[FFmpeg-devel] [PATCH] avfilter/vf_maskedmerge: add SIMD for maskedmerge with 8 bit depth input
Paul B Mahol
onemda at gmail.com
Thu Oct 1 19:25:57 CEST 2015
Signed-off-by: Paul B Mahol <onemda at gmail.com>
---
libavfilter/maskedmerge.h | 39 +++++++++++++++++++++
libavfilter/vf_maskedmerge.c | 33 ++++++------------
libavfilter/x86/Makefile | 2 ++
libavfilter/x86/vf_maskedmerge.asm | 66 +++++++++++++++++++++++++++++++++++
libavfilter/x86/vf_maskedmerge_init.c | 39 +++++++++++++++++++++
5 files changed, 156 insertions(+), 23 deletions(-)
create mode 100644 libavfilter/maskedmerge.h
create mode 100644 libavfilter/x86/vf_maskedmerge.asm
create mode 100644 libavfilter/x86/vf_maskedmerge_init.c
diff --git a/libavfilter/maskedmerge.h b/libavfilter/maskedmerge.h
new file mode 100644
index 0000000..b198e65
--- /dev/null
+++ b/libavfilter/maskedmerge.h
@@ -0,0 +1,39 @@
+/*
+ * Copyright (c) 2015 Paul B Mahol
+ *
+ * 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 "avfilter.h"
+#include "framesync.h"
+
+typedef struct MaskedMergeContext {
+ const AVClass *class;
+ int width[4], height[4];
+ int nb_planes;
+ int planes;
+ int half, depth;
+ FFFrameSync fs;
+
+ void (*maskedmerge)(const uint8_t *bsrc, int blinesize,
+ const uint8_t *osrc, int olinesize,
+ const uint8_t *msrc, int mlinesize,
+ uint8_t *dst, int dlinesize, int w, int h,
+ int half, int shift);
+} MaskedMergeContext;
+
+void ff_maskedmerge_init_x86(MaskedMergeContext *s);
diff --git a/libavfilter/vf_maskedmerge.c b/libavfilter/vf_maskedmerge.c
index 41bd039..f442bb7 100644
--- a/libavfilter/vf_maskedmerge.c
+++ b/libavfilter/vf_maskedmerge.c
@@ -23,24 +23,9 @@
#include "libavutil/opt.h"
#include "avfilter.h"
#include "formats.h"
-#include "framesync.h"
#include "internal.h"
#include "video.h"
-
-typedef struct MaskedMergeContext {
- const AVClass *class;
- int width[4], height[4];
- int nb_planes;
- int planes;
- int max, half, depth;
- FFFrameSync fs;
-
- void (*maskedmerge)(const uint8_t *bsrc, int blinesize,
- const uint8_t *osrc, int olinesize,
- const uint8_t *msrc, int mlinesize,
- uint8_t *dst, int dlinesize, int w, int h,
- int max, int half, int shift);
-} MaskedMergeContext;
+#include "maskedmerge.h"
#define OFFSET(x) offsetof(MaskedMergeContext, x)
#define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
@@ -114,7 +99,7 @@ static int process_frame(FFFrameSync *fs)
mask->data[p], mask->linesize[p],
out->data[p], out->linesize[p],
s->width[p], s->height[p],
- s->max, s->half, s->depth);
+ s->half, s->depth);
}
}
out->pts = av_rescale_q(base->pts, s->fs.time_base, outlink->time_base);
@@ -126,13 +111,13 @@ static void maskedmerge8(const uint8_t *bsrc, int blinesize,
const uint8_t *osrc, int olinesize,
const uint8_t *msrc, int mlinesize,
uint8_t *dst, int dlinesize, int w, int h,
- int max, int half, int shift)
+ int half, int shift)
{
int x, y;
for (y = 0; y < h; y++) {
for (x = 0; x < w; x++) {
- dst[x] = ((256 - msrc[x]) * bsrc[x] + msrc[x] * osrc[x] + 128) >> 8;
+ dst[x] = bsrc[x] + ((msrc[x] * (osrc[x] - bsrc[x]) + 128) >> 8);
}
dst += dlinesize;
@@ -146,7 +131,7 @@ static void maskedmerge16(const uint8_t *bbsrc, int blinesize,
const uint8_t *oosrc, int olinesize,
const uint8_t *mmsrc, int mlinesize,
uint8_t *ddst, int dlinesize, int w, int h,
- int max, int half, int shift)
+ int half, int shift)
{
const uint16_t *bsrc = (const uint16_t *)bbsrc;
const uint16_t *osrc = (const uint16_t *)oosrc;
@@ -156,7 +141,7 @@ static void maskedmerge16(const uint8_t *bbsrc, int blinesize,
for (y = 0; y < h; y++) {
for (x = 0; x < w; x++) {
- dst[x] = ((max - msrc[x]) * bsrc[x] + msrc[x] * osrc[x] + half) >> shift;
+ dst[x] = bsrc[x] + ((msrc[x] * (osrc[x] - bsrc[x]) + half) >> shift);
}
dst += dlinesize / 2;
@@ -183,14 +168,16 @@ static int config_input(AVFilterLink *inlink)
s->width[0] = s->width[3] = inlink->w;
s->depth = desc->comp[0].depth;
- s->max = 1 << s->depth;
- s->half = s->max / 2;
+ s->half = (1 << s->depth) / 2;
