Patchwork [v9,1/1] PPC64: Add libmvec SIMD double-precision cosine function.

login
register
mail settings
Submitter GT
Date April 2, 2019, 8:52 p.m.
Message ID <YEP8A5dSZWq5rkj5qvYnsR3jfq6UJTn3eZKHtLYlNvUjodQv2QqoN4lEHyrhifJ-GXnhQgVuob0c-bL97HeouvhLaiYPYjyZzDep6zG9Sz0=@protonmail.com>
Download mbox | patch
Permalink /patch/32124/
State New
Headers show

Comments

GT - April 2, 2019, 8:52 p.m.
Changes from prior version:

1. PPC64 libmvec vector math patches will, for now, be uploaded to branch
https://sourceware.org/git/?p=glibc.git;a=shortlog;h=refs/heads/tuliom/libmvec
That repository is managed by Tulio.
The plan is for libmvec patches to remain there until GCC implements sufficient
PPC64 libmvec support that allows testing GLIBC libmvec functionality in these
patches. Once that is done, the patches will be migrated to glibc master.

2. In response to https://sourceware.org/ml/libc-alpha/2019-03/msg00153.html,
we use the "-v <n>" option in patch generating command "git format-patch". As
previous versions were not similarly annotated, below is a list of all earlier
versions of this patch.

v1: https://sourceware.org/ml/libc-alpha/2019-02/msg00382.html
v2: https://sourceware.org/ml/libc-alpha/2019-02/msg00473.html
v3: https://sourceware.org/ml/libc-alpha/2019-02/msg00474.html
v4: https://sourceware.org/ml/libc-alpha/2019-02/msg00621.html
v5: https://sourceware.org/ml/libc-alpha/2019-02/msg00634.html
v6: https://sourceware.org/ml/libc-alpha/2019-03/msg00052.html
v7: https://sourceware.org/ml/libc-alpha/2019-03/msg00122.html
v8: https://sourceware.org/ml/libc-alpha/2019-03/msg00127.html

3. Also in response to https://sourceware.org/ml/libc-alpha/2019-03/msg00153.html,
and since the libmvec patches are not currently on glibc master, file
bits/math-vector.h is restored in the implementation. It was removed from the
previous patch with the goal being to prevent applications using the incomplete
libmvec implementation. That is now unnecessary as any user obtaining PPC64
libmvec from the branch given in (1) above is forewarned that the ABI is in flux.

4. Flags -mabi=altivec and -maltivec are added in the Makefile to account for
installed GCC having been configured without flags specifying the cpu as power8.
In that circumstance, building libmvec with the existing Makefile fails but GCC
helpfully suggests adding these flags. Doing so results in a successful libmvec
build. The behavior was observed on both Little-Endian and Big-Endian POWER8
VMs.

Patch

From 7b7dd6baf1b297a4d37ead820ed7cc1deb043527 Mon Sep 17 00:00:00 2001
From: Bert Tenjy <bert.tenjy@gmail.com>
Date: Wed, 6 Mar 2019 21:52:48 +0000
Subject: [PATCH v9 1/1] PPC64: Add libmvec SIMD double-precision cosine
 function.

[BZ #24205]

This is the 1st of 12 patches that will implement libmvec for PPC64 using
VSX hardware capabilities.

Implements double-precision cosine using VSX vector capability. Algorithm for
cosine is from x86_64 [commit #2193311288] adapted to PPC64.

Name-mangling exactly duplicates SSE ISA of the x86_64 ABI. The details are at
<https://sourceware.org/glibc/wiki/
libmvec?action=AttachFile&do=view&target=VectorABI.txt>

The patch has been tested on PPC64/POWER8 Little Endian and Big Endian. It is
tested using the framework created for libmvec on x86_64 which runs tests on
issuing 'make check'. Tests of the new vector cosine function all pass.

Library libmvec is built by default. To disable building it, pass flag
--disable-mathvec to the configure script.

A runtime check prevents vector tests running on systems lacking VSX hardware.

File vec_finite_alias.c is empty. Its existence ensures that libmvec_nonshared.a
is created. Otherwise, linking during testing of a newly-installed glibc fails.

