lavc/aarch64: add sbrdsp neon implementation

autocorrelate_c: 644.0
autocorrelate_neon: 420.0
hf_apply_noise_0_c: 1688.5
hf_apply_noise_0_neon: 1498.6
hf_apply_noise_1_c: 1691.2
hf_apply_noise_1_neon: 1500.6
hf_apply_noise_2_c: 1688.1
hf_apply_noise_2_neon: 1500.3
hf_apply_noise_3_c: 1696.6
hf_apply_noise_3_neon: 1502.2
hf_g_filt_c: 2117.8
hf_g_filt_neon: 1218.7
hf_gen_c: 4573.4
hf_gen_neon: 2461.0
neg_odd_64_c: 72.0
neg_odd_64_neon: 64.7
qmf_deint_bfly_c: 1107.6
qmf_deint_bfly_neon: 291.6
qmf_deint_neg_c: 210.4
qmf_deint_neg_neon: 107.4
qmf_post_shuffle_c: 163.0
qmf_post_shuffle_neon: 107.7
qmf_pre_shuffle_c: 120.5
qmf_pre_shuffle_neon: 110.7
sum64x5_c: 1361.6
sum64x5_neon: 435.4
sum_square_c: 1686.4
sum_square_neon: 787.2

3 files changed, 400 insertions, 1 deletions

diff --git a/libavcodec/aarch64/Makefile b/libavcodec/aarch64/Makefile
index d440b1b18a..72080c2dbb 100644
--- a/libavcodec/aarch64/Makefile
+++ b/libavcodec/aarch64/Makefile
@@ -11,7 +11,8 @@ OBJS-$(CONFIG_NEON_CLOBBER_TEST)        += aarch64/neontest.o
 OBJS-$(CONFIG_VIDEODSP)                 += aarch64/videodsp_init.o
 
 # decoders/encoders
-OBJS-$(CONFIG_AAC_DECODER)              += aarch64/aacpsdsp_init_aarch64.o
+OBJS-$(CONFIG_AAC_DECODER)              += aarch64/aacpsdsp_init_aarch64.o \
+                                           aarch64/sbrdsp_init_aarch64.o
 OBJS-$(CONFIG_DCA_DECODER)              += aarch64/synth_filter_init.o
 OBJS-$(CONFIG_RV40_DECODER)             += aarch64/rv40dsp_init_aarch64.o
 OBJS-$(CONFIG_VC1DSP)                   += aarch64/vc1dsp_init_aarch64.o
@@ -28,6 +29,7 @@ ARMV8-OBJS-$(CONFIG_VIDEODSP)           += aarch64/videodsp.o
 # NEON optimizations
 
 # subsystems
+NEON-OBJS-$(CONFIG_AAC_DECODER)         += aarch64/sbrdsp_neon.o
 NEON-OBJS-$(CONFIG_FFT)                 += aarch64/fft_neon.o
 NEON-OBJS-$(CONFIG_FMTCONVERT)          += aarch64/fmtconvert_neon.o
 NEON-OBJS-$(CONFIG_H264CHROMA)          += aarch64/h264cmc_neon.o
diff --git a/libavcodec/aarch64/sbrdsp_init_aarch64.c b/libavcodec/aarch64/sbrdsp_init_aarch64.c
new file mode 100644
index 0000000000..9c967990df
--- /dev/null
+++ b/libavcodec/aarch64/sbrdsp_init_aarch64.c
@@ -0,0 +1,70 @@
+/*
+ * 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/aarch64/cpu.h"
+#include "libavutil/attributes.h"
+#include "libavcodec/sbrdsp.h"
+
+void ff_sbr_sum64x5_neon(float *z);
+float ff_sbr_sum_square_neon(float (*x)[2], int n);
+void ff_sbr_neg_odd_64_neon(float *x);
+void ff_sbr_qmf_pre_shuffle_neon(float *z);
+void ff_sbr_qmf_post_shuffle_neon(float W[32][2], const float *z);
+void ff_sbr_qmf_deint_neg_neon(float *v, const float *src);
+void ff_sbr_qmf_deint_bfly_neon(float *v, const float *src0, const float *src1);
+void ff_sbr_hf_g_filt_neon(float (*Y)[2], const float (*X_high)[40][2],
+                           const float *g_filt, int m_max, intptr_t ixh);
+void ff_sbr_hf_gen_neon(float (*X_high)[2], const float (*X_low)[2],
+                        const float alpha0[2], const float alpha1[2],
+                        float bw, int start, int end);
+void ff_sbr_autocorrelate_neon(const float x[40][2], float phi[3][2][2]);
