swresample/aarch64: add ff_resample_common_apply_filter_{x4,x8}_{float,s16}_neon

3 files changed, 200 insertions, 2 deletions

diff --git a/libswresample/aarch64/Makefile b/libswresample/aarch64/Makefile
index 320ed67e82..5c34f8d949 100644
--- a/libswresample/aarch64/Makefile
+++ b/libswresample/aarch64/Makefile
@@ -1,5 +1,7 @@
-OBJS                             += aarch64/audio_convert_init.o
+OBJS                             += aarch64/audio_convert_init.o \
+                                    aarch64/resample_init.o
 
 OBJS-$(CONFIG_NEON_CLOBBER_TEST) += aarch64/neontest.o
 
-NEON-OBJS                        += aarch64/audio_convert_neon.o
+NEON-OBJS                        += aarch64/audio_convert_neon.o \
+                                    aarch64/resample.o
diff --git a/libswresample/aarch64/resample.S b/libswresample/aarch64/resample.S
new file mode 100644
index 0000000000..b3b1bb9fe1
--- /dev/null
+++ b/libswresample/aarch64/resample.S
@@ -0,0 +1,76 @@
+/*
+ * Copyright (c) 2017 Matthieu Bouron <matthieu.bouron gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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.
+ *
+ * FFmpeg 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 FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/aarch64/asm.S"
+
+function ff_resample_common_apply_filter_x4_float_neon, export=1
+    movi                v0.4S, #0                                      // accumulator
+1:  ld1                 {v1.4S}, [x1], #16                             // src[0..3]
+    ld1                 {v2.4S}, [x2], #16                             // filter[0..3]
+    fmla                v0.4S, v1.4S, v2.4S                            // accumulator += src[0..3] * filter[0..3]
+    subs                w3, w3, #4                                     // filter_length -= 4
+    b.gt                1b                                             // loop until filter_length
+    faddp               v0.4S, v0.4S, v0.4S                            // pair adding of the 4x32-bit accumulated values
+    faddp               v0.4S, v0.4S, v0.4S                            // pair adding of the 4x32-bit accumulated values
+    st1                 {v0.S}[0], [x0], #4                            // write accumulator
+    ret
+endfunc
+
+function ff_resample_common_apply_filter_x8_float_neon, export=1
+    movi                v0.4S, #0                                      // accumulator
+1:  ld1                 {v1.4S}, [x1], #16                             // src[0..3]
+    ld1                 {v2.4S}, [x2], #16                             // filter[0..3]
+    ld1                 {v3.4S}, [x1], #16                             // src[4..7]
+    ld1                 {v4.4S}, [x2], #16                             // filter[4..7]
+    fmla                v0.4S, v1.4S, v2.4S                            // accumulator += src[0..3] * filter[0..3]
+    fmla                v0.4S, v3.4S, v4.4S                            // accumulator += src[4..7] * filter[4..7]
+    subs                w3, w3, #8                                     // filter_length -= 8
+    b.gt                1b                                             // loop until filter_length
+    faddp               v0.4S, v0.4S, v0.4S                            // pair adding of the 4x32-bit accumulated values
+    faddp               v0.4S, v0.4S, v0.4S                            // pair adding of the 4x32-bit accumulated values
+    st1                 {v0.S}[0], [x0], #4                            // write accumulator
+endfunc
+
+function ff_resample_common_apply_filter_x4_s16_neon, export=1
+    movi                v0.4S, #0                                      // accumulator
+1:  ld1                 {v1.4H}, [x1], #8                              // src[0..3]
+    ld1                 {v2.4H}, [x2], #8                              // filter[0..3]
+    smlal               v0.4S, v1.4H, v2.4H                            // accumulator += src[0..3] * filter[0..3]
+    subs                w3, w3, #4                                     // filter_length -= 4
+    b.gt                1b                                             // loop until filter_length
+    addp                v0.4S, v0.4S, v0.4S                            // pair adding of the 4x32-bit accumulated values
+    addp                v0.4S, v0.4S, v0.4S                            // pair adding of the 4x32-bit accumulated values
+    st1                 {v0.S}[0], [x0], #4                            // write accumulator
+    ret
+endfunc
+
+function ff_resample_common_apply_filter_x8_s16_neon, export=1
+    movi                v0.4S, #0                                      // accumulator
+1:  ld1                 {v1.8H}, [x1], #16                             // src[0..7]
+    ld1                 {v2.8H}, [x2], #16                             // filter[0..7]
+    smlal               v0.4S, v1.4H, v2.4H                            // accumulator += src[0..3] * filter[0..3]
+    smlal2              v0.4S, v1.8H, v2.8H                            // accumulator += src[4..7] * filter[4..7]
+    subs                w3, w3, #8                                     // filter_length -= 8
+    b.gt                1b                                             // loop until filter_length
+    addp                v0.4S, v0.4S, v0.4S                            // pair adding of the 4x32-bit accumulated values
+    addp                v0.4S, v0.4S, v0.4S                            // pair adding of the 4x32-bit accumulated values
+    st1                 {v0.S}[0], [x0], #4                            // write accumulator
+    ret
+endfunc
diff --git a/libswresample/aarch64/resample_init.c b/libswresample/aarch64/resample_init.c
new file mode 100644
index 0000000000..d01ec18756
--- /dev/null
+++ b/libswresample/aarch64/resample_init.c
@@ -0,0 +1,120 @@
+/*
+ * Audio resampling
+ *
+ * Copyright (c) 2004-2012 Michael Niedermayer <michaelni@gmx.at>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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.
+ *
+ * FFmpeg 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 FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "config.h"
+
+#include "libavutil/cpu.h"
+#include "libavutil/avassert.h"
+
+#include "libavutil/aarch64/cpu.h"
+#include "libswresample/resample.h"
+
+#define DECLARE_RESAMPLE_COMMON_TEMPLATE(TYPE, DELEM, FELEM, FELEM2, OUT)                         \
+                                                                                                  \
+void ff_resample_common_apply_filter_x4_##TYPE##_neon(FELEM2 *acc, const DELEM *src,              \
+                                                      const FELEM *filter, int length);           \
