[FFmpeg-devel] [PATCH 1/3] VP8: neon loopfilter functions

Rob Clark rob
Thu Sep 16 20:28:31 CEST 2010


added:
 + vp8_v_loop_filter16y_neon
 + vp8_v_loop_filter8uv_neon
 + vp8_h_loop_filter16y_neon
 + vp8_h_loop_filter8uv_neon
 + vp8_v_loop_filter16y_inner_neon
 + vp8_h_loop_filter16y_inner_neon
 + vp8_v_loop_filter8uv_inner_neon
 + vp8_h_loop_filter8uv_inner_neon

---
 libavcodec/arm/Makefile          |    3 +
 libavcodec/arm/vp8dsp_init_arm.c |   65 +++++
 libavcodec/arm/vp8dsp_neon.S     |  475 ++++++++++++++++++++++++++++++++++++++
 libavcodec/vp8dsp.c              |    2 +
 libavcodec/vp8dsp.h              |    1 +
 5 files changed, 546 insertions(+), 0 deletions(-)
 create mode 100644 libavcodec/arm/vp8dsp_init_arm.c
 create mode 100644 libavcodec/arm/vp8dsp_neon.S

diff --git a/libavcodec/arm/Makefile b/libavcodec/arm/Makefile
index 4c30e0a..590b20a 100644
--- a/libavcodec/arm/Makefile
+++ b/libavcodec/arm/Makefile
@@ -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
@@ -50,6 +51,8 @@ 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/int_neon.o                \
diff --git a/libavcodec/arm/vp8dsp_init_arm.c b/libavcodec/arm/vp8dsp_init_arm.c
new file mode 100644
index 0000000..ab4600d
--- /dev/null
+++ b/libavcodec/arm/vp8dsp_init_arm.c
@@ -0,0 +1,65 @@
+/**
+ * VP8 compatible video decoder
+ *
+ * Copyright (C) 2010 Rob Clark <rob at ti.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 "libavcodec/vp8dsp.h"
+
+void vp8_v_loop_filter16y_neon(uint8_t *dst, int stride,
+        int flim_E, int flim_I, int hev_thresh);
+void vp8_h_loop_filter16y_neon(uint8_t *dst, int stride,
+        int flim_E, int flim_I, int hev_thresh);
+void vp8_v_loop_filter8uv_neon(uint8_t *dstU, uint8_t *dstV, int stride,
+        int flim_E, int flim_I, int hev_thresh);
+void vp8_h_loop_filter8uv_neon(uint8_t *dstU, uint8_t *dstV, int stride,
+        int flim_E, int flim_I, int hev_thresh);
+
+void vp8_v_loop_filter16y_inner_neon(uint8_t *dst, int stride,
+        int flim_E, int flim_I, int hev_thresh);
+void vp8_h_loop_filter16y_inner_neon(uint8_t *dst, int stride,
+        int flim_E, int flim_I, int hev_thresh);
+void vp8_v_loop_filter8uv_inner_neon(uint8_t *dstU, uint8_t *dstV, int stride,
+        int flim_E, int flim_I, int hev_thresh);
+void vp8_h_loop_filter8uv_inner_neon(uint8_t *dstU, uint8_t *dstV, int stride,
+        int flim_E, int flim_I, int hev_thresh);
+
+void vp8_v_loop_filter_simple_neon(uint8_t *dst, int stride, int flim);
+void vp8_h_loop_filter_simple_neon(uint8_t *dst, int stride, int flim);
+
+
+av_cold void ff_vp8dsp_init_arm(VP8DSPContext *dsp)
+{
+    if (HAVE_NEON) {
+        dsp->vp8_v_loop_filter16y = vp8_v_loop_filter16y_neon;
+        dsp->vp8_h_loop_filter16y = vp8_h_loop_filter16y_neon;
+        dsp->vp8_v_loop_filter8uv = vp8_v_loop_filter8uv_neon;
+        dsp->vp8_h_loop_filter8uv = vp8_h_loop_filter8uv_neon;
+
+        dsp->vp8_v_loop_filter16y_inner = vp8_v_loop_filter16y_inner_neon;
+        dsp->vp8_h_loop_filter16y_inner = vp8_h_loop_filter16y_inner_neon;
+        dsp->vp8_v_loop_filter8uv_inner = vp8_v_loop_filter8uv_inner_neon;
+        dsp->vp8_h_loop_filter8uv_inner = vp8_h_loop_filter8uv_inner_neon;
+
+#if 0
+        dsp->vp8_v_loop_filter_simple = vp8_v_loop_filter_simple_neon;
