[Speex-dev] [PATCH 3/5] resample: Add NEON optimized inner_product_single for fixed point
Jyri Sarha
oku at iki.fi
Thu Sep 1 05:30:29 PDT 2011
From: Jyri Sarha <jsarha at ti.com>
Semantics of inner_product_single have also been changed to contain
the final right shift and saturation so it can also be implemented in
the optimal way for the used platform. This change affects fixed point
calculations only.
I also added a new fixed point macro SATURATE32PSHR(x, shift, a). It
does pretty much the same thing as SATURATE32(PSHR32(x, shift), a),
but it avoids over flowing in rounding up phase in the rare occasion
where x has already been saturated. It should also be slightly faster.
---
libspeex/arch.h | 1 +
libspeex/fixed_generic.h | 4 ++
libspeex/resample.c | 10 ++++-
libspeex/resample_neon.h | 100 ++++++++++++++++++++++++++++++++++++++++++++++
4 files changed, 113 insertions(+), 2 deletions(-)
create mode 100644 libspeex/resample_neon.h
diff --git a/libspeex/arch.h b/libspeex/arch.h
index 3b47ed9..daa72a7 100644
--- a/libspeex/arch.h
+++ b/libspeex/arch.h
@@ -171,6 +171,7 @@ typedef float spx_word32_t;
#define VSHR32(a,shift) (a)
#define SATURATE16(x,a) (x)
#define SATURATE32(x,a) (x)
+#define SATURATE32PSHR(x,shift,a) (x)
#define PSHR(a,shift) (a)
#define SHR(a,shift) (a)
diff --git a/libspeex/fixed_generic.h b/libspeex/fixed_generic.h
index 3fb096e..0e012e9 100644
--- a/libspeex/fixed_generic.h
+++ b/libspeex/fixed_generic.h
@@ -52,6 +52,10 @@
#define SATURATE16(x,a) (((x)>(a) ? (a) : (x)<-(a) ? -(a) : (x)))
#define SATURATE32(x,a) (((x)>(a) ? (a) : (x)<-(a) ? -(a) : (x)))
+#define SATURATE32PSHR(x,shift,a) (((x)>=(SHL32(a,shift))) ? (a) : \
+ (x)<=-(SHL32(a,shift)) ? -(a) : \
+ (PSHR32(x, shift)))
+
#define SHR(a,shift) ((a) >> (shift))
#define SHL(a,shift) ((spx_word32_t)(a) << (shift))
#define PSHR(a,shift) (SHR((a)+((EXTEND32(1)<<((shift))>>1)),shift))
diff --git a/libspeex/resample.c b/libspeex/resample.c
index 3829715..7957c61 100644
--- a/libspeex/resample.c
+++ b/libspeex/resample.c
@@ -99,6 +99,10 @@ static void speex_free (void *ptr) {free(ptr);}
#include "resample_sse.h"
#endif
+#ifdef _USE_NEON
+#include "resample_neon.h"
+#endif
+
/* Numer of elements to allocate on the stack */
#ifdef VAR_ARRAYS
#define FIXED_STACK_ALLOC 8192
@@ -360,11 +364,12 @@ static int resampler_basic_direct_single(SpeexResamplerState *st, spx_uint32_t c
}
sum = accum[0] + accum[1] + accum[2] + accum[3];
*/
+ sum = SATURATE32PSHR(sum, 15, 32767);
#else
sum = inner_product_single(sinc, iptr, N);
#endif
- out[out_stride * out_sample++] = SATURATE32(PSHR32(sum, 15), 32767);
+ out[out_stride * out_sample++] = sum;
last_sample += int_advance;
samp_frac_num += frac_advance;
if (samp_frac_num >= den_rate)
@@ -470,12 +475,13 @@ static int resampler_basic_interpolate_single(SpeexResamplerState *st, spx_uint3
cubic_coef(frac, interp);
sum = MULT16_32_Q15(interp[0],SHR32(accum[0], 1)) + MULT16_32_Q15(interp[1],SHR32(accum[1], 1)) + MULT16_32_Q15(interp[2],SHR32(accum[2], 1)) + MULT16_32_Q15(interp[3],SHR32(accum[3], 1));
