x86-64: Add vector cosh/coshf implementation to libmvec

Implement vectorized cosh/coshf containing SSE, AVX, AVX2 and
AVX512 versions for libmvec as per vector ABI.  It also contains
accuracy and ABI tests for vector cosh/coshf with regenerated ulps.

Reviewed-by: H.J. Lu <hjl.tools@gmail.com>

1 files changed, 305 insertions, 0 deletions

diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_coshf4_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_s_coshf4_core_sse4.S
new file mode 100644
index 0000000000..506f6a4bd9
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/svml_s_coshf4_core_sse4.S
@@ -0,0 +1,305 @@
+/* Function coshf 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:
+ *
+ *   Compute cosh(x) as (exp(x)+exp(-x))/2,
+ *   where exp is calculated as
+ *   exp(M*ln2 + ln2*(j/2^k) + r) = 2^M * 2^(j/2^k) * exp(r)
+ *
+ *   Special cases:
+ *
+ *   cosh(NaN) = quiet NaN, and raise invalid exception
+ *   cosh(INF) = that INF
+ *   cosh(0)   = 1
+ *   cosh(x) overflows for big x and returns MAXLOG+log(2)
+ *
+ */
+
+/* Offsets for data table __svml_scosh_data_internal
+ */
+#define _sInvLn2                      	0
+#define _sLn2hi                       	16
+#define _sLn2lo                       	32
+#define _sSign                        	48
+#define _sShifter                     	64
+#define _iDomainRange                 	80
+#define _sPC1                         	96
+#define _sPC2                         	112
+#define _sPC3                         	128
+#define _sPC4                         	144
+#define _sPC5                         	160
+#define _sPC6                         	176
+#define _iHalf                        	192
+
+#include <sysdep.h>
+
+        .text
+	.section .text.sse4,"ax",@progbits
+ENTRY(_ZGVbN4v_coshf_sse4)
+        subq      $72, %rsp
+        cfi_def_cfa_offset(80)
+
+/*
+ *  Implementation
+ *  Abs argument
+ */
+        movups    _sSign+__svml_scosh_data_internal(%rip), %xmm1
+
+/*
+ *  Load argument
+ * dM = x/log(2) + RShifter
+ */
+        movups    _sInvLn2+__svml_scosh_data_internal(%rip), %xmm9
+        andnps    %xmm0, %xmm1
+        mulps     %xmm1, %xmm9
+
+/* Check for overflow\underflow  */
+        movaps    %xmm1, %xmm3
+        movups    _sShifter+__svml_scosh_data_internal(%rip), %xmm4
+        movups    _sLn2hi+__svml_scosh_data_internal(%rip), %xmm5
+        addps     %xmm4, %xmm9
+
+/*
+ *  R
+ * sN = sM - RShifter
+ */
+        movaps    %xmm9, %xmm6
+
+/*
+ *  G1,G2 2^N,2^(-N)
+ * iM now is an EXP(2^N)
+ */
+        pslld     $23, %xmm9
+        movups    _sLn2lo+__svml_scosh_data_internal(%rip), %xmm7
+        subps     %xmm4, %xmm6
+
+/* sR = sX - sN*Log2_hi */
+        mulps     %xmm6, %xmm5
+
+/* sR = (sX - sN*Log2_hi) - sN*Log2_lo */
+        mulps     %xmm6, %xmm7
+        movdqu    _iDomainRange+__svml_scosh_data_internal(%rip), %xmm2
+        pcmpgtd   %xmm2, %xmm3
+        pcmpeqd   %xmm1, %xmm2
+
+/*
+ * sinh(r) = r*((a1=1)+r^2*(a3+r^2*(a5+{v1 r^2*a7})))) = r + r*(r^2*(a3+r^2*(a5+r^2*a7))) ....
