Vector exp for x86_64 and tests.

Here is implementation of vectorized exp containing SSE, AVX,
AVX2 and AVX512 versions according to Vector ABI
<https://groups.google.com/forum/#!topic/x86-64-abi/LmppCfN1rZ4>.

    * bits/libm-simd-decl-stubs.h: Added stubs for exp.
    * math/bits/mathcalls.h: Added exp declaration with __MATHCALL_VEC.
    * sysdeps/unix/sysv/linux/x86_64/libmvec.abilist: New versions added.
    * sysdeps/x86/fpu/bits/math-vector.h: Added SIMD declaration and asm
    redirections for exp.
    * sysdeps/x86_64/fpu/Makefile (libmvec-support): Added new files.
    * sysdeps/x86_64/fpu/Versions: New versions added.
    * sysdeps/x86_64/fpu/libm-test-ulps: Regenerated.
    * sysdeps/x86_64/fpu/multiarch/Makefile (libmvec-sysdep_routines): Added
    build of SSE, AVX2 and AVX512 IFUNC versions.
    * sysdeps/x86_64/fpu/multiarch/svml_d_exp2_core.S: New file.
    * sysdeps/x86_64/fpu/multiarch/svml_d_exp2_core_sse4.S: New file.
    * sysdeps/x86_64/fpu/multiarch/svml_d_exp4_core.S: New file.
    * sysdeps/x86_64/fpu/multiarch/svml_d_exp4_core_avx2.S: New file.
    * sysdeps/x86_64/fpu/multiarch/svml_d_exp8_core.S: New file.
    * sysdeps/x86_64/fpu/multiarch/svml_d_exp8_core_avx512.S: New file.
    * sysdeps/x86_64/fpu/svml_d_exp2_core.S: New file.
    * sysdeps/x86_64/fpu/svml_d_exp4_core.S: New file.
    * sysdeps/x86_64/fpu/svml_d_exp4_core_avx.S: New file.
    * sysdeps/x86_64/fpu/svml_d_exp8_core.S: New file.
    * sysdeps/x86_64/fpu/svml_d_exp_data.S: New file.
    * sysdeps/x86_64/fpu/svml_d_exp_data.h: New file.
    * sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c: Added vector exp test.
    * sysdeps/x86_64/fpu/test-double-vlen2.c: Likewise.
    * sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c: Likewise.
    * sysdeps/x86_64/fpu/test-double-vlen4-avx2.c: Likewise.
    * sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c: Likewise.
    * sysdeps/x86_64/fpu/test-double-vlen4.c: Likewise.
    * sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c: Likewise.
    * sysdeps/x86_64/fpu/test-double-vlen8.c: Likewise.
    * NEWS: Mention addition of x86_64 vector exp.

7 files changed, 1010 insertions, 1 deletions

diff --git a/sysdeps/x86_64/fpu/multiarch/Makefile b/sysdeps/x86_64/fpu/multiarch/Makefile
index 5fc6ea3d23..d6355ae98d 100644
--- a/sysdeps/x86_64/fpu/multiarch/Makefile
+++ b/sysdeps/x86_64/fpu/multiarch/Makefile
@@ -62,5 +62,6 @@ libmvec-sysdep_routines += svml_d_cos2_core_sse4 svml_d_cos4_core_avx2 \
 			   svml_s_cosf16_core_avx512 svml_s_sinf4_core_sse4 \
 			   svml_s_sinf8_core_avx2 svml_s_sinf16_core_avx512 \
 			   svml_s_logf4_core_sse4 svml_s_logf8_core_avx2 \
-			   svml_s_logf16_core_avx512
+			   svml_s_logf16_core_avx512 svml_d_exp2_core_sse4 \
+			   svml_d_exp4_core_avx2 svml_d_exp8_core_avx512
 endif
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp2_core.S b/sysdeps/x86_64/fpu/multiarch/svml_d_exp2_core.S
new file mode 100644
index 0000000000..ef3dc49a1c
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp2_core.S
@@ -0,0 +1,38 @@
+/* Multiple versions of vectorized exp.
