This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
[RFC PATCH 1/2] Aarch64: Add simd exp/expf ABI symbols
- From: Szabolcs Nagy <Szabolcs dot Nagy at arm dot com>
- To: GNU C Library <libc-alpha at sourceware dot org>, Steve Ellcey <sellcey at marvell dot com>
- Cc: nd <nd at arm dot com>
- Date: Thu, 4 Jul 2019 11:09:39 +0000
- Subject: [RFC PATCH 1/2] Aarch64: Add simd exp/expf ABI symbols
The implementation is in assembly and just calls the scalar math code.
This ensures that old compiler without vector call abi support can
build libmvec. The abi is supported since GCC 9.1, the specification is
https://developer.arm.com/tools-and-software/server-and-hpc/arm-architecture-tools/arm-compiler-for-hpc/vector-function-abi
Vector functions require a STO_AARCH64_VARIANT_PCS marking in the
dynamic symbol table for lazy bound calls to work. This will be
missing in libmvec, which works because the marking only affects
the behaviour if there are calls to the symbols in the binary.
(TODO: detect .variant_pcs asm support and use the directive if available.)
Testing requires vector call abi support, which is detected.
Header declarations are not added yet, so the symbols will not be used
by the compiler: they are just added so the abi is in place which
enables backporting later. Currently we cannot add correct declarations
that only declare the specific symbols we provide: the OpenMP pragma
mechanism would declare both AdvSIMD and SVE variants.
(TODO: figure out the c and fortran header magic that works and backportable)
this is a bit late for 2.30, but wanted to show how i planed to add
libmvec symbols for backporting.
2019-07-04 Steve Ellcey <sellcey@marvell.com>
Szabolcs Nagy <szabolcs.nagy@arm.com>
* sysdeps/aarch64/configure.ac (build_mathvec): Enable.
(test-mathvec): Enable if ABI is supported.
* sysdeps/aarch64/configure: Regenerate.
* sysdeps/aarch64/fpu/Makefile
(libmvec-support): Add libmvec_double_vlen2_exp,
libmvec_float_vlen4_expf to list.
(libmvec-tests): Add double-vlen2, float-vlen4 to list.
(double-vlen2-funcs): Add new vector function name.
(float-vlen4-funcs): Add new vector function name.
* sysdeps/aarch64/fpu/Versions: New file.
* sysdeps/aarch64/fpu/libmvec_double_vlen2.h: New file.
* sysdeps/aarch64/fpu/libmvec_double_vlen2_exp.S: New file.
* sysdeps/aarch64/fpu/libmvec_float_vlen4.h: New file.
* sysdeps/aarch64/fpu/libmvec_float_vlen4_expf.S: New file.
* sysdeps/aarch64/fpu/test-double-vlen2-wrappers.c: New file.
* sysdeps/aarch64/fpu/test-float-vlen4-wrappers.c: New file.
* sysdeps/aarch64/libm-test-ulps (exp_vlen2): New entry.
(exp_vlen4): Likewise.
* sysdeps/unix/sysv/linux/aarch64/libmvec.abilist: New file.
---
sysdeps/aarch64/configure | 31 +++++++++
sysdeps/aarch64/configure.ac | 24 +++++++
sysdeps/aarch64/fpu/Makefile | 17 +++++
sysdeps/aarch64/fpu/Versions | 5 ++
sysdeps/aarch64/fpu/libmvec_double_vlen2.h | 59 +++++++++++++++++
.../aarch64/fpu/libmvec_double_vlen2_exp.S | 21 ++++++
sysdeps/aarch64/fpu/libmvec_float_vlen4.h | 65 +++++++++++++++++++
.../aarch64/fpu/libmvec_float_vlen4_expf.S | 21 ++++++
.../aarch64/fpu/test-double-vlen2-wrappers.c | 28 ++++++++
.../aarch64/fpu/test-float-vlen4-wrappers.c | 28 ++++++++
sysdeps/aarch64/libm-test-ulps | 6 ++
.../unix/sysv/linux/aarch64/libmvec.abilist | 2 +
12 files changed, 307 insertions(+)
create mode 100644 sysdeps/aarch64/fpu/Versions
create mode 100644 sysdeps/aarch64/fpu/libmvec_double_vlen2.h
create mode 100644 sysdeps/aarch64/fpu/libmvec_double_vlen2_exp.S
create mode 100644 sysdeps/aarch64/fpu/libmvec_float_vlen4.h