+                                                                                                  \
+void ff_resample_common_apply_filter_x8_##TYPE##_neon(FELEM2 *acc, const DELEM *src,              \
+                                                      const FELEM *filter, int length);           \
+                                                                                                  \
+static int ff_resample_common_##TYPE##_neon(ResampleContext *c, void *dest, const void *source,   \
+                                            int n, int update_ctx)                                \
+{                                                                                                 \
+    DELEM *dst = dest;                                                                            \
+    const DELEM *src = source;                                                                    \
+    int dst_index;                                                                                \
+    int index = c->index;                                                                         \
+    int frac = c->frac;                                                                           \
+    int sample_index = 0;                                                                         \
+    int x4_aligned_filter_length = c->filter_length & ~3;                                         \
+    int x8_aligned_filter_length = c->filter_length & ~7;                                         \
+                                                                                                  \
+    while (index >= c->phase_count) {                                                             \
+        sample_index++;                                                                           \
+        index -= c->phase_count;                                                                  \
+    }                                                                                             \
+                                                                                                  \
+    for (dst_index = 0; dst_index < n; dst_index++) {                                             \
+        FELEM *filter = ((FELEM *) c->filter_bank) + c->filter_alloc * index;                     \
+                                                                                                  \
+        FELEM2 val = 0;                                                                             \
+        int i = 0;                                                                                \
+        if (x8_aligned_filter_length >= 8) {                                                      \
+            ff_resample_common_apply_filter_x8_##TYPE##_neon(&val, &src[sample_index],            \
+                                                             filter, x8_aligned_filter_length);   \
+            i += x8_aligned_filter_length;                                                        \
+                                                                                                  \
+        } else if (x4_aligned_filter_length >= 4) {                                               \
+            ff_resample_common_apply_filter_x4_##TYPE##_neon(&val, &src[sample_index],            \
+                                                             filter, x4_aligned_filter_length);   \
+            i += x4_aligned_filter_length;                                                        \
+        }                                                                                         \
+        for (; i < c->filter_length; i++) {                                                       \
+            val += src[sample_index + i] * (FELEM2)filter[i];                                     \
+        }                                                                                         \
+        OUT(dst[dst_index], val);                                                                 \
+                                                                                                  \
+        frac  += c->dst_incr_mod;                                                                 \
+        index += c->dst_incr_div;                                                                 \
+        if (frac >= c->src_incr) {                                                                \
+            frac -= c->src_incr;                                                                  \
+            index++;                                                                              \
+        }                                                                                         \
+                                                                                                  \
+        while (index >= c->phase_count) {                                                         \
+            sample_index++;                                                                       \
+            index -= c->phase_count;                                                              \
+        }                                                                                         \
+    }                                                                                             \
+                                                                                                  \
+    if (update_ctx) {                                                                             \
+        c->frac = frac;                                                                           \
+        c->index = index;                                                                         \
+    }                                                                                             \
+                                                                                                  \
+    return sample_index;                                                                          \
+}                                                                                                 \
+
+#define OUT(d, v) d = v
+DECLARE_RESAMPLE_COMMON_TEMPLATE(float, float, float, float, OUT)
+#undef OUT
+
+#define OUT(d, v) (v) = ((v) + (1<<(14)))>>15; (d) = av_clip_int16(v)
+DECLARE_RESAMPLE_COMMON_TEMPLATE(s16, int16_t, int16_t, int32_t, OUT)
+#undef OUT
+
+av_cold void swri_resample_dsp_aarch64_init(ResampleContext *c)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (!have_neon(cpu_flags))
+        return;
+
+    switch(c->format) {
+    case AV_SAMPLE_FMT_FLTP:
+        c->dsp.resample_common = ff_resample_common_float_neon;
+        break;
+    case AV_SAMPLE_FMT_S16P:
+        c->dsp.resample_common = ff_resample_common_s16_neon;
+        break;
+    }
+}