+   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
+   <http://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include <init-arch.h>
+
+	.text
+ENTRY (_ZGVbN2v_exp)
+        .type   _ZGVbN2v_exp, @gnu_indirect_function
+        cmpl    $0, KIND_OFFSET+__cpu_features(%rip)
+        jne     1f
+        call    __init_cpu_features
+1:      leaq    _ZGVbN2v_exp_sse4(%rip), %rax
+        testl   $bit_SSE4_1, __cpu_features+CPUID_OFFSET+index_SSE4_1(%rip)
+        jz      2f
+        ret
+2:      leaq    _ZGVbN2v_exp_sse2(%rip), %rax
+        ret
+END (_ZGVbN2v_exp)
+libmvec_hidden_def (_ZGVbN2v_exp)
+
+#define _ZGVbN2v_exp _ZGVbN2v_exp_sse2
+#include "../svml_d_exp2_core.S"
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp2_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_d_exp2_core_sse4.S
new file mode 100644
index 0000000000..1f5445924a
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp2_core_sse4.S
@@ -0,0 +1,225 @@
+/* Function exp 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
+   <http://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include "svml_d_exp_data.h"
+
+	.text
+ENTRY (_ZGVbN2v_exp_sse4)
+/*
+   ALGORITHM DESCRIPTION:
+
+     Argument representation:
+     N = rint(X*2^k/ln2) = 2^k*M+j
+     X = N*ln2/2^k + r = M*ln2 + ln2*(j/2^k) + r
+     then -ln2/2^(k+1) < r < ln2/2^(k+1)
+     Alternatively:
+     N = trunc(X*2^k/ln2)
+     then 0 < r < ln2/2^k
+
+     Result calculation:
+     exp(X) = exp(M*ln2 + ln2*(j/2^k) + r)
+     = 2^M * 2^(j/2^k) * exp(r)
+     2^M is calculated by bit manipulation
+     2^(j/2^k) is stored in table
+     exp(r) is approximated by polynomial.
+
+     The table lookup is skipped if k = 0.  */
+
+        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, %xmm3
+        movq      __svml_dexp_data@GOTPCREL(%rip), %r8
+
+/* iAbsX = (int)(lX>>32), lX = *(longlong*)&X */
+        pshufd    $221, %xmm3, %xmm7
+        movups __dbInvLn2(%r8), %xmm0
+
+/* dK = X*dbInvLn2 */
+        mulpd     %xmm3, %xmm0
+        movq __iAbsMask(%r8), %xmm5
+        movq __iDomainRange(%r8), %xmm6
+
+/* iAbsX = iAbsX&iAbsMask */
+        pand      %xmm5, %xmm7
+
+/* iRangeMask = (iAbsX>iDomainRange) */
+        pcmpgtd   %xmm6, %xmm7
+
+/* Mask = iRangeMask?1:0, set mask for overflow/underflow */
+        movmskps  %xmm7, %eax
+
+/* dN = rint(X*2^k/Ln2) */
+        xorps     %xmm7, %xmm7
+        movups __dbLn2hi(%r8), %xmm5
+        movups __dbLn2lo(%r8), %xmm6
+        roundpd   $0, %xmm0, %xmm7
+
+/* dR = X - dN*dbLn2hi, dbLn2hi is 52-8-k hi bits of ln2/2^k */
+        mulpd     %xmm7, %xmm5
+
+/* dR = dR - dN*dbLn2lo, dbLn2lo is 40..94 bits of lo part of ln2/2^k */
+        mulpd     %xmm6, %xmm7
+        movups __dbShifter(%r8), %xmm4
+
+/* dM = X*dbInvLn2+dbShifter */
+        addpd     %xmm0, %xmm4
+        movaps    %xmm3, %xmm0
+        subpd     %xmm5, %xmm0
+        subpd     %xmm7, %xmm0
+        movups __dPC2(%r8), %xmm5
+
+/* exp(r) = b0+r*(b0+r*(b1+r*b2)) */
+        mulpd     %xmm0, %xmm5
+        addpd __dPC1(%r8), %xmm5
+        mulpd     %xmm0, %xmm5
+        movups __dPC0(%r8), %xmm6
+        addpd     %xmm6, %xmm5
+        mulpd     %xmm5, %xmm0
+        movdqu __lIndexMask(%r8), %xmm2
+
+/* lIndex = (*(longlong*)&dM)&lIndexMask, lIndex is the lower K bits of lM */