Glibc built with this patch was installed using the procedure outlined at
<https://sourceware.org/glibc/wiki/Testing/Builds>. Compiling against the new
library created a test executable which computes cosines using the vector
version of the function. The results are at most 2-ulps away from the scalar
cosine. That is expected and indicated in the comments describing the
algorithm - as obtained from x86_64 commit #2193311288.
---
 ChangeLog                                     | 21 +++++
 NEWS                                          | 12 +++
 sysdeps/powerpc/bits/math-vector.h            | 41 +++++++++
 sysdeps/powerpc/fpu/libm-test-ulps            |  3 +
 sysdeps/powerpc/powerpc64/fpu/Makefile        |  7 ++
 sysdeps/powerpc/powerpc64/fpu/Versions        |  5 ++
 sysdeps/powerpc/powerpc64/fpu/configure       |  6 ++
 sysdeps/powerpc/powerpc64/fpu/configure.ac    |  6 ++
 .../powerpc/powerpc64/fpu/math-tests-arch.h   | 34 +++++++
 .../powerpc/powerpc64/fpu/multiarch/Makefile  | 17 ++++
 .../multiarch/test-double-vlen2-wrappers.c    | 24 +++++
 .../powerpc64/fpu/multiarch/vec_d_cos2_vsx.c  | 88 +++++++++++++++++++
 .../powerpc64/fpu/multiarch/vec_d_trig_data.h | 60 +++++++++++++
 .../powerpc/powerpc64/fpu/vec_finite_alias.c  | 17 ++++
 .../linux/powerpc/powerpc64/libmvec.abilist   |  1 +
 15 files changed, 342 insertions(+)
 create mode 100644 sysdeps/powerpc/bits/math-vector.h
 create mode 100644 sysdeps/powerpc/powerpc64/fpu/Makefile
 create mode 100644 sysdeps/powerpc/powerpc64/fpu/Versions
 create mode 100644 sysdeps/powerpc/powerpc64/fpu/configure
 create mode 100644 sysdeps/powerpc/powerpc64/fpu/configure.ac
 create mode 100644 sysdeps/powerpc/powerpc64/fpu/math-tests-arch.h
 create mode 100644 sysdeps/powerpc/powerpc64/fpu/multiarch/test-double-vlen2-wrappers.c
 create mode 100644 sysdeps/powerpc/powerpc64/fpu/multiarch/vec_d_cos2_vsx.c
 create mode 100644 sysdeps/powerpc/powerpc64/fpu/multiarch/vec_d_trig_data.h
 create mode 100644 sysdeps/powerpc/powerpc64/fpu/vec_finite_alias.c
 create mode 100644 sysdeps/unix/sysv/linux/powerpc/powerpc64/libmvec.abilist

diff --git a/ChangeLog b/ChangeLog
index 8096175cc9..f6cfe0e150 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,24 @@ 
+2019-04-02    <bert.tenjy@gmail.com>
+
+	[BZ #24205]
+	* NEWS: Updated to note new SIMD vector cosine function.
+	* sysdeps/powerpc/bits/math-vector.h: New file.
+	* sysdeps/powerpc/fpu/libm-test-ulps (cos_vlen2): Regenerated.
+	* sysdeps/powerpc/powerpc64/fpu/Makefile: New file.
+	* sysdeps/powerpc/powerpc64/fpu/Versions: Likewise.
+	* sysdeps/powerpc/powerpc64/fpu/configure: Generated.
+	* sysdeps/powerpc/powerpc64/fpu/configure.ac: New file.
+	* sysdeps/powerpc/powerpc64/fpu/math-tests-arch.h: New file.
+	* sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile (libmvec-sysdep_routines)
+	(CFLAGS-vec_d_cos2_vsx.c, libmvec-tests, double-vlen2-funcs)
+	(double-vlen2-arch-ext-cflags): Added build of VSX vector cos function
+	and its tests.
+	* sysdeps/powerpc/powerpc64/fpu/multiarch/test-double-vlen2-wrappers.c: New file.
+	* sysdeps/powerpc/powerpc64/fpu/multiarch/vec_d_cos2_vsx.c: Likewise.
+	* sysdeps/powerpc/powerpc64/fpu/multiarch/vec_d_trig_data.h: Likewise.
+	* sysdeps/powerpc/powerpc64/fpu/vec_finite_alias.c: Likewise.
+	* sysdeps/unix/sysv/linux/powerpc/powerpc64/libmvec.abilist: Likewise.
+
 2019-02-26  Joseph Myers  <joseph@codesourcery.com>
 
 	* sysdeps/arm/sysdep.h (#if condition): Break lines before rather
diff --git a/NEWS b/NEWS
index 0a3b6c7a5a..22735e1d1d 100644
--- a/NEWS
+++ b/NEWS
@@ -4,11 +4,23 @@  See the end for copying conditions.
 
