VP8: ARM NEON optimisations for dsp functions
authorMans Rullgard <mans@mansr.com>
Wed, 2 Feb 2011 16:26:20 +0000 (16:26 +0000)
committerMans Rullgard <mans@mansr.com>
Mon, 7 Feb 2011 16:08:23 +0000 (16:08 +0000)
This adds NEON optimised versions of all functions in VP8DSPContext.
Based on initial work by Rob Clark.

Signed-off-by: Mans Rullgard <mans@mansr.com>
libavcodec/arm/Makefile
libavcodec/arm/vp8dsp_init_arm.c [new file with mode: 0644]
libavcodec/arm/vp8dsp_neon.S [new file with mode: 0644]
libavcodec/vp8dsp.c
libavcodec/vp8dsp.h

index 014456e..15269ea 100644 (file)
@@ -2,6 +2,7 @@ OBJS-$(CONFIG_DCA_DECODER)             += arm/dcadsp_init_arm.o         \
 
 OBJS-$(CONFIG_VP5_DECODER)             += arm/vp56dsp_init_arm.o
 OBJS-$(CONFIG_VP6_DECODER)             += arm/vp56dsp_init_arm.o
+OBJS-$(CONFIG_VP8_DECODER)             += arm/vp8dsp_init_arm.o
 
 OBJS-$(CONFIG_H264DSP)                 += arm/h264dsp_init_arm.o
 OBJS-$(CONFIG_H264PRED)                += arm/h264pred_init_arm.o
@@ -54,6 +55,8 @@ NEON-OBJS-$(CONFIG_VP5_DECODER)        += arm/vp56dsp_neon.o            \
 NEON-OBJS-$(CONFIG_VP6_DECODER)        += arm/vp56dsp_neon.o            \
                                           arm/vp3dsp_neon.o             \
 
+NEON-OBJS-$(CONFIG_VP8_DECODER)        += arm/vp8dsp_neon.o
+
 OBJS-$(HAVE_NEON)                      += arm/dsputil_init_neon.o       \
                                           arm/dsputil_neon.o            \
                                           arm/fmtconvert_neon.o         \
diff --git a/libavcodec/arm/vp8dsp_init_arm.c b/libavcodec/arm/vp8dsp_init_arm.c
new file mode 100644 (file)
index 0000000..c970ca5
--- /dev/null
@@ -0,0 +1,163 @@
+/**
+ * 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 <stdint.h>
+#include "libavcodec/vp8dsp.h"
+
+void ff_vp8_luma_dc_wht_neon(DCTELEM block[4][4][16], DCTELEM dc[16]);
+void ff_vp8_luma_dc_wht_dc_neon(DCTELEM block[4][4][16], DCTELEM dc[16]);
+
+void ff_vp8_idct_add_neon(uint8_t *dst, DCTELEM block[16], int stride);
+void ff_vp8_idct_dc_add_neon(uint8_t *dst, DCTELEM block[16], int stride);
+void ff_vp8_idct_dc_add4y_neon(uint8_t *dst, DCTELEM block[4][16], int stride);
+void ff_vp8_idct_dc_add4uv_neon(uint8_t *dst, DCTELEM block[4][16], int stride);
+
+void ff_vp8_v_loop_filter16_neon(uint8_t *dst, int stride,
+                                 int flim_E, int flim_I, int hev_thresh);
+void ff_vp8_h_loop_filter16_neon(uint8_t *dst, int stride,
+                                 int flim_E, int flim_I, int hev_thresh);
+void ff_vp8_v_loop_filter8uv_neon(uint8_t *dstU, uint8_t *dstV, int stride,
+                                  int flim_E, int flim_I, int hev_thresh);
+void ff_vp8_h_loop_filter8uv_neon(uint8_t *dstU, uint8_t *dstV, int stride,
+                                  int flim_E, int flim_I, int hev_thresh);
+
+void ff_vp8_v_loop_filter16_inner_neon(uint8_t *dst, int stride,
+                                       int flim_E, int flim_I, int hev_thresh);
+void ff_vp8_h_loop_filter16_inner_neon(uint8_t *dst, int stride,
+                                       int flim_E, int flim_I, int hev_thresh);
+void ff_vp8_v_loop_filter8uv_inner_neon(uint8_t *dstU, uint8_t *dstV,
+                                        int stride, int flim_E, int flim_I,
+                                        int hev_thresh);
+void ff_vp8_h_loop_filter8uv_inner_neon(uint8_t *dstU, uint8_t *dstV,
+                                        int stride, int flim_E, int flim_I,
+                                        int hev_thresh);
+
+void ff_vp8_v_loop_filter16_simple_neon(uint8_t *dst, int stride, int flim);
+void ff_vp8_h_loop_filter16_simple_neon(uint8_t *dst, int stride, int flim);
+
+
+#define VP8_MC(n)                                                       \
+    void ff_put_vp8_##n##_neon(uint8_t *dst, int dststride,             \
+                               uint8_t *src, int srcstride,             \
+                               int h, int x, int y)
+
+#define VP8_EPEL(w)                             \
+    VP8_MC(pixels ## w);                        \
+    VP8_MC(epel ## w ## _h4);                   \
+    VP8_MC(epel ## w ## _h6);                   \
+    VP8_MC(epel ## w ## _v4);                   \
+    VP8_MC(epel ## w ## _h4v4);                 \
+    VP8_MC(epel ## w ## _h6v4);                 \
+    VP8_MC(epel ## w ## _v6);                   \
+    VP8_MC(epel ## w ## _h4v6);                 \
+    VP8_MC(epel ## w ## _h6v6)
+
+VP8_EPEL(16);
+VP8_EPEL(8);
+VP8_EPEL(4);
+
+VP8_MC(bilin16_h);
+VP8_MC(bilin16_v);
+VP8_MC(bilin16_hv);
+VP8_MC(bilin8_h);
+VP8_MC(bilin8_v);
+VP8_MC(bilin8_hv);
+VP8_MC(bilin4_h);
+VP8_MC(bilin4_v);
+VP8_MC(bilin4_hv);
+
+av_cold void ff_vp8dsp_init_arm(VP8DSPContext *dsp)
+{
+    if (HAVE_NEON) {
+        dsp->vp8_luma_dc_wht    = ff_vp8_luma_dc_wht_neon;
+        dsp->vp8_luma_dc_wht_dc = ff_vp8_luma_dc_wht_dc_neon;
+
+        dsp->vp8_idct_add       = ff_vp8_idct_add_neon;
+        dsp->vp8_idct_dc_add    = ff_vp8_idct_dc_add_neon;
+        dsp->vp8_idct_dc_add4y  = ff_vp8_idct_dc_add4y_neon;
+        dsp->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_neon;
+
+        dsp->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16_neon;
+        dsp->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16_neon;
+        dsp->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_neon;
+        dsp->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_neon;
+
+        dsp->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16_inner_neon;
+        dsp->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16_inner_neon;
+        dsp->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_neon;
+        dsp->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_neon;
+
+        dsp->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter16_simple_neon;
+        dsp->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter16_simple_neon;
+
+        dsp->put_vp8_epel_pixels_tab[0][0][0] = ff_put_vp8_pixels16_neon;
+        dsp->put_vp8_epel_pixels_tab[0][0][2] = ff_put_vp8_epel16_h6_neon;
+        dsp->put_vp8_epel_pixels_tab[0][2][0] = ff_put_vp8_epel16_v6_neon;
+        dsp->put_vp8_epel_pixels_tab[0][2][2] = ff_put_vp8_epel16_h6v6_neon;
+
+        dsp->put_vp8_epel_pixels_tab[1][0][0] = ff_put_vp8_pixels8_neon;
+        dsp->put_vp8_epel_pixels_tab[1][0][1] = ff_put_vp8_epel8_h4_neon;
+        dsp->put_vp8_epel_pixels_tab[1][0][2] = ff_put_vp8_epel8_h6_neon;
+        dsp->put_vp8_epel_pixels_tab[1][1][0] = ff_put_vp8_epel8_v4_neon;
+        dsp->put_vp8_epel_pixels_tab[1][1][1] = ff_put_vp8_epel8_h4v4_neon;
+        dsp->put_vp8_epel_pixels_tab[1][1][2] = ff_put_vp8_epel8_h6v4_neon;
+        dsp->put_vp8_epel_pixels_tab[1][2][0] = ff_put_vp8_epel8_v6_neon;
+        dsp->put_vp8_epel_pixels_tab[1][2][1] = ff_put_vp8_epel8_h4v6_neon;
+        dsp->put_vp8_epel_pixels_tab[1][2][2] = ff_put_vp8_epel8_h6v6_neon;
+
+        dsp->put_vp8_epel_pixels_tab[2][0][0] = ff_put_vp8_pixels4_neon;
+        dsp->put_vp8_epel_pixels_tab[2][0][1] = ff_put_vp8_epel4_h4_neon;
+        dsp->put_vp8_epel_pixels_tab[2][0][2] = ff_put_vp8_epel4_h6_neon;
+        dsp->put_vp8_epel_pixels_tab[2][1][0] = ff_put_vp8_epel4_v4_neon;
+        dsp->put_vp8_epel_pixels_tab[2][1][1] = ff_put_vp8_epel4_h4v4_neon;
+        dsp->put_vp8_epel_pixels_tab[2][1][2] = ff_put_vp8_epel4_h6v4_neon;
+        dsp->put_vp8_epel_pixels_tab[2][2][0] = ff_put_vp8_epel4_v6_neon;
+        dsp->put_vp8_epel_pixels_tab[2][2][1] = ff_put_vp8_epel4_h4v6_neon;
+        dsp->put_vp8_epel_pixels_tab[2][2][2] = ff_put_vp8_epel4_h6v6_neon;
+
+        dsp->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_neon;
+        dsp->put_vp8_bilinear_pixels_tab[0][0][1] = ff_put_vp8_bilin16_h_neon;
+        dsp->put_vp8_bilinear_pixels_tab[0][0][2] = ff_put_vp8_bilin16_h_neon;
+        dsp->put_vp8_bilinear_pixels_tab[0][1][0] = ff_put_vp8_bilin16_v_neon;
+        dsp->put_vp8_bilinear_pixels_tab[0][1][1] = ff_put_vp8_bilin16_hv_neon;
+        dsp->put_vp8_bilinear_pixels_tab[0][1][2] = ff_put_vp8_bilin16_hv_neon;
+        dsp->put_vp8_bilinear_pixels_tab[0][2][0] = ff_put_vp8_bilin16_v_neon;
+        dsp->put_vp8_bilinear_pixels_tab[0][2][1] = ff_put_vp8_bilin16_hv_neon;
+        dsp->put_vp8_bilinear_pixels_tab[0][2][2] = ff_put_vp8_bilin16_hv_neon;
+
+        dsp->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_neon;
+        dsp->put_vp8_bilinear_pixels_tab[1][0][1] = ff_put_vp8_bilin8_h_neon;
+        dsp->put_vp8_bilinear_pixels_tab[1][0][2] = ff_put_vp8_bilin8_h_neon;
+        dsp->put_vp8_bilinear_pixels_tab[1][1][0] = ff_put_vp8_bilin8_v_neon;
+        dsp->put_vp8_bilinear_pixels_tab[1][1][1] = ff_put_vp8_bilin8_hv_neon;
+        dsp->put_vp8_bilinear_pixels_tab[1][1][2] = ff_put_vp8_bilin8_hv_neon;
+        dsp->put_vp8_bilinear_pixels_tab[1][2][0] = ff_put_vp8_bilin8_v_neon;
+        dsp->put_vp8_bilinear_pixels_tab[1][2][1] = ff_put_vp8_bilin8_hv_neon;
+        dsp->put_vp8_bilinear_pixels_tab[1][2][2] = ff_put_vp8_bilin8_hv_neon;
+
+        dsp->put_vp8_bilinear_pixels_tab[2][0][0] = ff_put_vp8_pixels4_neon;
+        dsp->put_vp8_bilinear_pixels_tab[2][0][1] = ff_put_vp8_bilin4_h_neon;
+        dsp->put_vp8_bilinear_pixels_tab[2][0][2] = ff_put_vp8_bilin4_h_neon;
+        dsp->put_vp8_bilinear_pixels_tab[2][1][0] = ff_put_vp8_bilin4_v_neon;
+        dsp->put_vp8_bilinear_pixels_tab[2][1][1] = ff_put_vp8_bilin4_hv_neon;
+        dsp->put_vp8_bilinear_pixels_tab[2][1][2] = ff_put_vp8_bilin4_hv_neon;
+        dsp->put_vp8_bilinear_pixels_tab[2][2][0] = ff_put_vp8_bilin4_v_neon;
+        dsp->put_vp8_bilinear_pixels_tab[2][2][1] = ff_put_vp8_bilin4_hv_neon;
+        dsp->put_vp8_bilinear_pixels_tab[2][2][2] = ff_put_vp8_bilin4_hv_neon;
+    }
+}
diff --git a/libavcodec/arm/vp8dsp_neon.S b/libavcodec/arm/vp8dsp_neon.S
new file mode 100644 (file)
index 0000000..01c3959
--- /dev/null
@@ -0,0 +1,1910 @@
+/**
+ * VP8 NEON optimisations
+ *
+ * Copyright (c) 2010 Rob Clark <rob@ti.com>
+ * Copyright (c) 2011 Mans Rullgard <mans@mansr.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 "asm.S"
+
+function ff_vp8_luma_dc_wht_neon, export=1
+        vld1.16         {q0-q1},  [r1,:128]
+        vmov.i16        q15, #0
+
+        vadd.i16        d4,  d0,  d3
+        vadd.i16        d6,  d1,  d2
+        vst1.16         {q15},    [r1,:128]!
