quad 0x3fc999999998741e, 0x3fc999999998741e, 0x3fc999999998741e, 0x3fc999999998741e + .quad 0xbfd555555555554d, 0xbfd555555555554d, 0xbfd555555555554d, 0xbfd555555555554d + .align 32 + .type __svml_datan_data_internal_avx512,@object + .size __svml_datan_data_internal_avx512,.-__svml_datan_data_internal_avx512 diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan8_core-avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_atan8_core-avx2.S new file mode 100644 index 0000000..723734e --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan8_core-avx2.S @@ -0,0 +1,20 @@ +/* AVX2 version of vectorized atan, 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 + <https://www.gnu.org/licenses/>. */ + +#define _ZGVeN8v_atan _ZGVeN8v_atan_avx2_wrapper +#include "../svml_d_atan8_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan8_core.c b/sysdeps/x86_64/fpu/multiarch/svml_d_atan8_core.c new file mode 100644 index 0000000..e97a41b --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan8_core.c @@ -0,0 +1,27 @@ +/* Multiple versions of vectorized atan, 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 + <https://www.gnu.org/licenses/>. */ + +#define SYMBOL_NAME _ZGVeN8v_atan +#include "ifunc-mathvec-avx512-skx.h" + +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); + +#ifdef SHARED +__hidden_ver1 (_ZGVeN8v_atan, __GI__ZGVeN8v_atan, __redirect__ZGVeN8v_atan) + __attribute__ ((visibility ("hidden"))); +#endif diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_atan8_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_d_atan8_core_avx512.S new file mode 100644 index 0000000..fa6cb47 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_atan8_core_avx512.S @@ -0,0 +1,213 @@ +/* Function atan 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: + * + * For 0.0 <= x <= 7.0/16.0: atan(x) = atan(0.0) + atan(s), where s=(x-0.0)/(1.0+0.0*x) + * For 7.0/16.0 <= x <= 11.0/16.0: atan(x) = atan(0.5) + atan(s), where s=(x-0.5)/(1.0+0.5*x) + * For 11.0/16.0 <= x <= 19.0/16.0: atan(x) = atan(1.0) + atan(s), where s=(x-1.0)/(1.0+1.0*x) + * For 19.0/16.0 <= x <= 39.0/16.0: atan(x) = atan(1.5) + atan(s), where s=(x-1.5)/(1.0+1.5*x) + * For 39.0/16.0 <= x <= inf : atan(x) = atan(inf) + atan(s), where s=-1.0/x + * Where atan(s) ~= s+s^3*Poly11(s^2) on interval |s|<7.0/0.16. + * + */ + +/* Offsets for data table __svml_datan_data_internal_avx512 + */ +#define AbsMask 0 +#define Shifter 64 +#define MaxThreshold 128 +#define MOne 192 +#define One 256 +#define LargeX 320 +#define Zero 384 +#define Tbl_H 448 +#define dIndexMed 704 +#define Pi2 768 +#define coeff_1 832 +#define coeff_2 896 +#define coeff_3 960 +#define coeff_4 1024 +#define coeff_5 1088 +#define coeff_6 1152 + +#include <sysdep.h> + + .text + .section .text.evex512,"ax",@progbits +ENTRY(_ZGVeN8v_atan_skx) + vmovups Shifter+__svml_datan_data_internal_avx512(%rip), %zmm4 + vmovups MaxThreshold+__svml_datan_data_internal_avx512(%rip), %zmm3 + vmovups One+__svml_datan_data_internal_avx512(%rip), %zmm9 + +/* saturate X range */ + vmovups LargeX+__svml_datan_data_internal_avx512(%rip), %zmm7 + vandpd __svml_datan_data_internal_avx512(%rip), %zmm0, %zmm8 + +/* R+Rl = DiffX/Y */ + vbroadcastsd .FLT_10(%rip), %zmm15 + vaddpd {rn-sae}, %zmm4, %zmm8, %zmm2 + vxorpd %zmm0, %zmm8, %zmm1 + vcmppd $29, {sae}, %zmm3, %zmm8, %k2 + +/* round to 2 bits after binary point */ + vreducepd $40, {sae}, %zmm8, %zmm6 + vsubpd {rn-sae}, %zmm4, %zmm2, %zmm5 + +/* + * if|X|>=MaxThreshold, set DiffX=-1 + * VMSUB(D, DiffX, LargeMask, Zero, One); + */ + vblendmpd MOne+__svml_datan_data_internal_avx512(%rip), %zmm6, %zmm10{%k2} + vfmadd231pd {rn-sae}, %zmm8, %zmm5, %zmm9 + vmovups dIndexMed+__svml_datan_data_internal_avx512(%rip), %zmm5 + +/* table lookup sequence */ + vmovups Tbl_H+__svml_datan_data_internal_avx512(%rip), %zmm6 + vgetmantpd $0, {sae}, %zmm10, %zmm14 + vgetexppd {sae}, %zmm10, %zmm11 + vmovups coeff_5+__svml_datan_data_internal_avx512(%rip), %zmm10 + +/* + * if|X|>=MaxThreshold, set Y=X + * VMADD(D, Y, LargeMask, X, Zero); + */ + vminpd {sae}, %zmm8, %zmm7, %zmm9{%k2} + vcmppd $29, {sae}, %zmm5, %zmm2, %k1 + vmovups Tbl_H+128+__svml_datan_data_internal_avx512(%rip), %zmm7 + vmovups coeff_1+__svml_datan_data_internal_avx512(%rip), %zmm8 + vgetmantpd $0, {sae}, %zmm9, %zmm3 + vgetexppd {sae}, %zmm9, %zmm12 + vmovups coeff_3+__svml_datan_data_internal_avx512(%rip), %zmm9 + vpermt2pd Tbl_H+64+__svml_datan_data_internal_avx512(%rip), %zmm2, %zmm6 + vsubpd {rn-sae}, %zmm12, %zmm11, %zmm4 + vpermt2pd Tbl_H+192+__svml_datan_data_internal_avx512(%rip), %zmm2, %zmm7 + vrcp14pd %zmm3, %zmm13 + vmovups coeff_4+__svml_datan_data_internal_avx512(%rip), %zmm12 + vmovups coeff_6+__svml_datan_data_internal_avx512(%rip), %zmm11 + vblendmpd %zmm7, %zmm6, %zmm2{%k1} + vmulpd {rn-sae}, %zmm13, %zmm14, %zmm0 + vfnmadd231pd {rn-sae}, %zmm3, %zmm13, %zmm15 + vfnmadd213pd {rn-sae}, %zmm14, %zmm0, %zmm3 + vfmadd213pd {rn-sae}, %zmm15, %zmm15, %zmm15 + vfmadd213pd {rn-sae}, %zmm13, %zmm13, %zmm15 + vfmadd213pd {rn-sae}, %zmm0, %zmm15, %zmm3 + vscalefpd {rn-sae}, %zmm4, %zmm3, %zmm0 + +/* set table value to Pi/2 for large X */ + vblendmpd Pi2+__svml_datan_data_internal_avx512(%rip), %zmm2, %zmm3{%k2} + vmovups coeff_2+__svml_datan_data_internal_avx512(%rip), %zmm2 + +/* polynomial evaluation */ + vmulpd {rn-sae}, %zmm0, %zmm0, %zmm14 + vmulpd {rn-sae}, %zmm14, %zmm14, %zmm13 + vmulpd {rn-sae}, %zmm0, %zmm14, %zmm15 + vfmadd231pd {rn-sae}, %zmm14, %zmm8, %zmm2 + vfmadd231pd {rn-sae}, %zmm14, %zmm9, %zmm12 + vfmadd213pd {rn-sae}, %zmm11, %zmm10, %zmm14 + vfmadd213pd {rn-sae}, %zmm12, %zmm13, %zmm2 + vfmadd213pd {rn-sae}, %zmm14, %zmm13, %zmm2 + vfmadd213pd {rn-sae}, %zmm0, %zmm15, %zmm2 + vaddpd {rn-sae}, %zmm3, %zmm2, %zmm0 + vxorpd %zmm1, %zmm0, %zmm0 + ret + +END(_ZGVeN8v_atan_skx) + + .section .rodata, "a" + .align 64 + +#ifdef __svml_datan_data_internal_avx512_typedef +typedef unsigned int VUINT32; +typedef struct { + __declspec(align(64)) VUINT32 AbsMask[8][2]; + __declspec(align(64)) VUINT32 Shifter[8][2]; + __declspec(align(64)) VUINT32 MaxThreshold[8][2]; + __declspec(align(64)) VUINT32 MOne[8][2]; + __declspec(align(64)) VUINT32 One[8][2]; + __declspec(align(64)) VUINT32 LargeX[8][2]; + __declspec(align(64)) VUINT32 Zero[8][2]; + __declspec(align(64)) VUINT32 Tbl_H[32][2]; + __declspec(align(64)) VUINT32 dIndexMed[8][2]; + __declspec(align(64)) VUINT32 Pi2[8][2]; + __declspec(align(64)) VUINT32 coeff[6][8][2]; + } __svml_datan_data_internal_avx512; +#endif +__svml_datan_data_internal_avx512: + /*== AbsMask ==*/ + .quad 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff + /*== Shifter ==*/ + .align 64 + .quad 0x4318000000000000, 0x4318000000000000, 0x4318000000000000, 0x4318000000000000, 0x4318000000000000, 0x4318000000000000, 0x4318000000000000, 0x4318000000000000 + /*== MaxThreshold ==*/ + .align 64 + .quad 0x401f800000000000, 0x401f800000000000, 0x401f800000000000, 0x401f800000000000, 0x401f800000000000, 0x401f800000000000, 0x401f800000000000, 0x401f800000000000 + /*== MOne ==*/ + .align 64 + .quad 0xbff0000000000000, 0xbff0000000000000, 0xbff0000000000000, 0xbff0000000000000, 0xbff0000000000000, 0xbff0000000000000, 0xbff0000000000000, 0xbff0000000000000 + /*== One ==*/ + .align 64 + .quad 0x3ff0000000000000, 0x3ff0000000000000, 0x3ff0000000000000, 0x3ff0000000000000, 0x3ff0000000000000, 0x3ff0000000000000, 0x3ff0000000000000, 0x3ff0000000000000 + /*== LargeX ==*/ + .align 64 + .quad 0x47f0000000000000, 0x47f0000000000000, 0x47f0000000000000, 0x47f0000000000000, 0x47f0000000000000, 0x47f0000000000000, 0x47f0000000000000, 0x47f0000000000000 + /*== Zero ==*/ + .align 64 + .quad 0x0000000000000000, 0x0000000000000000, 0x0000000000000000, 0x0000000000000000, 0x0000000000000000, 0x0000000000000000, 0x0000000000000000, 0x0000000000000000 + /*== Tbl_H ==*/ + .align 64 + .quad 0x0000000000000000, 0x3fcf5b75f92c80dd + .quad 0x3fddac670561bb4f, 0x3fe4978fa3269ee1 + .quad 0x3fe921fb54442d18, 0x3fecac7c57846f9e + .quad 0x3fef730bd281f69b, 0x3ff0d38f2c5ba09f + .quad 0x3ff1b6e192ebbe44, 0x3ff270ef55a53a25 + .quad 0x3ff30b6d796a4da8, 0x3ff38d6a6ce13353 + .quad 0x3ff3fc176b7a8560, 0x3ff45b54837351a0 + .quad 0x3ff4ae10fc6589a5, 0x3ff4f68dea672617 + .quad 0x3ff5368c951e9cfd, 0x3ff56f6f33a3e6a7 + .quad 0x3ff5a25052114e60, 0x3ff5d013c41adabd + .quad 0x3ff5f97315254857, 0x3ff61f06c6a92b89 + .quad 0x3ff6414d44094c7c, 0x3ff660b02c736a06 + .quad 0x3ff67d8863bc99bd, 0x3ff698213a9d5053 + .quad 0x3ff6b0bae830c070, 0x3ff6c78c7edeb195 + .quad 0x3ff6dcc57bb565fd, 0x3ff6f08f07435fec + .quad 0x3ff7030cf9403197, 0x3ff7145eac2088a4 + /*== dIndexMed ==*/ + .align 64 + .quad 0x4318000000000010, 0x4318000000000010, 0x4318000000000010, 0x4318000000000010, 0x4318000000000010, 0x4318000000000010, 0x4318000000000010, 0x4318000000000010 + /*== Pi2 ==*/ + .align 64 + .quad 0x3ff921fb54442d18, 0x3ff921fb54442d18, 0x3ff921fb54442d18, 0x3ff921fb54442d18, 0x3ff921fb54442d18, 0x3ff921fb54442d18, 0x3ff921fb54442d18, 0x3ff921fb54442d18 + /*== coeff6 ==*/ + .align 64 + .quad 0x3fb2e9b9f5c4fe97, 0x3fb2e9b9f5c4fe97, 0x3fb2e9b9f5c4fe97, 0x3fb2e9b9f5c4fe97, 0x3fb2e9b9f5c4fe97, 0x3fb2e9b9f5c4fe97, 0x3fb2e9b9f5c4fe97, 0x3fb2e9b9f5c4fe97 + .quad 0xbfb74257c46790cc, 0xbfb74257c46790cc, 0xbfb74257c46790cc, 0xbfb74257c46790cc, 0xbfb74257c46790cc, 0xbfb74257c46790cc, 0xbfb74257c46790cc, 0xbfb74257c46790cc + .quad 0x3fbc71bfeff916a0, 0x3fbc71bfeff916a0, 0x3fbc71bfeff916a0, 0x3fbc71bfeff916a0, 0x3fbc71bfeff916a0, 0x3fbc71bfeff916a0, 0x3fbc71bfeff916a0, 0x3fbc71bfeff916a0 + .quad 0xbfc249248eef04da, 0xbfc249248eef04da, 0xbfc249248eef04da, 0xbfc249248eef04da, 0xbfc249248eef04da, 0xbfc249248eef04da, 0xbfc249248eef04da, 0xbfc249248eef04da + .quad 0x3fc999999998741e, 0x3fc999999998741e, 0x3fc999999998741e, 0x3fc999999998741e, 0x3fc999999998741e, 0x3fc999999998741e, 0x3fc999999998741e, 0x3fc999999998741e + .quad 0xbfd555555555554d, 0xbfd555555555554d, 0xbfd555555555554d, 0xbfd555555555554d, 0xbfd555555555554d, 0xbfd555555555554d, 0xbfd555555555554d, 0xbfd555555555554d + .align 64 + .type __svml_datan_data_internal_avx512,@object + .size __svml_datan_data_internal_avx512,.