+        dsp->vp8_h_loop_filter_simple = vp8_h_loop_filter_simple_neon;
+#endif
+    }
+}
diff --git a/libavcodec/arm/vp8dsp_neon.S b/libavcodec/arm/vp8dsp_neon.S
new file mode 100644
index 0000000..f1d5de2
--- /dev/null
+++ b/libavcodec/arm/vp8dsp_neon.S
@@ -0,0 +1,475 @@
+/**
+ * VP8 compatible video decoder
+ *
+ * Copyright (C) 2010 Rob Clark <rob at ti.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"
+
+@ Register layout:
+@   P3..Q3 -> q0..q7
+@   flim_E -> q14
+@   flim_I -> q15
+@   hev_thresh -> r12 (ip)
+@
+.macro  vp8_loop_filter, inner=0
+        @ calculate hev and normal_limit:
+        vabd.u8         q12, q2,  q3            @ abs(P1-P0)
+        vabd.u8         q13, q5,  q4            @ abs(Q1-Q0)
+        vcle.u8         q8,  q12, q15           @ abs(P1-P0) <= flim_I
+        vcle.u8         q9,  q13, q15           @ abs(Q1-Q0) <= flim_I
+        vabd.u8         q10, q0,  q1            @ abs(P3-P2)
+        vabd.u8         q11, q1,  q2            @ abs(P2-P1)
+        vand            q8,  q8,  q9
+        vcle.u8         q10, q10, q15           @ abs(P3-P2) <= flim_I
+        vcle.u8         q11, q11, q15           @ abs(P2-P1) <= flim_I
+        vand            q8,  q8,  q10
+        vand            q8,  q8,  q11
+        vabd.u8         q10, q7,  q6            @ abs(Q3-Q2)
+        vabd.u8         q11, q6,  q5            @ abs(Q2-Q1)
+        vcle.u8         q10, q10, q15           @ abs(Q3-Q2) <= flim_I
+        vcle.u8         q11, q11, q15           @ abs(Q2-Q1) <= flim_I
+        vand            q8,  q8,  q10
+        vand            q8,  q8,  q11
+        vabd.u8         q9,  q3,  q4            @ abs(P0-Q0)
+        vabd.u8         q10, q2,  q5            @ abs(P1-Q1)
+        vqadd.u8        q9,  q9,  q9            @ abs(P0-Q0) * 2
+        vshr.u8         q10, q10, #1            @ abs(P1-Q1) / 2
+        vqadd.u8        q11, q9,  q10           @ (abs(P0-Q0)*2) + (abs(P1-Q1)/2)
+        vcle.u8         q11, q11, q14           @ (abs(P0-Q0)*2) + (abs(P1-Q1)/2) <= flim_E
+        vdup.8          q15, r12                @ hev_thresh
+        vand            q8,  q8,  q11
+        vcgt.u8         q12, q12, q15           @ abs(P1-P0) > hev_thresh
+        vcgt.u8         q13, q13, q15           @ abs(Q1-Q0) > hev_thresh
+        vorr            q9,  q12, q13
+
+        vmov.i8         q13, #0x80
+
+        @ at this point:
+        @   q8: normal_limit
+        @   q9: hev
+
+        @ convert to signed value:
+    .if ! \inner
+        veor            q1,  q1,  q13           @ PS2 = P2 ^ 0x80
+    .endif
+        veor            q2,  q2,  q13           @ PS1 = P1 ^ 0x80
+        veor            q3,  q3,  q13           @ PS0 = P0 ^ 0x80
+        veor            q4,  q4,  q13           @ QS0 = Q0 ^ 0x80
+        veor            q5,  q5,  q13           @ QS1 = Q1 ^ 0x80
+    .if ! \inner
+        veor            q6,  q6,  q13           @ QS2 = Q2 ^ 0x80
+    .endif
+
+        vmov.i16        q12, #3
+        vsubl.s8        q10, d8,  d6            @ QS0 - PS0
+        vsubl.s8        q11, d9,  d7            @   (widened to 16bit)
+        vmul.i16        q10, q10, q12           @ w = 3 * (QS0 - PS0)
+        vmul.i16        q11, q11, q12
+
+        vmov.i8         q14, #4
+        vmov.i8         q15, #3
+