+        vsub.i16        d7,  d1,  d2
+        vsub.i16        d5,  d0,  d3
+        vst1.16         {q15},    [r1,:128]
+        vadd.i16        q0,  q2,  q3
+        vsub.i16        q1,  q2,  q3
+
+        vmov.i16        q8, #3
+
+        vtrn.32         d0,  d2
+        vtrn.32         d1,  d3
+        vtrn.16         d0,  d1
+        vtrn.16         d2,  d3
+
+        vadd.i16        d0,  d0,  d16
+
+        vadd.i16        d4,  d0,  d3
+        vadd.i16        d6,  d1,  d2
+        vsub.i16        d7,  d1,  d2
+        vsub.i16        d5,  d0,  d3
+        vadd.i16        q0,  q2,  q3
+        vsub.i16        q1,  q2,  q3
+
+        vshr.s16        q0,  q0,  #3
+        vshr.s16        q1,  q1,  #3
+
+        mov             r3,  #32
+        vst1.16         {d0[0]},  [r0,:16], r3
+        vst1.16         {d1[0]},  [r0,:16], r3
+        vst1.16         {d2[0]},  [r0,:16], r3
+        vst1.16         {d3[0]},  [r0,:16], r3
+        vst1.16         {d0[1]},  [r0,:16], r3
+        vst1.16         {d1[1]},  [r0,:16], r3
+        vst1.16         {d2[1]},  [r0,:16], r3
+        vst1.16         {d3[1]},  [r0,:16], r3
+        vst1.16         {d0[2]},  [r0,:16], r3
+        vst1.16         {d1[2]},  [r0,:16], r3
+        vst1.16         {d2[2]},  [r0,:16], r3
+        vst1.16         {d3[2]},  [r0,:16], r3
+        vst1.16         {d0[3]},  [r0,:16], r3
+        vst1.16         {d1[3]},  [r0,:16], r3
+        vst1.16         {d2[3]},  [r0,:16], r3
+        vst1.16         {d3[3]},  [r0,:16], r3
+
+        bx              lr
+endfunc
+
+function ff_vp8_luma_dc_wht_dc_neon, export=1
+        ldrsh           r2,  [r1]
+        mov             r3,  #0
+        add             r2,  r2,  #3
+        strh            r3,  [r1]
+        asr             r2,  r2,  #3
+    .rept 16
+        strh            r2,  [r0], #32
+    .endr
+        bx              lr
+endfunc
+
+function ff_vp8_idct_add_neon, export=1
+        vld1.16         {q0-q1},  [r1,:128]
+        movw            r3,  #20091
+        movt            r3,  #35468/2
+        vdup.32         d4,  r3
+
+        vmull.s16       q12, d1,  d4[0]
+        vmull.s16       q13, d3,  d4[0]
+        vqdmulh.s16     d20, d1,  d4[1]
+        vqdmulh.s16     d23, d3,  d4[1]
+        vshrn.s32       d21, q12, #16
+        vshrn.s32       d22, q13, #16
+        vadd.s16        d21, d21, d1
+        vadd.s16        d22, d22, d3
+
+        vadd.s16        d16, d0,  d2
+        vsub.s16        d17, d0,  d2
+        vadd.s16        d18, d21, d23
+        vsub.s16        d19, d20, d22
+        vadd.s16        q0,  q8,  q9
+        vsub.s16        q1,  q8,  q9
+
+        vtrn.32         d0,  d3
+        vtrn.32         d1,  d2
+        vtrn.16         d0,  d1
+        vtrn.16         d3,  d2
+
+        vmov.i16        q15, #0
+        vmull.s16       q12, d1,  d4[0]
+        vst1.16         {q15},    [r1,:128]!
+        vmull.s16       q13, d2,  d4[0]
+        vst1.16         {q15},    [r1,:128]
+        vqdmulh.s16     d21, d1,  d4[1]
+        vqdmulh.s16     d23, d2,  d4[1]
+        vshrn.s32       d20, q12, #16
+        vshrn.s32       d22, q13, #16
+        vadd.i16        d20, d20, d1
+        vadd.i16        d22, d22, d2
+
+        vadd.i16        d16, d0,  d3
+        vsub.i16        d17, d0,  d3
+        vadd.i16        d18, d20, d23
+        vld1.32         {d20[]},  [r0,:32], r2
+        vsub.i16        d19, d21, d22
+        vld1.32         {d22[]},  [r0,:32], r2
+        vadd.s16        q0,  q8,  q9
+        vld1.32         {d23[]},  [r0,:32], r2
+        vsub.s16        q1,  q8,  q9
+        vld1.32         {d21[]},  [r0,:32], r2
+        vrshr.s16       q0,  q0,  #3
+        vtrn.32         q10, q11
+        vrshr.s16       q1,  q1,  #3
+
+        sub             r0,  r0,  r2,  lsl #2
+
+        vtrn.32         d0,  d3
+        vtrn.32         d1,  d2
+        vtrn.16         d0,  d1
+        vtrn.16         d3,  d2
+
+        vaddw.u8        q0,  q0,  d20
+        vaddw.u8        q1,  q1,  d21
+        vqmovun.s16     d0,  q0
+        vqmovun.s16     d1,  q1
+
+        vst1.32         {d0[0]},  [r0,:32], r2
+        vst1.32         {d0[1]},  [r0,:32], r2
+        vst1.32         {d1[1]},  [r0,:32], r2
+        vst1.32         {d1[0]},  [r0,:32], r2
+
+        bx              lr
+endfunc
+
+function ff_vp8_idct_dc_add_neon, export=1
+        mov             r3,  #0
+        ldrsh           r12, [r1]
+        strh            r3,  [r1]
+        vdup.16         q1,  r12
+        vrshr.s16       q1,  q1,  #3
+        vld1.32         {d0[]},   [r0,:32], r2
+        vld1.32         {d1[]},   [r0,:32], r2
+        vld1.32         {d0[1]},  [r0,:32], r2
+        vld1.32         {d1[1]},  [r0,:32], r2
+        vaddw.u8        q2,  q1,  d0
+        vaddw.u8        q3,  q1,  d1
+        sub             r0,  r0,  r2, lsl #2
+        vqmovun.s16     d0,  q2
+        vqmovun.s16     d1,  q3
+        vst1.32         {d0[0]},  [r0,:32], r2
+        vst1.32         {d1[0]},  [r0,:32], r2
+        vst1.32         {d0[1]},  [r0,:32], r2
+        vst1.32         {d1[1]},  [r0,:32], r2
+        bx              lr
+endfunc
+
+function ff_vp8_idct_dc_add4uv_neon, export=1
+        vmov.i16        d0,  #0
+        mov             r3,  #32
+        vld1.16         {d16[]},  [r1,:16]
+        vst1.16         {d0[0]},  [r1,:16], r3
+        vld1.16         {d17[]},  [r1,:16]
+        vst1.16         {d0[0]},  [r1,:16], r3
+        vld1.16         {d18[]},  [r1,:16]
+        vst1.16         {d0[0]},  [r1,:16], r3
+        vld1.16         {d19[]},  [r1,:16]
+        vst1.16         {d0[0]},  [r1,:16], r3
+        mov             r3,  r0
+        vrshr.s16       q8,  q8,  #3            @ dc >>= 3
+        vld1.8          {d0},     [r0,:64], r2
+        vrshr.s16       q9,  q9,  #3
+        vld1.8          {d1},     [r0,:64], r2
+        vaddw.u8        q10, q8,  d0
+        vld1.8          {d2},     [r0,:64], r2
+        vaddw.u8        q0,  q8,  d1
+        vld1.8          {d3},     [r0,:64], r2
+        vaddw.u8        q11, q8,  d2
+        vld1.8          {d4},     [r0,:64], r2
+        vaddw.u8        q1,  q8,  d3
+        vld1.8          {d5},     [r0,:64], r2
+        vaddw.u8        q12, q9,  d4
+        vld1.8          {d6},     [r0,:64], r2
+        vaddw.u8        q2,  q9,  d5
+        vld1.8          {d7},     [r0,:64], r2
+        vaddw.u8        q13, q9,  d6
+        vqmovun.s16     d20, q10
+        vaddw.u8        q3,  q9,  d7
+        vqmovun.s16     d21, q0
+        vqmovun.s16     d22, q11
+        vst1.8          {d20},    [r3,:64], r2
+        vqmovun.s16     d23, q1
+        vst1.8          {d21},    [r3,:64], r2
+        vqmovun.s16     d24, q12
+        vst1.8          {d22},    [r3,:64], r2
+        vqmovun.s16     d25, q2
+        vst1.8          {d23},    [r3,:64], r2
+        vqmovun.s16     d26, q13
+        vst1.8          {d24},    [r3,:64], r2
+        vqmovun.s16     d27, q3
+        vst1.8          {d25},    [r3,:64], r2
+        vst1.8          {d26},    [r3,:64], r2
+        vst1.8          {d27},    [r3,:64], r2
+
+        bx              lr
+endfunc
+
+function ff_vp8_idct_dc_add4y_neon, export=1
+        vmov.i16        d0,  #0
+        mov             r3,  #32
+        vld1.16         {d16[]},  [r1,:16]
+        vst1.16         {d0[0]},  [r1,:16], r3
+        vld1.16         {d17[]},  [r1,:16]
+        vst1.16         {d0[0]},  [r1,:16], r3
+        vld1.16         {d18[]},  [r1,:16]
+        vst1.16         {d0[0]},  [r1,:16], r3
+        vld1.16         {d19[]},  [r1,:16]
+        vst1.16         {d0[0]},  [r1,:16], r3
+        vrshr.s16       q8,  q8,  #3            @ dc >>= 3
+        vld1.8          {q0},     [r0,:128], r2
+        vrshr.s16       q9,  q9,  #3
+        vld1.8          {q1},     [r0,:128], r2
+        vaddw.u8        q10, q8,  d0
+        vld1.8          {q2},     [r0,:128], r2
+        vaddw.u8        q0,  q9,  d1
+        vld1.8          {q3},     [r0,:128], r2
+        vaddw.u8        q11, q8,  d2
+        vaddw.u8        q1,  q9,  d3
+        vaddw.u8        q12, q8,  d4
+        vaddw.u8        q2,  q9,  d5
+        vaddw.u8        q13, q8,  d6
+        vaddw.u8        q3,  q9,  d7
+        sub             r0,  r0,  r2,  lsl #2
+        vqmovun.s16     d20, q10
+        vqmovun.s16     d21, q0
+        vqmovun.s16     d22, q11
+        vqmovun.s16     d23, q1
+        vqmovun.s16     d24, q12
+        vst1.8          {q10},    [r0,:128], r2
+        vqmovun.s16     d25, q2
+        vst1.8          {q11},    [r0,:128], r2
+        vqmovun.s16     d26, q13
+        vst1.8          {q12},    [r0,:128], r2
+        vqmovun.s16     d27, q3
+        vst1.8          {q13},    [r0,:128], r2
+
+        bx              lr
+endfunc
+
+@ Register layout:
+@   P3..Q3 -> q0..q7
+@   flim_E -> q14
+@   flim_I -> q15
+@   hev_thresh -> r12
+@
+.macro  vp8_loop_filter, inner=0, simple=0
+    .if \simple
+        vabd.u8         q9,  q3,  q4            @ abs(P0-Q0)
+        vabd.u8         q15, q2,  q5            @ abs(P1-Q1)