+        movdqa    %xmm2, %xmm1
+
+/* lM = (*(longlong*)&dM)&(~lIndexMask) */
+        pandn     %xmm4, %xmm2
+        pand      %xmm4, %xmm1
+
+/* lM = lM<<(52-K), 2^M */
+        psllq     $42, %xmm2
+
+/* table lookup for dT[j] = 2^(j/2^k) */
+        movd      %xmm1, %edx
+        pextrw    $4, %xmm1, %ecx
+        addpd     %xmm0, %xmm6
+        shll      $3, %edx
+        shll      $3, %ecx
+        movq      (%r8,%rdx), %xmm0
+        andl      $3, %eax
+        movhpd    (%r8,%rcx), %xmm0
+
+/* 2^(j/2^k) * exp(r) */
+        mulpd     %xmm6, %xmm0
+
+/* multiply by 2^M through integer add */
+        paddq     %xmm2, %xmm0
+        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    %xmm3, 192(%rsp)
+        movups    %xmm0, 256(%rsp)
+        je        .LBL_1_2
+
+        xorb      %cl, %cl
+        xorl      %edx, %edx
+        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      %cl, %r12b
+        movq      %r13, 160(%rsp)
+        cfi_offset_rel_rsp (13, 160)
+        movl      %eax, %r13d
+        movq      %r14, 152(%rsp)
+        cfi_offset_rel_rsp (14, 152)
+        movl      %edx, %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
+        shlq      $4, %r15
+        movsd     200(%rsp,%r15), %xmm0
+
+        call      exp@PLT
+
+        movsd     %xmm0, 264(%rsp,%r15)
+        jmp       .LBL_1_8
+
+.LBL_1_12:
+        movzbl    %r12b, %r15d
+        shlq      $4, %r15
+        movsd     192(%rsp,%r15), %xmm0
+
+        call      exp@PLT
+
+        movsd     %xmm0, 256(%rsp,%r15)
+        jmp       .LBL_1_7
+
+END (_ZGVbN2v_exp_sse4)
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp4_core.S b/sysdeps/x86_64/fpu/multiarch/svml_d_exp4_core.S
new file mode 100644
index 0000000000..7f2ebdef67
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp4_core.S
@@ -0,0 +1,38 @@
+/* Multiple versions of vectorized exp.
+   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
+   <http://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include <init-arch.h>
+
+	.text
+ENTRY (_ZGVdN4v_exp)
+        .type   _ZGVdN4v_exp, @gnu_indirect_function
+        cmpl    $0, KIND_OFFSET+__cpu_features(%rip)
+        jne     1f
+        call    __init_cpu_features
+1:      leaq    _ZGVdN4v_exp_avx2(%rip), %rax
+        testl   $bit_AVX2_Usable, __cpu_features+FEATURE_OFFSET+index_AVX2_Usable(%rip)
+        jz      2f
+        ret
+2:      leaq    _ZGVdN4v_exp_sse_wrapper(%rip), %rax
+        ret
+END (_ZGVdN4v_exp)
+libmvec_hidden_def (_ZGVdN4v_exp)
+
+#define _ZGVdN4v_exp _ZGVdN4v_exp_sse_wrapper
+#include "../svml_d_exp4_core.S"
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp4_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_exp4_core_avx2.S
new file mode 100644
index 0000000000..a34e267433
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp4_core_avx2.S
@@ -0,0 +1,212 @@
+/* Function exp 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
+   <http://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include "svml_d_exp_data.h"
+
+	.text
+ENTRY (_ZGVdN4v_exp_avx2)
+/*
+   ALGORITHM DESCRIPTION:
+
+     Argument representation:
+     N = rint(X*2^k/ln2) = 2^k*M+j
+     X = N*ln2/2^k + r = M*ln2 + ln2*(j/2^k) + r
+     then -ln2/2^(k+1) < r < ln2/2^(k+1)
+     Alternatively:
