x86-64: Add vector exp10/exp10f implementation to libmvec

Implement vectorized exp10/exp10f containing SSE, AVX, AVX2 and
AVX512 versions for libmvec as per vector ABI.  It also contains
accuracy and ABI tests for vector exp10/exp10f with regenerated ulps.

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

1 files changed, 311 insertions, 0 deletions

diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core_sse4.S
new file mode 100644
index 0000000000..879592b789
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/svml_s_exp10f4_core_sse4.S
@@ -0,0 +1,311 @@
+/* Function exp10f vectorized with SSE4.
+   Copyright (C) 2021 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   https://www.gnu.org/licenses/.  */
+
+/*
+ * ALGORITHM DESCRIPTION:
+ *
+ *   exp10(x)  = 2^x/log10(2) = 2^n * (1 + T[j]) * (1 + P(y))
+ *   where
+ *        x = m*log10(2)/K + y,  y in [-log10(2)/K..log10(2)/K]
+ *        m = n*K + j,           m,n,j - signed integer, j in [-K/2..K/2]
+ *
+ *        values of 2^j/K are tabulated
+ *
+ *        P(y) is a minimax polynomial approximation of exp10(x)-1
+ *        on small interval [-log10(2)/K..log10(2)/K]
+ *
+ *  Special cases:
+ *
+ *   exp10(NaN)  = NaN
+ *   exp10(+INF) = +INF
+ *   exp10(-INF) = 0
+ *   exp10(x)    = 1 for subnormals
+ *   For IEEE float
+ *     if x >  38.5318412780761720 then exp10f(x) overflow
+ *     if x < -45.4555282592773440 then exp10f(x) underflow
+ *
+ */
+
+/* Offsets for data table __svml_sexp10_data_internal
+ */
+#define _sT                           	0
+#define _sLg2_10                      	128
+#define _sShifter                     	144
+#define _sInvLg2_10hi                 	160
+#define _sInvLg2_10lo                 	176
+#define _sPC0                         	192
+#define _sPC1                         	208
+#define _sPC2                         	224
+#define _iIndexMask                   	240
+#define _iAbsMask                     	256
+#define _iDomainRange                 	272
+
+#include <sysdep.h>
+
+        .text
+	.section .text.sse4,"ax",@progbits
+ENTRY(_ZGVbN4v_exp10f_sse4)
+        subq      $72, %rsp
+        cfi_def_cfa_offset(80)
+        movaps    %xmm0, %xmm4
+
+/*  Load arument  */
+        movups    _sLg2_10+__svml_sexp10_data_internal(%rip), %xmm2
+        lea       __svml_sexp10_data_internal(%rip), %r8
+        mulps     %xmm4, %xmm2
+        movups    _sShifter+__svml_sexp10_data_internal(%rip), %xmm5
+
+/*  R  */
+        movups    _sInvLg2_10hi+__svml_sexp10_data_internal(%rip), %xmm14
+        addps     %xmm5, %xmm2
+        movaps    %xmm2, %xmm1
+        movups    _sInvLg2_10lo+__svml_sexp10_data_internal(%rip), %xmm15
+        subps     %xmm5, %xmm1
+        mulps     %xmm1, %xmm14
+        movaps    %xmm4, %xmm5
+        mulps     %xmm1, %xmm15
+        subps     %xmm14, %xmm5
+
+/*
+ *  Polynomial
+ * exp10 = 2^N*(Tj+Tj*poly)
+ * poly(sN) = {1+later} a0+a1*sR
+ */
+        movups    _sPC2+__svml_sexp10_data_internal(%rip), %xmm1
+        subps     %xmm15, %xmm5
+        mulps     %xmm5, %xmm1
+        movdqu    _iIndexMask+__svml_sexp10_data_internal(%rip), %xmm3
+
+/*  Index and lookup  */
+        movdqa    %xmm3, %xmm10
+
+/* remove index bits */
+        pandn     %xmm2, %xmm3
+        pand      %xmm2, %xmm10
+
+/*  2^N  */
+        pslld     $18, %xmm3
+
+/* iIndex *= sizeof(S); */
+        pslld     $2, %xmm10
+        addps     _sPC1+__svml_sexp10_data_internal(%rip), %xmm1