+        vqadd.u8        q9,  q9,  q9            @ abs(P0-Q0) * 2
+        vshr.u8         q10, q15, #1            @ abs(P1-Q1) / 2
+        vqadd.u8        q11, q9,  q10           @ (abs(P0-Q0)*2) + (abs(P1-Q1)/2)
+        vmov.i8         q13, #0x80
+        vcle.u8         q8,  q11, q14           @ (abs(P0-Q0)*2) + (abs(P1-Q1)/2) <= flim
+    .else
+        @ calculate hev and normal_limit:
+        vabd.u8         q12, q2,  q3            @ abs(P1-P0)
+        vabd.u8         q13, q5,  q4            @ abs(Q1-Q0)
+        vabd.u8         q10, q0,  q1            @ abs(P3-P2)
+        vabd.u8         q11, q1,  q2            @ abs(P2-P1)
+        vcle.u8         q8,  q12, q15           @ abs(P1-P0) <= flim_I
+        vcle.u8         q9,  q13, q15           @ abs(Q1-Q0) <= flim_I
+        vcle.u8         q10, q10, q15           @ abs(P3-P2) <= flim_I
+        vcle.u8         q11, q11, q15           @ abs(P2-P1) <= flim_I
+        vand            q8,  q8,  q9
+        vabd.u8         q9,  q7,  q6            @ abs(Q3-Q2)
+        vand            q8,  q8,  q11
+        vabd.u8         q11, q6,  q5            @ abs(Q2-Q1)
+        vand            q8,  q8,  q10
+        vcle.u8         q10, q9,  q15           @ abs(Q3-Q2) <= flim_I
+        vcle.u8         q11, q11, q15           @ abs(Q2-Q1) <= flim_I
+        vabd.u8         q9,  q3,  q4            @ abs(P0-Q0)
+        vabd.u8         q15, q2,  q5            @ abs(P1-Q1)
+        vand            q8,  q8,  q10
+        vqadd.u8        q9,  q9,  q9            @ abs(P0-Q0) * 2
+        vand            q8,  q8,  q11
+        vshr.u8         q10, q15, #1            @ abs(P1-Q1) / 2
+        vdup.8          q15, r12                @ hev_thresh
+        vqadd.u8        q11, q9,  q10           @ (abs(P0-Q0)*2) + (abs(P1-Q1)/2)
+        vcgt.u8         q12, q12, q15           @ abs(P1-P0) > hev_thresh
+        vcle.u8         q11, q11, q14           @ (abs(P0-Q0)*2) + (abs(P1-Q1)/2) <= flim_E
+        vcgt.u8         q14, q13, q15           @ abs(Q1-Q0) > hev_thresh
+        vand            q8,  q8,  q11
+        vmov.i8         q13, #0x80
+        vorr            q9,  q12, q14
+    .endif
+
+        @ at this point:
+        @   q8: normal_limit
+        @   q9: hev
+
+        @ convert to signed value:
+        veor            q3,  q3,  q13           @ PS0 = P0 ^ 0x80
+        veor            q4,  q4,  q13           @ QS0 = Q0 ^ 0x80
+
+        vmov.i16        q12, #3
+        vsubl.s8        q10, d8,  d6            @ QS0 - PS0
+        vsubl.s8        q11, d9,  d7            @   (widened to 16bit)
+        veor            q2,  q2,  q13           @ PS1 = P1 ^ 0x80
+        veor            q5,  q5,  q13           @ QS1 = Q1 ^ 0x80
+        vmul.i16        q10, q10, q12           @ w = 3 * (QS0 - PS0)
+        vmul.i16        q11, q11, q12
+
+        vqsub.s8        q12, q2,  q5            @ clamp(PS1-QS1)
+        vmov.i8         q14, #4
+        vmov.i8         q15, #3
+    .if \inner
+        vand            q12, q12, q9            @ if(hev) w += clamp(PS1-QS1)
+    .endif
+        vaddw.s8        q10, q10, d24           @ w += clamp(PS1-QS1)
+        vaddw.s8        q11, q11, d25
+        vqmovn.s16      d20, q10                @ narrow result back into q10
+        vqmovn.s16      d21, q11
+    .if !\inner && !\simple
+        veor            q1,  q1,  q13           @ PS2 = P2 ^ 0x80
+        veor            q6,  q6,  q13           @ QS2 = Q2 ^ 0x80
+    .endif
+        vand            q10, q10, q8            @ w &= normal_limit
+
+        @ registers used at this point..
+        @   q0 -> P3  (don't corrupt)
+        @   q1-q6 -> PS2-QS2
+        @   q7 -> Q3  (don't corrupt)
+        @   q9 -> hev
+        @   q10 -> w
+        @   q13 -> #0x80
+        @   q14 -> #4
+        @   q15 -> #3
+        @   q8, q11, q12 -> unused
+
+        @ filter_common:   is4tap==1
+        @   c1 = clamp(w + 4) >> 3;
+        @   c2 = clamp(w + 3) >> 3;
+        @   Q0 = s2u(QS0 - c1);
+        @   P0 = s2u(PS0 + c2);
+
+    .if \simple
+        vqadd.s8        q11, q10, q14           @ c1 = clamp((w&hev)+4)
+        vqadd.s8        q12, q10, q15           @ c2 = clamp((w&hev)+3)
+        vshr.s8         q11, q11, #3            @ c1 >>= 3
+        vshr.s8         q12, q12, #3            @ c2 >>= 3
+        vqsub.s8        q4,  q4,  q11           @ QS0 = clamp(QS0-c1)
+        vqadd.s8        q3,  q3,  q12           @ PS0 = clamp(PS0+c2)
+        veor            q4,  q4,  q13           @ Q0 = QS0 ^ 0x80
+        veor            q3,  q3,  q13           @ P0 = PS0 ^ 0x80
+        veor            q5,  q5,  q13           @ Q1 = QS1 ^ 0x80
+        veor            q2,  q2,  q13           @ P1 = PS1 ^ 0x80
+    .elseif \inner
+        @ the !is4tap case of filter_common, only used for inner blocks
+        @   c3 = ((c1&~hev) + 1) >> 1;
+        @   Q1 = s2u(QS1 - c3);
+        @   P1 = s2u(PS1 + c3);
+        vqadd.s8        q11, q10, q14           @ c1 = clamp((w&hev)+4)
+        vqadd.s8        q12, q10, q15           @ c2 = clamp((w&hev)+3)
+        vshr.s8         q11, q11, #3            @ c1 >>= 3
+        vshr.s8         q12, q12, #3            @ c2 >>= 3
+        vqsub.s8        q4,  q4,  q11           @ QS0 = clamp(QS0-c1)
+        vqadd.s8        q3,  q3,  q12           @ PS0 = clamp(PS0+c2)
+        vbic            q11, q11, q9            @ c1 & ~hev
+        veor            q4,  q4,  q13           @ Q0 = QS0 ^ 0x80
+        vrshr.s8        q11, q11, #1            @ c3 >>= 1
+        veor            q3,  q3,  q13           @ P0 = PS0 ^ 0x80
+        vqsub.s8        q5,  q5,  q11           @ QS1 = clamp(QS1-c3)
+        vqadd.s8        q2,  q2,  q11           @ PS1 = clamp(PS1+c3)
+        veor            q5,  q5,  q13           @ Q1 = QS1 ^ 0x80
+        veor            q2,  q2,  q13           @ P1 = PS1 ^ 0x80
+    .else
+        vand            q12, q10, q9            @ w & hev
+        vqadd.s8        q11, q12, q14           @ c1 = clamp((w&hev)+4)
+        vqadd.s8        q12, q12, q15           @ c2 = clamp((w&hev)+3)
+        vshr.s8         q11, q11, #3            @ c1 >>= 3
+        vshr.s8         q12, q12, #3            @ c2 >>= 3
+        vbic            q10, q10, q9            @ w &= ~hev
+        vqsub.s8        q4,  q4,  q11           @ QS0 = clamp(QS0-c1)
+        vqadd.s8        q3,  q3,  q12           @ PS0 = clamp(PS0+c2)
+
+        @ filter_mbedge:
+        @   a = clamp((27*w + 63) >> 7);
+        @   Q0 = s2u(QS0 - a);
+        @   P0 = s2u(PS0 + a);
+        @   a = clamp((18*w + 63) >> 7);
+        @   Q1 = s2u(QS1 - a);
+        @   P1 = s2u(PS1 + a);
+        @   a = clamp((9*w + 63) >> 7);
+        @   Q2 = s2u(QS2 - a);
+        @   P2 = s2u(PS2 + a);
+        vmov.i16        q9,  #63
+        vshll.s8        q14, d20, #3
+        vshll.s8        q15, d21, #3
+        vaddw.s8        q14, q14, d20
+        vaddw.s8        q15, q15, d21
+        vadd.s16        q8,  q9,  q14
+        vadd.s16        q9,  q9,  q15           @  9*w + 63
+        vadd.s16        q11, q8,  q14
+        vadd.s16        q12, q9,  q15           @ 18*w + 63
+        vadd.s16        q14, q11, q14
+        vadd.s16        q15, q12, q15           @ 27*w + 63
+        vqshrn.s16      d16, q8,  #7
+        vqshrn.s16      d17, q9,  #7            @ clamp(( 9*w + 63)>>7)
+        vqshrn.s16      d22, q11, #7
+        vqshrn.s16      d23, q12, #7            @ clamp((18*w + 63)>>7)
+        vqshrn.s16      d28, q14, #7
+        vqshrn.s16      d29, q15, #7            @ clamp((27*w + 63)>>7)
+        vqadd.s8        q1,  q1,  q8            @ PS2 = clamp(PS2+a)
+        vqsub.s8        q6,  q6,  q8            @ QS2 = clamp(QS2-a)
+        vqadd.s8        q2,  q2,  q11           @ PS1 = clamp(PS1+a)
+        vqsub.s8        q5,  q5,  q11           @ QS1 = clamp(QS1-a)
+        vqadd.s8        q3,  q3,  q14           @ PS0 = clamp(PS0+a)
+        vqsub.s8        q4,  q4,  q14           @ QS0 = clamp(QS0-a)
+        veor            q3,  q3,  q13           @ P0 = PS0 ^ 0x80
+        veor            q4,  q4,  q13           @ Q0 = QS0 ^ 0x80
+        veor            q2,  q2,  q13           @ P1 = PS1 ^ 0x80
+        veor            q5,  q5,  q13           @ Q1 = QS1 ^ 0x80
+        veor            q1,  q1,  q13           @ P2 = PS2 ^ 0x80
+        veor            q6,  q6,  q13           @ Q2 = QS2 ^ 0x80
+    .endif
+.endm
+
+.macro transpose8x16matrix
+        vtrn.32         q0,   q4
+        vtrn.32         q1,   q5
+        vtrn.32         q2,   q6
+        vtrn.32         q3,   q7
+
+        vtrn.16         q0,   q2
+        vtrn.16         q1,   q3
+        vtrn.16         q4,   q6
+        vtrn.16         q5,   q7
+
+        vtrn.8          q0,   q1
+        vtrn.8          q2,   q3
+        vtrn.8          q4,   q5
+        vtrn.8          q6,   q7
+.endm
+
+.macro  vp8_v_loop_filter16 name, inner=0, simple=0
+function ff_vp8_v_loop_filter16\name\()_neon, export=1
+        vpush           {q4-q7}
+        sub             r0,  r0,  r1,  lsl #1+!\simple
+
+        @ Load pixels:
+    .if !\simple
