deinterlace: arm64 NEON merge asm

Approximately factor 2 faster.

Also adds build system support / cpu "detection" for arm64 neon.
Advanced SIMD (neon) is mandatory for general purpose ARMv8-a CPU so the
CPU feature detection is a constant 1.

Signed-off-by: Felix Paul Kühne <fkuehne@videolan.org>

configure.ac

@@ -1432,6 +1432,26 @@ asm volatile("vqmovun.s64 d0, q1":::"d0");
 ])
 AM_CONDITIONAL(HAVE_NEON, [test "${ac_cv_arm_neon}" = "yes"])
 
+AC_ARG_ENABLE(arm64,
+  [AS_HELP_STRING([--disable-arm64],
+    [disable arm 64-bit optimizations (default auto)])],, [
+  AS_IF([test "${host_cpu}" = "aarch64"], [enable_arm64="yes"] ,[enable_arm64="no"])
+])
+AS_IF([test "${enable_arm64}" != "no"], [
+  AC_CACHE_CHECK([if $CCAS groks ARM 64 SIMD assembly], [ac_cv_arm64], [
+    AC_COMPILE_IFELSE([
+      AC_LANG_PROGRAM(,[[
+asm volatile("uhadd v0.8b, v0.8b, v1.8b":::"v0");
+]])
+    ], [
+      ac_cv_arm64="yes"
+    ], [
+      ac_cv_arm64="no"
+    ])
+  ])
+])
+AM_CONDITIONAL(HAVE_ARM64, [test "${ac_cv_arm64}" = "yes"])
+
 
 AC_ARG_ENABLE(altivec,
   [AS_HELP_STRING([--disable-altivec],

include/vlc_cpu.h

@@ -178,6 +178,8 @@ VLC_API unsigned vlc_CPU(void);
 
 # elif defined (__aarch64__)
 #  define HAVE_FPU 1
+// NEON is mandatory for general purpose ARMv8-a CPUs
+#  define vlc_CPU_ARM64_NEON() (1)
 
 # elif defined (__sparc__)
 #  define HAVE_FPU 1

modules/video_filter/Makefile.am

@@ -124,6 +124,10 @@ if HAVE_NEON
 libdeinterlace_plugin_la_SOURCES += video_filter/deinterlace/merge_arm.S
 libdeinterlace_plugin_la_CFLAGS += -DCAN_COMPILE_ARM
 endif
+if HAVE_ARM64
+libdeinterlace_plugin_la_SOURCES += video_filter/deinterlace/merge_arm64.S
+libdeinterlace_plugin_la_CFLAGS += -DCAN_COMPILE_ARM64
+endif
 video_filter_LTLIBRARIES += libdeinterlace_plugin.la
 
 libdynamicoverlay_plugin_la_SOURCES = \

modules/video_filter/deinterlace/deinterlace.c

@@ -699,6 +699,11 @@ notsupp:
     if( vlc_CPU_ARMv6() )
         p_sys->pf_merge = pixel_size == 1 ? merge8_armv6 : merge16_armv6;
     else
+#endif
+#if defined(CAN_COMPILE_ARM64)
+    if( vlc_CPU_ARM64_NEON() )
+        p_sys->pf_merge = pixel_size == 1 ? merge8_arm64_neon : merge16_arm64_neon;
+    else
 #endif
     {
         p_sys->pf_merge = pixel_size == 1 ? Merge8BitGeneric : Merge16BitGeneric;

modules/video_filter/deinterlace/merge.h

@@ -172,6 +172,15 @@ void merge8_armv6 (void *, const void *, const void *, size_t);
 void merge16_armv6 (void *, const void *, const void *, size_t);
 #endif
 