+        vqsub.s8        q12, q2,  q5            @ clamp(PS1-QS1)
+    .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
+        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 \inner
+        @ filter_common is unconditional in inner blocks
+        vmov            q12, q10
+    .else
+        vand            q12, q10, q9            @ w & hev
+    .endif
+        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
+        vqsub.s8        q4,  q4,  q11           @ QS0 = clamp(QS0-c1)
+        vqadd.s8        q3,  q3,  q12           @ PS0 = clamp(PS0+c2)
+
+    .if \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);
+        vbic            q11, q11, q9            @ c1 & ~hev
+        vmov.i8         q15, #1
+        vadd.s8         q11, q11, q15           @ c3 = ((c1&~hev) + 1)
+        vshr.s8         q11, q11, #1            @ c3 >>= 1
+        vqsub.s8        q5,  q5,  q11           @ QS1 = clamp(QS1-c3)
+        vqadd.s8        q2,  q2,  q11           @ PS1 = clamp(PS1+c3)
+    .else
+
+        @ filter_mbedge:
+        @   a = clamp((27*w + 63) >> 7);
+        @   Q0 = s2u(QS0 - a);
+        @   P0 = s2u(PS0 + a);
+        vmov.i16        q14, #63
+        vbic            q10, q10, q9            @ w &= ~hev
+        vmov.i8         d30, #27
+        vmov            q11, q14
+        vmov            q12, q14
+        vmlal.s8        q11, d20, d30           @ 27*w + 63
+        vmlal.s8        q12, d21, d30
+        vqshrn.s16      d22, q11, #7            @ a = clamp((27*w + 63)>>7)
+        vqshrn.s16      d23, q12, #7
+        vqsub.s8        q4,  q4,  q11           @ QS0 = clamp(QS0-a)
+        vqadd.s8        q3,  q3,  q11           @ PS0 = clamp(PS0+a)
+        @   a = clamp((18*w + 63) >> 7);
+        @   Q1 = s2u(QS1 - a);
+        @   P1 = s2u(PS1 + a);
+        vmov.i8         d30, #18
+        vmov            q11, q14
+        vmov            q12, q14
+        vmlal.s8        q11, d20, d30           @ 18*w + 63
+        vmlal.s8        q12, d21, d30
+        vqshrn.s16      d22, q11, #7            @ a = clamp((18*w + 63)>>7)
+        vqshrn.s16      d23, q12, #7
+        vqsub.s8        q5,  q5,  q11           @ QS1 = clamp(QS1-a)
+        vqadd.s8        q2,  q2,  q11           @ PS1 = clamp(PS1+a)
+        @   a = clamp((9*w + 63) >> 7);
+        @   Q2 = s2u(QS2 - a);
+        @   P2 = s2u(PS2 + a);
+        vmov.i8         d30, #9
+        vmov            q11, q14
+        vmlal.s8        q11, d20, d30           @ 9*w + 63
+        vmlal.s8        q14, d21, d30
+        vqshrn.s16      d22, q11, #7            @ a = clamp((9*w + 63)>>7)
+        vqshrn.s16      d23, q14, #7
+        vqsub.s8        q6,  q6,  q11           @ QS2 = clamp(QS2-a)
+        vqadd.s8        q1,  q1,  q11           @ PS2 = clamp(PS2+a)
+    .endif
+
+        @ convert back to unsigned value:
+    .if ! \inner
+        veor            q1,  q1,  q13           @ P2 = PS2 ^ 0x80
+    .endif
+        veor            q2,  q2,  q13           @ P1 = PS1 ^ 0x80
+        veor            q3,  q3,  q13           @ P0 = PS0 ^ 0x80
+        veor            q4,  q4,  q13           @ Q0 = QS0 ^ 0x80
+        veor            q5,  q5,  q13           @ Q1 = QS1 ^ 0x80
+    .if ! \inner
+        veor            q6,  q6,  q13           @ Q2 = QS2 ^ 0x80