+        ldr             r12, [sp, #64]          @ hev_thresh
+        vld1.8          {q0},     [r0,:128], r1 @ P3
+        vld1.8          {q1},     [r0,:128], r1 @ P2
+    .endif
+        vld1.8          {q2},     [r0,:128], r1 @ P1
+        vld1.8          {q3},     [r0,:128], r1 @ P0
+        vld1.8          {q4},     [r0,:128], r1 @ Q0
+        vld1.8          {q5},     [r0,:128], r1 @ Q1
+    .if !\simple
+        vld1.8          {q6},     [r0,:128], r1 @ Q2
+        vld1.8          {q7},     [r0,:128]     @ Q3
+        vdup.8          q15, r3                 @ flim_I
+    .endif
+        vdup.8          q14, r2                 @ flim_E
+
+        vp8_loop_filter inner=\inner, simple=\simple
+
+        @ back up to P2:  dst -= stride * 6
+        sub             r0,  r0,  r1,  lsl #2
+    .if !\simple
+        sub             r0,  r0,  r1,  lsl #1
+
+        @ Store pixels:
+        vst1.8          {q1},     [r0,:128], r1 @ P2
+    .endif
+        vst1.8          {q2},     [r0,:128], r1 @ P1
+        vst1.8          {q3},     [r0,:128], r1 @ P0
+        vst1.8          {q4},     [r0,:128], r1 @ Q0
+        vst1.8          {q5},     [r0,:128], r1 @ Q1
+    .if !\simple
+        vst1.8          {q6},     [r0,:128]     @ Q2
+    .endif
+
+        vpop            {q4-q7}
+        bx              lr
+endfunc
+.endm
+
+vp8_v_loop_filter16
+vp8_v_loop_filter16 _inner,  inner=1
+vp8_v_loop_filter16 _simple, simple=1
+
+.macro  vp8_v_loop_filter8uv name, inner=0
+function ff_vp8_v_loop_filter8uv\name\()_neon, export=1
+        vpush           {q4-q7}
+        sub             r0,  r0,  r2,  lsl #2
+        sub             r1,  r1,  r2,  lsl #2
+        ldr             r12, [sp, #64]          @ flim_I
+
+        @ Load pixels:
+        vld1.8          {d0},     [r0,:64], r2  @ P3
+        vld1.8          {d1},     [r1,:64], r2  @ P3
+        vld1.8          {d2},     [r0,:64], r2  @ P2
+        vld1.8          {d3},     [r1,:64], r2  @ P2
+        vld1.8          {d4},     [r0,:64], r2  @ P1
+        vld1.8          {d5},     [r1,:64], r2  @ P1
+        vld1.8          {d6},     [r0,:64], r2  @ P0
+        vld1.8          {d7},     [r1,:64], r2  @ P0
+        vld1.8          {d8},     [r0,:64], r2  @ Q0
+        vld1.8          {d9},     [r1,:64], r2  @ Q0
+        vld1.8          {d10},    [r0,:64], r2  @ Q1
+        vld1.8          {d11},    [r1,:64], r2  @ Q1
+        vld1.8          {d12},    [r0,:64], r2  @ Q2
+        vld1.8          {d13},    [r1,:64], r2  @ Q2
+        vld1.8          {d14},    [r0,:64]      @ Q3
+        vld1.8          {d15},    [r1,:64]      @ Q3
+
+        vdup.8          q14, r3                 @ flim_E
+        vdup.8          q15, r12                @ flim_I
+        ldr             r12, [sp, #68]          @ hev_thresh
+
+        vp8_loop_filter inner=\inner
+
+        @ back up to P2:  u,v -= stride * 6
+        sub             r0,  r0,  r2,  lsl #2
+        sub             r1,  r1,  r2,  lsl #2
+        sub             r0,  r0,  r2,  lsl #1
+        sub             r1,  r1,  r2,  lsl #1
+
+        @ Store pixels:
+        vst1.8          {d2},     [r0,:64], r2  @ P2
+        vst1.8          {d3},     [r1,:64], r2  @ P2
+        vst1.8          {d4},     [r0,:64], r2  @ P1
+        vst1.8          {d5},     [r1,:64], r2  @ P1
+        vst1.8          {d6},     [r0,:64], r2  @ P0
+        vst1.8          {d7},     [r1,:64], r2  @ P0
+        vst1.8          {d8},     [r0,:64], r2  @ Q0
+        vst1.8          {d9},     [r1,:64], r2  @ Q0
+        vst1.8          {d10},    [r0,:64], r2  @ Q1
+        vst1.8          {d11},    [r1,:64], r2  @ Q1
+        vst1.8          {d12},    [r0,:64]      @ Q2
+        vst1.8          {d13},    [r1,:64]      @ Q2
+
+        vpop            {q4-q7}
+        bx              lr
+endfunc
+.endm
+
+vp8_v_loop_filter8uv
+vp8_v_loop_filter8uv _inner, inner=1
+
+.macro  vp8_h_loop_filter16 name, inner=0, simple=0
+function ff_vp8_h_loop_filter16\name\()_neon, export=1
+        vpush           {q4-q7}
+        sub             r0,  r0,  #4
+    .if !\simple
+        ldr             r12, [sp, #64]          @ hev_thresh
+    .endif
+
+        @ Load pixels:
+        vld1.8          {d0},     [r0], r1      @ load first 8-line src data
+        vld1.8          {d2},     [r0], r1
+        vld1.8          {d4},     [r0], r1
+        vld1.8          {d6},     [r0], r1
+        vld1.8          {d8},     [r0], r1
+        vld1.8          {d10},    [r0], r1
+        vld1.8          {d12},    [r0], r1
+        vld1.8          {d14},    [r0], r1
+        vld1.8          {d1},     [r0], r1      @ load second 8-line src data
+        vld1.8          {d3},     [r0], r1
+        vld1.8          {d5},     [r0], r1
+        vld1.8          {d7},     [r0], r1
+        vld1.8          {d9},     [r0], r1
+        vld1.8          {d11},    [r0], r1
+        vld1.8          {d13},    [r0], r1
+        vld1.8          {d15},    [r0], r1
+
+        transpose8x16matrix
+
+        vdup.8          q14, r2                 @ flim_E
+    .if !\simple
+        vdup.8          q15, r3                 @ flim_I
+    .endif
+
+        vp8_loop_filter inner=\inner, simple=\simple
+
+        sub             r0,  r0,  r1, lsl #4    @ backup 16 rows
+
+        transpose8x16matrix
+
+        @ Store pixels:
+        vst1.8          {d0},     [r0],     r1
+        vst1.8          {d2},     [r0],     r1
+        vst1.8          {d4},     [r0],     r1
+        vst1.8          {d6},     [r0],     r1
+        vst1.8          {d8},     [r0],     r1
+        vst1.8          {d10},    [r0],     r1
+        vst1.8          {d12},    [r0],     r1
+        vst1.8          {d14},    [r0],     r1
+        vst1.8          {d1},     [r0],     r1
+        vst1.8          {d3},     [r0],     r1
+        vst1.8          {d5},     [r0],     r1
+        vst1.8          {d7},     [r0],     r1
+        vst1.8          {d9},     [r0],     r1
+        vst1.8          {d11},    [r0],     r1
+        vst1.8          {d13},    [r0],     r1
+        vst1.8          {d15},    [r0]
+
+        vpop            {q4-q7}
+        bx              lr
+endfunc
+.endm
+
+vp8_h_loop_filter16
+vp8_h_loop_filter16 _inner,  inner=1
+vp8_h_loop_filter16 _simple, simple=1
+
+.macro  vp8_h_loop_filter8uv name, inner=0
+function ff_vp8_h_loop_filter8uv\name\()_neon, export=1
+        vpush           {q4-q7}
+        sub             r0,  r0,  #4
+        sub             r1,  r1,  #4
+        ldr             r12, [sp, #64]          @ flim_I
+
+        @ Load pixels:
+        vld1.8          {d0},     [r0], r2      @ load u
+        vld1.8          {d1},     [r1], r2      @ load v
+        vld1.8          {d2},     [r0], r2
+        vld1.8          {d3},     [r1], r2
+        vld1.8          {d4},     [r0], r2
+        vld1.8          {d5},     [r1], r2
+        vld1.8          {d6},     [r0], r2
+        vld1.8          {d7},     [r1], r2
+        vld1.8          {d8},     [r0], r2
+        vld1.8          {d9},     [r1], r2
+        vld1.8          {d10},    [r0], r2
+        vld1.8          {d11},    [r1], r2
+        vld1.8          {d12},    [r0], r2
+        vld1.8          {d13},    [r1], r2
+        vld1.8          {d14},    [r0], r2
+        vld1.8          {d15},    [r1], r2
+
+        transpose8x16matrix
+
+        vdup.8          q14, r3                 @ flim_E
+        vdup.8          q15, r12                @ flim_I
+        ldr             r12, [sp, #68]          @ hev_thresh
+
+        vp8_loop_filter inner=\inner
+
+        sub             r0,  r0,  r2, lsl #3    @ backup u 8 rows
+        sub             r1,  r1,  r2, lsl #3    @ backup v 8 rows
+
+        transpose8x16matrix
+
+        @ Store pixels:
+        vst1.8          {d0},     [r0], r2
+        vst1.8          {d1},     [r1], r2
+        vst1.8          {d2},     [r0], r2
+        vst1.8          {d3},     [r1], r2
+        vst1.8          {d4},     [r0], r2
+        vst1.8          {d5},     [r1], r2
+        vst1.8          {d6},     [r0], r2
+        vst1.8          {d7},     [r1], r2
+        vst1.8          {d8},     [r0], r2
+        vst1.8          {d9},     [r1], r2
+        vst1.8          {d10},    [r0], r2
+        vst1.8          {d11},    [r1], r2
+        vst1.8          {d12},    [r0], r2
+        vst1.8          {d13},    [r1], r2
+        vst1.8          {d14},    [r0]
+        vst1.8          {d15},    [r1]
+
+        vpop            {q4-q7}
+        bx              lr
+endfunc
+.endm
+
+vp8_h_loop_filter8uv
+vp8_h_loop_filter8uv _inner, inner=1
+
+function ff_put_vp8_pixels16_neon, export=1
+        ldr             r12, [sp, #0]           @ h
+1:
+        subs            r12, r12, #4
+        vld1.8          {q0},     [r2], r3
+        vld1.8          {q1},     [r2], r3
+        vld1.8          {q2},     [r2], r3
+        vld1.8          {q3},     [r2], r3
+        vst1.8          {q0},     [r0,:128], r1
+        vst1.8          {q1},     [r0,:128], r1
+        vst1.8          {q2},     [r0,:128], r1
+        vst1.8          {q3},     [r0,:128], r1
