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.

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.

Reviewed-by: Tulio Magno Quites Machado Filho <tuliom@linux.ibm.com>

12 files changed, 295 insertions, 0 deletions

diff --git a/NEWS b/NEWS
index 77631ca707..ef0843fd91 100644
--- a/NEWS
+++ b/NEWS
@@ -257,6 +257,17 @@ 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>
+
 * Unicode 12.1.0 Support: Character encoding, character type info, and
   transliteration tables are all updated to Unicode 12.1.0, using
   generator scripts contributed by Mike FABIAN (Red Hat).
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/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
new file mode 100644
index 0000000000..b56e756f80
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile
@@ -0,0 +1,16 @@
+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/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