 Please send GNU C library bug reports via <https://sourceware.org/bugzilla/>
 using `glibc' in the "product" field.
+
 
 Version 2.30
 
 Major new features:
 
+* Start of implementing vector math library libmvec on PPC64/POWER8.
+  The double-precision cosine now has a vector version.
+  GCC support for auto-vectorization of functions on PPC64 is not yet
+  available. Until that is done, the new vector math functions are
+  inaccessible to applications.
+  Library libmvec is built by default for PPC64. Disable its creation by
+  passing flag --disable-mathvec to configure.
+  The library ABI specification is x86_64 Vector Function ABI.
+  More information on libmvec including a link to the ABI document is at:
+  <https://sourceware.org/glibc/wiki/libmvec>
+
 * The dynamic linker accepts the --preload argument to preload shared
   objects, in addition to the LD_PRELOAD environment variable.
 
diff --git a/sysdeps/powerpc/bits/math-vector.h b/sysdeps/powerpc/bits/math-vector.h
new file mode 100644
index 0000000000..78d9db64bf
--- /dev/null
+++ b/sysdeps/powerpc/bits/math-vector.h
@@ -0,0 +1,41 @@ 
+/* Platform-specific SIMD declarations of math functions.
+   Copyright (C) 2019 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/>.  */
+
+#ifndef _MATH_H
+# error "Never include <bits/math-vector.h> directly;\
+ include <math.h> instead."
+#endif
+
+/* Get default empty definitions for simd declarations.  */
+#include <bits/libm-simd-decl-stubs.h>
+
+#if defined __PPC64__ && defined __FAST_MATH__
+# if defined _OPENMP && _OPENMP >= 201307
+/* OpenMP case.  */
+#  define __DECL_SIMD_PPC64 _Pragma ("omp declare simd notinbranch")
+# elif __GNUC_PREREQ (6,0)
+/* W/o OpenMP use GCC 6.* __attribute__ ((__simd__)).  */
+#  define __DECL_SIMD_PPC64 __attribute__ ((__simd__ ("notinbranch")))
+# endif
+
+# ifdef __DECL_SIMD_PPC64
+#  undef __DECL_SIMD_cos
+#  define __DECL_SIMD_cos __DECL_SIMD_PPC64
+
+# endif
+#endif
diff --git a/sysdeps/powerpc/fpu/libm-test-ulps b/sysdeps/powerpc/fpu/libm-test-ulps
index 1eec27c1dc..d392b135a7 100644
--- a/sysdeps/powerpc/fpu/libm-test-ulps
+++ b/sysdeps/powerpc/fpu/libm-test-ulps
@@ -1311,6 +1311,9 @@  ifloat128: 2
 ildouble: 5
 ldouble: 5
 