+    .endif
+.endm
+
+@ macro to transpose q0..q7 for h filter variants
+.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
+
+@ Filter top edge:
+@
+ at void vp8_v_loop_filter16y_(inner_)neon(uint8_t *dst, int stride,
+@        int flim_E, int flim_I, int hev_thresh)
+@{
+.macro vp8_v_loop_filter16y, inner
+        vpush           {q4-q7}
+        sub             r0,  r0,  r1,  lsl #2   @ subtract 4 rows
+        ldr             r12, [sp, #64]          @ load hev_thresh from stack
+
+        @ Load pixels:
+        vld1.8          {q0},     [r0], r1      @ P3
+        vld1.8          {q1},     [r0], r1      @ P2
+        vld1.8          {q2},     [r0], r1      @ P1
+        vld1.8          {q3},     [r0], r1      @ P0
+        vld1.8          {q4},     [r0], r1      @ Q0
+        vld1.8          {q5},     [r0], r1      @ Q1
+        vld1.8          {q6},     [r0], r1      @ Q2
+        vld1.8          {q7},     [r0]          @ Q3
+
+        vdup.8          q14, r2                 @ flim_E
+        vdup.8          q15, r3                 @ flim_I
+
+        vp8_loop_filter inner=\inner
+
+        @ back up to P2:  dst -= stride * 6
+        sub             r0,  r0,  r1,  lsl #2
+        sub             r0,  r0,  r1,  lsl #1
+
+        @ Store pixels:
+        vst1.8          {q1},     [r0], r1      @ P2
+        vst1.8          {q2},     [r0], r1      @ P1
+        vst1.8          {q3},     [r0], r1      @ P0
+        vst1.8          {q4},     [r0], r1      @ Q0
+        vst1.8          {q5},     [r0], r1      @ Q1
+        vst1.8          {q6},     [r0]          @ Q2
+
+        vpop            {q4-q7}
+        bx              lr
+.endm
+@}
+
+function vp8_v_loop_filter16y_neon, export=1
+        vp8_v_loop_filter16y inner=0
+endfunc
+function vp8_v_loop_filter16y_inner_neon, export=1
+        vp8_v_loop_filter16y inner=1
+endfunc
+
+ at void vp8_v_loop_filter8uv_(inner_)neon(uint8_t *dstU, uint8_t *dstV, int stride,
+@        int flim_E, int flim_I, int hev_thresh)
+@{
+.macro vp8_v_loop_filter8uv, inner
+        vpush           {q4-q7}
+        sub             r0,  r0,  r2,  lsl #2   @ subtract 4 rows from u
+        sub             r1,  r1,  r2,  lsl #2   @ subtract 4 rows from v
+        ldr             r12, [sp, #64]          @ load flim_I from stack
+
+        @ Load pixels:
+        vld1.8          {d0},     [r0], r2      @ P3
+        vld1.8          {d1},     [r1], r2      @ P3
+        vld1.8          {d2},     [r0], r2      @ P2
+        vld1.8          {d3},     [r1], r2      @ P2
+        vld1.8          {d4},     [r0], r2      @ P1
+        vld1.8          {d5},     [r1], r2      @ P1
+        vld1.8          {d6},     [r0], r2      @ P0
+        vld1.8          {d7},     [r1], r2      @ P0
+        vld1.8          {d8},     [r0], r2      @ Q0
+        vld1.8          {d9},     [r1], r2      @ Q0
+        vld1.8          {d10},    [r0], r2      @ Q1
+        vld1.8          {d11},    [r1], r2      @ Q1
+        vld1.8          {d12},    [r0], r2      @ Q2
+        vld1.8          {d13},    [r1], r2      @ Q2
+        vld1.8          {d14},    [r0]          @ Q3
+        vld1.8          {d15},    [r1]          @ 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 -= stride * 6
+        sub             r0,  r0,  r2,  lsl #2