+void ff_sbr_hf_apply_noise_0_neon(float Y[64][2], const float *s_m,
+                                  const float *q_filt, int noise,
+                                  int kx, int m_max);
+void ff_sbr_hf_apply_noise_1_neon(float Y[64][2], const float *s_m,
+                                  const float *q_filt, int noise,
+                                  int kx, int m_max);
+void ff_sbr_hf_apply_noise_2_neon(float Y[64][2], const float *s_m,
+                                  const float *q_filt, int noise,
+                                  int kx, int m_max);
+void ff_sbr_hf_apply_noise_3_neon(float Y[64][2], const float *s_m,
+                                  const float *q_filt, int noise,
+                                  int kx, int m_max);
+
+av_cold void ff_sbrdsp_init_aarch64(SBRDSPContext *s)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (have_neon(cpu_flags)) {
+        s->sum64x5 = ff_sbr_sum64x5_neon;
+        s->sum_square = ff_sbr_sum_square_neon;
+        s->neg_odd_64 = ff_sbr_neg_odd_64_neon;
+        s->qmf_pre_shuffle = ff_sbr_qmf_pre_shuffle_neon;
+        s->qmf_post_shuffle = ff_sbr_qmf_post_shuffle_neon;
+        s->qmf_deint_neg = ff_sbr_qmf_deint_neg_neon;
+        s->qmf_deint_bfly = ff_sbr_qmf_deint_bfly_neon;
+        s->hf_g_filt = ff_sbr_hf_g_filt_neon;
+        s->hf_gen = ff_sbr_hf_gen_neon;
+        s->autocorrelate = ff_sbr_autocorrelate_neon;
+        s->hf_apply_noise[0] = ff_sbr_hf_apply_noise_0_neon;
+        s->hf_apply_noise[1] = ff_sbr_hf_apply_noise_1_neon;
+        s->hf_apply_noise[2] = ff_sbr_hf_apply_noise_2_neon;
+        s->hf_apply_noise[3] = ff_sbr_hf_apply_noise_3_neon;
+    }
+}
diff --git a/libavcodec/aarch64/sbrdsp_neon.S b/libavcodec/aarch64/sbrdsp_neon.S
new file mode 100644
index 0000000000..d1d79b749c
--- /dev/null
+++ b/libavcodec/aarch64/sbrdsp_neon.S
@@ -0,0 +1,327 @@
+/*
+ * 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"
+
+const factors, align=4
+        .float 1.0, -1.0, 1.0, -1.0
+endconst
+
+const phi_noise_0, align=4
+        .float 1.0, 0.0, 1.0, 0.0
+endconst
+
+const phi_noise_1, align=4
+        .float 0.0,  1.0,  0.0, -1.0
+        .float 0.0, -1.0,  0.0,  1.0
+endconst
+
+const phi_noise_2, align=4
+        .float -1.0, 0.0, -1.0, 0.0
+endconst
+
+const phi_noise_3, align=4
+        .float 0.0, -1.0,  0.0,  1.0
+        .float 0.0,  1.0,  0.0, -1.0
+endconst
+
+function ff_sbr_sum64x5_neon, export=1
+        add             x1, x0, #64*4
+        add             x2, x0, #128*4
+        add             x3, x0, #192*4
+        add             x4, x0, #256*4
+        mov             x5, #64
+1:      ld1             {v0.4S}, [x0]
+        ld1             {v1.4S}, [x1], #16
+        fadd            v0.4S, v0.4S, v1.4S
+        ld1             {v2.4S}, [x2], #16
+        fadd            v0.4S, v0.4S, v2.4S
+        ld1             {v3.4S}, [x3], #16
+        fadd            v0.4S, v0.4S, v3.4S
+        ld1             {v4.4S}, [x4], #16
+        fadd            v0.4S, v0.4S, v4.4S
+        st1             {v0.4S}, [x0], #16
+        subs            x5, x5, #4
+        b.gt            1b
+        ret
+endfunc
+
+function ff_sbr_sum_square_neon, export=1