if (desc->comp[0].depth == 8)
s->maskedmerge = maskedmerge8;
else
s->maskedmerge = maskedmerge16;
+ if (ARCH_X86)
+ ff_maskedmerge_init_x86(s);
+
return 0;
}
diff --git a/libavfilter/x86/Makefile b/libavfilter/x86/Makefile
index 5382027..db072cd 100644
--- a/libavfilter/x86/Makefile
+++ b/libavfilter/x86/Makefile
@@ -4,6 +4,7 @@ OBJS-$(CONFIG_GRADFUN_FILTER) += x86/vf_gradfun_init.o
OBJS-$(CONFIG_HQDN3D_FILTER) += x86/vf_hqdn3d_init.o
OBJS-$(CONFIG_IDET_FILTER) += x86/vf_idet_init.o
OBJS-$(CONFIG_INTERLACE_FILTER) += x86/vf_interlace_init.o
+OBJS-$(CONFIG_MASKEDMERGE_FILTER) += x86/vf_maskedmerge_init.o
OBJS-$(CONFIG_NOISE_FILTER) += x86/vf_noise.o
OBJS-$(CONFIG_PP7_FILTER) += x86/vf_pp7_init.o
OBJS-$(CONFIG_PSNR_FILTER) += x86/vf_psnr_init.o
@@ -20,6 +21,7 @@ YASM-OBJS-$(CONFIG_GRADFUN_FILTER) += x86/vf_gradfun.o
YASM-OBJS-$(CONFIG_HQDN3D_FILTER) += x86/vf_hqdn3d.o
YASM-OBJS-$(CONFIG_IDET_FILTER) += x86/vf_idet.o
YASM-OBJS-$(CONFIG_INTERLACE_FILTER) += x86/vf_interlace.o
+YASM-OBJS-$(CONFIG_MASKEDMERGE_FILTER) += x86/vf_maskedmerge.o
YASM-OBJS-$(CONFIG_PP7_FILTER) += x86/vf_pp7.o
YASM-OBJS-$(CONFIG_PSNR_FILTER) += x86/vf_psnr.o
YASM-OBJS-$(CONFIG_PULLUP_FILTER) += x86/vf_pullup.o
diff --git a/libavfilter/x86/vf_maskedmerge.asm b/libavfilter/x86/vf_maskedmerge.asm
new file mode 100644
index 0000000..462674a
--- /dev/null
+++ b/libavfilter/x86/vf_maskedmerge.asm
@@ -0,0 +1,66 @@
+;*****************************************************************************
+;* x86-optimized functions for maskedmerge filter
+;*
+;* Copyright (C) 2015 Paul B Mahol
+;*
+;* This file is part of FFmpeg.
+;*
+;* FFmpeg is free software; you can redistribute it and/or
+;* modify it under the terms of the GNU Lesser General Public
+;* License as published by the Free Software Foundation; either
+;* version 2.1 of the License, or (at your option) any later version.
+;*
+;* FFmpeg is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+;* Lesser General Public License for more details.
+;*
+;* You should have received a copy of the GNU Lesser General Public
+;* License along with FFmpeg; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;*****************************************************************************
+
+%include "libavutil/x86/x86util.asm"
+
+SECTION_RODATA
+
+pw_128: times 8 dw 128
+pw_256: times 8 dw 256
+
+SECTION .text
+
+INIT_XMM sse2
+cglobal maskedmerge8, 10, 11, 3, 0, bsrc, blinesize, osrc, olinesize, msrc, mlinesize, dst, dlinesize, w, h
+ mova m7, [pw_128]
+ pxor m6, m6
+.nextrow:
+ mov r10q, 0
+ %define x r10q
+
+ .loop:
+ movh m0, [bsrcq + x]
+ movh m1, [osrcq + x]
+ movh m3, [msrcq + x]
+ mova m4, [pw_256]
+ punpcklbw m0, m6
+ punpcklbw m1, m6
+ punpcklbw m3, m6
+ psubw m4, m3
+ pmullw m4, m0
+ pmullw m1, m3
+ paddw m1, m4
+ paddw m1, m7
+ psrlw m1, 8
+ packuswb m1, m1
+ movh [dstq + x], m1
+ add r10q, mmsize / 2
+ cmp r10q, wq
+ jl .loop
+
+ lea bsrcq, [bsrcq+blinesizeq]
+ lea osrcq, [osrcq+olinesizeq]
+ lea msrcq, [msrcq+mlinesizeq]
+ lea dstq, [dstq+dlinesizeq]
+ sub hd, 1
+ jg .nextrow
+REP_RET
diff --git a/libavfilter/x86/vf_maskedmerge_init.c b/libavfilter/x86/vf_maskedmerge_init.c
new file mode 100644
index 0000000..6d317f9
--- /dev/null
+++ b/libavfilter/x86/vf_maskedmerge_init.c
@@ -0,0 +1,39 @@
+/*
+ * Copyright (c) 2015 Paul B Mahol
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/attributes.h"
+#include "libavutil/cpu.h"
+#include "libavutil/x86/cpu.h"
+#include "libavfilter/maskedmerge.h"
+
+void ff_maskedmerge8_sse2(const uint8_t *bsrc, int blinesize,
+ const uint8_t *osrc, int olinesize,
+ const uint8_t *msrc, int mlinesize,
+ uint8_t *dst, int dlinesize, int w, int h,
+ int half, int shift);
+
+av_cold void ff_maskedmerge_init_x86(MaskedMergeContext *s)
+{
+ int cpu_flags = av_get_cpu_flags();
+
+ if (ARCH_X86_64 && EXTERNAL_SSE2(cpu_flags) && s->depth == 8) {
+ s->maskedmerge = ff_maskedmerge8_sse2;
+ }
+}
--
1.9.1
More information about the ffmpeg-devel
mailing list