aboutsummaryrefslogtreecommitdiffstats
path: root/Transceiver52M/arm
diff options
context:
space:
mode:
authorThomas Tsou <tom@tsou.cc>2013-11-09 02:29:55 -0500
committerThomas Tsou <tom@tsou.cc>2013-11-15 23:34:59 -0500
commit0a3dc4c21028f36f02528c9a486de229f786b272 (patch)
tree86770439e6accc183d3f04c9e3f46f7e5b4d866c /Transceiver52M/arm
parentacc22fa3ffe30f28d62c41f62152b1f48fa595b4 (diff)
Transceiver52M: Add NEON complex-complex multiply
Complex-complex block multiples are used for phase rotation of bursts. Optimization targeted from perf profiling. Signed-off-by: Thomas Tsou <tom@tsou.cc>
Diffstat (limited to 'Transceiver52M/arm')
-rw-r--r--Transceiver52M/arm/Makefile.am4
-rw-r--r--Transceiver52M/arm/mult.c56
-rw-r--r--Transceiver52M/arm/mult_neon.S42
3 files changed, 101 insertions, 1 deletions
diff --git a/Transceiver52M/arm/Makefile.am b/Transceiver52M/arm/Makefile.am
index 6d34daa..0b959be 100644
--- a/Transceiver52M/arm/Makefile.am
+++ b/Transceiver52M/arm/Makefile.am
@@ -17,5 +17,7 @@ libarch_la_SOURCES = \
convolve.c \
convolve_neon.S \
scale.c \
- scale_neon.S
+ scale_neon.S \
+ mult.c \
+ mult_neon.S
endif
diff --git a/Transceiver52M/arm/mult.c b/Transceiver52M/arm/mult.c
new file mode 100644
index 0000000..245be50
--- /dev/null
+++ b/Transceiver52M/arm/mult.c
@@ -0,0 +1,56 @@
+/*
+ * NEON scaling
+ * Copyright (C) 2012,2013 Thomas Tsou <tom@tsou.cc>
+ *
+ * This 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.
+ *
+ * This 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 this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <malloc.h>
+#include <string.h>
+#include <mult.h>
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+void neon_cmplx_mul_4n(float *, float *, float *, int);
+
+static void cmplx_mul_ps(float *out, float *a, float *b, int len)
+{
+ float ai, aq, bi, bq;
+
+ for (int i = 0; i < len; i++) {
+ ai = a[2 * i + 0];
+ aq = a[2 * i + 1];
+
+ bi = b[2 * i + 0];
+ bq = b[2 * i + 1];
+
+ out[2 * i + 0] = ai * bi - aq * bq;
+ out[2 * i + 1] = ai * bq + aq * bi;
+ }
+}
+
+void mul_complex(float *out, float *a, float *b, int len)
+{
+#ifdef HAVE_NEON
+ if (len % 4)
+ cmplx_mul_ps(out, a, b, len);
+ else
+ neon_cmplx_mul_4n(out, a, b, len >> 2);
+#else
+ cmplx_mul_ps(out, a, b, len);
+#endif
+}
diff --git a/Transceiver52M/arm/mult_neon.S b/Transceiver52M/arm/mult_neon.S
new file mode 100644
index 0000000..162846e
--- /dev/null
+++ b/Transceiver52M/arm/mult_neon.S
@@ -0,0 +1,42 @@
+/*
+ * NEON complex multiplication
+ * Copyright (C) 2012,2013 Thomas Tsou <tom@tsou.cc>
+ *
+ * This 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.
+ *
+ * This 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 this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+ .syntax unified
+ .text
+ .align 2
+ .global neon_cmplx_mul_4n
+ .type neon_cmplx_mul_4n, %function
+neon_cmplx_mul_4n:
+ vpush {q4-q7}
+.loop_mul:
+ vld2.32 {q0-q1}, [r1]!
+ vld2.32 {q2-q3}, [r2]!
+ vmul.f32 q4, q0, q2
+ vmul.f32 q5, q1, q3
+ vmul.f32 q6, q0, q3
+ vmul.f32 q7, q2, q1
+ vsub.f32 q8, q4, q5
+ vadd.f32 q9, q6, q7
+ vst2.32 {q8-q9}, [r0]!
+ subs r3, #1
+ bne .loop_mul
+ vpop {q4-q7}
+ bx lr
+ .size neon_cmplx_mul_4n, .-neon_cmplx_mul_4n
+ .section .note.GNU-stack,"",%progbits