+#if defined(CAN_COMPILE_ARM64)
+/**
+ * ARM64 NEON routine to blend pixels from two picture lines.
+ */
+void merge8_arm64_neon (void *, const void *, const void *, size_t);
+void merge16_arm64_neon (void *, const void *, const void *, size_t);
+
+#endif
+
 /*****************************************************************************
  * EndMerge routines
  *****************************************************************************/

modules/video_filter/deinterlace/merge_arm64.S

+ //*****************************************************************************
+ // merge_arm64.S : ARM64 NEON mean
+ //*****************************************************************************
+ // Copyright (C) 2009-2012 Rémi Denis-Courmont
+ // Copyright (C) 2016-	   Janne Grunau
+ //
+ // This program 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.
+ //
+ // This program 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 this program; if not, write to the Free Software Foundation,
+ // Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
+ //****************************************************************************/
+
+	.text
+
+#define	DEST	x0
+#define	SRC1	x1
+#define	SRC2	x2
+#define	SIZE	x3
+
+	.align 2
+	.global merge8_arm64_neon
+	.type	merge8_arm64_neon, %function
+	// NOTE: Offset and pitch must be multiple of 16-bytes in VLC.
+merge8_arm64_neon:
+	ands		x5, SIZE, #~63
+	b.eq		2f
+	mov		x10, #64
+	add		x11, SRC1, #32
+        add             x12, SRC2, #32
+1:
+	ld1		{v0.16b,v1.16b}, [SRC1], x10
+	ld1		{v4.16b,v5.16b}, [SRC2], x10
+	ld1		{v2.16b,v3.16b}, [x11], x10
+	uhadd		v0.16b, v0.16b, v4.16b
+	ld1		{v6.16b,v7.16b}, [x12], x10
+	subs		x5,  x5,  #64
+	uhadd		v0.16b, v0.16b, v4.16b
+	uhadd		v1.16b, v1.16b, v5.16b
+	uhadd		v2.16b, v2.16b, v6.16b
+	uhadd		v3.16b, v3.16b, v7.16b
+	st1		{v0.16b,v1.16b}, [DEST], #32
+	st1		{v2.16b,v3.16b}, [DEST], #32
+	b.gt		1b
+2:
+	tbz		SIZE, #32, 3f
+	ld1		{v0.16b,v1.16b}, [SRC1], #32
+	ld1		{v4.16b,v5.16b}, [SRC2], #32
+	uhadd		v0.16b, v0.16b, v4.16b
+	uhadd		v1.16b, v1.16b, v5.16b
+	st1		{v0.16b,v1.16b}, [DEST], #32
+3:
+	tbz		SIZE, #16, 4f
+	ld1		{v0.16b},  [SRC1]
+	ld1		{v4.16b},  [SRC2]
+	uhadd		v0.16b, v0.16b, v4.16b
+	st1		{v0.16b},  [DEST]
+4:
+	ret
+
+	.align 2
+	.global merge16_arm64_neon
+	.type	merge16_arm64_neon, %function
+merge16_arm64_neon:
+	ands		x5, SIZE, #~63
+	b.eq		2f
+1:
+	ld1		{v0.8h,v1.8h}, [SRC1], #32
+	ld1		{v4.8h,v5.8h}, [SRC2], #32
+	ld1		{v2.8h,v3.8h}, [SRC1], #32
+	uhadd		v0.8h,  v0.8h,  v4.8h
+	ld1		{v6.8h,v7.8h}, [SRC2], #32
+	uhadd		v1.8h,  v1.8h,  v5.8h
+	uhadd		v2.8h,  v2.8h,  v6.8h
+	uhadd		v3.8h,  v3.8h,  v7.8h
+	st1		{v0.8h,v1.8h}, [DEST], #32
+	st1		{v2.8h,v3.8h}, [DEST], #32
+	subs		x5,  x5,  #64
+	b.gt		1b
+2:
+	tbz		SIZE, #32, 3f
+	ld1		{v0.8h,v1.8h}, [SRC1], #32
+	ld1		{v4.8h,v5.8h}, [SRC2], #32
+	uhadd		v0.8h,  v0.8h,  v4.8h
+	uhadd		v1.8h,  v1.8h,  v5.8h
+	st1		{v0.8h,v1.8h}, [DEST], #32
+3:
+	tbz		SIZE, #16, 4f
+	ld1		{v0.8h},  [SRC1]
+	ld1		{v4.8h},  [SRC2]
+	uhadd		v0.8h,  v0.8h,  v4.8h
+	st1		{v0.8h},  [DEST]
+4:
+	ret