+Function: "cos_vlen2":
+double: 2
+
 Function: "cosh":
 double: 1
 float: 1
diff --git a/sysdeps/powerpc/powerpc64/fpu/Makefile b/sysdeps/powerpc/powerpc64/fpu/Makefile
new file mode 100644
index 0000000000..21dc67ff73
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/fpu/Makefile
@@ -0,0 +1,7 @@ 
+ifeq ($(subdir),mathvec)
+libmvec-support += vec_finite_alias
+
+CFLAGS-vec_finite_alias.c += -mvsx
+
+libmvec-static-only-routines = vec_finite_alias
+endif
diff --git a/sysdeps/powerpc/powerpc64/fpu/Versions b/sysdeps/powerpc/powerpc64/fpu/Versions
new file mode 100644
index 0000000000..9a3e1211cc
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/fpu/Versions
@@ -0,0 +1,5 @@ 
+libmvec {
+  GLIBC_2.30 {
+    _ZGVbN2v_cos;
+  }
+}
diff --git a/sysdeps/powerpc/powerpc64/fpu/configure b/sysdeps/powerpc/powerpc64/fpu/configure
new file mode 100644
index 0000000000..1b2bb78165
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/fpu/configure
@@ -0,0 +1,6 @@ 
+# This file is generated from configure.ac by Autoconf.  DO NOT EDIT!
+ # Local configure fragment for sysdeps/powerpc/powerpc64/fpu/.
+
+if test "$enable_mathvec" != "no" ; then
+  build_mathvec=yes
+fi
diff --git a/sysdeps/powerpc/powerpc64/fpu/configure.ac b/sysdeps/powerpc/powerpc64/fpu/configure.ac
new file mode 100644
index 0000000000..d2ef5c8739
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/fpu/configure.ac
@@ -0,0 +1,6 @@ 
+GLIBC_PROVIDES dnl See aclocal.m4 in the top level source directory.
+# Local configure fragment for sysdeps/powerpc/powerpc64/fpu/.
+
+if test "$enable_mathvec" != "no" ; then
+  build_mathvec=yes
+fi
diff --git a/sysdeps/powerpc/powerpc64/fpu/math-tests-arch.h b/sysdeps/powerpc/powerpc64/fpu/math-tests-arch.h
new file mode 100644
index 0000000000..74fac298d2
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/fpu/math-tests-arch.h
@@ -0,0 +1,34 @@ 
+/* Runtime architecture check for math tests. PPC64 version.
+   Copyright (C) 2014-2019 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/>.  */
+
+#if defined REQUIRE_VSX
+
+# include "init-arch.h"
+
+# define INIT_ARCH_EXT
+# define CHECK_ARCH_EXT                                        \
+  do                                                           \
+    {                                                          \
+      INIT_ARCH()                                              \
+      if (!(hwcap & PPC_FEATURE_HAS_VSX )) return;             \
+    }                                                          \
+  while (0)
+
+#else
+# include <sysdeps/generic/math-tests-arch.h>
+#endif
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile b/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile
index 39b557604c..d713b61431 100644
--- a/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile
+++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile
@@ -42,3 +42,20 @@  CFLAGS-e_hypotf-power7.c = -mcpu=power7
 CFLAGS-s_modf-ppc64.c += -fsignaling-nans
 CFLAGS-s_modff-ppc64.c += -fsignaling-nans
 endif
+
+ifeq ($(subdir),mathvec)
+libmvec-sysdep_routines += vec_d_cos2_vsx
+CFLAGS-vec_d_cos2_vsx.c += -mabi=altivec -maltivec -mvsx
+endif
+
+# Variables for libmvec tests.
+ifeq ($(subdir),math)
+ifeq ($(build-mathvec),yes)
+libmvec-tests += double-vlen2
+
+double-vlen2-funcs = cos
+
+double-vlen2-arch-ext-cflags = -mabi=altivec -maltivec -mvsx -DREQUIRE_VSX
+CFLAGS-test-double-vlen2-wrappers.c += -mabi=altivec -maltivec -mvsx -DREQUIRE_VSX
+endif
+endif
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/test-double-vlen2-wrappers.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/test-double-vlen2-wrappers.c
new file mode 100644
index 0000000000..17e2cc0724
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/test-double-vlen2-wrappers.c
@@ -0,0 +1,24 @@ 
+/* Wrapper part of tests for VSX ISA versions of vector math functions.
+   Copyright (C) 2019 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 "test-double-vlen2.h"
+#include <altivec.h>
+
+#define VEC_TYPE vector double
+
+VECTOR_WRAPPER (WRAPPER_NAME (cos), _ZGVbN2v_cos)
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/vec_d_cos2_vsx.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/vec_d_cos2_vsx.c
new file mode 100644
index 0000000000..e1467a78b1
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/vec_d_cos2_vsx.c
@@ -0,0 +1,88 @@ 
+/* Function cos vectorized with VSX.
+   Copyright (C) 2019 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 <math.h>
+#include "vec_d_trig_data.h"
+
+vector double
+_ZGVbN2v_cos (vector double x)