+     N = trunc(X*2^k/ln2)
+     then 0 < r < ln2/2^k
+
+     Result calculation:
+     exp(X) = exp(M*ln2 + ln2*(j/2^k) + r)
+     = 2^M * 2^(j/2^k) * exp(r)
+     2^M is calculated by bit manipulation
+     2^(j/2^k) is stored in table
+     exp(r) is approximated by polynomial
+
+     The table lookup is skipped if k = 0.  */
+
+        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_dexp_data@GOTPCREL(%rip), %rax
+        vmovdqa   %ymm0, %ymm2
+        vmovupd __dbInvLn2(%rax), %ymm3
+        vmovupd __dbShifter(%rax), %ymm1
+        vmovupd __lIndexMask(%rax), %ymm4
+
+/* dM = X*dbInvLn2+dbShifter, dbInvLn2 = 2^k/Ln2 */
+        vfmadd213pd %ymm1, %ymm2, %ymm3
+
+/* iAbsX = (int)(lX>>32), lX = *(longlong*)&X */
+        vextracti128 $1, %ymm2, %xmm5
+        vshufps   $221, %xmm5, %xmm2, %xmm6
+
+/* iAbsX = iAbsX&iAbsMask */
+        vandps __iAbsMask(%rax), %xmm6, %xmm7
+
+/* dN = dM-dbShifter, dN = rint(X*2^k/Ln2) */
+        vsubpd    %ymm1, %ymm3, %ymm6
+
+/* iRangeMask = (iAbsX>iDomainRange) */
+        vpcmpgtd __iDomainRange(%rax), %xmm7, %xmm0
+        vmovupd __dbLn2hi(%rax), %ymm1
+        vmovupd __dPC0(%rax), %ymm7
+
+/* Mask = iRangeMask?1:0, set mask for overflow/underflow */
+        vmovmskps %xmm0, %ecx
+        vmovupd __dPC2(%rax), %ymm0
+
+/* dR = X - dN*dbLn2hi, dbLn2hi is 52-8-k hi bits of ln2/2^k */
+        vmovdqa   %ymm2, %ymm5
+        vfnmadd231pd %ymm6, %ymm1, %ymm5
+
+/* dR = dR - dN*dbLn2lo, dbLn2lo is 40..94 bits of lo part of ln2/2^k */
+        vfnmadd132pd __dbLn2lo(%rax), %ymm5, %ymm6
+
+/* exp(r) = b0+r*(b0+r*(b1+r*b2)) */
+        vfmadd213pd __dPC1(%rax), %ymm6, %ymm0
+        vfmadd213pd %ymm7, %ymm6, %ymm0
+        vfmadd213pd %ymm7, %ymm6, %ymm0
+
+/* lIndex = (*(longlong*)&dM)&lIndexMask, lIndex is the lower K bits of lM */
+        vandps    %ymm4, %ymm3, %ymm1
+
+/* table lookup for dT[j] = 2^(j/2^k) */
+        vxorpd    %ymm6, %ymm6, %ymm6
+        vpcmpeqd  %ymm5, %ymm5, %ymm5
+        vgatherqpd %ymm5, (%rax,%ymm1,8), %ymm6
+
+/* lM = (*(longlong*)&dM)&(~lIndexMask) */
+        vpandn    %ymm3, %ymm4, %ymm3
+
+/* 2^(j/2^k) * exp(r) */
+        vmulpd    %ymm0, %ymm6, %ymm0
+
+/* lM = lM<<(52-K), 2^M */
+        vpsllq    $42, %ymm3, %ymm4
+
+/* multiply by 2^M through integer add */
+        vpaddq    %ymm4, %ymm0, %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
+        vmovupd   %ymm2, 320(%rsp)
+        vmovupd   %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
+        vmovupd   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
+        shlq      $4, %r15
+        vmovsd    328(%rsp,%r15), %xmm0
+        vzeroupper
+
+        call      exp@PLT
+
+        vmovsd    %xmm0, 392(%rsp,%r15)
+        jmp       .LBL_1_8
+
+.LBL_1_12:
+        movzbl    %r12b, %r15d
+        shlq      $4, %r15
+        vmovsd    320(%rsp,%r15), %xmm0
+        vzeroupper
+
+        call      exp@PLT
+
+        vmovsd    %xmm0, 384(%rsp,%r15)
+        jmp       .LBL_1_7
+
+END (_ZGVdN4v_exp_avx2)
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp8_core.S b/sysdeps/x86_64/fpu/multiarch/svml_d_exp8_core.S
new file mode 100644
index 0000000000..8f837fbfb9
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp8_core.S
@@ -0,0 +1,39 @@
+/* Multiple versions of vectorized exp.
