[FFmpeg-cvslog] vf_ssim: x86 simd for ssim_4x4xN and ssim_endN.

Ronald S. Bultje git at videolan.org
Tue Jul 14 05:07:34 CEST 2015


ffmpeg | branch: master | Ronald S. Bultje <rsbultje at gmail.com> | Sun Jul 12 19:33:06 2015 -0400| [dfc58584b4e5c35d6aebc5d21223bf714024f0ea] | committer: Michael Niedermayer

vf_ssim: x86 simd for ssim_4x4xN and ssim_endN.

Both are 2-2.5x faster than their C counterpart.

Reviewed-by: Paul B Mahol <onemda at gmail.com>
Reviewed-by: James Almer <jamrial at gmail.com>
Signed-off-by: Michael Niedermayer <michael at niedermayer.cc>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=dfc58584b4e5c35d6aebc5d21223bf714024f0ea
---

 libavfilter/ssim.h             |   36 ++++++++
 libavfilter/vf_ssim.c          |   26 ++++--
 libavfilter/x86/Makefile       |    2 +
 libavfilter/x86/vf_ssim.asm    |  191 ++++++++++++++++++++++++++++++++++++++++
 libavfilter/x86/vf_ssim_init.c |   38 ++++++++
 5 files changed, 284 insertions(+), 9 deletions(-)

diff --git a/libavfilter/ssim.h b/libavfilter/ssim.h
new file mode 100644
index 0000000..cd3a6ee
--- /dev/null
+++ b/libavfilter/ssim.h
@@ -0,0 +1,36 @@
+/*
+ * Copyright (c) 2015 Ronald S. Bultje <rsbultje at gmail.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
+ */
+
+#ifndef LIBAVFILTER_SSIM_H
+#define LIBAVFILTER_SSIM_H
+
+#include <stddef.h>
+#include <stdint.h>
+
+typedef struct SSIMDSPContext {
+    void (*ssim_4x4_line)(const uint8_t *buf, ptrdiff_t buf_stride,
+                          const uint8_t *ref, ptrdiff_t ref_stride,
+                          int (*sums)[4], int w);
+    float (*ssim_end_line)(const int (*sum0)[4], const int (*sum1)[4], int w);
+} SSIMDSPContext;
+
+void ff_ssim_init_x86(SSIMDSPContext *dsp);
+
+#endif /* LIBAVFILTER_SSIM_H */
diff --git a/libavfilter/vf_ssim.c b/libavfilter/vf_ssim.c
index f7a259e..b5a61ee 100644
--- a/libavfilter/vf_ssim.c
+++ b/libavfilter/vf_ssim.c
@@ -42,6 +42,7 @@
 #include "drawutils.h"
 #include "formats.h"
 #include "internal.h"
+#include "ssim.h"
 #include "video.h"
 
 typedef struct SSIMContext {
@@ -59,6 +60,7 @@ typedef struct SSIMContext {
     int planeheight[4];
     int *temp;
     int is_rgb;
+    SSIMDSPContext dsp;
 } SSIMContext;
 
 #define OFFSET(x) offsetof(SSIMContext, x)
@@ -85,8 +87,8 @@ static void set_meta(AVDictionary **metadata, const char *key, char comp, float
     }
 }
 
-static void ssim_4x4xn(const uint8_t *main, int main_stride,
-                       const uint8_t *ref, int ref_stride,
+static void ssim_4x4xn(const uint8_t *main, ptrdiff_t main_stride,
+                       const uint8_t *ref, ptrdiff_t ref_stride,
                        int (*sums)[4], int width)
 {
     int x, y, z;
@@ -132,7 +134,7 @@ static float ssim_end1(int s1, int s2, int ss, int s12)
          / ((float)(fs1 * fs1 + fs2 * fs2 + ssim_c1) * (float)(vars + ssim_c2));
 }
 
-static float ssim_endn(int (*sum0)[4], int (*sum1)[4], int width)
+static float ssim_endn(const int (*sum0)[4], const int (*sum1)[4], int width)
 {
     float ssim = 0.0;
     int i;
@@ -145,7 +147,8 @@ static float ssim_endn(int (*sum0)[4], int (*sum1)[4], int width)
     return ssim;
 }
 
-static float ssim_plane(uint8_t *main, int main_stride,
+static float ssim_plane(SSIMDSPContext *dsp,
+                        uint8_t *main, int main_stride,
                         uint8_t *ref, int ref_stride,
                         int width, int height, void *temp)
 {
@@ -160,12 +163,12 @@ static float ssim_plane(uint8_t *main, int main_stride,
     for (y = 1; y < height; y++) {
         for (; z <= y; z++) {
             FFSWAP(void*, sum0, sum1);
-            ssim_4x4xn(&main[4 * z * main_stride], main_stride,
-                       &ref[4 * z * ref_stride], ref_stride,
-                       sum0, width);
+            dsp->ssim_4x4_line(&main[4 * z * main_stride], main_stride,
+                               &ref[4 * z * ref_stride], ref_stride,
+                               sum0, width);
         }
 