-__svml_datan_data_internal_avx512 + .align 8 + +.FLT_10: + .long 0x00000000,0x3ff00000 + .type .FLT_10,@object + .size .FLT_10,8 diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atanf16_core-avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf16_core-avx2.S new file mode 100644 index 0000000..27623cd --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf16_core-avx2.S @@ -0,0 +1,20 @@ +/* AVX2 version of vectorized atanf. + 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/>. */ + +#define _ZGVeN16v_atanf _ZGVeN16v_atanf_avx2_wrapper +#include "../svml_s_atanf16_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atanf16_core.c b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf16_core.c new file mode 100644 index 0000000..940de26 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf16_core.c @@ -0,0 +1,28 @@ +/* Multiple versions of vectorized atanf, 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 + <https://www.gnu.org/licenses/>. */ + +#define SYMBOL_NAME _ZGVeN16v_atanf +#include "ifunc-mathvec-avx512-skx.h" + +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); + +#ifdef SHARED +__hidden_ver1 (_ZGVeN16v_atanf, __GI__ZGVeN16v_atanf, + __redirect__ZGVeN16v_atanf) + __attribute__ ((visibility ("hidden"))); +#endif diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atanf16_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf16_core_avx512.S new file mode 100644 index 0000000..4a37f03 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf16_core_avx512.S @@ -0,0 +1,174 @@ +/* Function atanf 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: + * + * For 0.0 <= x <= 7.0/16.0: atan(x) = atan(0.0) + atan(s), where s=(x-0.0)/(1.0+0.0*x) + * For 7.0/16.0 <= x <= 11.0/16.0: atan(x) = atan(0.5) + atan(s), where s=(x-0.5)/(1.0+0.5*x) + * For 11.0/16.0 <= x <= 19.0/16.0: atan(x) = atan(1.0) + atan(s), where s=(x-1.0)/(1.0+1.0*x) + * For 19.0/16.0 <= x <= 39.0/16.0: atan(x) = atan(1.5) + atan(s), where s=(x-1.5)/(1.0+1.5*x) + * For 39.0/16.0 <= x <= inf : atan(x) = atan(inf) + atan(s), where s=-1.0/x + * Where atan(s) ~= s+s^3*Poly11(s^2) on interval |s|<7.0/0.16. + * + */ + +/* Offsets for data table __svml_satan_data_internal_avx512 + */ +#define AbsMask 0 +#define Shifter 64 +#define MaxThreshold 128 +#define MOne 192 +#define One 256 +#define LargeX 320 +#define Zero 384 +#define Tbl_H 448 +#define Pi2 576 +#define coeff_1 640 +#define coeff_2 704 +#define coeff_3 768 + +#include <sysdep.h> + + .text + .section .text.exex512,"ax",@progbits +ENTRY(_ZGVeN16v_atanf_skx) + vandps __svml_satan_data_internal_avx512(%rip), %zmm0, %zmm7 + vmovups MaxThreshold+__svml_satan_data_internal_avx512(%rip), %zmm3 + vmovups One+__svml_satan_data_internal_avx512(%rip), %zmm8 + +/* round to 2 bits after binary point */ + vreduceps $40, {sae}, %zmm7, %zmm5 + +/* saturate X range */ + vmovups LargeX+__svml_satan_data_internal_avx512(%rip), %zmm6 + vmovups Shifter+__svml_satan_data_internal_avx512(%rip), %zmm2 + vcmpps $29, {sae}, %zmm3, %zmm7, %k1 + +/* table lookup sequence */ + vmovups Tbl_H+__svml_satan_data_internal_avx512(%rip), %zmm3 + vsubps {rn-sae}, %zmm5, %zmm7, %zmm4 + vaddps {rn-sae}, %zmm2, %zmm7, %zmm1 + vxorps %zmm0, %zmm7, %zmm0 + vfmadd231ps {rn-sae}, %zmm7, %zmm4, %zmm8 + vmovups coeff_2+__svml_satan_data_internal_avx512(%rip), %zmm4 + +/* if|X|>=MaxThreshold, set DiffX=-1 */ + vblendmps MOne+__svml_satan_data_internal_avx512(%rip), %zmm5, %zmm9{%k1} + vmovups coeff_3+__svml_satan_data_internal_avx512(%rip), %zmm5 + +/* if|X|>=MaxThreshold, set Y=X */ + vminps {sae}, %zmm7, %zmm6, %zmm8{%k1} + +/* R+Rl = DiffX/Y */ + vgetmantps $0, {sae}, %zmm9, %zmm12 + vgetexpps {sae}, %zmm9, %zmm10 + vpermt2ps Tbl_H+64+__svml_satan_data_internal_avx512(%rip), %zmm1, %zmm3 + vgetmantps $0, {sae}, %zmm8, %zmm15 + vgetexpps {sae}, %zmm8, %zmm11 + vmovups coeff_1+__svml_satan_data_internal_avx512(%rip), %zmm1 + +/* set table value to Pi/2 for large X */ + vblendmps Pi2+__svml_satan_data_internal_avx512(%rip), %zmm3, %zmm9{%k1} + vrcp14ps %zmm15, %zmm13 + vsubps {rn-sae}, %zmm11, %zmm10, %zmm2 + vmulps {rn-sae}, %zmm13, %zmm12, %zmm14 + vfnmadd213ps {rn-sae}, %zmm12, %zmm14, %zmm15 + vfmadd213ps {rn-sae}, %zmm14, %zmm13, %zmm15 + vscalefps {rn-sae}, %zmm2, %zmm15, %zmm7 + +/* polynomial evaluation */ + vmulps {rn-sae}, %zmm7, %zmm7, %zmm8 + vmulps {rn-sae}, %zmm7, %zmm8, %zmm6 + vfmadd231ps {rn-sae}, %zmm8, %zmm1, %zmm4 + vfmadd213ps {rn-sae}, %zmm5, %zmm4, %zmm8 + vfmadd213ps {rn-sae}, %zmm7, %zmm6, %zmm8 + vaddps {rn-sae}, %zmm9, %zmm8, %zmm10 + vxorps %zmm0, %zmm10, %zmm0 + ret + +END(_ZGVeN16v_atanf_skx) + + .section .rodata, "a" + .align 64 + +#ifdef __svml_satan_data_internal_avx512_typedef +typedef unsigned int VUINT32; +typedef struct { + __declspec(align(64)) VUINT32 AbsMask[16][1]; + __declspec(align(64)) VUINT32 Shifter[16][1]; + __declspec(align(64)) VUINT32 MaxThreshold[16][1]; + __declspec(align(64)) VUINT32 MOne[16][1]; + __declspec(align(64)) VUINT32 One[16][1]; + __declspec(align(64)) VUINT32 LargeX[16][1]; + __declspec(align(64)) VUINT32 Zero[16][1]; + __declspec(align(64)) VUINT32 Tbl_H[32][1]; + __declspec(align(64)) VUINT32 Pi2[16][1]; + __declspec(align(64)) VUINT32 coeff[3][16][1]; + } __svml_satan_data_internal_avx512; +#endif +__svml_satan_data_internal_avx512: + /*== AbsMask ==*/ + .