+   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
+   <http://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include <init-arch.h>
+
+	.text
+ENTRY (_ZGVeN8v_exp)
+        .type   _ZGVeN8v_exp, @gnu_indirect_function
+        cmpl    $0, KIND_OFFSET+__cpu_features(%rip)
+        jne     1
+        call    __init_cpu_features
+1:      leaq    _ZGVeN8v_exp_skx(%rip), %rax
+        testl   $bit_AVX512DQ_Usable, __cpu_features+FEATURE_OFFSET+index_AVX512DQ_Usable(%rip)
+        jnz     3
+2:      leaq    _ZGVeN8v_exp_knl(%rip), %rax
+        testl   $bit_AVX512F_Usable, __cpu_features+FEATURE_OFFSET+index_AVX512F_Usable(%rip)
+        jnz     3
+        leaq    _ZGVeN8v_exp_avx2_wrapper(%rip), %rax
+3:      ret
+END (_ZGVeN8v_exp)
+
+#define _ZGVeN8v_exp _ZGVeN8v_exp_avx2_wrapper
+#include "../svml_d_exp8_core.S"
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp8_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_d_exp8_core_avx512.S
new file mode 100644
index 0000000000..049a7e49cd
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp8_core_avx512.S
@@ -0,0 +1,456 @@
+/* Function exp 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
+   <http://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include "svml_d_exp_data.h"
+#include "svml_d_wrapper_impl.h"
+
+	.text
+ENTRY (_ZGVeN8v_exp_knl)
+#ifndef HAVE_AVX512_ASM_SUPPORT
+WRAPPER_IMPL_AVX512 _ZGVdN4v_exp
+#else
+/*
+   ALGORITHM DESCRIPTION:
+
+     Argument representation:
+     N = rint(X*2^k/ln2) = 2^k*M+j
+     X = N*ln2/2^k + r = M*ln2 + ln2*(j/2^k) + r
+     then -ln2/2^(k+1) < r < ln2/2^(k+1)
+     Alternatively:
+     N = trunc(X*2^k/ln2)
+     then 0 < r < ln2/2^k
+
+     Result calculation:
+     exp(X) = exp(M*ln2 + ln2*(j/2^k) + r)
+     = 2^M * 2^(j/2^k) * exp(r)
+     2^M is calculated by bit manipulation
+     2^(j/2^k) is stored in table
+     exp(r) is approximated by polynomial
+
+     The table lookup is skipped if k = 0.  */
+
+        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_dexp_data@GOTPCREL(%rip), %rax
+
+/* dR = X - dN*dbLn2hi, dbLn2hi is 52-8-k hi bits of ln2/2^k */
+        vmovaps   %zmm0, %zmm8
+
+/* iAbsX = (int)(lX>>32), lX = *(longlong*)&X */
+        vpsrlq    $32, %zmm0, %zmm1
+
+/* iAbsX = iAbsX&iAbsMask */
+        movl      $255, %edx
+        vpmovqd   %zmm1, %ymm2
+        kmovw     %edx, %k2
+
+/* iRangeMask = (iAbsX>iDomainRange) */
+        movl      $-1, %ecx
+
+/* table lookup for dT[j] = 2^(j/2^k) */
+        vpxord    %zmm11, %zmm11, %zmm11
+        vmovups __dbInvLn2(%rax), %zmm5
+        vmovups __dbLn2hi(%rax), %zmm7