+        movi            v0.4S, #0
+1:      ld1             {v1.4S}, [x0], #16
+        fmla            v0.4S, v1.4S, v1.4S
+        subs            w1, w1, #2
+        b.gt            1b
+        faddp           v0.4S, v0.4S, v0.4S
+        faddp           v0.4S, v0.4S, v0.4S
+        ret
+endfunc
+
+function ff_sbr_neg_odd_64_neon, export=1
+        mov             x1, x0
+        movi            v5.4S, #1<<7, lsl #24
+        ld2             {v0.4S, v1.4S}, [x0], #32
+        eor             v1.16B, v1.16B, v5.16B
+        ld2             {v2.4S, v3.4S}, [x0], #32
+.rept 3
+        st2             {v0.4S, v1.4S}, [x1], #32
+        eor             v3.16B, v3.16B, v5.16B
+        ld2             {v0.4S, v1.4S}, [x0], #32
+        st2             {v2.4S, v3.4S}, [x1], #32
+        eor             v1.16B, v1.16B, v5.16B
+        ld2             {v2.4S, v3.4S}, [x0], #32
+.endr
+        eor             v3.16B, v3.16B, v5.16B
+        st2             {v0.4S, v1.4S}, [x1], #32
+        st2             {v2.4S, v3.4S}, [x1], #32
+        ret
+endfunc
+
+function ff_sbr_qmf_pre_shuffle_neon, export=1
+        add             x1, x0, #60*4
+        add             x2, x0, #64*4
+        mov             x3, #-16
+        mov             x4, #-4
+        movi            v6.4S, #1<<7, lsl #24
+        ld1             {v0.2S}, [x0], #8
+        st1             {v0.2S}, [x2], #8
+.rept 7
+        ld1             {v1.4S}, [x1], x3
+        ld1             {v2.4S}, [x0], #16
+        eor             v1.16B, v1.16B, v6.16B
+        rev64           v1.4S, v1.4S
+        ext             v1.16B, v1.16B, v1.16B, #8
+        st2             {v1.4S, v2.4S}, [x2], #32
+.endr
+        add             x1, x1, #8
+        ld1             {v1.2S}, [x1], x4
+        ld1             {v2.2S}, [x0], #8
+        ld1             {v1.S}[3], [x1]
+        ld1             {v2.S}[2], [x0]
+        eor             v1.16B, v1.16B, v6.16B
+        rev64           v1.4S, v1.4S
+        st2             {v1.2S, v2.2S}, [x2], #16
+        st2             {v1.S, v2.S}[2], [x2]
+        ret
+endfunc
+
+function ff_sbr_qmf_post_shuffle_neon, export=1
+        add             x2, x1, #60*4
+        mov             x3, #-16
+        mov             x4, #32
+        movi            v6.4S, #1<<7, lsl #24
+1:      ld1             {v0.4S}, [x2], x3
+        ld1             {v1.4S}, [x1], #16
+        eor             v0.16B, v0.16B, v6.16B
+        rev64           v0.4S, v0.4S
+        ext             v0.16B, v0.16B, v0.16B, #8
+        st2             {v0.4S, v1.4S}, [x0], #32
+        subs            x4, x4, #4
+        b.gt            1b
+        ret
+endfunc
+
+function ff_sbr_qmf_deint_neg_neon, export=1
+        add             x1, x1, #56*4
+        add             x2, x0, #60*4
+        mov             x3, #-32
+        mov             x4, #32
+        movi            v2.4S, #1<<7, lsl #24
+1:      ld2             {v0.4S, v1.4S}, [x1], x3
+        eor             v0.16B, v0.16B, v2.16B
+        rev64           v1.4S, v1.4S
+        ext             v1.16B, v1.16B, v1.16B, #8
+        st1             {v0.4S}, [x2]
+        st1             {v1.4S}, [x0], #16