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff + /*== Shifter ==*/ + .align 64 + .long 0x4a000000, 0x4a000000, 0x4a000000, 0x4a000000, 0x4a000000, 0x4a000000, 0x4a000000, 0x4a000000, 0x4a000000, 0x4a000000, 0x4a000000, 0x4a000000, 0x4a000000, 0x4a000000, 0x4a000000, 0x4a000000 + /*== MaxThreshold ==*/ + .align 64 + .long 0x40F80000, 0x40F80000, 0x40F80000, 0x40F80000, 0x40F80000, 0x40F80000, 0x40F80000, 0x40F80000, 0x40F80000, 0x40F80000, 0x40F80000, 0x40F80000, 0x40F80000, 0x40F80000, 0x40F80000, 0x40F80000 + /*== MOne ==*/ + .align 64 + .long 0xbf800000, 0xbf800000, 0xbf800000, 0xbf800000, 0xbf800000, 0xbf800000, 0xbf800000, 0xbf800000, 0xbf800000, 0xbf800000, 0xbf800000, 0xbf800000, 0xbf800000, 0xbf800000, 0xbf800000, 0xbf800000 + /*== One ==*/ + .align 64 + .long 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 + /*== LargeX ==*/ + .align 64 + .long 0x4f800000, 0x4f800000, 0x4f800000, 0x4f800000, 0x4f800000, 0x4f800000, 0x4f800000, 0x4f800000, 0x4f800000, 0x4f800000, 0x4f800000, 0x4f800000, 0x4f800000, 0x4f800000, 0x4f800000, 0x4f800000 + /*== Zero ==*/ + .align 64 + .long 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000 + /*== Tbl_H ==*/ + .align 64 + .long 0x00000000, 0x3e7adbb0 + .long 0x3eed6338, 0x3f24bc7d + .long 0x3f490fdb, 0x3f6563e3 + .long 0x3f7b985f, 0x3f869c79 + .long 0x3f8db70d, 0x3f93877b + .long 0x3f985b6c, 0x3f9c6b53 + .long 0x3f9fe0bb, 0x3fa2daa4 + .long 0x3fa57088, 0x3fa7b46f + .long 0x3fa9b465, 0x3fab7b7a + .long 0x3fad1283, 0x3fae809e + .long 0x3fafcb99, 0x3fb0f836 + .long 0x3fb20a6a, 0x3fb30581 + .long 0x3fb3ec43, 0x3fb4c10a + .long 0x3fb585d7, 0x3fb63c64 + .long 0x3fb6e62c, 0x3fb78478 + .long 0x3fb81868, 0x3fb8a2f5 + /*== Pi2 ==*/ + .align 64 + .long 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB + /*== coeff3 ==*/ + .align 64 + .long 0xbe0fa8de, 0xbe0fa8de, 0xbe0fa8de, 0xbe0fa8de, 0xbe0fa8de, 0xbe0fa8de, 0xbe0fa8de, 0xbe0fa8de, 0xbe0fa8de, 0xbe0fa8de, 0xbe0fa8de, 0xbe0fa8de, 0xbe0fa8de, 0xbe0fa8de, 0xbe0fa8de, 0xbe0fa8de + .long 0x3e4cc8e2, 0x3e4cc8e2, 0x3e4cc8e2, 0x3e4cc8e2, 0x3e4cc8e2, 0x3e4cc8e2, 0x3e4cc8e2, 0x3e4cc8e2, 0x3e4cc8e2, 0x3e4cc8e2, 0x3e4cc8e2, 0x3e4cc8e2, 0x3e4cc8e2, 0x3e4cc8e2, 0x3e4cc8e2, 0x3e4cc8e2 + .long 0xbeaaaaaa, 0xbeaaaaaa, 0xbeaaaaaa, 0xbeaaaaaa, 0xbeaaaaaa, 0xbeaaaaaa, 0xbeaaaaaa, 0xbeaaaaaa, 0xbeaaaaaa, 0xbeaaaaaa, 0xbeaaaaaa, 0xbeaaaaaa, 0xbeaaaaaa, 0xbeaaaaaa, 0xbeaaaaaa, 0xbeaaaaaa + .align 64 + .type __svml_satan_data_internal_avx512,@object + .size __svml_satan_data_internal_avx512,.-__svml_satan_data_internal_avx512 diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atanf4_core-sse2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf4_core-sse2.S new file mode 100644 index 0000000..fe81170 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf4_core-sse2.S @@ -0,0 +1,20 @@ +/* SSE2 version of vectorized atanf, 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 + <https://www.gnu.org/licenses/>. */ + +#define _ZGVbN4v_atanf _ZGVbN4v_atanf_sse2 +#include "../svml_s_atanf4_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atanf4_core.c b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf4_core.c new file mode 100644 index 0000000..975ece6 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf4_core.c @@ -0,0 +1,28 @@ +/* Multiple versions of vectorized atanf, 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 + <https://www.gnu.org/licenses/>. */ + +#define SYMBOL_NAME _ZGVbN4v_atanf +#include "ifunc-mathvec-sse4_1.h" + +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); + +#ifdef SHARED +__hidden_ver1 (_ZGVbN4v_atanf, __GI__ZGVbN4v_atanf, + __redirect__ZGVbN4v_atanf) + __attribute__ ((visibility ("hidden"))); +#endif diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atanf4_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf4_core_sse4.S new file mode 100644 index 0000000..c58a894 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf4_core_sse4.S @@ -0,0 +1,164 @@ +/* Function atanf 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: + * + * For 0.0 <= x <= 7.0/16.0: atan(x) = atan(0.0) + atan(s), where s=(x-0.0)/(1.0+0.0*x) + * For 7.0/16.0 <= x <= 11.0/16.0: atan(x) = atan(0.5) + atan(s), where s=(x-0.5)/(1.0+0.5*x) + * For 11.0/16.0 <= x <= 19.0/16.0: atan(x) = atan(1.0) + atan(s), where s=(x-1.0)/(1.0+1.0*x) + * For 19.0/16.0 <= x <= 39.0/16.0: atan(x) = atan(1.5) + atan(s), where s=(x-1.5)/(1.0+1.5*x) + * For 39.0/16.0 <= x <= inf : atan(x) = atan(inf) + atan(s), where s=-1.0/x + * Where atan(s) ~= s+s^3*Poly11(s^2) on interval |s|<7.0/0.16. + * + */ + +/* Offsets for data table __svml_satan_data_internal + */ +#define _sSIGN_MASK 0 +#define _sABS_MASK 16 +#define _sONE 32 +#define _sPIO2 48 +#define _sPC8 64 +#define _sPC7 80 +#define _sPC6 96 +#define _sPC5 112 +#define _sPC4 128 +#define _sPC3 144 +#define _sPC2 160 +#define _sPC1 176 +#define _sPC0 192 + +#include <sysdep.h> + + .text + .section .text.sse4,"ax",@progbits +ENTRY(_ZGVbN4v_atanf_sse4) +/* + * To use minps\maxps operations for argument reduction + * uncomment _AT_USEMINMAX_ definition + * Declarations + * Variables + * Constants + */ + movups _sABS_MASK+__svml_satan_data_internal(%rip), %xmm2 + +/* + * 1) If x>1, then r=-1/x, PIO2=Pi/2 + * 2) If -1<=x<=1, then r=x, PIO2=0 + * 3) If x<-1, then r=-1/x, PIO2=-Pi/2 + */ + movups _sONE+__svml_satan_data_internal(%rip), %xmm1 + andps %xmm0, %xmm2 + movaps %xmm2, %xmm9 + movaps %xmm1, %xmm3 + cmpleps %xmm1, %xmm9 + maxps %xmm2, %xmm3 + minps %xmm2, %xmm1 + divps %xmm3, %xmm1 + movups __svml_satan_data_internal(%rip), %xmm4 + movaps %xmm9, %xmm10 + andps %xmm4, %xmm0 + andnps %xmm4, %xmm9 + pxor %xmm0, %xmm9 + pxor %xmm1, %xmm9 + +/* Polynomial. */ + movaps %xmm9, %xmm8 + mulps %xmm9, %xmm8 + movaps %xmm8, %xmm7 + mulps %xmm8, %xmm7 + movups _sPC8+__svml_satan_data_internal(%rip), %xmm6 + mulps %xmm7, %xmm6 + movups _sPC7+__svml_satan_data_internal(%rip), %xmm5 + mulps %xmm7, %xmm5 + addps _sPC6+__svml_satan_data_internal(%rip), %xmm6 + mulps %xmm7, %xmm6 + addps _sPC5+__svml_satan_data_internal(%rip), %xmm5 + mulps %xmm7, %xmm5 + addps _sPC4+__svml_satan_data_internal(%rip), %xmm6 + mulps %xmm7, %xmm6 + addps _sPC3+__svml_satan_data_internal(%rip), %xmm5 + mulps %xmm5, %xmm7 + addps _sPC2+__svml_satan_data_internal(%rip), %xmm6 + mulps %xmm8, %xmm6 + addps _sPC1+__svml_satan_data_internal(%rip), %xmm7 + andnps _sPIO2+__svml_satan_data_internal(%rip), %xmm10 + addps %xmm6, %xmm7 + mulps %xmm7, %xmm8 + pxor %xmm0, %xmm10 + addps _sPC0+__svml_satan_data_internal(%rip), %xmm8 + +/* Reconstruction. */ + mulps %xmm8, %xmm9 + addps %xmm9, %xmm10 + movaps %xmm10, %xmm0 + ret + +END(_ZGVbN4v_atanf_sse4) + + .section .rodata, "a" + .align 16 + +#ifdef __svml_satan_data_internal_typedef +typedef unsigned int VUINT32; +typedef struct { + __declspec(align(16)) VUINT32 _sSIGN_MASK[4][1]; + __declspec(align(16)) VUINT32 _sABS_MASK[4][1]; + __declspec(align(16)) VUINT32 _sONE[4][1]; + __declspec(align(16)) VUINT32 _sPIO2[4][1]; + __declspec(align(16)) VUINT32 _sPC8[4][1]; + __declspec(align(16)) VUINT32 _sPC7[4][1]; + __declspec(align(16)) VUINT32 _sPC6[4][1]; + __declspec(align(16)) VUINT32 _sPC5[4][1]; + __declspec(align(16)) VUINT32 _sPC4[4][1]; + __declspec(align(16)) VUINT32 _sPC3[4][1]; + __declspec(align(16)) VUINT32 _sPC2[4][1]; + __declspec(align(16)) VUINT32 _sPC1[4][1]; + __declspec(align(16)) VUINT32 _sPC0[4][1]; +} __svml_satan_data_internal; +#endif +__svml_satan_data_internal: + .long 0x80000000, 0x80000000, 0x80000000, 0x80000000 //_sSIGN_MASK + .align 16 + .long 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF //_sABS_MASK + .align 16 + .long 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 //_sONE + .align 16 + .long 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB //_sPIO2 + .align 16 + .long 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0 //_sPC8 + .align 16 + .long 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631 //_sPC7 + .align 16 + .long 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384 //_sPC6 + .align 16 + .long 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629 //_sPC5 + .align 16 + .long 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474 //_sPC4 + .align 16 + .long 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8 //_sPC3 + .align 16 + .long 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F //_sPC2 + .align 16 + .long 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49 //_sPC1 + .align 16 + .long 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 //_sPC0 + .align 16 + .type __svml_satan_data_internal,@object + .size __svml_satan_data_internal,.-__svml_satan_data_internal diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atanf8_core-sse.S b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf8_core-sse.S new file mode 100644 index 0000000..1652a8f --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf8_core-sse.S @@ -0,0 +1,20 @@ +/* SSE version of vectorized atanf, 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 + <https://www.gnu.org/licenses/>. */ + +#define _ZGVdN8v_atanf _ZGVdN8v_atanf_sse_wrapper +#include "../svml_s_atanf8_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atanf8_core.c b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf8_core.c new file mode 100644 index 0000000..733d8c3 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf8_core.c @@ -0,0 +1,28 @@ +/* Multiple versions of vectorized atanf, 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 + <https://www.gnu.org/licenses/>. */ + +#define SYMBOL_NAME _ZGVdN8v_atanf +#include "ifunc-mathvec-avx2.h" + +libc_ifunc_redirected (REDIRECT_NAME, SYMBOL_NAME, IFUNC_SELECTOR ()); + +#ifdef SHARED +__hidden_ver1 (_ZGVdN8v_atanf, __GI__ZGVdN8v_atanf, + __redirect__ZGVdN8v_atanf) + __attribute__ ((visibility ("hidden"))); +#endif diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_atanf8_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf8_core_avx2.S new file mode 100644 index 0000000..e333f97 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_atanf8_core_avx2.S @@ -0,0 +1,148 @@ +/* Function atanf 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: + * + * For 0.0 <= x <= 7.0/16.0: atan(x) = atan(0.0) + atan(s), where s=(x-0.0)/(1.0+0.0*x) + * For 7.0/16.0 <= x <= 11.0/16.0: atan(x) = atan(0.5) + atan(s), where s=(x-0.5)/(1.0+0.5*x) + * For 11.0/16.0 <= x <= 19.0/16.0: atan(x) = atan(1.0) + atan(s), where s=(x-1.0)/(1.0+1.0*x) + * For 19.0/16.0 <= x <= 39.0/16.0: atan(x) = atan(1.5) + atan(s), where s=(x-1.5)/(1.0+1.5*x) + * For 39.0/16.0 <= x <= inf : atan(x) = atan(inf) + atan(s), where s=-1.0/x + * Where atan(s) ~= s+s^3*Poly11(s^2) on interval |s|<7.0/0.16. + * + */ + +/* Offsets for data table __svml_satan_data_internal + */ +#define _sSIGN_MASK 0 +#define _sABS_MASK 32 +#define _sONE 64 +#define _sPIO2 96 +#define _sPC8 128 +#define _sPC7 160 +#define _sPC6 192 +#define _sPC5 224 +#define _sPC4 256 +#define _sPC3 288 +#define _sPC2 320 +#define _sPC1 352 +#define _sPC0 384 + +#include <sysdep.h> + + .text + .section .text.avx2,"ax",@progbits +ENTRY(_ZGVdN8v_atanf_avx2) +/* + * 1) If x>1, then r=-1/x, PIO2=Pi/2 + * 2) If -1<=x<=1, then r=x, PIO2=0 + * 3) If x<-1, then r=-1/x, PIO2=-Pi/2 + */ + vmovups _sONE+__svml_satan_data_internal(%rip), %ymm2 + vmovups __svml_satan_data_internal(%rip), %ymm7 + vmovups _sPC7+__svml_satan_data_internal(%rip), %ymm13 + +/* + * To use minps\maxps operations for argument reduction + * uncomment _AT_USEMINMAX_ definition + * Declarations + * Variables + * Constants + */ + vandps _sABS_MASK+__svml_satan_data_internal(%rip), %ymm0, %ymm3 + vmaxps %ymm3, %ymm2, %ymm5 + vminps %ymm3, %ymm2, %ymm4 + vcmple_oqps %ymm2, %ymm3, %ymm6 + vdivps %ymm5, %ymm4, %ymm11 + vandps %ymm7, %ymm0, %ymm9 + vandnps %ymm7, %ymm6, %ymm8 + vxorps %ymm9, %ymm8, %ymm10 + vxorps %ymm11, %ymm10, %ymm15 + +/* Polynomial. */ + vmulps %ymm15, %ymm15, %ymm14 + vmovups _sPC8+__svml_satan_data_internal(%rip), %ymm0 + vmulps %ymm14, %ymm14, %ymm12 + vfmadd213ps _sPC6+__svml_satan_data_internal(%rip), %ymm12, %ymm0 + vfmadd213ps _sPC5+__svml_satan_data_internal(%rip), %ymm12, %ymm13 + vfmadd213ps _sPC4+__svml_satan_data_internal(%rip), %ymm12, %ymm0 + vfmadd213ps _sPC3+__svml_satan_data_internal(%rip), %ymm12, %ymm13 + vfmadd213ps _sPC2+__svml_satan_data_internal(%rip), %ymm12, %ymm0 + vfmadd213ps _sPC1+__svml_satan_data_internal(%rip), %ymm12, %ymm13 + vfmadd213ps %ymm13, %ymm14, %ymm0 + vfmadd213ps _sPC0+__svml_satan_data_internal(%rip), %ymm14, %ymm0 + vandnps _sPIO2+__svml_satan_data_internal(%rip), %ymm6, %ymm1 + vxorps %ymm9, %ymm1, %ymm1 + +/* Reconstruction. */ + vfmadd213ps %ymm1, %ymm15, %ymm0 + ret + +END(_ZGVdN8v_atanf_avx2) + + .section .rodata, "a" + .align 32 + +#ifdef __svml_satan_data_internal_typedef +typedef unsigned int VUINT32; +typedef struct { + __declspec(align(32)) VUINT32 _sSIGN_MASK[8][1]; + __declspec(align(32)) VUINT32 _sABS_MASK[8][1]; + __declspec(align(32)) VUINT32 _sONE[8][1]; + __declspec(align(32)) VUINT32 _sPIO2[8][1]; + __declspec(align(32)) VUINT32 _sPC8[8][1]; + __declspec(align(32)) VUINT32 _sPC7[8][1]; + __declspec(align(32)) VUINT32 _sPC6[8][1]; + __declspec(align(32)) VUINT32 _sPC5[8][1]; + __declspec(align(32)) VUINT32 _sPC4[8][1]; + __declspec(align(32)) VUINT32 _sPC3[8][1]; + __declspec(align(32)) VUINT32 _sPC2[8][1]; + __declspec(align(32)) VUINT32 _sPC1[8][1]; + __declspec(align(32)) VUINT32 _sPC0[8][1]; +} __svml_satan_data_internal; +#endif +__svml_satan_data_internal: + .long 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000 //_sSIGN_MASK + .align 32 + .long 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF //_sABS_MASK + .align 32 + .long 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 //_sONE + .align 32 + .long 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB //_sPIO2 + .align 32 + .long 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0, 0x3B322CC0 //_sPC8 + .align 32 + .long 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631, 0xBC7F2631 //_sPC7 + .align 32 + .long 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384, 0x3D2BC384 //_sPC6 + .align 32 + .long 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629, 0xBD987629 //_sPC5 + .align 32 + .long 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474, 0x3DD96474 //_sPC4 + .align 32 + .long 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8, 0xBE1161F8 //_sPC3 + .align 32 + .long 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F, 0x3E4CB79F //_sPC2 + .align 32 + .long 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49, 0xBEAAAA49 //_sPC1 + .align 32 + .long 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 //_sPC0 + .align 32 + .type __svml_satan_data_internal,@object + .size __svml_satan_data_internal,.