libomsocoding: NEON viterbi acceleration

configure flag required to enable this: --enable-neon

Although autodetection according to __ARM_NEON would work because this
is only defined if the fpu is neon neon-fp16 neon-vfpv3 neon-vfpv4
neon-fp-armv8 crypto-neon-fp-armv8 doing that would lead to a unknown
performance impact, so it needs to be enabled manually.

Speedup is about ~1.3-1.5 on a unspecified single core Cortex A9. This
requires handling a special case for RACH with len 14 which is far too
short for neon and would actually incur a performance penalty of 25%.

Related: OS#4585
Change-Id: I58ff2cb4ce3514f43390ff0a2121f81e6a4983b5

1 files changed, 110 insertions, 0 deletions

diff --git a/src/conv_acc_neon.c b/src/conv_acc_neon.c
new file mode 100644
index 00000000..72449468
--- /dev/null
+++ b/src/conv_acc_neon.c
@@ -0,0 +1,110 @@
+/*! \file conv_acc_neon.c
+ * Accelerated Viterbi decoder implementation
+ * for architectures with only NEON available. */
+/*
+ * (C) 2020 by sysmocom - s.f.m.c. GmbH
+ * Author: Eric Wild
+ *
+ * All Rights Reserved
+ *
+ * SPDX-License-Identifier: GPL-2.0+
+ *
+ * This program 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.
+ *
+ * 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU 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.
+ */
+
+#include <stdlib.h>
+#include <stdint.h>
+#include <malloc.h>
+#include "config.h"
+
+#if defined(HAVE_NEON)
+#include <arm_neon.h>
+#endif
+
+/* align req is 16 on android because google was confused, 8 on sane platforms */
+#define NEON_ALIGN 8
+
+#include <conv_acc_neon_impl.h>
+
+/* Aligned Memory Allocator
+ * NEON requires 8-byte memory alignment. We store relevant trellis values
+ * (accumulated sums, outputs, and path decisions) as 16 bit signed integers
+ * so the allocated memory is casted as such.
+ */
+__attribute__ ((visibility("hidden")))
+int16_t *osmo_conv_neon_vdec_malloc(size_t n)
+{
+	return (int16_t *) memalign(NEON_ALIGN, sizeof(int16_t) * n);
+}
+
+__attribute__ ((visibility("hidden")))
+void osmo_conv_neon_vdec_free(int16_t *ptr)
+{
+	free(ptr);
+}
+
+__attribute__ ((visibility("hidden")))
+void osmo_conv_neon_metrics_k5_n2(const int8_t *val, const int16_t *out,
+	int16_t *sums, int16_t *paths, int norm)
+{
+	const int16_t _val[4] = { val[0], val[1], val[0], val[1] };
+
+	_neon_metrics_k5_n2(_val, out, sums, paths, norm);
+}
+
+__attribute__ ((visibility("hidden")))
+void osmo_conv_neon_metrics_k5_n3(const int8_t *val, const int16_t *out,
+	int16_t *sums, int16_t *paths, int norm)
+{
+	const int16_t _val[4] = { val[0], val[1], val[2], 0 };
+
+	_neon_metrics_k5_n4(_val, out, sums, paths, norm);
+}
+
+__attribute__ ((visibility("hidden")))
+void osmo_conv_neon_metrics_k5_n4(const int8_t *val, const int16_t *out,
+	int16_t *sums, int16_t *paths, int norm)
+{
+	const int16_t _val[4] = { val[0], val[1], val[2], val[3] };
+
+	_neon_metrics_k5_n4(_val, out, sums, paths, norm);
+}
+
+__attribute__ ((visibility("hidden")))
+void osmo_conv_neon_metrics_k7_n2(const int8_t *val, const int16_t *out,
+	int16_t *sums, int16_t *paths, int norm)
+{
+	const int16_t _val[4] = { val[0], val[1], val[0], val[1] };
+
+	_neon_metrics_k7_n2(_val, out, sums, paths, norm);
+}
+
+__attribute__ ((visibility("hidden")))
+void osmo_conv_neon_metrics_k7_n3(const int8_t *val, const int16_t *out,
+	int16_t *sums, int16_t *paths, int norm)
+{
+	const int16_t _val[4] = { val[0], val[1], val[2], 0 };
+
+	_neon_metrics_k7_n4(_val, out, sums, paths, norm);
+}
+
+__attribute__ ((visibility("hidden")))
+void osmo_conv_neon_metrics_k7_n4(const int8_t *val, const int16_t *out,
+	int16_t *sums, int16_t *paths, int norm)
+{
+	const int16_t _val[4] = { val[0], val[1], val[2], val[3] };
+
+	_neon_metrics_k7_n4(_val, out, sums, paths, norm);
+}