+        bgt             1b
+        bx              lr
+endfunc
+
+function ff_put_vp8_pixels8_neon, export=1
+        ldr             r12, [sp, #0]           @ h
+1:
+        subs            r12, r12, #4
+        vld1.8          {d0},     [r2], r3
+        vld1.8          {d1},     [r2], r3
+        vld1.8          {d2},     [r2], r3
+        vld1.8          {d3},     [r2], r3
+        vst1.8          {d0},     [r0,:64], r1
+        vst1.8          {d1},     [r0,:64], r1
+        vst1.8          {d2},     [r0,:64], r1
+        vst1.8          {d3},     [r0,:64], r1
+        bgt             1b
+        bx              lr
+endfunc
+
+function ff_put_vp8_pixels4_neon, export=1
+        ldr             r12, [sp, #0]           @ h
+        push            {r4-r6,lr}
+1:
+        subs            r12, r12, #4
+        ldr             r4,       [r2], r3
+        ldr             r5,       [r2], r3
+        ldr             r6,       [r2], r3
+        ldr             lr,       [r2], r3
+        str             r4,       [r0], r1
+        str             r5,       [r0], r1
+        str             r6,       [r0], r1
+        str             lr,       [r0], r1
+        bgt             1b
+        pop             {r4-r6,pc}
+endfunc
+
+/* 4/6-tap 8th-pel MC */
+
+.macro  vp8_epel8_h6    d,   a,   b
+        vext.8          d27, \a,  \b,  #1
+        vmovl.u8        q8,  \a
+        vext.8          d28, \a,  \b,  #2
+        vmovl.u8        q9,  d27
+        vext.8          d29, \a,  \b,  #3
+        vmovl.u8        q10, d28
+        vext.8          d30, \a,  \b,  #4
+        vmovl.u8        q11, d29
+        vext.8          d31, \a,  \b,  #5
+        vmovl.u8        q12, d30
+        vmul.u16        q10, q10, d0[2]
+        vmovl.u8        q13, d31
+        vmul.u16        q11, q11, d0[3]
+        vmls.u16        q10, q9,  d0[1]
+        vmls.u16        q11, q12, d1[0]
+        vmla.u16        q10, q8,  d0[0]
+        vmla.u16        q11, q13, d1[1]
+        vqadd.s16       q11, q10, q11
+        vqrshrun.s16    \d,  q11, #7
+.endm
+
+.macro  vp8_epel16_h6   d0,  d1,  s0,  s1,  s2,  q0,  q1
+        vext.8          q14, \q0, \q1, #3
+        vext.8          q15, \q0, \q1, #4
+        vmovl.u8        q11, d28
+        vmovl.u8        q14, d29
+        vext.8          q3,  \q0, \q1, #2
+        vmovl.u8        q12, d30
+        vmovl.u8        q15, d31
+        vext.8          q8,  \q0, \q1, #1
+        vmovl.u8        q10, d6
+        vmovl.u8        q3,  d7
+        vext.8          q2,  \q0, \q1, #5
+        vmovl.u8        q13, d4
+        vmovl.u8        q2,  d5
+        vmovl.u8        q9,  d16
+        vmovl.u8        q8,  d17
+        vmul.u16        q11, q11, d0[3]
+        vmul.u16        q10, q10, d0[2]
+        vmul.u16        q3,  q3,  d0[2]
+        vmul.u16        q14, q14, d0[3]
+        vmls.u16        q11, q12, d1[0]
+        vmovl.u8        q12, \s0
+        vmovl.u8        q1,  \s1
+        vmls.u16        q10, q9,  d0[1]
+        vmls.u16        q3,  q8,  d0[1]
+        vmls.u16        q14, q15, d1[0]
+        vmla.u16        q10, q12, d0[0]
+        vmla.u16        q11, q13, d1[1]
+        vmla.u16        q3,  q1,  d0[0]
+        vmla.u16        q14, q2,  d1[1]
+        vqadd.s16       q11, q10, q11
+        vqadd.s16       q14, q3,  q14
+        vqrshrun.s16    \d0, q11, #7
+        vqrshrun.s16    \d1, q14, #7
+.endm
+
+.macro  vp8_epel8_v6    d0,  s0,  s1,  s2,  s3,  s4,  s5
+        vmovl.u8        q10, \s2
+        vmovl.u8        q11, \s3
+        vmovl.u8        q9,  \s1
+        vmovl.u8        q12, \s4
+        vmovl.u8        q8,  \s0
+        vmovl.u8        q13, \s5
+        vmul.u16        q10, q10, d0[2]
+        vmul.u16        q11, q11, d0[3]
+        vmls.u16        q10, q9,  d0[1]
+        vmls.u16        q11, q12, d1[0]
+        vmla.u16        q10, q8,  d0[0]
+        vmla.u16        q11, q13, d1[1]
+        vqadd.s16       q11, q10, q11
+        vqrshrun.s16    \d0, q11, #7
+.endm
+
+.macro  vp8_epel8_v6_y2 d0, d1, s0, s1, s2, s3, s4, s5, s6
+        vmovl.u8        q10, \s0
+        vmovl.u8        q11, \s3
+        vmovl.u8        q14, \s6
+        vmovl.u8        q9,  \s1
+        vmovl.u8        q12, \s4
+        vmovl.u8        q8,  \s2
+        vmovl.u8        q13, \s5
+        vmul.u16        q10, q10, d0[0]
+        vmul.u16        q15, q11, d0[3]
+        vmul.u16        q11, q11, d0[2]
+        vmul.u16        q14, q14, d1[1]
+        vmls.u16        q10, q9,  d0[1]
+        vmls.u16        q15, q12, d1[0]
+        vmls.u16        q11, q8,  d0[1]
+        vmls.u16        q14, q13, d1[0]
+        vmla.u16        q10, q8,  d0[2]
+        vmla.u16        q15, q13, d1[1]
+        vmla.u16        q11, q9,  d0[0]
+        vmla.u16        q14, q12, d0[3]
+        vqadd.s16       q15, q10, q15
+        vqadd.s16       q14, q11, q14
+        vqrshrun.s16    \d0, q15, #7
+        vqrshrun.s16    \d1, q14, #7
+.endm
+
+.macro  vp8_epel8_h4    d,   a,   b
+        vext.8          d28, \a,  \b,  #1
+        vmovl.u8        q9,  \a
+        vext.8          d29, \a,  \b,  #2
+        vmovl.u8        q10, d28
+        vext.8          d30, \a,  \b,  #3
+        vmovl.u8        q11, d29
+        vmovl.u8        q12, d30
+        vmul.u16        q10, q10, d0[2]
+        vmul.u16        q11, q11, d0[3]
+        vmls.u16        q10, q9,  d0[1]
+        vmls.u16        q11, q12, d1[0]
+        vqadd.s16       q11, q10, q11
+        vqrshrun.s16    \d,  q11, #7
+.endm
+
+.macro  vp8_epel8_v4_y2 d0,  d1,  s0,  s1,  s2,  s3,  s4
+        vmovl.u8        q9,  \s0
+        vmovl.u8        q10, \s1
+        vmovl.u8        q11, \s2
+        vmovl.u8        q12, \s3
+        vmovl.u8        q13, \s4
+        vmul.u16        q8,  q10, d0[2]
+        vmul.u16        q14, q11, d0[3]
+        vmul.u16        q11, q11, d0[2]
+        vmul.u16        q15, q12, d0[3]
+        vmls.u16        q8,  q9,  d0[1]
+        vmls.u16        q14, q12, d1[0]
+        vmls.u16        q11, q10, d0[1]
+        vmls.u16        q15, q13, d1[0]
+        vqadd.s16       q8,  q8,  q14
+        vqadd.s16       q11, q11, q15
+        vqrshrun.s16    \d0, q8,  #7
+        vqrshrun.s16    \d1, q11, #7
+.endm
+
+function ff_put_vp8_epel16_v6_neon, export=1
+        sub             r2,  r2,  r3,  lsl #1
+        push            {r4,lr}
+        vpush           {d8-d15}
+
+        ldr             r4,  [sp, #80]          @ my
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #72]          @ h
+        add             r4,  lr,  r4, lsl #4
+        vld1.16         {q0},     [r4,:128]
+1:
+        vld1.8          {d2-d3},  [r2], r3
+        vld1.8          {d4-d5},  [r2], r3
+        vld1.8          {d6-d7},  [r2], r3
+        vld1.8          {d8-d9},  [r2], r3
+        vld1.8          {d10-d11},[r2], r3
+        vld1.8          {d12-d13},[r2], r3
+        vld1.8          {d14-d15},[r2]
+        sub             r2,  r2,  r3,  lsl #2
+
+        vp8_epel8_v6_y2 d2,  d4,  d2,  d4,  d6,  d8,  d10, d12, d14
+        vp8_epel8_v6_y2 d3,  d5,  d3,  d5,  d7,  d9,  d11, d13, d15
+
+        vst1.8          {d2-d3},  [r0,:128], r1
+        vst1.8          {d4-d5},  [r0,:128], r1
+        subs            r12, r12, #2
+        bne             1b
+
+        vpop            {d8-d15}
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel16_h6_neon, export=1
+        sub             r2,  r2,  #2
+        push            {r4,lr}
+
+        ldr             r4,  [sp, #12]          @ mx
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        vld1.16         {q0},     [r4,:128]
+1:
+        vld1.8          {d2-d4},  [r2], r3
+
+        vp8_epel16_h6   d2,  d3,  d2,  d3,  d4,  q1,  q2
+
+        vst1.8          {d2-d3}, [r0,:128], r1
+        subs            r12, r12, #1
+        bne             1b
+
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel16_h6v6_neon, export=1
+        sub             r2,  r2,  r3,  lsl #1
+        sub             r2,  r2,  #2
+        push            {r4,lr}
+        vpush           {d8-d9}
+
+        @ first pass (horizontal):
+        ldr             r4,  [sp, #28]          @ mx
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #24]          @ h
+        add             r4,  lr,  r4, lsl #4
+        sub             sp,  sp,  #336+16
+        vld1.16         {q0},     [r4,:128]
+        add             lr,  sp,  #15
+        add             r12, r12, #5
+        bic             lr,  lr,  #15
+1:
+        vld1.8          {d2,d3,d4}, [r2], r3
+
+        vp8_epel16_h6   d2,  d3,  d2,  d3,  d4,  q1,  q2
+
+        vst1.8          {d2-d3}, [lr,:128]!
+        subs            r12, r12, #1
+        bne             1b
+
+        @ second pass (vertical):
+        ldr             r4,  [sp, #336+16+32]   @ my
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #336+16+24]   @ h
+        add             r4,  lr,  r4, lsl #4
+        add             lr,  sp,  #15
+        vld1.16         {q0},     [r4,:128]
+        bic             lr,  lr,  #15
+2:
+        vld1.8          {d2-d5},  [lr,:128]!