-__svml_satan_data_internal diff --git a/sysdeps/x86_64/fpu/svml_d_atan2_core.S b/sysdeps/x86_64/fpu/svml_d_atan2_core.S new file mode 100644 index 0000000..e86d5b7 --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_d_atan2_core.S @@ -0,0 +1,29 @@ +/* Function atan 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 + <https://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include "svml_d_wrapper_impl.h" + + .text +ENTRY (_ZGVbN2v_atan) +WRAPPER_IMPL_SSE2 atan +END (_ZGVbN2v_atan) + +#ifndef USE_MULTIARCH + libmvec_hidden_def (_ZGVbN2v_atan) +#endif diff --git a/sysdeps/x86_64/fpu/svml_d_atan4_core.S b/sysdeps/x86_64/fpu/svml_d_atan4_core.S new file mode 100644 index 0000000..eb11fd2 --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_d_atan4_core.S @@ -0,0 +1,29 @@ +/* Function atan 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 + <https://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include "svml_d_wrapper_impl.h" + + .text +ENTRY (_ZGVdN4v_atan) +WRAPPER_IMPL_AVX _ZGVbN2v_atan +END (_ZGVdN4v_atan) + +#ifndef USE_MULTIARCH + libmvec_hidden_def (_ZGVdN4v_atan) +#endif diff --git a/sysdeps/x86_64/fpu/svml_d_atan4_core_avx.S b/sysdeps/x86_64/fpu/svml_d_atan4_core_avx.S new file mode 100644 index 0000000..b83a4be --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_d_atan4_core_avx.S @@ -0,0 +1,25 @@ +/* Function atan 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 + <https://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include "svml_d_wrapper_impl.h" + + .text +ENTRY (_ZGVcN4v_atan) +WRAPPER_IMPL_AVX _ZGVbN2v_atan +END (_ZGVcN4v_atan) diff --git a/sysdeps/x86_64/fpu/svml_d_atan8_core.S b/sysdeps/x86_64/fpu/svml_d_atan8_core.S new file mode 100644 index 0000000..9685a32 --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_d_atan8_core.S @@ -0,0 +1,25 @@ +/* Function atan 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 + <https://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include "svml_d_wrapper_impl.h" + + .text +ENTRY (_ZGVeN8v_atan) +WRAPPER_IMPL_AVX512 _ZGVdN4v_atan +END (_ZGVeN8v_atan) diff --git a/sysdeps/x86_64/fpu/svml_s_atanf16_core.S b/sysdeps/x86_64/fpu/svml_s_atanf16_core.S new file mode 100644 index 0000000..f82d242 --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_s_atanf16_core.S @@ -0,0 +1,25 @@ +/* Function atanf 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 + <https://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include "svml_s_wrapper_impl.h" + + .text +ENTRY (_ZGVeN16v_atanf) +WRAPPER_IMPL_AVX512 _ZGVdN8v_atanf +END (_ZGVeN16v_atanf) diff --git a/sysdeps/x86_64/fpu/svml_s_atanf4_core.S b/sysdeps/x86_64/fpu/svml_s_atanf4_core.S new file mode 100644 index 0000000..6b8c4d9 --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_s_atanf4_core.S @@ -0,0 +1,29 @@ +/* Function atanf 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 + <https://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include "svml_s_wrapper_impl.h" + + .text +ENTRY (_ZGVbN4v_atanf) +WRAPPER_IMPL_SSE2 atanf +END (_ZGVbN4v_atanf) + +#ifndef USE_MULTIARCH + libmvec_hidden_def (_ZGVbN4v_atanf) +#endif diff --git a/sysdeps/x86_64/fpu/svml_s_atanf8_core.S b/sysdeps/x86_64/fpu/svml_s_atanf8_core.S new file mode 100644 index 0000000..315681f --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_s_atanf8_core.S @@ -0,0 +1,29 @@ +/* Function atanf 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 + <https://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include "svml_s_wrapper_impl.h" + + .text +ENTRY (_ZGVdN8v_atanf) +WRAPPER_IMPL_AVX _ZGVbN4v_atanf +END (_ZGVdN8v_atanf) + +#ifndef USE_MULTIARCH + libmvec_hidden_def (_ZGVdN8v_atanf) +#endif diff --git a/sysdeps/x86_64/fpu/svml_s_atanf8_core_avx.S b/sysdeps/x86_64/fpu/svml_s_atanf8_core_avx.S new file mode 100644 index 0000000..b9cd502 --- /dev/null +++ b/sysdeps/x86_64/fpu/svml_s_atanf8_core_avx.S @@ -0,0 +1,25 @@ +/* Function atanf 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 + <https://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include "svml_s_wrapper_impl.h" + + .text +ENTRY (_ZGVcN8v_atanf) +WRAPPER_IMPL_AVX _ZGVbN4v_atanf +END (_ZGVcN8v_atanf) diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-atan-avx.c b/sysdeps/x86_64/fpu/test-double-libmvec-atan-avx.c new file mode 100644 index 0000000..0f7176a --- /dev/null +++ b/sysdeps/x86_64/fpu/test-double-libmvec-atan-avx.c @@ -0,0 +1 @@ +#include "test-double-libmvec-atan.c" diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-atan-avx2.c b/sysdeps/x86_64/fpu/test-double-libmvec-atan-avx2.c new file mode 100644 index 0000000..0f7176a --- /dev/null +++ b/sysdeps/x86_64/fpu/test-double-libmvec-atan-avx2.c @@ -0,0 +1 @@ +#include "test-double-libmvec-atan.c" diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-atan-avx512f.c b/sysdeps/x86_64/fpu/test-double-libmvec-atan-avx512f.c new file mode 100644 index 0000000..0f7176a --- /dev/null +++ b/sysdeps/x86_64/fpu/test-double-libmvec-atan-avx512f.c @@ -0,0 +1 @@ +#include "test-double-libmvec-atan.c" diff --git a/sysdeps/x86_64/fpu/test-double-libmvec-atan.c b/sysdeps/x86_64/fpu/test-double-libmvec-atan.c new file mode 100644 index 0000000..982687b --- /dev/null +++ b/sysdeps/x86_64/fpu/test-double-libmvec-atan.c @@ -0,0 +1,3 @@ +#define LIBMVEC_TYPE double +#define LIBMVEC_FUNC atan +#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 0abc7d2..467c913 100644 --- a/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c +++ b/sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c @@ -28,6 +28,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (log), _ZGVbN2v_log) VECTOR_WRAPPER (WRAPPER_NAME (exp), _ZGVbN2v_exp) VECTOR_WRAPPER_ff (WRAPPER_NAME (pow), _ZGVbN2vv_pow) VECTOR_WRAPPER (WRAPPER_NAME (acos), _ZGVbN2v_acos) +VECTOR_WRAPPER (WRAPPER_NAME (atan), _ZGVbN2v_atan) #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 dda093b..b72a7de 100644 --- a/sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c +++ b/sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c @@ -31,6 +31,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (log), _ZGVdN4v_log) VECTOR_WRAPPER (WRAPPER_NAME (exp), _ZGVdN4v_exp) VECTOR_WRAPPER_ff (WRAPPER_NAME (pow), _ZGVdN4vv_pow) VECTOR_WRAPPER (WRAPPER_NAME (acos), _ZGVdN4v_acos) +VECTOR_WRAPPER (WRAPPER_NAME (atan), _ZGVdN4v_atan) #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 f323046..d2434df 100644 --- a/sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c +++ b/sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c @@ -28,6 +28,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (log), _ZGVcN4v_log) VECTOR_WRAPPER (WRAPPER_NAME (exp), _ZGVcN4v_exp) VECTOR_WRAPPER_ff (WRAPPER_NAME (pow), _ZGVcN4vv_pow) VECTOR_WRAPPER (WRAPPER_NAME (acos), _ZGVcN4v_acos) +VECTOR_WRAPPER (WRAPPER_NAME (atan), _ZGVcN4v_atan) #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 cf9f52f..f7aaf81 100644 --- a/sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c +++ b/sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c @@ -28,6 +28,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (log), _ZGVeN8v_log) VECTOR_WRAPPER (WRAPPER_NAME (exp), _ZGVeN8v_exp) VECTOR_WRAPPER_ff (WRAPPER_NAME (pow), _ZGVeN8vv_pow) VECTOR_WRAPPER (WRAPPER_NAME (acos), _ZGVeN8v_acos) +VECTOR_WRAPPER (WRAPPER_NAME (atan), _ZGVeN8v_atan) #ifndef __ILP32__ # define VEC_INT_TYPE __m512i diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-atanf-avx.c b/sysdeps/x86_64/fpu/test-float-libmvec-atanf-avx.c new file mode 100644 index 0000000..9251c65 --- /dev/null +++ b/sysdeps/x86_64/fpu/test-float-libmvec-atanf-avx.c @@ -0,0 +1 @@ +#include "test-float-libmvec-atanf.c" diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-atanf-avx2.c b/sysdeps/x86_64/fpu/test-float-libmvec-atanf-avx2.c new file mode 100644 index 0000000..9251c65 --- /dev/null +++ b/sysdeps/x86_64/fpu/test-float-libmvec-atanf-avx2.c @@ -0,0 +1 @@ +#include "test-float-libmvec-atanf.c" diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-atanf-avx512f.c b/sysdeps/x86_64/fpu/test-float-libmvec-atanf-avx512f.c new file mode 100644 index 0000000..9251c65 --- /dev/null +++ b/sysdeps/x86_64/fpu/test-float-libmvec-atanf-avx512f.c @@ -0,0 +1 @@ +#include "test-float-libmvec-atanf.c" diff --git a/sysdeps/x86_64/fpu/test-float-libmvec-atanf.c b/sysdeps/x86_64/fpu/test-float-libmvec-atanf.c new file mode 100644 index 0000000..2a8ab87 --- /dev/null +++ b/sysdeps/x86_64/fpu/test-float-libmvec-atanf.c @@ -0,0 +1,3 @@ +#define LIBMVEC_TYPE float +#define LIBMVEC_FUNC atanf +#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 abbd3ed..af769c5 100644 --- a/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c +++ b/sysdeps/x86_64/fpu/test-float-vlen16-wrappers.c @@ -28,6 +28,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (logf), _ZGVeN16v_logf) VECTOR_WRAPPER (WRAPPER_NAME (expf), _ZGVeN16v_expf) VECTOR_WRAPPER_ff (WRAPPER_NAME (powf), _ZGVeN16vv_powf) VECTOR_WRAPPER (WRAPPER_NAME (acosf), _ZGVeN16v_acosf) +VECTOR_WRAPPER (WRAPPER_NAME (atanf), _ZGVeN16v_atanf) #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 8a24027..76e61d2 100644 --- a/sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c +++ b/sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c @@ -28,6 +28,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (logf), _ZGVbN4v_logf) VECTOR_WRAPPER (WRAPPER_NAME (expf), _ZGVbN4v_expf) VECTOR_WRAPPER_ff (WRAPPER_NAME (powf), _ZGVbN4vv_powf) VECTOR_WRAPPER (WRAPPER_NAME (acosf), _ZGVbN4v_acosf) +VECTOR_WRAPPER (WRAPPER_NAME (atanf), _ZGVbN4v_atanf) #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 aff0442..5e27eaa 100644 --- a/sysdeps/x86_64/fpu/test-float-vlen8-avx2-wrappers.c +++ b/sysdeps/x86_64/fpu/test-float-vlen8-avx2-wrappers.c @@ -31,6 +31,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (logf), _ZGVdN8v_logf) VECTOR_WRAPPER (WRAPPER_NAME (expf), _ZGVdN8v_expf) VECTOR_WRAPPER_ff (WRAPPER_NAME (powf), _ZGVdN8vv_powf) VECTOR_WRAPPER (WRAPPER_NAME (acosf), _ZGVdN8v_acosf) +VECTOR_WRAPPER (WRAPPER_NAME (atanf), _ZGVdN8v_atanf) /* 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 913584d..28daf79 100644 --- a/sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c +++ b/sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c @@ -28,6 +28,7 @@ VECTOR_WRAPPER (WRAPPER_NAME (logf), _ZGVcN8v_logf) VECTOR_WRAPPER (WRAPPER_NAME (expf), _ZGVcN8v_expf) VECTOR_WRAPPER_ff (WRAPPER_NAME (powf), _ZGVcN8vv_powf) VECTOR_WRAPPER (WRAPPER_NAME (acosf), _ZGVcN8v_acosf) +VECTOR_WRAPPER (WRAPPER_NAME (atanf), _ZGVcN8v_atanf) #define VEC_INT_TYPE __m128i |