+        kxnorw    %k3, %k3, %k3
+
+/* dM = X*dbInvLn2+dbShifter, dbInvLn2 = 2^k/Ln2 */
+        vfmadd213pd __dbShifter(%rax), %zmm0, %zmm5
+        vmovups __dPC2(%rax), %zmm12
+
+/* dN = dM-dbShifter, dN = rint(X*2^k/Ln2) */
+        vsubpd __dbShifter(%rax), %zmm5, %zmm9
+        vmovups __lIndexMask(%rax), %zmm4
+        vfnmadd231pd %zmm9, %zmm7, %zmm8
+        vpandd __iAbsMask(%rax), %zmm2, %zmm2{%k2}
+
+/* lIndex = (*(longlong*)&dM)&lIndexMask, lIndex is the lower K bits of lM */
+        vpandq    %zmm4, %zmm5, %zmm10
+        vgatherqpd (%rax,%zmm10,8), %zmm11{%k3}
+        vpcmpgtd __iDomainRange(%rax), %zmm2, %k1{%k2}
+
+/* lM = (*(longlong*)&dM)&(~lIndexMask) */
+        vpandnq   %zmm5, %zmm4, %zmm6
+        vpbroadcastd %ecx, %zmm3{%k1}{z}
+
+/* lM = lM<<(52-K), 2^M */
+        vpsllq    $42, %zmm6, %zmm14
+
+/* dR = dR - dN*dbLn2lo, dbLn2lo is 40..94 bits of lo part of ln2/2^k */
+        vfnmadd132pd __dbLn2lo(%rax), %zmm8, %zmm9
+
+/* Mask = iRangeMask?1:0, set mask for overflow/underflow */
+        vptestmd  %zmm3, %zmm3, %k0{%k2}
+
+/* exp(r) = b0+r*(b0+r*(b1+r*b2)) */
+        vfmadd213pd __dPC1(%rax), %zmm9, %zmm12
+        kmovw     %k0, %ecx
+        movzbl    %cl, %ecx
+        vfmadd213pd __dPC0(%rax), %zmm9, %zmm12
+        vfmadd213pd __dPC0(%rax), %zmm9, %zmm12
+
+/* 2^(j/2^k) * exp(r) */
+        vmulpd    %zmm12, %zmm11, %zmm13
+
+/* multiply by 2^M through integer add */
+        vpaddq    %zmm14, %zmm13, %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
+        shlq      $4, %r15
+        vmovsd    1160(%rsp,%r15), %xmm0
+        call      exp@PLT
+        vmovsd    %xmm0, 1224(%rsp,%r15)
+        jmp       .LBL_1_8
+
+.LBL_1_12:
+        movzbl    %r12b, %r15d
+        shlq      $4, %r15
+        vmovsd    1152(%rsp,%r15), %xmm0
+        call      exp@PLT
+        vmovsd    %xmm0, 1216(%rsp,%r15)
+        jmp       .LBL_1_7
+#endif
+END (_ZGVeN8v_exp_knl)
+
+ENTRY (_ZGVeN8v_exp_skx)
+#ifndef HAVE_AVX512_ASM_SUPPORT
+WRAPPER_IMPL_AVX512 _ZGVdN4v_exp
+#else
+/*
+   ALGORITHM DESCRIPTION:
+
+     Argument representation:
+     N = rint(X*2^k/ln2) = 2^k*M+j
+     X = N*ln2/2^k + r = M*ln2 + ln2*(j/2^k) + r
+     then -ln2/2^(k+1) < r < ln2/2^(k+1)
+     Alternatively:
+     N = trunc(X*2^k/ln2)
+     then 0 < r < ln2/2^k
+
+     Result calculation:
+     exp(X) = exp(M*ln2 + ln2*(j/2^k) + r)
+     = 2^M * 2^(j/2^k) * exp(r)
+     2^M is calculated by bit manipulation
+     2^(j/2^k) is stored in table
+     exp(r) is approximated by polynomial
+
+     The table lookup is skipped if k = 0.  */