+        sub             x2, x2, #16
+        subs            x4, x4, #4
+        b.gt            1b
+        ret
+endfunc
+
+function ff_sbr_qmf_deint_bfly_neon, export=1
+        add             x2, x2, #60*4
+        add             x3, x0, #124*4
+        mov             x4, #64
+        mov             x5, #-16
+1:      ld1             {v0.4S}, [x1], #16
+        ld1             {v1.4S}, [x2], x5
+        rev64           v2.4S, v0.4S
+        ext             v2.16B, v2.16B, v2.16B, #8
+        rev64           v3.4S, v1.4S
+        ext             v3.16B, v3.16B, v3.16B, #8
+        fadd            v1.4S, v1.4S, v2.4S
+        fsub            v0.4S, v0.4S, v3.4S
+        st1             {v0.4S}, [x0], #16
+        st1             {v1.4S}, [x3], x5
+        subs            x4, x4, #4
+        b.gt            1b
+        ret
+endfunc
+
+function ff_sbr_hf_gen_neon, export=1
+        sxtw            x4, w4
+        sxtw            x5, w5
+        movrel          x6, factors
+        ld1             {v7.4S}, [x6]
+        dup             v1.4S, v0.S[0]
+        mov             v2.8B, v1.8B
+        mov             v2.S[2], v7.S[0]
+        mov             v2.S[3], v7.S[0]
+        fmul            v1.4S, v1.4S, v2.4S
+        ld1             {v0.D}[0], [x3]
+        ld1             {v0.D}[1], [x2]
+        fmul            v0.4S, v0.4S, v1.4S
+        fmul            v1.4S, v0.4S, v7.4S
+        rev64           v0.4S, v0.4S
+        sub             x7, x5, x4
+        add             x0, x0, x4, lsl #3
+        add             x1, x1, x4, lsl #3
+        sub             x1, x1, #16
+1:      ld1             {v2.4S}, [x1], #16
+        ld1             {v3.2S}, [x1]
+        fmul            v4.4S, v2.4S, v1.4S
+        fmul            v5.4S, v2.4S, v0.4S
+        faddp           v4.4S, v4.4S, v4.4S
+        faddp           v5.4S, v5.4S, v5.4S
+        faddp           v4.4S, v4.4S, v4.4S
+        faddp           v5.4S, v5.4S, v5.4S
+        mov             v4.S[1], v5.S[0]
+        fadd            v4.2S, v4.2S, v3.2S
+        st1             {v4.2S}, [x0], #8
+        sub             x1, x1, #8
+        subs            x7, x7, #1
+        b.gt            1b
+        ret
+endfunc
+
+function ff_sbr_hf_g_filt_neon, export=1
+        sxtw            x3, w3
+        sxtw            x4, w4
+        mov             x5, #40*2*4
+        add             x1, x1, x4, lsl #3
+1:      ld1             {v0.2S}, [x1], x5
+        ld1             {v1.S}[0], [x2], #4
+        fmul            v2.4S, v0.4S, v1.S[0]
+        st1             {v2.2S}, [x0], #8
+        subs            x3, x3, #1
+        b.gt            1b
+        ret
+endfunc
+
+function ff_sbr_autocorrelate_neon, export=1
+        mov             x2, #38
+        movrel          x3, factors
+        ld1             {v0.4S}, [x3]
+        movi            v1.4S, #0
+        movi            v2.4S, #0
+        movi            v3.4S, #0
+        ld1             {v4.2S}, [x0], #8
+        ld1             {v5.2S}, [x0], #8
+        fmul            v16.2S, v4.2S, v4.2S
+        fmul            v17.2S, v5.2S, v4.S[0]
+        fmul            v18.2S, v5.2S, v4.S[1]