+        sub             r0,  r0,  r2,  lsl #1
+
+        @ back up to P2:  v -= stride * 6
+        sub             r1,  r1,  r2,  lsl #2
+        sub             r1,  r1,  r2,  lsl #1
+
+        @ Store pixels:
+        vst1.8          {d2},     [r0], r2      @ P2
+        vst1.8          {d3},     [r1], r2      @ P2
+        vst1.8          {d4},     [r0], r2      @ P1
+        vst1.8          {d5},     [r1], r2      @ P1
+        vst1.8          {d6},     [r0], r2      @ P0
+        vst1.8          {d7},     [r1], r2      @ P0
+        vst1.8          {d8},     [r0], r2      @ Q0
+        vst1.8          {d9},     [r1], r2      @ Q0
+        vst1.8          {d10},    [r0], r2      @ Q1
+        vst1.8          {d11},    [r1], r2      @ Q1
+        vst1.8          {d12},    [r0]          @ Q2
+        vst1.8          {d13},    [r1]          @ Q2
+
+        vpop            {q4-q7}
+        bx              lr
+.endm
+@}
+
+function vp8_v_loop_filter8uv_neon, export=1
+        vp8_v_loop_filter8uv inner=0
+endfunc
+function vp8_v_loop_filter8uv_inner_neon, export=1
+        vp8_v_loop_filter8uv inner=1
+endfunc
+
+ at void vp8_v_loop_filter_simple_neon(uint8_t *dst, int stride, int flim)
+@{
+@}
+
+@ Filter left edge:
+@
+ at void vp8_h_loop_filter16y_(inner_)neon(uint8_t *dst, int stride,
+@        int flim_E, int flim_I, int hev_thresh)
+@{
+.macro vp8_h_loop_filter16y, inner
+        vpush           {q4-q7}
+        sub             r0,  r0,  #4            @ subtract 4 cols
+        ldr             r12, [sp, #64]          @ load hev_thresh from stack
+
+        @ 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
+        vdup.8          q15, r3                 @ flim_I
+
+        vp8_loop_filter inner=\inner
+
+        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
+.endm
+@}
+
+function vp8_h_loop_filter16y_neon, export=1
+        vp8_h_loop_filter16y inner=0
+endfunc
+function vp8_h_loop_filter16y_inner_neon, export=1
+        vp8_h_loop_filter16y inner=1
+endfunc
+
+ at void vp8_h_loop_filter8uv_(inner_)neon(uint8_t *dstU, uint8_t *dstV, int stride,
+@        int flim_E, int flim_I, int hev_thresh)
+@{
+.macro vp8_h_loop_filter8uv, inner
+        vpush           {q4-q7}
+        sub             r0,  r0,  #4            @ subtract 4 cols from u
+        sub             r1,  r1,  #4            @ subtract 4 rows from v
+        ldr             r12, [sp, #64]          @ load flim_I from stack
+
+        @ 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
+.endm
+@}
+
+function vp8_h_loop_filter8uv_neon, export=1
+        vp8_h_loop_filter8uv inner=0
+endfunc
+function vp8_h_loop_filter8uv_inner_neon, export=1
+        vp8_h_loop_filter8uv inner=1
+endfunc
+
+ at void vp8_h_loop_filter_simple_neon(uint8_t *dst, int stride, int flim)
+@{
+@}
+
diff --git a/libavcodec/vp8dsp.c b/libavcodec/vp8dsp.c
index b8cf0b2..168b515 100644
--- a/libavcodec/vp8dsp.c
+++ b/libavcodec/vp8dsp.c
@@ -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);
 }
diff --git a/libavcodec/vp8dsp.h b/libavcodec/vp8dsp.h
index ee5c7ec..e8757ab 100644
--- a/libavcodec/vp8dsp.h
+++ b/libavcodec/vp8dsp.h
@@ -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 */
-- 
1.7.1.1




More information about the ffmpeg-devel mailing list