create mode 100644 sysdeps/aarch64/fpu/libmvec_float_vlen4_expf.S
create mode 100644 sysdeps/aarch64/fpu/test-double-vlen2-wrappers.c
create mode 100644 sysdeps/aarch64/fpu/test-float-vlen4-wrappers.c
create mode 100644 sysdeps/unix/sysv/linux/aarch64/libmvec.abilist
diff --git a/sysdeps/aarch64/configure b/sysdeps/aarch64/configure
index 5bd355a691..df15cdb02a 100644
--- a/sysdeps/aarch64/configure
+++ b/sysdeps/aarch64/configure
@@ -172,3 +172,34 @@ else
config_vars="$config_vars
default-abi = lp64"
fi
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for pcs attribute support" >&5
+$as_echo_n "checking for pcs attribute support... " >&6; }
+if ${libc_cv_gcc_pcs_attribute+:} false; then :
+ $as_echo_n "(cached) " >&6
+else
+ cat > conftest.c <<EOF
+__attribute__((aarch64_vector_pcs)) extern void foo (void);
+EOF
+libc_cv_gcc_pcs_attribute=no
+if ${CC-cc} -c -Wall -Werror conftest.c -o conftest.o 1>&5 \
+ 2>&5 ; then
+ libc_cv_gcc_pcs_attribute=yes
+fi
+rm -f conftest*
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libc_cv_gcc_pcs_attribute" >&5
+$as_echo "$libc_cv_gcc_pcs_attribute" >&6; }
+
+# Enable libmvec by default.
+if test x"$build_mathvec" = xnotset; then
+ build_mathvec=yes
+fi
+
+# Only test libmvec if the compiler supports aarch64_vector_pcs.
+if test x"$build_mathvec" = xyes; then
+ if test $libc_cv_gcc_pcs_attribute = yes; then
+ config_vars="$config_vars
+test-mathvec = yes"
+ fi
+fi
diff --git a/sysdeps/aarch64/configure.ac b/sysdeps/aarch64/configure.ac
index 7851dd4dac..eab411cad4 100644
--- a/sysdeps/aarch64/configure.ac
+++ b/sysdeps/aarch64/configure.ac
@@ -20,3 +20,27 @@ if test $libc_cv_aarch64_be = yes; then
else
LIBC_CONFIG_VAR([default-abi], [lp64])
fi
+
+AC_CACHE_CHECK([for pcs attribute support],
+ libc_cv_gcc_pcs_attribute, [dnl
+cat > conftest.c <<EOF
+__attribute__((aarch64_vector_pcs)) extern void foo (void);
+EOF
+libc_cv_gcc_pcs_attribute=no
+if ${CC-cc} -c -Wall -Werror conftest.c -o conftest.o 1>&AS_MESSAGE_LOG_FD \
+ 2>&AS_MESSAGE_LOG_FD ; then
+ libc_cv_gcc_pcs_attribute=yes
+fi
+rm -f conftest*])
+
+# Enable libmvec by default.
+if test x"$build_mathvec" = xnotset; then
+ build_mathvec=yes
+fi
+
+# Only test libmvec if the compiler supports aarch64_vector_pcs.
+if test x"$build_mathvec" = xyes; then
+ if test $libc_cv_gcc_pcs_attribute = yes; then
+ LIBC_CONFIG_VAR([test-mathvec], [yes])
+ fi
+fi
diff --git a/sysdeps/aarch64/fpu/Makefile b/sysdeps/aarch64/fpu/Makefile
index 4a182bd6d6..220b664323 100644
--- a/sysdeps/aarch64/fpu/Makefile
+++ b/sysdeps/aarch64/fpu/Makefile
@@ -12,3 +12,20 @@ CFLAGS-s_fmaxf.c += -ffinite-math-only
CFLAGS-s_fmin.c += -ffinite-math-only
CFLAGS-s_fminf.c += -ffinite-math-only
endif
+
+ifeq ($(subdir),mathvec)
+libmvec-support += \
+ libmvec_double_vlen2_exp \
+ libmvec_float_vlen4_expf \
+
+endif
+
+ifeq ($(subdir),math)
+ifeq ($(build-mathvec),yes)
+double-vlen2-funcs = exp
+float-vlen4-funcs = exp
+ifeq ($(test-mathvec),yes)
+libmvec-tests += double-vlen2 float-vlen4
+endif
+endif
+endif
diff --git a/sysdeps/aarch64/fpu/Versions b/sysdeps/aarch64/fpu/Versions
new file mode 100644
index 0000000000..da36f3c495
--- /dev/null
+++ b/sysdeps/aarch64/fpu/Versions
@@ -0,0 +1,5 @@
+libmvec {
+ GLIBC_2.30 {
+ _ZGVnN2v_exp; _ZGVnN4v_expf;
+ }
+}
diff --git a/sysdeps/aarch64/fpu/libmvec_double_vlen2.h b/sysdeps/aarch64/fpu/libmvec_double_vlen2.h
new file mode 100644
index 0000000000..383980d6ef
--- /dev/null
+++ b/sysdeps/aarch64/fpu/libmvec_double_vlen2.h
@@ -0,0 +1,59 @@
+/* Double-precision 2 element vector function template.
