[vlc-devel] [PATCH 04/16] Add ARM NEON support
glenvt18
glenvt18 at gmail.com
Fri Jun 26 13:19:58 CEST 2015
---
configure.ac | 6 ++++++
src/Makefile.am | 2 +-
src/dvbcsa_bs.h | 3 +++
src/dvbcsa_bs_neon.h | 56 ++++++++++++++++++++++++++++++++++++++++++++++++++++
test/testbsops.c | 29 +++++++++++++++++++++++++++
5 files changed, 95 insertions(+), 1 deletion(-)
create mode 100644 src/dvbcsa_bs_neon.h
diff --git a/configure.ac b/configure.ac
index ed8a1ad..cefdf8a 100644
--- a/configure.ac
+++ b/configure.ac
@@ -14,6 +14,7 @@ AC_ARG_ENABLE(uint64, AC_HELP_STRING(--enable-uint64, [Use native 64 bits intege
AC_ARG_ENABLE(mmx, AC_HELP_STRING(--enable-mmx, [Use MMX for bitslice]), mmx_debug=$enableval, enable_mmx=no)
AC_ARG_ENABLE(sse2, AC_HELP_STRING(--enable-sse2, [Use SSE2 for bitslice]), sse2_debug=$enableval, enable_sse2=no)
AC_ARG_ENABLE(altivec, AC_HELP_STRING(--enable-altivec, [Use AltiVec for bitslice]), altivec_debug=$enableval, enable_altivec=no)
+AC_ARG_ENABLE(neon, AC_HELP_STRING(--enable-neon, [Use NEON for bitslice]), neon_debug=$enableval, enable_neon=no)
AM_INIT_AUTOMAKE(libdvbcsa, 1.1.0)
AC_CONFIG_HEADERS(config.h)
@@ -46,6 +47,11 @@ elif test "$enable_altivec" = "yes" ; then
AC_DEFINE(DVBCSA_USE_ALTIVEC, 1, Using AltiVec bitslice.)
GCC_CFLAGS="$GCC_CFLAGS -maltivec -mabi=altivec"
+elif test "$enable_neon" = "yes" ; then
+ transpose_128=yes
+ AC_DEFINE(DVBCSA_USE_NEON, 1, Using NEON bitslice.)
+ GCC_CFLAGS="$GCC_CFLAGS -mfpu=neon"
+
elif test "$enable_uint32" = "yes" ; then
transpose_32=yes
AC_DEFINE(DVBCSA_USE_UINT32, 1, Using 32 bits integer bitslice.)
diff --git a/src/Makefile.am b/src/Makefile.am
index dec4f55..3bad07a 100644
--- a/src/Makefile.am
+++ b/src/Makefile.am
@@ -7,7 +7,7 @@ libdvbcsa_la_SOURCES = dvbcsa_algo.c dvbcsa_block.c dvbcsa_bs_algo.c \
dvbcsa_bs_block.c dvbcsa_bs_key.c dvbcsa_bs_stream.c \
dvbcsa_stream.c dvbcsa_bs.h dvbcsa_pv.h dvbcsa_bs_uint64.h \
dvbcsa_bs_uint32.h dvbcsa_bs_mmx.h dvbcsa_bs_sse.h \
- dvbcsa_bs_altivec.h dvbcsa_bs_transpose.c dvbcsa_key.c \
+ dvbcsa_bs_altivec.h dvbcsa_bs_neon.h dvbcsa_bs_transpose.c dvbcsa_key.c \
dvbcsa_bs_stream_kernel.inc dvbcsa_bs_stream_kernel.h
if TRANSPOSE_128
diff --git a/src/dvbcsa_bs.h b/src/dvbcsa_bs.h
index 75cabc9..7145048 100644
--- a/src/dvbcsa_bs.h
+++ b/src/dvbcsa_bs.h
@@ -43,6 +43,9 @@
#elif defined(DVBCSA_USE_ALTIVEC)
# include "dvbcsa_bs_altivec.h"
+#elif defined(DVBCSA_USE_NEON)
+# include "dvbcsa_bs_neon.h"
+
#else
# error No dvbcsa word size defined
#endif
diff --git a/src/dvbcsa_bs_neon.h b/src/dvbcsa_bs_neon.h
new file mode 100644
index 0000000..7bf0156
--- /dev/null
+++ b/src/dvbcsa_bs_neon.h
@@ -0,0 +1,56 @@
+/*
+
+ This file is part of libdvbcsa.
+
+ libdvbcsa is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published
+ by the Free Software Foundation; either version 2 of the License,
+ or (at your option) any later version.