+        movd      %xmm10, %edx
+        pshufd    $1, %xmm10, %xmm7
+        pshufd    $2, %xmm10, %xmm9
+        pshufd    $3, %xmm10, %xmm11
+        movd      %xmm7, %ecx
+        movd      %xmm9, %esi
+        movd      %xmm11, %edi
+
+/* Check for overflow\underflow  */
+        movdqu    _iAbsMask+__svml_sexp10_data_internal(%rip), %xmm6
+        pand      %xmm4, %xmm6
+        mulps     %xmm1, %xmm5
+        movslq    %edx, %rdx
+        addps     _sPC0+__svml_sexp10_data_internal(%rip), %xmm5
+        movslq    %ecx, %rcx
+        movslq    %esi, %rsi
+        movslq    %edi, %rdi
+        movd      (%r8,%rdx), %xmm0
+        movd      (%r8,%rcx), %xmm8
+        movd      (%r8,%rsi), %xmm13
+        movd      (%r8,%rdi), %xmm12
+        punpckldq %xmm8, %xmm0
+        punpckldq %xmm12, %xmm13
+        punpcklqdq %xmm13, %xmm0
+
+/* Tj_l+Tj_h*poly */
+        mulps     %xmm0, %xmm5
+        pcmpgtd   _iDomainRange+__svml_sexp10_data_internal(%rip), %xmm6
+        addps     %xmm5, %xmm0
+        movmskps  %xmm6, %eax
+
+/* quick mul 2^N */
+        paddd     %xmm3, %xmm0
+
+/*  Finish   */
+        testl     %eax, %eax
+
+/* Go to special inputs processing branch */
+        jne       L(SPECIAL_VALUES_BRANCH)
+                                # LOE rbx rbp r12 r13 r14 r15 eax xmm0 xmm4
+
+/* Restore registers
+ * and exit the function
+ */
+
+L(EXIT):
+        addq      $72, %rsp
+        cfi_def_cfa_offset(8)
+        ret
+        cfi_def_cfa_offset(80)
+
+/* Branch to process
+ * special inputs
+ */
+
+L(SPECIAL_VALUES_BRANCH):
+        movups    %xmm4, 32(%rsp)
+        movups    %xmm0, 48(%rsp)
+                                # LOE rbx rbp r12 r13 r14 r15 eax
+
+        xorl      %edx, %edx
+        movq      %r12, 16(%rsp)
+        cfi_offset(12, -64)
+        movl      %edx, %r12d
+        movq      %r13, 8(%rsp)
+        cfi_offset(13, -72)
+        movl      %eax, %r13d
+        movq      %r14, (%rsp)
+        cfi_offset(14, -80)
+                                # LOE rbx rbp r15 r12d r13d
+
+/* Range mask
+ * bits check
+ */
+
+L(RANGEMASK_CHECK):
+        btl       %r12d, %r13d
+
+/* Call scalar math function */
+        jc        L(SCALAR_MATH_CALL)
+                                # LOE rbx rbp r15 r12d r13d
+
+/* Special inputs
+ * processing loop
+ */
+
+L(SPECIAL_VALUES_LOOP):
+        incl      %r12d
+        cmpl      $4, %r12d
+
+/* Check bits in range mask */
+        jl        L(RANGEMASK_CHECK)
+                                # LOE rbx rbp r15 r12d r13d
+
+        movq      16(%rsp), %r12
+        cfi_restore(12)
+        movq      8(%rsp), %r13
+        cfi_restore(13)
+        movq      (%rsp), %r14
+        cfi_restore(14)
+        movups    48(%rsp), %xmm0
+
+/* Go to exit */
+        jmp       L(EXIT)
+        cfi_offset(12, -64)
+        cfi_offset(13, -72)
+        cfi_offset(14, -80)
+                                # LOE rbx rbp r12 r13 r14 r15 xmm0
+
+/* Scalar math fucntion call
+ * to process special input
+ */
+
+L(SCALAR_MATH_CALL):
+        movl      %r12d, %r14d
+        movss     32(%rsp,%r14,4), %xmm0
+        call      exp10f@PLT
+                                # LOE rbx rbp r14 r15 r12d r13d xmm0
+
+        movss     %xmm0, 48(%rsp,%r14,4)
+
+/* Process special inputs in loop */
+        jmp       L(SPECIAL_VALUES_LOOP)
+                                # LOE rbx rbp r15 r12d r13d
+END(_ZGVbN4v_exp10f_sse4)
+
+        .section .rodata, "a"
+        .align 16