+ 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 <sysdep.h>
+
+ENTRY (VECTOR_FUNCTION)
+ stp x29, x30, [sp, -288]!
+ cfi_adjust_cfa_offset (288)
+ cfi_rel_offset (x29, 0)
+ cfi_rel_offset (x30, 8)
+ mov x29, sp
+ stp q8, q9, [sp, 16]
+ stp q10, q11, [sp, 48]
+ stp q12, q13, [sp, 80]
+ stp q14, q15, [sp, 112]
+ stp q16, q17, [sp, 144]
+ stp q18, q19, [sp, 176]
+ stp q20, q21, [sp, 208]
+ stp q22, q23, [sp, 240]
+
+ // Use per lane load/store to avoid endianness issues.
+ str q0, [sp, 272]
+ ldr d0, [sp, 272]
+ bl SCALAR_FUNCTION
+ str d0, [sp, 272]
+ ldr d0, [sp, 280]
+ bl SCALAR_FUNCTION
+ str d0, [sp, 280]
+ ldr q0, [sp, 272]
+
+ ldp q8, q9, [sp, 16]
+ ldp q10, q11, [sp, 48]
+ ldp q12, q13, [sp, 80]
+ ldp q14, q15, [sp, 112]
+ ldp q16, q17, [sp, 144]
+ ldp q18, q19, [sp, 176]
+ ldp q20, q21, [sp, 208]
+ ldp q22, q23, [sp, 240]
+ ldp x29, x30, [sp], 288
+ cfi_adjust_cfa_offset (288)
+ cfi_restore (x29)
+ cfi_restore (x30)
+ ret
+END (VECTOR_FUNCTION)
diff --git a/sysdeps/aarch64/fpu/libmvec_double_vlen2_exp.S b/sysdeps/aarch64/fpu/libmvec_double_vlen2_exp.S
new file mode 100644
index 0000000000..644405cc4f
--- /dev/null
+++ b/sysdeps/aarch64/fpu/libmvec_double_vlen2_exp.S
@@ -0,0 +1,21 @@
+/* Double-precision 2 element vector e^x function.
+ 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/>. */
+
+#define SCALAR_FUNCTION exp
+#define VECTOR_FUNCTION _ZGVnN2v_exp
+#include "libmvec_double_vlen2.h"
diff --git a/sysdeps/aarch64/fpu/libmvec_float_vlen4.h b/sysdeps/aarch64/fpu/libmvec_float_vlen4.h
new file mode 100644
index 0000000000..2450309c13
--- /dev/null
+++ b/sysdeps/aarch64/fpu/libmvec_float_vlen4.h
@@ -0,0 +1,65 @@
+/* Single-precision 4 element vector function template.
+ 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 <sysdep.h>
+
+ENTRY (VECTOR_FUNCTION)
+ stp x29, x30, [sp, -288]!
+ cfi_adjust_cfa_offset (288)
+ cfi_rel_offset (x29, 0)
+ cfi_rel_offset (x30, 8)
+ mov x29, sp
+ stp q8, q9, [sp, 16]
+ stp q10, q11, [sp, 48]
+ stp q12, q13, [sp, 80]
+ stp q14, q15, [sp, 112]
+ stp q16, q17, [sp, 144]
+ stp q18, q19, [sp, 176]
+ stp q20, q21, [sp, 208]
+ stp q22, q23, [sp, 240]
+
+ // Use per lane load/store to avoid endianness issues.