+
+ libdvbcsa 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
+ General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with libdvbcsa; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+ 02111-1307 USA
+
+ Based on FFdecsa, Copyright (C) 2003-2004 fatih89r
+
+ (c) 2006-2008 Alexandre Becoulet <alexandre.becoulet at free.fr>
+
+*/
+
+#ifndef DVBCSA_NEON_H_
+#define DVBCSA_NEON_H_
+
+# include <arm_neon.h>
+
+typedef uint64x2_t dvbcsa_bs_word_t;
+
+#define BS_BATCH_SIZE 128
+#define BS_BATCH_BYTES 16
+
+#define BS_VAL(n, m) vcombine_u64((uint64_t)(m), (uint64_t)(n))
+#define BS_VAL64(n) vdupq_n_u64(0x##n##ULL)
+#define BS_VAL32(n) vreinterpretq_u64_u32(vdupq_n_u32(0x##n))
+#define BS_VAL16(n) vreinterpretq_u64_u16(vdupq_n_u16(0x##n))
+#define BS_VAL8(n) vreinterpretq_u64_u8(vdupq_n_u8(0x##n))
+
+#define BS_AND(a, b) vandq_u64 ((a), (b))
+#define BS_OR(a, b) vorrq_u64 ((a), (b))
+#define BS_XOR(a, b) veorq_u64 ((a), (b))
+#define BS_NOT(a) vreinterpretq_u64_u8(vmvnq_u8(vreinterpretq_u8_u64(a)))
+
+#define BS_SHL(a, n) vshlq_n_u64 ((a), n)
+#define BS_SHR(a, n) vshrq_n_u64 ((a), n)
+#define BS_SHL8(a, n) BS_SHL(a, 8 * (n))
+#define BS_SHR8(a, n) BS_SHR(a, 8 * (n))
+#define BS_EXTRACT8(a, n) (((uint8_t*)&(a))[n])
+
+#define BS_EMPTY()
+
+#endif
+
diff --git a/test/testbsops.c b/test/testbsops.c
index 50748ab..a8588bd 100644
--- a/test/testbsops.c
+++ b/test/testbsops.c
@@ -248,17 +248,30 @@ main (void)
vec_testeq(b, c);
a = BS_VAL(0xaaaaaaaaaaaaaaaaLL, 0x5555555555555555LL);
+ /* there is no neon instruction to shift a whole register */
+#ifdef DVBCSA_USE_NEON
+ b = BS_VAL(0xaaaaaaaaaaaaaa00LL, 0x5555555555555500LL);
+#else
b = BS_VAL(0xaaaaaaaaaaaaaa55LL, 0x5555555555555500LL);
+#endif
c = BS_SHL8(a, 1);
vec_testeq(b, c);
a = BS_VAL(0xffffffffffffffffLL, 0xffffffffffffffffLL);
+#ifdef DVBCSA_USE_NEON
+ b = BS_VAL(0xffffffffffffff00LL, 0xffffffffffffff00LL);
+#else
b = BS_VAL(0xffffffffffffffffLL, 0xffffffffffffff00LL);
+#endif
c = BS_SHL8(a, 1);
vec_testeq(b, c);
a = BS_VAL(0xffffffffffffffffLL, 0xffffffffffffffffLL);
+#ifdef DVBCSA_USE_NEON
+ b = BS_VAL(0xffffffff00000000LL, 0xffffffff00000000LL);
+#else
b = BS_VAL(0xffffffffffffffffLL, 0xffffffff00000000LL);
+#endif
c = BS_SHL8(a, 4);
vec_testeq(b, c);
@@ -270,22 +283,38 @@ main (void)
vec_testeq(b, c);
a = BS_VAL(0xaaaaaaaaaaaaaaaaLL, 0x5555555555555555LL);
+#ifdef DVBCSA_USE_NEON
+ b = BS_VAL(0x00aaaaaaaaaaaaaaLL, 0x0055555555555555LL);
+#else
b = BS_VAL(0x00aaaaaaaaaaaaaaLL, 0xaa55555555555555LL);
+#endif
c = BS_SHR8(a, 1);
vec_testeq(b, c);
a = BS_VAL(0xffffffffffffffffLL, 0xffffffffffffffffLL);
+#ifdef DVBCSA_USE_NEON
+ b = BS_VAL(0x00ffffffffffffffLL, 0x00ffffffffffffffLL);
+#else
b = BS_VAL(0x00ffffffffffffffLL, 0xffffffffffffffffLL);
+#endif
c = BS_SHR8(a, 1);
vec_testeq(b, c);
a = BS_VAL(0xffffffffffffffffLL, 0xffffffffffffffffLL);
+#ifdef DVBCSA_USE_NEON
+ b = BS_VAL(0x00000000ffffffffLL, 0x00000000ffffffffLL);
+#else
b = BS_VAL(0x00000000ffffffffLL, 0xffffffffffffffffLL);
+#endif
c = BS_SHR8(a, 4);
vec_testeq(b, c);
a = BS_VAL(0xffffffffffffffffLL, 0xffffffffffffffffLL);
+#ifdef DVBCSA_USE_NEON
+ b = BS_VAL(0x000000ffffffffffLL, 0x000000ffffffffffLL);
+#else
b = BS_VAL(0x000000ffffffffffLL, 0xffffffffffffffffLL);
+#endif
c = BS_SHR8(a, 3);
vec_testeq(b, c);
--
1.9.1
More information about the vlc-devel
mailing list