+{
+
+  /*
+   ARGUMENT RANGE REDUCTION:
+   Add Pi/2 to argument: X' = X+Pi/2.  */
+  vector double x_prime = (vector double) __d_half_pi + x;
+
+  /* Get absolute argument value: X' = |X'|.  */
+  vector double abs_x_prime = vec_abs (x_prime);
+
+  /* Y = X'*InvPi + RS : right shifter add.  */
+  vector double y = (x_prime * __d_inv_pi) + __d_rshifter;
+
+  /* Check for large arguments path.  */
+  vector bool long long large_in = vec_cmpgt (abs_x_prime, __d_rangeval);
+
+  /* N = Y - RS : right shifter sub.  */
+  vector double n = y - __d_rshifter;
+
+  /* SignRes = Y<<63 : shift LSB to MSB place for result sign.  */
+  vector double sign_res = (vector double) vec_sl ((vector long long) y,
+						   (vector unsigned long long)
+						   vec_splats (63));
+
+  /* N = N - 0.5.  */
+  n = n - __d_one_half;
+
+  /* R = X - N*Pi1.  */
+  vector double r = x - (n * __d_pi1_fma);
+
+  /* R = R - N*Pi2.  */
+  r = r - (n * __d_pi2_fma);
+
+  /* R = R - N*Pi3.  */
+  r = r - (n * __d_pi3_fma);
+
+  /* R2 = R*R.  */
+  vector double r2 = r * r;
+
+  /* Poly = C3+R2*(C4+R2*(C5+R2*(C6+R2*C7))).  */
+  vector double poly = r2 * __d_coeff7 + __d_coeff6;
+  poly = poly * r2 + __d_coeff5;
+  poly = poly * r2 + __d_coeff4;
+  poly = poly * r2 + __d_coeff3;
+
+  /* Poly = R+R*(R2*(C1+R2*(C2+R2*Poly))).  */
+  poly = poly * r2 + __d_coeff2;
+  poly = poly * r2 + __d_coeff1;
+  poly = poly * r2 * r + r;
+
+  /*
+     RECONSTRUCTION:
+     Final sign setting: Res = Poly^SignRes.  */
+  vector double out
+    = (vector double) ((vector long long) poly ^ (vector long long) sign_res);
+
+  if (large_in[0] != 0)
+    out[0] = cos (x[0]);
+
+  if (large_in[1] != 0)
+    out[1] = cos (x[1]);
+
+  return out;
+
+}
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/vec_d_trig_data.h b/sysdeps/powerpc/powerpc64/fpu/multiarch/vec_d_trig_data.h
new file mode 100644
index 0000000000..ecd47d258f
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/vec_d_trig_data.h
@@ -0,0 +1,60 @@ 
+/* Constants used in polynomial approximations for vectorized sin, cos,
+   and sincos functions.
+   Copyright (C) 2019 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/>.  */
+
+#ifndef D_TRIG_DATA_H
+#define D_TRIG_DATA_H
+
+#include <altivec.h>
+
+/* PI/2.  */
+const vector double __d_half_pi  = {0x1.921fb54442d18p+0, 0x1.921fb54442d18p+0};
+
+/* Inverse PI.  */
+const vector double __d_inv_pi   = {0x1.45f306dc9c883p-2, 0x1.45f306dc9c883p-2};
+
+/* Right-shifter constant.  */
+const vector double __d_rshifter = {0x1.8p+52, 0x1.8p+52};
+
+/* Working range threshold.  */
+const vector double __d_rangeval = {0x1p+23, 0x1p+23};
+
+/* One-half.  */
+const vector double __d_one_half = {0x1p-1, 0x1p-1};
+
+/* Range reduction PI-based constants if FMA available:
+   PI high part (FMA available).  */
+const vector double __d_pi1_fma = {0x1.921fb54442d18p+1, 0x1.921fb54442d18p+1};
+
+/* PI mid part  (FMA available).  */
+const vector double __d_pi2_fma = {0x1.1a62633145c06p-53, 0x1.1a62633145c06p-53};
+
+/* PI low part  (FMA available).  */
+const vector double __d_pi3_fma
+= {0x1.c1cd129024e09p-106,0x1.c1cd129024e09p-106};
+
+/* Polynomial coefficients (relative error 2^(-52.115)).  */
+const vector double __d_coeff7 = {-0x1.9f0d60811aac8p-41,-0x1.9f0d60811aac8p-41};
+const vector double __d_coeff6 = {0x1.60e6857a2f22p-33,0x1.60e6857a2f22p-33};
+const vector double __d_coeff5 = {-0x1.ae63546002231p-26,-0x1.ae63546002231p-26};
+const vector double __d_coeff4 = {0x1.71de38030feap-19,0x1.71de38030feap-19};
+const vector double __d_coeff3 = {-0x1.a01a019a5b86dp-13,-0x1.a01a019a5b86dp-13};
+const vector double __d_coeff2 = {0x1.111111110a4a8p-7,0x1.111111110a4a8p-7};
+const vector double __d_coeff1 = {-0x1.55555555554a7p-3,-0x1.55555555554a7p-3};
+
+#endif /* D_TRIG_DATA_H.  */
diff --git a/sysdeps/powerpc/powerpc64/fpu/vec_finite_alias.c b/sysdeps/powerpc/powerpc64/fpu/vec_finite_alias.c
new file mode 100644
index 0000000000..5fd180fbeb
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/fpu/vec_finite_alias.c
@@ -0,0 +1,17 @@ 
+/* A temporary workaround to force creation of libmvec_nonshared.a.
+   Copyright (C) 2019 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/>.  */
diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc64/libmvec.abilist b/sysdeps/unix/sysv/linux/powerpc/powerpc64/libmvec.abilist
new file mode 100644
index 0000000000..656ce0541f
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/powerpc/powerpc64/libmvec.abilist
@@ -0,0 +1 @@ 
+GLIBC_2.30 _ZGVbN2v_cos F
-- 
2.20.1