+ * sSinh_r = (a3+r^2*a5)
+ */
+        movups    _sPC5+__svml_scosh_data_internal(%rip), %xmm10
+        por       %xmm2, %xmm3
+
+/*
+ * sinh(X) = sG2 + sG1*sinh(dR) + sG2*sR2*(a2+sR2*(a4+a6*sR2)
+ * sOut = (a4 +a6*sR2)
+ */
+        movups    _sPC6+__svml_scosh_data_internal(%rip), %xmm11
+        subps     %xmm5, %xmm1
+        movmskps  %xmm3, %edx
+        movdqu    _iHalf+__svml_scosh_data_internal(%rip), %xmm8
+        subps     %xmm7, %xmm1
+
+/* sR2 = sR^2,shaffled */
+        movaps    %xmm1, %xmm13
+        movdqa    %xmm8, %xmm2
+        mulps     %xmm1, %xmm13
+        paddd     %xmm9, %xmm2
+        mulps     %xmm13, %xmm10
+        psubd     %xmm9, %xmm8
+        mulps     %xmm13, %xmm11
+        addps     _sPC3+__svml_scosh_data_internal(%rip), %xmm10
+        addps     _sPC4+__svml_scosh_data_internal(%rip), %xmm11
+
+/* sSinh_r = r^2*(a3+r^2*a5) */
+        mulps     %xmm13, %xmm10
+
+/* sOut = a2+sR2*(a4+a6*sR2) */
+        mulps     %xmm13, %xmm11
+
+/* sSinh_r = r + r*(r^2*(a3+r^2*a5)) */
+        mulps     %xmm1, %xmm10
+        addps     _sPC2+__svml_scosh_data_internal(%rip), %xmm11
+        addps     %xmm10, %xmm1
+
+/* sOut = sR2*(a2+sR2*(a4+a6*sR2) */
+        mulps     %xmm11, %xmm13
+
+/* sG1 = 2^(N-1)-2^(-N-1) */
+        movdqa    %xmm2, %xmm12
+
+/* sG2 = 2^(N-1)+2^(-N-1) */
+        addps     %xmm8, %xmm2
+        subps     %xmm8, %xmm12
+
+/* sOut = sG2*sR2*(a2+sR2*(a4+a6*sR2) */
+        mulps     %xmm2, %xmm13
+
+/* sOut = sG1*sinh(dR)+sG2*sR2*(a2+sR2*(a4+a6*sR2) */
+        mulps     %xmm1, %xmm12
+        addps     %xmm12, %xmm13
+
+/* sOut = sG2 + sG1*sinh(dR) + sG2*sR2*(a2+sR2*(a4+a6*sR2) */
+        addps     %xmm13, %xmm2
+
+/*  Ret H  */
+        testl     %edx, %edx
+
+/* Go to special inputs processing branch */
+        jne       L(SPECIAL_VALUES_BRANCH)
+                                # LOE rbx rbp r12 r13 r14 r15 edx xmm0 xmm2
+
+/* Restore registers
+ * and exit the function
+ */
+
+L(EXIT):
+        movaps    %xmm2, %xmm0
+        addq      $72, %rsp
+        cfi_def_cfa_offset(8)
+        ret
+        cfi_def_cfa_offset(80)
+
+/* Branch to process
+ * special inputs
+ */
+
+L(SPECIAL_VALUES_BRANCH):
+        movups    %xmm0, 32(%rsp)
+        movups    %xmm2, 48(%rsp)
+                                # LOE rbx rbp r12 r13 r14 r15 edx
+
+        xorl      %eax, %eax
+        movq      %r12, 16(%rsp)
+        cfi_offset(12, -64)
+        movl      %eax, %r12d
+        movq      %r13, 8(%rsp)
+        cfi_offset(13, -72)
+        movl      %edx, %r13d
+        movq      %r14, (%rsp)
+        cfi_offset(14, -80)
+                                # LOE rbx rbp r15 r12d r13d
+
+/* Range mask
+ * bits check
+ */
+
+L(RANGEMASK_CHECK):
+        btl       %r12d, %r13d
+
+/* Call scalar math function */
+        jc        L(SCALAR_MATH_CALL)
+                                # LOE rbx rbp r15 r12d r13d
+
+/* Special inputs
+ * processing loop
+ */
+
+L(SPECIAL_VALUES_LOOP):
+        incl      %r12d
+        cmpl      $4, %r12d
+
+/* Check bits in range mask */
+        jl        L(RANGEMASK_CHECK)
+                                # LOE rbx rbp r15 r12d r13d
+
+        movq      16(%rsp), %r12
+        cfi_restore(12)
+        movq      8(%rsp), %r13
+        cfi_restore(13)
