[FFmpeg-devel] [PATCH] ARMv6 optimised VP3 loop filter
David Conrad
lessen42
Sat Feb 13 11:01:05 CET 2010
10% faster overall on a 1g ipod touch
---
libavcodec/Makefile | 3 +
libavcodec/arm/dsputil_init_armv6.c | 8 ++
libavcodec/arm/vp3dsp_armv6.S | 126 +++++++++++++++++++++++++++++++++++
3 files changed, 137 insertions(+), 0 deletions(-)
create mode 100644 libavcodec/arm/vp3dsp_armv6.S
diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index 0b58459..6108d30 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -610,9 +610,12 @@ OBJS-$(HAVE_ARMV5TE) += arm/dsputil_init_armv5te.o \
arm/mpegvideo_armv5te_s.o \
arm/simple_idct_armv5te.o \
+ARMV6-OBJS-$(CONFIG_VP3_DECODER) += arm/vp3dsp_armv6.o
+
OBJS-$(HAVE_ARMV6) += arm/dsputil_init_armv6.o \
arm/dsputil_armv6.o \
arm/simple_idct_armv6.o \
+ $(ARMV6-OBJS-yes)
OBJS-$(HAVE_ARMVFP) += arm/dsputil_vfp.o \
arm/dsputil_init_vfp.o \
diff --git a/libavcodec/arm/dsputil_init_armv6.c b/libavcodec/arm/dsputil_init_armv6.c
index 3209062..2d7a354 100644
--- a/libavcodec/arm/dsputil_init_armv6.c
+++ b/libavcodec/arm/dsputil_init_armv6.c
@@ -70,6 +70,9 @@ int ff_sse16_armv6(void *s, uint8_t *blk1, uint8_t *blk2,
int ff_pix_norm1_armv6(uint8_t *pix, int line_size);
int ff_pix_sum_armv6(uint8_t *pix, int line_size);
+void ff_vp3_v_loop_filter_armv6(uint8_t *, int, int *);
+void ff_vp3_h_loop_filter_armv6(uint8_t *, int, int *);
+
void av_cold ff_dsputil_init_armv6(DSPContext* c, AVCodecContext *avctx)
{
if (!avctx->lowres && (avctx->idct_algo == FF_IDCT_AUTO ||
@@ -118,4 +121,9 @@ void av_cold ff_dsputil_init_armv6(DSPContext* c, AVCodecContext *avctx)
c->pix_norm1 = ff_pix_norm1_armv6;
c->pix_sum = ff_pix_sum_armv6;
+
+ if (CONFIG_VP3_DECODER) {
+ c->vp3_v_loop_filter = ff_vp3_v_loop_filter_armv6;
+ c->vp3_h_loop_filter = ff_vp3_h_loop_filter_armv6;
+ }
}
diff --git a/libavcodec/arm/vp3dsp_armv6.S b/libavcodec/arm/vp3dsp_armv6.S
new file mode 100644
index 0000000..3d751e6
--- /dev/null
+++ b/libavcodec/arm/vp3dsp_armv6.S
@@ -0,0 +1,126 @@
+/*
+ * Copyright (c) 2010 David Conrad
+ *
+ * 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"
+
+.text
+
+// ((p0 - p3) + 3*(p2 - p1) + 4) >> 3 has range -127, 128
+// so we instead calculate the negative, so it fits in signed 8 bits
+
+function ff_vp3_v_loop_filter_armv6, export=1
+ push {r4-r10,lr}
+ sub r0, r0, r1
+ ldr r2, [r2, #129*4]
+ ldr r3, =0x030003
+ ldr ip, =0xff00ff
+ mov lr, #8
+1:
+ ldr r4, [r0,-r1]
+ ldr r5, [r0]
+ ldr r6, [r0, r1]
+ ldr r7, [r0, r1, lsl #1]
+ subs lr, lr, #4
+
+ uxtb16 r10, r4
+ uxtb16 r8, r5
+ uxtb16 r9, r6
+ ssub16 r8, r8, r9
+ sadd16 r9, r8, r8
+ sadd16 r8, r8, r9 // 3*(p1-p2)
+ uxtb16 r9, r7
+ ssub16 r9, r9, r10
+ sadd16 r8, r8, r9
+ sadd16 r8, r8, r3 // (p3 - p0) + 3*(p1 - p2) + 3
+ uxtb16 r9, r5, ror #8
+ and r8, ip, r8, lsr #3
+
+ uxtb16 r10, r6, ror #8
+ ssub16 r9, r9, r10
+ sadd16 r10, r9, r9
+ sadd16 r9, r9, r10
+ uxtb16 r4, r4, ror #8
+ uxtb16 r7, r7, ror #8
+ ssub16 r4, r7, r4
+ sadd16 r4, r4, r9
+ sadd16 r4, r4, r3
+ mov r10, #0
+ and r4, ip, r4, lsr #3
+ orr r8, r8, r4, lsl #8 // 4 filter values
+
+ ssub8 r9, r10, r8
+ sel r9, r9, r8 // abs(filter_value)
+ uqsub8 r7, r2, r9
+ uqsub8 r9, r7, r9
+ usub8 r9, r7, r9
+// add and sub, then select the right one based on the original sign
+ uqsub8 r7, r6, r9
+ uqadd8 r6, r6, r9
+ uqadd8 r4, r5, r9
+ uqsub8 r5, r5, r9
+ ssub8 r10, r8, r10
+ sel r6, r6, r7
+ sel r5, r5, r4
+ str r6, [r0, r1]
+ str r5, [r0], #4
+ bgt 1b
+ pop {r4-r10,pc}
+.endfunc
+
+function ff_vp3_h_loop_filter_armv6, export=1
+ push {r4-r10,lr}
+ sub ip, r0, #1
+ sub r0, r0, #2
+ ldr r3, =0x010003
+ mov r10, #4
+ mov lr, #8
+1:
+ ldr r4, [r0], r1
+ ldr r7, [r0], r1
+ subs lr, lr, #2
+
+ uxtb16 r5, r4, ror #8
+ uxtb16 r6, r4, ror #16
+ ssub16 r6, r6, r5 // p0-p3 p2-p1
+ smlad r6, r6, r3, r10 // (p0-p3) + 3*(p2-p1) + 4
+
+ uxtb16 r8, r7, ror #8
+ uxtb16 r9, r7, ror #16
+ ssub16 r9, r9, r8
+ smlad r9, r9, r3, r10
+
+.macro filter p1 p2 r
+ asr \r, \r, #3
+ ldr \r, [r2, \r, lsl #2] // filter value
+ uxtb \p2, \p1, ror #16 // p2
+ uxtb \p1, \p1, ror #8 // p1
+ sub \p2, \p2, \r
+ add \p1, \p1, \r
+ usat \p2, #8, \p2
+ usat \p1, #8, \p1
+ strb \p2, [ip, #1]
+ strb \p1, [ip], r1
+.endm
+
+ filter r4, r5, r6
+ filter r7, r8, r9
+ bgt 1b
+ pop {r4-r10,pc}
+.endfunc
--
1.6.6
More information about the ffmpeg-devel
mailing list