+ str q0, [sp, 272]
+ ldr s0, [sp, 272]
+ bl SCALAR_FUNCTION
+ str s0, [sp, 272]
+ ldr s0, [sp, 276]
+ bl SCALAR_FUNCTION
+ str s0, [sp, 276]
+ ldr s0, [sp, 280]
+ bl SCALAR_FUNCTION
+ str s0, [sp, 280]
+ ldr s0, [sp, 284]
+ bl SCALAR_FUNCTION
+ str s0, [sp, 284]
+ ldr q0, [sp, 272]
+
+ ldp q8, q9, [sp, 16]
+ ldp q10, q11, [sp, 48]
+ ldp q12, q13, [sp, 80]
+ ldp q14, q15, [sp, 112]
+ ldp q16, q17, [sp, 144]
+ ldp q18, q19, [sp, 176]
+ ldp q20, q21, [sp, 208]
+ ldp q22, q23, [sp, 240]
+ ldp x29, x30, [sp], 288
+ cfi_adjust_cfa_offset (288)
+ cfi_restore (x29)
+ cfi_restore (x30)
+ ret
+END (VECTOR_FUNCTION)
diff --git a/sysdeps/aarch64/fpu/libmvec_float_vlen4_expf.S b/sysdeps/aarch64/fpu/libmvec_float_vlen4_expf.S
new file mode 100644
index 0000000000..ab76ea0c77
--- /dev/null
+++ b/sysdeps/aarch64/fpu/libmvec_float_vlen4_expf.S
@@ -0,0 +1,21 @@
+/* Single-precision 4 element vector e^x function.
+ 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/>. */
+
+#define SCALAR_FUNCTION expf
+#define VECTOR_FUNCTION _ZGVnN4v_expf
+#include "libmvec_float_vlen4.h"
diff --git a/sysdeps/aarch64/fpu/test-double-vlen2-wrappers.c b/sysdeps/aarch64/fpu/test-double-vlen2-wrappers.c
new file mode 100644
index 0000000000..6c6c44d6b5
--- /dev/null
+++ b/sysdeps/aarch64/fpu/test-double-vlen2-wrappers.c
@@ -0,0 +1,28 @@
+/* Wrapper part of tests for aarch64 double 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 <arm_neon.h>
+#include "test-double-vlen2.h"
+
+#define VEC_TYPE float64x2_t
+
+/* Hack: VECTOR_WRAPPER declares the vector function without the pcs attribute,
+ placing it here happens to work, should be fixed in test-math-vector.h. */
+__attribute__ ((aarch64_vector_pcs))
+
+VECTOR_WRAPPER (WRAPPER_NAME (exp), _ZGVnN2v_exp)
diff --git a/sysdeps/aarch64/fpu/test-float-vlen4-wrappers.c b/sysdeps/aarch64/fpu/test-float-vlen4-wrappers.c
new file mode 100644
index 0000000000..5117633f1f
--- /dev/null
+++ b/sysdeps/aarch64/fpu/test-float-vlen4-wrappers.c
@@ -0,0 +1,28 @@
+/* Wrapper part of tests for float aarch64 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 <arm_neon.h>
+#include "test-float-vlen4.h"
+
+#define VEC_TYPE float32x4_t
+
+/* Hack: VECTOR_WRAPPER declares the vector function without the pcs attribute,
+ placing it here happens to work, should be fixed in test-math-vector.h. */
+__attribute__ ((aarch64_vector_pcs))
+
+VECTOR_WRAPPER (WRAPPER_NAME (expf), _ZGVnN4v_expf)
diff --git a/sysdeps/aarch64/libm-test-ulps b/sysdeps/aarch64/libm-test-ulps
index 585e5bbce7..1ed4af9e55 100644
--- a/sysdeps/aarch64/libm-test-ulps
+++ b/sysdeps/aarch64/libm-test-ulps
@@ -1601,6 +1601,12 @@ float: 1
idouble: 1
ifloat: 1
+Function: "exp_vlen2":
+double: 1
+
+Function: "exp_vlen4":
+float: 1
+
Function: "expm1":
double: 1
float: 1
diff --git a/sysdeps/unix/sysv/linux/aarch64/libmvec.abilist b/sysdeps/unix/sysv/linux/aarch64/libmvec.abilist
new file mode 100644
index 0000000000..9e178253f7
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/aarch64/libmvec.abilist
@@ -0,0 +1,2 @@
+GLIBC_2.30 _ZGVnN2v_exp F
+GLIBC_2.30 _ZGVnN4v_expf F