+        vld1.8          {d6-d9},  [lr,:128]!
+        vld1.8          {d28-d31},[lr,:128]
+        sub             lr,  lr,  #48
+
+        vp8_epel8_v6    d2, d2, d4, d6, d8, d28, d30
+        vp8_epel8_v6    d3, d3, d5, d7, d9, d29, d31
+
+        vst1.8          {d2-d3}, [r0,:128], r1
+        subs            r12, r12, #1
+        bne             2b
+
+        add             sp,  sp,  #336+16
+        vpop            {d8-d9}
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel8_v6_neon, export=1
+        sub             r2,  r2,  r3,  lsl #1
+        push            {r4,lr}
+
+        ldr             r4,  [sp, #16]          @ my
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        vld1.16         {q0},     [r4,:128]
+1:
+        vld1.8          {d2},  [r2], r3
+        vld1.8          {d3},  [r2], r3
+        vld1.8          {d4},  [r2], r3
+        vld1.8          {d5},  [r2], r3
+        vld1.8          {d6},  [r2], r3
+        vld1.8          {d7},  [r2], r3
+        vld1.8          {d28}, [r2]
+
+        sub             r2,  r2,  r3,  lsl #2
+
+        vp8_epel8_v6_y2 d2,  d3,  d2,  d3,  d4,  d5,  d6,  d7,  d28
+
+        vst1.8          {d2}, [r0,:64], r1
+        vst1.8          {d3}, [r0,:64], r1
+        subs            r12, r12, #2
+        bne             1b
+
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel8_h6_neon, export=1
+        sub             r2,  r2,  #2
+        push            {r4,lr}
+
+        ldr             r4,  [sp, #12]          @ mx
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        vld1.16         {q0},     [r4,:128]
+1:
+        vld1.8          {d2,d3}, [r2], r3
+
+        vp8_epel8_h6    d2,  d2,  d3
+
+        vst1.8          {d2}, [r0,:64], r1
+        subs            r12, r12, #1
+        bne             1b
+
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel8_h6v6_neon, export=1
+        sub             r2,  r2,  r3,  lsl #1
+        sub             r2,  r2,  #2
+        push            {r4,lr}
+
+        @ first pass (horizontal):
+        ldr             r4,  [sp, #12]          @ mx
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        sub             sp,  sp,  #168+16
+        vld1.16         {q0},     [r4,:128]
+        add             lr,  sp,  #15
+        add             r12, r12, #5
+        bic             lr,  lr,  #15
+1:
+        vld1.8          {d2,d3}, [r2], r3
+
+        vp8_epel8_h6    d2,  d2,  d3
+
+        vst1.8          {d2}, [lr,:64]!
+        subs            r12, r12, #1
+        bne             1b
+
+        @ second pass (vertical):
+        ldr             r4,  [sp, #168+16+16]   @ my
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #168+16+8]    @ h
+        add             r4,  lr,  r4, lsl #4
+        add             lr,  sp,  #15
+        vld1.16         {q0},     [r4,:128]
+        bic             lr,  lr,  #15
+2:
+        vld1.8          {d2-d5},  [lr,:128]!
+        vld1.8          {d6-d7},  [lr,:128]!
+        vld1.8          {d30},    [lr,:64]
+        sub             lr,  lr,  #32
+
+        vp8_epel8_v6_y2 d2,  d3,  d2,  d3,  d4,  d5,  d6,  d7,  d30
+
+        vst1.8          {d2}, [r0,:64], r1
+        vst1.8          {d3}, [r0,:64], r1
+        subs            r12, r12, #2
+        bne             2b
+
+        add             sp,  sp,  #168+16
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel8_v4_neon, export=1
+        sub             r2,  r2,  r3
+        push            {r4,lr}
+
+        ldr             r4,  [sp, #16]          @ my
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        vld1.16         {q0},     [r4,:128]
+1:
+        vld1.8          {d2},     [r2], r3
+        vld1.8          {d3},     [r2], r3
+        vld1.8          {d4},     [r2], r3
+        vld1.8          {d5},     [r2], r3
+        vld1.8          {d6},     [r2]
+        sub             r2,  r2,  r3,  lsl #1
+
+        vp8_epel8_v4_y2 d2,  d3,  d2,  d3,  d4,  d5,  d6
+
+        vst1.8          {d2}, [r0,:64], r1
+        vst1.8          {d3}, [r0,:64], r1
+        subs            r12, r12, #2
+        bne             1b
+
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel8_h4_neon, export=1
+        sub             r2,  r2,  #1
+        push            {r4,lr}
+
+        ldr             r4,  [sp, #12]          @ mx
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        vld1.16         {q0},     [r4,:128]
+1:
+        vld1.8          {d2,d3}, [r2], r3
+
+        vp8_epel8_h4    d2,  d2,  d3
+
+        vst1.8          {d2}, [r0,:64], r1
+        subs            r12, r12, #1
+        bne             1b
+
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel8_h4v4_neon, export=1
+        sub             r2,  r2,  r3
+        sub             r2,  r2,  #1
+        push            {r4,lr}
+
+        @ first pass (horizontal):
+        ldr             r4,  [sp, #12]          @ mx
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        sub             sp,  sp,  #168+16
+        vld1.16         {q0},     [r4,:128]
+        add             lr,  sp,  #15
+        add             r12, r12, #3
+        bic             lr,  lr,  #15
+1:
+        vld1.8          {d2,d3}, [r2], r3
+
+        vp8_epel8_h4    d2,  d2,  d3
+
+        vst1.8          {d2}, [lr,:64]!
+        subs            r12, r12, #1
+        bne             1b
+
+        @ second pass (vertical):
+        ldr             r4,  [sp, #168+16+16]   @ my
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #168+16+8]    @ h
+        add             r4,  lr,  r4, lsl #4
+        add             lr,  sp,  #15
+        vld1.16         {q0},     [r4,:128]
+        bic             lr,  lr,  #15
+2:
+        vld1.8          {d2-d5},  [lr,:128]!
+        vld1.8          {d6},     [lr,:64]
+        sub             lr,  lr,  #16
+
+        vp8_epel8_v4_y2 d2,  d3,  d2,  d3,  d4,  d5,  d6
+
+        vst1.8          {d2},     [r0,:64], r1
+        vst1.8          {d3},     [r0,:64], r1
+        subs            r12, r12, #2
+        bne             2b
+
+        add             sp,  sp,  #168+16
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel8_h6v4_neon, export=1
+        sub             r2,  r2,  r3
+        sub             r2,  r2,  #2
+        push            {r4,lr}
+
+        @ first pass (horizontal):
+        ldr             r4,  [sp, #12]          @ mx
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        sub             sp,  sp,  #168+16
+        vld1.16         {q0},     [r4,:128]
+        add             lr,  sp,  #15
+        add             r12, r12, #3
+        bic             lr,  lr,  #15
+1:
+        vld1.8          {d2,d3}, [r2], r3
+
+        vp8_epel8_h6    d2,  d2,  d3
+
+        vst1.8          {d2}, [lr,:64]!
+        subs            r12, r12, #1
+        bne             1b
+
+        @ second pass (vertical):
+        ldr             r4,  [sp, #168+16+16]   @ my
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #168+16+8]    @ h
+        add             r4,  lr,  r4, lsl #4
+        add             lr,  sp,  #15
+        vld1.16         {q0},     [r4,:128]
+        bic             lr,  lr,  #15
+2:
+        vld1.8          {d2-d5},  [lr,:128]!
+        vld1.8          {d6},     [lr,:64]
+        sub             lr,  lr,  #16
+
+        vp8_epel8_v4_y2 d2,  d3,  d2,  d3,  d4,  d5,  d6
+
+        vst1.8          {d2},     [r0,:64], r1
+        vst1.8          {d3},     [r0,:64], r1
+        subs            r12, r12, #2
+        bne             2b
+
+        add             sp,  sp,  #168+16
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel8_h4v6_neon, export=1
+        sub             r2,  r2,  r3,  lsl #1
+        sub             r2,  r2,  #1
+        push            {r4,lr}
+
+        @ first pass (horizontal):
+        ldr             r4,  [sp, #12]          @ mx
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        sub             sp,  sp,  #168+16
+        vld1.16         {q0},     [r4,:128]
+        add             lr,  sp,  #15
+        add             r12, r12, #5
+        bic             lr,  lr,  #15
+1:
+        vld1.8          {d2,d3}, [r2], r3
+
+        vp8_epel8_h4    d2,  d2,  d3
+
+        vst1.8          {d2}, [lr,:64]!
+        subs            r12, r12, #1
+        bne             1b
+
+        @ second pass (vertical):
+        ldr             r4,  [sp, #168+16+16]   @ my
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #168+16+8]    @ h
+        add             r4,  lr,  r4, lsl #4
+        add             lr,  sp,  #15
+        vld1.16         {q0},     [r4,:128]
+        bic             lr,  lr,  #15
+2:
+        vld1.8          {d2-d5},  [lr,:128]!
+        vld1.8          {d6-d7},  [lr,:128]!
+        vld1.8          {d30},    [lr,:64]
+        sub             lr,  lr,  #32
+
+        vp8_epel8_v6_y2 d2,  d3,  d2,  d3,  d4,  d5,  d6,  d7,  d30
+
+        vst1.8          {d2}, [r0,:64], r1
+        vst1.8          {d3}, [r0,:64], r1
+        subs            r12, r12, #2
+        bne             2b
+
+        add             sp,  sp,  #168+16
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel4_v6_neon, export=1
+        sub             r2,  r2,  r3,  lsl #1
+        push            {r4,lr}
+
+        ldr             r4,  [sp, #16]          @ my
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        vld1.16         {q0},     [r4,:128]
+1:
+        vld1.32         {d2[]},   [r2], r3
+        vld1.32         {d3[]},   [r2], r3
+        vld1.32         {d4[]},   [r2], r3
+        vld1.32         {d5[]},   [r2], r3
+        vld1.32         {d6[]},   [r2], r3
+        vld1.32         {d7[]},   [r2], r3
+        vld1.32         {d28[]},  [r2]
+        sub             r2,  r2,  r3,  lsl #2
+        vld1.32         {d2[1]},  [r2], r3
+        vld1.32         {d3[1]},  [r2], r3
+        vld1.32         {d4[1]},  [r2], r3
+        vld1.32         {d5[1]},  [r2], r3
+        vld1.32         {d6[1]},  [r2], r3
+        vld1.32         {d7[1]},  [r2], r3
+        vld1.32         {d28[1]}, [r2]
+        sub             r2,  r2,  r3,  lsl #2
+
+        vp8_epel8_v6_y2 d2,  d3,  d2,  d3,  d4,  d5,  d6,  d7,  d28
+
+        vst1.32         {d2[0]},  [r0,:32], r1
+        vst1.32         {d3[0]},  [r0,:32], r1
+        vst1.32         {d2[1]},  [r0,:32], r1
+        vst1.32         {d3[1]},  [r0,:32], r1
+        subs            r12, r12, #4
+        bne             1b
+
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel4_h6_neon, export=1
+        sub             r2,  r2,  #2
+        push            {r4,lr}
+
+        ldr             r4,  [sp, #12]          @ mx
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        vld1.16         {q0},     [r4,:128]
+1:
+        vld1.8          {q1},     [r2], r3
+        vp8_epel8_h6    d2,  d2,  d3
+        vst1.32         {d2[0]},  [r0,:32], r1
+        subs            r12, r12, #1
+        bne             1b
+
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel4_h6v6_neon, export=1
+        sub             r2,  r2,  r3,  lsl #1
+        sub             r2,  r2,  #2
+        push            {r4,lr}
+
+        ldr             r4,  [sp, #12]          @ mx
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        sub             sp,  sp,  #52+16
+        vld1.16         {q0},     [r4,:128]
+        add             lr,  sp,  #15
+        add             r12, r12, #5
+        bic             lr,  lr,  #15
+1:
+        vld1.8          {q1},     [r2], r3
+        vp8_epel8_h6    d2,  d2,  d3
+        vst1.32         {d2[0]},  [lr,:32]!