+
+#ifdef __svml_sexp10_data_internal_typedef
+typedef unsigned int VUINT32;
+typedef struct
+{
+        __declspec(align(16)) VUINT32 _sT[(1<<5)][1];
+        __declspec(align(16)) VUINT32 _sLg2_10[4][1];
+        __declspec(align(16)) VUINT32 _sShifter[4][1];
+        __declspec(align(16)) VUINT32 _sInvLg2_10hi[4][1];
+        __declspec(align(16)) VUINT32 _sInvLg2_10lo[4][1];
+        __declspec(align(16)) VUINT32 _sPC0[4][1];
+        __declspec(align(16)) VUINT32 _sPC1[4][1];
+        __declspec(align(16)) VUINT32 _sPC2[4][1];
+        __declspec(align(16)) VUINT32 _iIndexMask[4][1];
+        __declspec(align(16)) VUINT32 _iAbsMask[4][1];
+        __declspec(align(16)) VUINT32 _iDomainRange[4][1];
+} __svml_sexp10_data_internal;
+#endif
+__svml_sexp10_data_internal:
+        /*== _sT ==*/
+        .long 0x3f800000  // 2^( 0 /32 )
+        .long 0x3f82cd87  // 2^( 1 /32 )
+        .long 0x3f85aac3  // 2^( 2 /32 )
+        .long 0x3f88980f  // 2^( 3 /32 )
+        .long 0x3f8b95c2  // 2^( 4 /32 )
+        .long 0x3f8ea43a  // 2^( 5 /32 )
+        .long 0x3f91c3d3  // 2^( 6 /32 )
+        .long 0x3f94f4f0  // 2^( 7 /32 )
+        .long 0x3f9837f0  // 2^( 8 /32 )
+        .long 0x3f9b8d3a  // 2^( 9 /32 )
+        .long 0x3f9ef532  // 2^( 10/32 )
+        .long 0x3fa27043  // 2^( 11/32 )
+        .long 0x3fa5fed7  // 2^( 12/32 )
+        .long 0x3fa9a15b  // 2^( 13/32 )
+        .long 0x3fad583f  // 2^( 14/32 )
+        .long 0x3fb123f6  // 2^( 15/32 )
+        .long 0x3fb504f3  // 2^( 16/32 )
+        .long 0x3fb8fbaf  // 2^( 17/32 )
+        .long 0x3fbd08a4  // 2^( 18/32 )
+        .long 0x3fc12c4d  // 2^( 19/32 )
+        .long 0x3fc5672a  // 2^( 20/32 )
+        .long 0x3fc9b9be  // 2^( 21/32 )
+        .long 0x3fce248c  // 2^( 22/32 )
+        .long 0x3fd2a81e  // 2^( 23/32 )
+        .long 0x3fd744fd  // 2^( 24/32 )
+        .long 0x3fdbfbb8  // 2^( 25/32 )
+        .long 0x3fe0ccdf  // 2^( 26/32 )
+        .long 0x3fe5b907  // 2^( 27/32 )
+        .long 0x3feac0c7  // 2^( 28/32 )
+        .long 0x3fefe4ba  // 2^( 29/32 )
+        .long 0x3ff5257d  // 2^( 30/32 )
+        .long 0x3ffa83b3  // 2^( 31/32 )
+        .align 16
+        .long 0x42d49a78, 0x42d49a78, 0x42d49a78, 0x42d49a78  /* _sLg2_10*2^K   */
+        .align 16
+        .long 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000  /* _sShifter) */
+        .align 16
+        .long 0x3c1a2000, 0x3c1a2000, 0x3c1a2000, 0x3c1a2000  /* _sInvLg2_10hi/2^K hi (24-K-7) bits*/
+        .align 16
+        .long 0x341a84fc, 0x341a84fc, 0x341a84fc, 0x341a84fc  /* _sInvLg2_10lo/2^K  lo bits */
+        // otherwise exp10(0) won't produce exact 1.0
+        .align 16
+        .long 0x2fecc868, 0x2fecc868, 0x2fecc868, 0x2fecc868  /* _sPC0 */
+        .align 16
+        .long 0x40135e1b, 0x40135e1b, 0x40135e1b, 0x40135e1b  /* _sPC1 */
+        .align 16
+        .long 0x4029a8d2, 0x4029a8d2, 0x4029a8d2, 0x4029a8d2  /* _sPC2 */
+        .align 16
+        .long 0x0000001f, 0x0000001f, 0x0000001f, 0x0000001f  /* _iIndexMask =(2^K-1)*/
+        //common
+        .align 16
+        .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff   /* _iAbsMask */
+        .align 16
+        .long 0x4217b818, 0x4217b818, 0x4217b818, 0x4217b818   /* _iDomainRange=-log10(max_denormal=0x007fffff) RZ */
+        .align 16
+        .type	__svml_sexp10_data_internal,@object
+        .size	__svml_sexp10_data_internal,.-__svml_sexp10_data_internal