+
+        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_dexp_data@GOTPCREL(%rip), %rax
+
+/* table lookup for dT[j] = 2^(j/2^k) */
+        kxnorw    %k1, %k1, %k1
+
+/* iAbsX = (int)(lX>>32), lX = *(longlong*)&X */
+        vpsrlq    $32, %zmm0, %zmm1
+        vmovups __dbInvLn2(%rax), %zmm7
+        vmovups __dbShifter(%rax), %zmm5
+        vmovups __lIndexMask(%rax), %zmm6
+        vmovups __dbLn2hi(%rax), %zmm9
+        vmovups __dPC0(%rax), %zmm12
+
+/* dM = X*dbInvLn2+dbShifter, dbInvLn2 = 2^k/Ln2 */
+        vfmadd213pd %zmm5, %zmm0, %zmm7
+        vpmovqd   %zmm1, %ymm2
+
+/* dN = dM-dbShifter, dN = rint(X*2^k/Ln2) */
+        vsubpd    %zmm5, %zmm7, %zmm11
+
+/* iAbsX = iAbsX&iAbsMask */
+        vpand __iAbsMask(%rax), %ymm2, %ymm3
+
+/* dR = X - dN*dbLn2hi, dbLn2hi is 52-8-k hi bits of ln2/2^k */
+        vmovaps   %zmm0, %zmm10
+        vfnmadd231pd %zmm11, %zmm9, %zmm10
+        vmovups __dPC2(%rax), %zmm9
+
+/* dR = dR - dN*dbLn2lo, dbLn2lo is 40..94 bits of lo part of ln2/2^k */
+        vfnmadd132pd __dbLn2lo(%rax), %zmm10, %zmm11
+
+/* exp(r) = b0+r*(b0+r*(b1+r*b2)) */
+        vfmadd213pd __dPC1(%rax), %zmm11, %zmm9
+        vfmadd213pd %zmm12, %zmm11, %zmm9
+        vfmadd213pd %zmm12, %zmm11, %zmm9
+
+/* iRangeMask = (iAbsX>iDomainRange) */
+        vpcmpgtd __iDomainRange(%rax), %ymm3, %ymm4
+
+/* Mask = iRangeMask?1:0, set mask for overflow/underflow */
+        vmovmskps %ymm4, %ecx
+
+/* lIndex = (*(longlong*)&dM)&lIndexMask, lIndex is the lower K bits of lM */
+        vpandq    %zmm6, %zmm7, %zmm13
+        vpmovqd   %zmm13, %ymm14
+        vpxord    %zmm15, %zmm15, %zmm15
+        vgatherdpd (%rax,%ymm14,8), %zmm15{%k1}
+
+/* 2^(j/2^k) * exp(r) */
+        vmulpd    %zmm9, %zmm15, %zmm10
+
+/* lM = (*(longlong*)&dM)&(~lIndexMask) */
+        vpandnq   %zmm7, %zmm6, %zmm8
+
+/* lM = lM<<(52-K), 2^M */
+        vpsllq    $42, %zmm8, %zmm1
+
+/* multiply by 2^M through integer add */
+        vpaddq    %zmm1, %zmm10, %zmm1
+        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
+        shlq      $4, %r15
+        vmovsd    1160(%rsp,%r15), %xmm0
+        vzeroupper
+        vmovsd    1160(%rsp,%r15), %xmm0
+        call      exp@PLT
+        vmovsd    %xmm0, 1224(%rsp,%r15)
+        jmp       .LBL_2_8
+
+.LBL_2_12:
+        movzbl    %r12b, %r15d
+        shlq      $4, %r15
+        vmovsd    1152(%rsp,%r15), %xmm0
+        vzeroupper
+        vmovsd    1152(%rsp,%r15), %xmm0
+        call      exp@PLT
+        vmovsd    %xmm0, 1216(%rsp,%r15)
+        jmp       .LBL_2_7
+
+#endif
+END (_ZGVeN8v_exp_skx)