+        subs            r12, r12, #1
+        bne             1b
+
+        ldr             r4,  [sp, #52+16+16]    @ my
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #52+16+8]     @ h
+        add             r4,  lr,  r4, lsl #4
+        add             lr,  sp,  #15
+        vld1.16         {q0},     [r4,:128]
+        bic             lr,  lr,  #15
+2:
+        vld1.8          {d2-d3},  [lr,:128]!
+        vld1.8          {d6},     [lr,:64]!
+        vld1.32         {d28[]},  [lr,:32]
+        sub             lr,  lr,  #16
+        vld1.8          {d4-d5},  [lr]!
+        vld1.8          {d7},     [lr,:64]!
+        vld1.32         {d28[1]}, [lr,:32]
+        sub             lr,  lr,  #16
+        vtrn.32         q1,  q2
+        vtrn.32         d6,  d7
+        vp8_epel8_v6_y2 d2,  d3,  d2,  d4,  d3,  d5,  d6,  d7,  d28
+        vst1.32         {d2[0]},  [r0,:32], r1
+        vst1.32         {d3[0]},  [r0,:32], r1
+        vst1.32         {d2[1]},  [r0,:32], r1
+        vst1.32         {d3[1]},  [r0,:32], r1
+        subs            r12, r12, #4
+        bne             2b
+
+        add             sp,  sp,  #52+16
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel4_h4v6_neon, export=1
+        sub             r2,  r2,  r3,  lsl #1
+        sub             r2,  r2,  #1
+        push            {r4,lr}
+
+        ldr             r4,  [sp, #12]          @ mx
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        sub             sp,  sp,  #52+16
+        vld1.16         {q0},     [r4,:128]
+        add             lr,  sp,  #15
+        add             r12, r12, #5
+        bic             lr,  lr,  #15
+1:
+        vld1.8          {d2},     [r2], r3
+        vp8_epel8_h4    d2,  d2,  d2
+        vst1.32         {d2[0]},  [lr,:32]!
+        subs            r12, r12, #1
+        bne             1b
+
+        ldr             r4,  [sp, #52+16+16]    @ my
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #52+16+8]     @ h
+        add             r4,  lr,  r4, lsl #4
+        add             lr,  sp,  #15
+        vld1.16         {q0},     [r4,:128]
+        bic             lr,  lr,  #15
+2:
+        vld1.8          {d2-d3},  [lr,:128]!
+        vld1.8          {d6},     [lr,:64]!
+        vld1.32         {d28[]},  [lr,:32]
+        sub             lr,  lr,  #16
+        vld1.8          {d4-d5},  [lr]!
+        vld1.8          {d7},     [lr,:64]!
+        vld1.32         {d28[1]}, [lr,:32]
+        sub             lr,  lr,  #16
+        vtrn.32         q1,  q2
+        vtrn.32         d6,  d7
+        vp8_epel8_v6_y2 d2,  d3,  d2,  d4,  d3,  d5,  d6,  d7,  d28
+        vst1.32         {d2[0]},  [r0,:32], r1
+        vst1.32         {d3[0]},  [r0,:32], r1
+        vst1.32         {d2[1]},  [r0,:32], r1
+        vst1.32         {d3[1]},  [r0,:32], r1
+        subs            r12, r12, #4
+        bne             2b
+
+        add             sp,  sp,  #52+16
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel4_h6v4_neon, export=1
+        sub             r2,  r2,  r3
+        sub             r2,  r2,  #2
+        push            {r4,lr}
+
+        ldr             r4,  [sp, #12]          @ mx
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        sub             sp,  sp,  #44+16
+        vld1.16         {q0},     [r4,:128]
+        add             lr,  sp,  #15
+        add             r12, r12, #3
+        bic             lr,  lr,  #15
+1:
+        vld1.8          {q1},     [r2], r3
+        vp8_epel8_h6    d2,  d2,  d3
+        vst1.32         {d2[0]},  [lr,:32]!
+        subs            r12, r12, #1
+        bne             1b
+
+        ldr             r4,  [sp, #44+16+16]    @ my
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #44+16+8]     @ h
+        add             r4,  lr,  r4, lsl #4
+        add             lr,  sp,  #15
+        vld1.16         {q0},     [r4,:128]
+        bic             lr,  lr,  #15
+2:
+        vld1.8          {d2-d3},  [lr,:128]!
+        vld1.32         {d6[]},   [lr,:32]
+        sub             lr,  lr,  #8
+        vld1.8          {d4-d5},  [lr]!
+        vld1.32         {d6[1]},  [lr,:32]
+        sub             lr,  lr,  #8
+        vtrn.32         q1,  q2
+        vp8_epel8_v4_y2 d2,  d3,  d2,  d4,  d3,  d5,  d6
+        vst1.32         {d2[0]},  [r0,:32], r1
+        vst1.32         {d3[0]},  [r0,:32], r1
+        vst1.32         {d2[1]},  [r0,:32], r1
+        vst1.32         {d3[1]},  [r0,:32], r1
+        subs            r12, r12, #4
+        bne             2b
+
+        add             sp,  sp,  #44+16
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel4_h4_neon, export=1
+        sub             r2,  r2,  #1
+        push            {r4,lr}
+
+        ldr             r4,  [sp, #12]          @ mx
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        vld1.16         {q0},     [r4,:128]
+1:
+        vld1.8          {d2},     [r2], r3
+        vp8_epel8_h4    d2,  d2,  d2
+        vst1.32         {d2[0]},  [r0,:32], r1
+        subs            r12, r12, #1
+        bne             1b
+
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel4_v4_neon, export=1
+        sub             r2,  r2,  r3
+        push            {r4,lr}
+
+        ldr             r4,  [sp, #16]          @ my
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        vld1.16         {q0},     [r4,:128]
+1:
+        vld1.32         {d2[]},   [r2], r3
+        vld1.32         {d3[]},   [r2], r3
+        vld1.32         {d4[]},   [r2], r3
+        vld1.32         {d5[]},   [r2], r3
+        vld1.32         {d6[]},   [r2]
+        sub             r2,  r2,  r3,  lsl #1
+        vld1.32         {d2[1]},  [r2], r3
+        vld1.32         {d3[1]},  [r2], r3
+        vld1.32         {d4[1]},  [r2], r3
+        vld1.32         {d5[1]},  [r2], r3
+        vld1.32         {d6[1]},  [r2]
+        sub             r2,  r2,  r3,  lsl #1
+
+        vp8_epel8_v4_y2 d2,  d3,  d2,  d3,  d4,  d5,  d6
+
+        vst1.32         {d2[0]},  [r0,:32], r1
+        vst1.32         {d3[0]},  [r0,:32], r1
+        vst1.32         {d2[1]},  [r0,:32], r1
+        vst1.32         {d3[1]},  [r0,:32], r1
+        subs            r12, r12, #4
+        bne             1b
+
+        pop             {r4,pc}
+endfunc
+
+function ff_put_vp8_epel4_h4v4_neon, export=1
+        sub             r2,  r2,  r3
+        sub             r2,  r2,  #1
+        push            {r4,lr}
+
+        ldr             r4,  [sp, #12]          @ mx
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #8]           @ h
+        add             r4,  lr,  r4, lsl #4
+        sub             sp,  sp,  #44+16
+        vld1.16         {q0},     [r4,:128]
+        add             lr,  sp,  #15
+        add             r12, r12, #3
+        bic             lr,  lr,  #15
+1:
+        vld1.8          {d2},     [r2], r3
+        vp8_epel8_h4    d2,  d2,  d3
+        vst1.32         {d2[0]},  [lr,:32]!
+        subs            r12, r12, #1
+        bne             1b
+
+        ldr             r4,  [sp, #44+16+16]    @ my
+        movrel          lr,  subpel_filters-16
+        ldr             r12, [sp, #44+16+8]     @ h
+        add             r4,  lr,  r4, lsl #4
+        add             lr,  sp,  #15
+        vld1.16         {q0},     [r4,:128]
+        bic             lr,  lr,  #15
+2:
+        vld1.8          {d2-d3},  [lr,:128]!
+        vld1.32         {d6[]},   [lr,:32]
+        sub             lr,  lr,  #8
+        vld1.8          {d4-d5},  [lr]!