+ sum = SATURATE32PSHR(sum, 15, 32767);
#else
cubic_coef(frac, interp);
sum = interpolate_product_single(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp);
#endif
- out[out_stride * out_sample++] = SATURATE32(PSHR32(sum, 14), 32767);
+ out[out_stride * out_sample++] = sum;
last_sample += int_advance;
samp_frac_num += frac_advance;
if (samp_frac_num >= den_rate)
diff --git a/libspeex/resample_neon.h b/libspeex/resample_neon.h
new file mode 100644
index 0000000..ba93e41
--- /dev/null
+++ b/libspeex/resample_neon.h
@@ -0,0 +1,100 @@
+/* Copyright (C) 2007-2008 Jean-Marc Valin
+ * Copyright (C) 2008 Thorvald Natvig
+ * Copyright (C) 2011 Texas Instruments
+ * author Jyri Sarha
+ */
+/**
+ @file resample_neon.h
+ @brief Resampler functions (NEON version)
+*/
+/*
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions
+ are met:
+
+ - Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+
+ - Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+
+ - Neither the name of the Xiph.org Foundation nor the names of its
+ contributors may be used to endorse or promote products derived from
+ this software without specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR
+ CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+ EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#include <arm_neon.h>
+
+#ifdef FIXED_POINT
+#define OVERRIDE_INNER_PRODUCT_SINGLE
+/* Only works when len % 4 == 0 */
+static inline int32_t inner_product_single(const int16_t *a, const int16_t *b, unsigned int len)
+{
+ int32_t ret;
+ uint32_t remainder = len % 16;
+ len = len - remainder;
+
+ asm volatile (" cmp %[len], #0\n"
+ " bne 1f\n"
+ " vld1.16 {d16}, [%[b]]!\n"
+ " vld1.16 {d20}, [%[a]]!\n"
+ " subs %[remainder], %[remainder], #4\n"
+ " vmull.s16 q0, d16, d20\n"
+ " beq 5f\n"
+ " b 4f\n"
+ "1:"
+ " vld1.16 {d16, d17, d18, d19}, [%[b]]!\n"
+ " vld1.16 {d20, d21, d22, d23}, [%[a]]!\n"
+ " subs %[len], %[len], #16\n"
+ " vmull.s16 q0, d16, d20\n"
+ " vmlal.s16 q0, d17, d21\n"
+ " vmlal.s16 q0, d18, d22\n"
+ " vmlal.s16 q0, d19, d23\n"
+ " beq 3f\n"
+ "2:"
+ " vld1.16 {d16, d17, d18, d19}, [%[b]]!\n"
+ " vld1.16 {d20, d21, d22, d23}, [%[a]]!\n"
+ " subs %[len], %[len], #16\n"
+ " vmlal.s16 q0, d16, d20\n"
+ " vmlal.s16 q0, d17, d21\n"
+ " vmlal.s16 q0, d18, d22\n"
+ " vmlal.s16 q0, d19, d23\n"
+ " bne 2b\n"
+ "3:"
+ " cmp %[remainder], #0\n"
+ " beq 5f\n"
+ "4:"
+ " vld1.16 {d16}, [%[b]]!\n"
+ " vld1.16 {d20}, [%[a]]!\n"
+ " subs %[remainder], %[remainder], #4\n"
+ " vmlal.s16 q0, d16, d20\n"
+ " bne 4b\n"
+ "5:"
+ " vaddl.s32 q0, d0, d1\n"
+ " vadd.s64 d0, d0, d1\n"
+ " vqmovn.s64 d0, q0\n"
+ " vqrshrn.s32 d0, q0, #15\n"
+ " vmov.s16 %[ret], d0[0]\n"
+ : [ret] "=&r" (ret), [a] "+r" (a), [b] "+r" (b),
+ [len] "+r" (len), [remainder] "+r" (remainder)
+ :
+ : "cc", "q0",
+ "d16", "d17", "d18", "d19",
+ "d20", "d21", "d22", "d23");
+
+ return ret;
+}
+#endif
--
1.7.4.1
More information about the Speex-dev
mailing list