+        movq      (%rsp), %r14
+        cfi_restore(14)
+        movups    48(%rsp), %xmm2
+
+/* Go to exit */
+        jmp       L(EXIT)
+        cfi_offset(12, -64)
+        cfi_offset(13, -72)
+        cfi_offset(14, -80)
+                                # LOE rbx rbp r12 r13 r14 r15 xmm2
+
+/* Scalar math fucntion call
+ * to process special input
+ */
+
+L(SCALAR_MATH_CALL):
+        movl      %r12d, %r14d
+        movss     32(%rsp,%r14,4), %xmm0
+        call      coshf@PLT
+                                # LOE rbx rbp r14 r15 r12d r13d xmm0
+
+        movss     %xmm0, 48(%rsp,%r14,4)
+
+/* Process special inputs in loop */
+        jmp       L(SPECIAL_VALUES_LOOP)
+                                # LOE rbx rbp r15 r12d r13d
+END(_ZGVbN4v_coshf_sse4)
+
+        .section .rodata, "a"
+        .align 16
+
+#ifdef __svml_scosh_data_internal_typedef
+typedef unsigned int VUINT32;
+typedef struct
+{
+        __declspec(align(16)) VUINT32 _sInvLn2[4][1];
+        __declspec(align(16)) VUINT32 _sLn2hi[4][1];
+        __declspec(align(16)) VUINT32 _sLn2lo[4][1];
+        __declspec(align(16)) VUINT32 _sSign[4][1];
+        __declspec(align(16)) VUINT32 _sShifter[4][1];
+        __declspec(align(16)) VUINT32 _iDomainRange[4][1];
+        __declspec(align(16)) VUINT32 _sPC1[4][1];
+        __declspec(align(16)) VUINT32 _sPC2[4][1];
+        __declspec(align(16)) VUINT32 _sPC3[4][1];
+        __declspec(align(16)) VUINT32 _sPC4[4][1];
+        __declspec(align(16)) VUINT32 _sPC5[4][1];
+        __declspec(align(16)) VUINT32 _sPC6[4][1];
+        __declspec(align(16)) VUINT32 _iHalf[4][1];
+} __svml_scosh_data_internal;
+#endif
+__svml_scosh_data_internal:
+        .long 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B       /* _sInvLn2  */  //k=0
+        .align 16
+        .long 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000       /* _sLn2hi   */
+        .align 16
+        .long 0x3805fdf4, 0x3805fdf4, 0x3805fdf4, 0x3805fdf4       /* _sLn2lo   */
+        .align 16
+        .long 0x80000000, 0x80000000, 0x80000000, 0x80000000       /* _sSign    */
+        .align 16
+        .long 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000           /* _sShifter */
+        .align 16
+        .long 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E           /* _iDomainRange */
+        .align 16
+        .long 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000         /* _sPC1=1  */
+        .align 16
+        .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000         /* _sPC2  */
+        .align 16
+        .long 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57         /* _sPC3  */
+        .align 16
+        .long 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72         /* _sPC4  */
+        .align 16
+        .long 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461         /* _sPC5  */
+        .align 16
+        .long 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3         /* _sPC6  */
+        // Integer constants
+        .align 16
+        .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000 /* _iHalf*/
+        .align 16
+        .type	__svml_scosh_data_internal,@object
+        .size	__svml_scosh_data_internal,.-__svml_scosh_data_internal