+        vld1.32         {d6[1]},  [lr,:32]
+        sub             lr,  lr,  #8
+        vtrn.32         q1,  q2
+        vp8_epel8_v4_y2 d2,  d3,  d2,  d4,  d3,  d5,  d6
+        vst1.32         {d2[0]},  [r0,:32], r1
+        vst1.32         {d3[0]},  [r0,:32], r1
+        vst1.32         {d2[1]},  [r0,:32], r1
+        vst1.32         {d3[1]},  [r0,:32], r1
+        subs            r12, r12, #4
+        bne             2b
+
+        add             sp,  sp,  #44+16
+        pop             {r4,pc}
+endfunc
+
+@ note: worst case sum of all 6-tap filter values * 255 is 0x7f80 so 16 bit
+@ arithmatic can be used to apply filters
+const   subpel_filters, align=4
+        .short     0,   6, 123,  12,   1,   0,   0,   0
+        .short     2,  11, 108,  36,   8,   1,   0,   0
+        .short     0,   9,  93,  50,   6,   0,   0,   0
+        .short     3,  16,  77,  77,  16,   3,   0,   0
+        .short     0,   6,  50,  93,   9,   0,   0,   0
+        .short     1,   8,  36, 108,  11,   2,   0,   0
+        .short     0,   1,  12, 123,   6,   0,   0,   0
+endconst
+
+/* Bilinear MC */
+
+function ff_put_vp8_bilin16_h_neon, export=1
+        ldr             r3,  [sp, #4]           @ mx
+        rsb             r12, r3,  #8
+        vdup.8          d0,  r3
+        vdup.8          d1,  r12
+        ldr             r12, [sp]               @ h
+1:
+        subs            r12, r12, #2
+        vld1.8          {d2-d4},  [r2], r1
+        vext.8          q2,  q1,  q2,  #1
+        vmull.u8        q8,  d2,  d1
+        vmlal.u8        q8,  d4,  d0
+        vld1.8          {d18-d20},[r2], r1
+        vmull.u8        q3,  d3,  d1
+        vmlal.u8        q3,  d5,  d0
+        vext.8          q10, q9,  q10, #1
+        vmull.u8        q11, d18, d1
+        vmlal.u8        q11, d20, d0
+        vmull.u8        q12, d19, d1
+        vmlal.u8        q12, d21, d0
+        vrshrn.u16      d4,  q8,  #3
+        vrshrn.u16      d5,  q3,  #3
+        vrshrn.u16      d6,  q11, #3
+        vrshrn.u16      d7,  q12, #3
+        vst1.8          {q2},     [r0,:128], r1
+        vst1.8          {q3},     [r0,:128], r1
+        bgt             1b
+
+        bx              lr
+endfunc
+
+function ff_put_vp8_bilin16_v_neon, export=1
+        ldr             r3,  [sp, #8]           @ my
+        rsb             r12, r3,  #8
+        vdup.8          d0,  r3
+        vdup.8          d1,  r12
+        ldr             r12, [sp]               @ h
+        vld1.8          {q1},     [r2], r1
+1:
+        subs            r12, r12, #2
+        vld1.8          {q2},     [r2], r1
+        vmull.u8        q3,  d2,  d1
+        vmlal.u8        q3,  d4,  d0
+        vmull.u8        q8,  d3,  d1
+        vmlal.u8        q8,  d5,  d0
+        vld1.8          {q1},     [r2], r1
+        vmull.u8        q9,  d4,  d1
+        vmlal.u8        q9,  d2,  d0
+        vmull.u8        q10, d5,  d1
+        vmlal.u8        q10, d3,  d0
+        vrshrn.u16      d4,  q3,  #3
+        vrshrn.u16      d5,  q8,  #3
+        vrshrn.u16      d6,  q9,  #3
+        vrshrn.u16      d7,  q10, #3
+        vst1.8          {q2},     [r0,:128], r1
+        vst1.8          {q3},     [r0,:128], r1
+        bgt             1b
+
+        bx              lr
+endfunc
+
+function ff_put_vp8_bilin16_hv_neon, export=1
+        ldr             r3,  [sp, #4]           @ mx
+        rsb             r12, r3,  #8
+        vdup.8          d0,  r3
+        vdup.8          d1,  r12
+        ldr             r3,  [sp, #8]           @ my
+        rsb             r12, r3,  #8
+        vdup.8          d2,  r3
+        vdup.8          d3,  r12
+        ldr             r12, [sp]               @ h
+
+        vld1.8          {d4-d6},  [r2], r1
+        vext.8          q3,  q2,  q3,  #1
+        vmull.u8        q8,  d4,  d1
+        vmlal.u8        q8,  d6,  d0
+        vmull.u8        q9,  d5,  d1
+        vmlal.u8        q9,  d7,  d0
+        vrshrn.u16      d4,  q8,  #3
+        vrshrn.u16      d5,  q9,  #3
+1:
+        subs            r12, r12, #2
+        vld1.8          {d18-d20},[r2], r1
+        vext.8          q10, q9,  q10, #1
+        vmull.u8        q11, d18, d1
+        vmlal.u8        q11, d20, d0
+        vld1.8          {d26-d28},[r2], r1
+        vmull.u8        q12, d19, d1
+        vmlal.u8        q12, d21, d0
+        vext.8          q14, q13, q14, #1
+        vmull.u8        q8,  d26, d1
+        vmlal.u8        q8,  d28, d0
+        vmull.u8        q9,  d27, d1
+        vmlal.u8        q9,  d29, d0
+        vrshrn.u16      d6,  q11, #3
+        vrshrn.u16      d7,  q12, #3
+        vmull.u8        q12, d4,  d3
+        vmlal.u8        q12, d6,  d2
+        vmull.u8        q15, d5,  d3
+        vmlal.u8        q15, d7,  d2
+        vrshrn.u16      d4,  q8,  #3
+        vrshrn.u16      d5,  q9,  #3
+        vmull.u8        q10, d6,  d3
+        vmlal.u8        q10, d4,  d2
+        vmull.u8        q11, d7,  d3
+        vmlal.u8        q11, d5,  d2
+        vrshrn.u16      d24, q12, #3
+        vrshrn.u16      d25, q15, #3
+        vst1.8          {q12},    [r0,:128], r1
+        vrshrn.u16      d20, q10, #3
+        vrshrn.u16      d21, q11, #3
+        vst1.8          {q10},    [r0,:128], r1
+        bgt             1b
+
+        bx              lr
+endfunc
+
+function ff_put_vp8_bilin8_h_neon, export=1
+        ldr             r3,  [sp, #4]           @ mx
+        rsb             r12, r3,  #8
+        vdup.8          d0,  r3
+        vdup.8          d1,  r12
+        ldr             r12, [sp]               @ h
+1:
+        subs            r12, r12, #2
+        vld1.8          {q1},     [r2], r1
+        vext.8          d3,  d2,  d3,  #1
+        vmull.u8        q2,  d2,  d1
+        vmlal.u8        q2,  d3,  d0
+        vld1.8          {q3},     [r2], r1
+        vext.8          d7,  d6,  d7,  #1
+        vmull.u8        q8,  d6,  d1
+        vmlal.u8        q8,  d7,  d0
+        vrshrn.u16      d4,  q2,  #3
+        vrshrn.u16      d16, q8,  #3
+        vst1.8          {d4},     [r0,:64], r1
+        vst1.8          {d16},    [r0,:64], r1
+        bgt             1b
+
+        bx              lr
+endfunc
+
+function ff_put_vp8_bilin8_v_neon, export=1
+        ldr             r3,  [sp, #8]           @ my
+        rsb             r12, r3,  #8
+        vdup.8          d0,  r3
+        vdup.8          d1,  r12
+        ldr             r12, [sp]               @ h
+        vld1.8          {d2},     [r2], r1
+1:
+        subs            r12, r12, #2
+        vld1.8          {d3},     [r2], r1
+        vmull.u8        q2,  d2,  d1
+        vmlal.u8        q2,  d3,  d0
+        vld1.8          {d2},     [r2], r1
+        vmull.u8        q3,  d3,  d1
+        vmlal.u8        q3,  d2,  d0
+        vrshrn.u16      d4,  q2,  #3
+        vrshrn.u16      d6,  q3,  #3
+        vst1.8          {d4},     [r0,:64], r1
+        vst1.8          {d6},     [r0,:64], r1
+        bgt             1b
+
+        bx              lr
+endfunc
+
+function ff_put_vp8_bilin8_hv_neon, export=1
+        ldr             r3,  [sp, #4]           @ mx
+        rsb             r12, r3,  #8
+        vdup.8          d0,  r3
+        vdup.8          d1,  r12
+        ldr             r3,  [sp, #8]           @ my
+        rsb             r12, r3,  #8
+        vdup.8          d2,  r3
+        vdup.8          d3,  r12
+        ldr             r12, [sp]               @ h
+
+        vld1.8          {q2},     [r2], r1
+        vext.8          d5,  d4,  d5,  #1
+        vmull.u8        q9,  d4,  d1
+        vmlal.u8        q9,  d5,  d0
+        vrshrn.u16      d22, q9,  #3
+1:
+        subs            r12, r12, #2
+        vld1.8          {q3},     [r2], r1
+        vext.8          d7,  d6,  d7,  #1
+        vmull.u8        q8,  d6,  d1
+        vmlal.u8        q8,  d7,  d0
+        vld1.8          {q2},     [r2], r1
+        vext.8          d5,  d4,  d5,  #1
+        vmull.u8        q9,  d4,  d1
+        vmlal.u8        q9,  d5,  d0
+        vrshrn.u16      d16, q8,  #3
+        vmull.u8        q10, d22, d3
+        vmlal.u8        q10, d16, d2
+        vrshrn.u16      d22, q9,  #3
+        vmull.u8        q12, d16, d3
+        vmlal.u8        q12, d22, d2
+        vrshrn.u16      d20, q10, #3
+        vst1.8          {d20},    [r0,:64], r1
+        vrshrn.u16      d23, q12, #3
+        vst1.8          {d23},    [r0,:64], r1
+        bgt             1b
+
+        bx              lr
+endfunc
+
+function ff_put_vp8_bilin4_h_neon, export=1
+        ldr             r3,  [sp, #4]           @ mx
+        rsb             r12, r3,  #8
+        vdup.8          d0,  r3
+        vdup.8          d1,  r12
+        ldr             r12, [sp]               @ h
+1:
+        subs            r12, r12, #2
+        vld1.8          {d2},     [r2], r1
+        vext.8          d3,  d2,  d3,  #1
+        vld1.8          {d6},     [r2], r1
+        vext.8          d7,  d6,  d7,  #1
+        vtrn.32         q1,  q3
+        vmull.u8        q2,  d2,  d1
+        vmlal.u8        q2,  d3,  d0
+        vrshrn.u16      d4,  q2,  #3
+        vst1.32         {d4[0]},  [r0,:32], r1
+        vst1.32         {d4[1]}, [r0,:32], r1
+        bgt             1b
+
+        bx              lr
+endfunc
+
+function ff_put_vp8_bilin4_v_neon, export=1
+        ldr             r3,  [sp, #8]           @ my
+        rsb             r12, r3,  #8
+        vdup.8          d0,  r3
+        vdup.8          d1,  r12
+        ldr             r12, [sp]               @ h
+        vld1.32         {d2[]},   [r2], r1
+1:
+        vld1.32         {d3[]},   [r2]
+        vld1.32         {d2[1]},  [r2], r1
+        vld1.32         {d3[1]},  [r2], r1
+        vmull.u8        q2,  d2,  d1
+        vmlal.u8        q2,  d3,  d0
+        vtrn.32         d3,  d2
+        vrshrn.u16      d4,  q2,  #3
+        vst1.32         {d4[0]},  [r0,:32], r1
+        vst1.32         {d4[1]},  [r0,:32], r1
+        subs            r12, r12, #2
+        bgt             1b
+
+        bx              lr
+endfunc
+
+function ff_put_vp8_bilin4_hv_neon, export=1
+        ldr             r3,  [sp, #4]           @ mx
+        rsb             r12, r3,  #8
+        vdup.8          d0,  r3
+        vdup.8          d1,  r12
+        ldr             r3,  [sp, #8]           @ my
+        rsb             r12, r3,  #8
+        vdup.8          d2,  r3
+        vdup.8          d3,  r12
+        ldr             r12, [sp]               @ h
+
+        vld1.8          {d4},     [r2], r1
+        vext.8          d5,  d4,  d4,  #1
+        vmull.u8        q9,  d4,  d1
+        vmlal.u8        q9,  d5,  d0
+        vrshrn.u16      d22, q9,  #3
+1:
+        subs            r12, r12, #2
+        vld1.8          {d6},     [r2], r1
+        vext.8          d7,  d6,  d6,  #1
+        vld1.8          {d4},     [r2], r1
+        vext.8          d5,  d4,  d4,  #1
+        vtrn.32         q3,  q2
+        vmull.u8        q8,  d6,  d1
+        vmlal.u8        q8,  d7,  d0
+        vrshrn.u16      d16, q8,  #3
+        vmull.u8        q10, d16, d2
+        vtrn.32         d22, d16
+        vmlal.u8        q10, d22, d3
+        vrev64.32       d22, d16
+        vrshrn.u16      d20, q10, #3
+        vst1.32         {d20[0]}, [r0,:32], r1
+        vst1.32         {d20[1]}, [r0,:32], r1
+        bgt             1b
+
+        bx              lr
+endfunc
index 30699ce..4391a7e 100644 (file)
@@ -523,4 +523,6 @@ av_cold void ff_vp8dsp_init(VP8DSPContext *dsp)
         ff_vp8dsp_init_x86(dsp);
     if (HAVE_ALTIVEC)
         ff_vp8dsp_init_altivec(dsp);
+    if (ARCH_ARM)
+        ff_vp8dsp_init_arm(dsp);
 }
index ee5c7ec..e8757ab 100644 (file)
@@ -79,5 +79,6 @@ void ff_put_vp8_pixels4_c(uint8_t *dst, uint8_t *src, int stride, int h, int x,
 void ff_vp8dsp_init(VP8DSPContext *c);
 void ff_vp8dsp_init_x86(VP8DSPContext *c);
 void ff_vp8dsp_init_altivec(VP8DSPContext *c);
+void ff_vp8dsp_init_arm(VP8DSPContext *c);
 
 #endif /* AVCODEC_VP8DSP_H */