+1:      ld1             {v5.D}[1], [x0], #8
+        fmla            v1.2S, v4.2S, v4.2S
+        fmla            v2.4S, v5.4S, v4.S[0]
+        fmla            v3.4S, v5.4S, v4.S[1]
+        mov             v4.D[0], v5.D[0]
+        mov             v5.D[0], v5.D[1]
+        subs            x2, x2, #1
+        b.gt            1b
+        fmul            v19.2S, v4.2S, v4.2S
+        fmul            v20.2S, v5.2S, v4.S[0]
+        fmul            v21.2S, v5.2S, v4.S[1]
+        fadd            v22.4S, v2.4S, v20.4S
+        fsub            v22.4S, v22.4S, v17.4S
+        fadd            v23.4S, v3.4S, v21.4S
+        fsub            v23.4S, v23.4S, v18.4S
+        rev64           v23.4S, v23.4S
+        fmul            v23.4S, v23.4S, v0.4S
+        fadd            v22.4S, v22.4S, v23.4S
+        st1             {v22.4S}, [x1], #16
+        fadd            v23.2S, v1.2S, v19.2S
+        fsub            v23.2S, v23.2S, v16.2S
+        faddp           v23.2S, v23.2S, v23.2S
+        st1             {v23.S}[0], [x1]
+        add             x1, x1, #8
+        rev64           v3.2S, v3.2S
+        fmul            v3.2S, v3.2S, v0.2S
+        fadd            v2.2S, v2.2S, v3.2S
+        st1             {v2.2S}, [x1]
+        add             x1, x1, #16
+        faddp           v1.2S, v1.2S, v1.2S
+        st1             {v1.S}[0], [x1]
+        ret
+endfunc
+
+.macro apply_noise_common
+        sxtw            x3, w3
+        sxtw            x5, w5
+        movrel          x7, X(ff_sbr_noise_table)
+        add             x3, x3, #1
+1:      and             x3, x3, #0x1ff
+        add             x8, x7, x3, lsl #3
+        add             x3, x3, #2
+        ld1             {v2.4S}, [x0]
+        ld1             {v3.2S}, [x1], #8
+        ld1             {v4.2S}, [x2], #8
+        ld1             {v5.4S}, [x8]
+        mov             v6.16B, v2.16B
+        zip1            v3.4S, v3.4S, v3.4S
+        zip1            v4.4S, v4.4S, v4.4S
+        fmla            v6.4S, v1.4S, v3.4S
+        fmla            v2.4S, v5.4S, v4.4S
+        fcmeq           v7.4S, v3.4S, #0.0
+        bif             v2.16B, v6.16B, v7.16B
+        st1             {v2.4S}, [x0], #16
+        subs            x5, x5, #2
+        b.gt            1b
+.endm
+
+function ff_sbr_hf_apply_noise_0_neon, export=1
+        movrel          x9, phi_noise_0
+        ld1             {v1.4S}, [x9]
+        apply_noise_common
+        ret
+endfunc
+
+function ff_sbr_hf_apply_noise_1_neon, export=1
+        movrel          x9, phi_noise_1
+        and             x4, x4, #1
+        add             x9, x9, x4, lsl #4
+        ld1             {v1.4S}, [x9]
+        apply_noise_common
+        ret
+endfunc
+
+function ff_sbr_hf_apply_noise_2_neon, export=1
+        movrel          x9, phi_noise_2
+        ld1             {v1.4S}, [x9]
+        apply_noise_common
+        ret
+endfunc
+
+function ff_sbr_hf_apply_noise_3_neon, export=1
+        movrel          x9, phi_noise_3
+        and             x4, x4, #1
+        add             x9, x9, x4, lsl #4
+        ld1             {v1.4S}, [x9]
+        apply_noise_common
+        ret
+endfunc