-        ssim += ssim_endn(sum0, sum1, width - 1);
+        ssim += dsp->ssim_end_line(sum0, sum1, width - 1);
     }
 
     return ssim / ((height - 1) * (width - 1));
@@ -187,7 +190,7 @@ static AVFrame *do_ssim(AVFilterContext *ctx, AVFrame *main,
     s->nb_frames++;
 
     for (i = 0; i < s->nb_components; i++) {
-        c[i] = ssim_plane(main->data[i], main->linesize[i],
+        c[i] = ssim_plane(&s->dsp, main->data[i], main->linesize[i],
                           ref->data[i], ref->linesize[i],
                           s->planewidth[i], s->planeheight[i], s->temp);
         ssimv += s->coefs[i] * c[i];
@@ -294,6 +297,11 @@ static int config_input_ref(AVFilterLink *inlink)
     if (!s->temp)
         return AVERROR(ENOMEM);
 
+    s->dsp.ssim_4x4_line = ssim_4x4xn;
+    s->dsp.ssim_end_line = ssim_endn;
+    if (ARCH_X86)
+        ff_ssim_init_x86(&s->dsp);
+
     return 0;
 }
 
diff --git a/libavfilter/x86/Makefile b/libavfilter/x86/Makefile
index 49f45b6..61be8c6 100644
--- a/libavfilter/x86/Makefile
+++ b/libavfilter/x86/Makefile
@@ -8,6 +8,7 @@ OBJS-$(CONFIG_NOISE_FILTER)                  += x86/vf_noise.o
 OBJS-$(CONFIG_PP7_FILTER)                    += x86/vf_pp7_init.o
 OBJS-$(CONFIG_PULLUP_FILTER)                 += x86/vf_pullup_init.o
 OBJS-$(CONFIG_SPP_FILTER)                    += x86/vf_spp.o
+OBJS-$(CONFIG_SSIM_FILTER)                   += x86/vf_ssim_init.o
 OBJS-$(CONFIG_TINTERLACE_FILTER)             += x86/vf_tinterlace_init.o
 OBJS-$(CONFIG_VOLUME_FILTER)                 += x86/af_volume_init.o
 OBJS-$(CONFIG_YADIF_FILTER)                  += x86/vf_yadif_init.o
@@ -19,6 +20,7 @@ YASM-OBJS-$(CONFIG_IDET_FILTER)              += x86/vf_idet.o
 YASM-OBJS-$(CONFIG_INTERLACE_FILTER)         += x86/vf_interlace.o
 YASM-OBJS-$(CONFIG_PP7_FILTER)               += x86/vf_pp7.o
 YASM-OBJS-$(CONFIG_PULLUP_FILTER)            += x86/vf_pullup.o
+YASM-OBJS-$(CONFIG_SSIM_FILTER)              += x86/vf_ssim.o
 YASM-OBJS-$(CONFIG_TINTERLACE_FILTER)        += x86/vf_interlace.o
 YASM-OBJS-$(CONFIG_VOLUME_FILTER)            += x86/af_volume.o
 YASM-OBJS-$(CONFIG_YADIF_FILTER)             += x86/vf_yadif.o x86/yadif-16.o x86/yadif-10.o
diff --git a/libavfilter/x86/vf_ssim.asm b/libavfilter/x86/vf_ssim.asm
new file mode 100644
index 0000000..be6a4d3
--- /dev/null
+++ b/libavfilter/x86/vf_ssim.asm
@@ -0,0 +1,191 @@
+;*****************************************************************************
+;* x86-optimized functions for ssim filter
+;*
+;* Copyright (C) 2015 Ronald S. Bultje <rsbultje at gmail.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 "libavutil/x86/x86util.asm"
+
+SECTION_RODATA
+
+pw_1: times 8 dw 1
+ssim_c1: times 4 dd 416 ;(.01*.01*255*255*64 + .5)
+ssim_c2: times 4 dd 235963 ;(.03*.03*255*255*64*63 + .5)
+
+SECTION .text
+
+%if ARCH_X86_64
+
+INIT_XMM ssse3
+cglobal ssim_4x4_line, 6, 8, 16, buf, buf_stride, ref, ref_stride, sums, w, buf_stride3, ref_stride3
+    lea     ref_stride3q, [ref_strideq*3]
+    lea     buf_stride3q, [buf_strideq*3]
+    pxor              m7, m7
+    mova             m15, [pw_1]
+
+.loop:
+    movh              m0, [bufq+buf_strideq*0]  ; a1
+    movh              m1, [refq+ref_strideq*0]  ; b1
+    movh              m2, [bufq+buf_strideq*1]  ; a2
+    movh              m3, [refq+ref_strideq*1]  ; b2
+    punpcklbw         m0, m7                    ; s1 [word]
+    punpcklbw         m1, m7                    ; s2 [word]
+    punpcklbw         m2, m7                    ; s1 [word]
+    punpcklbw         m3, m7                    ; s2 [word]
+    pmaddwd           m4, m0, m0                ; a1 * a1
+    pmaddwd           m5, m1, m1                ; b1 * b1
+    pmaddwd           m8, m2, m2                ; a2 * a2
+    pmaddwd           m9, m3, m3                ; b2 * b2
+    paddd             m4, m5                    ; ss
+    paddd             m8, m9                    ; ss
+    pmaddwd           m6, m0, m1                ; a1 * b1 = ss12
+    pmaddwd           m5, m2, m3                ; a2 * b2 = ss12
+    paddw             m0, m2
+    paddw             m1, m3
+    paddd             m6, m5                    ; s12
+    paddd             m4, m8                    ; ss
+
+    movh              m2, [bufq+buf_strideq*2]  ; a3
+    movh              m3, [refq+ref_strideq*2]  ; b3
+    movh              m5, [bufq+buf_stride3q]   ; a4
+    movh              m8, [refq+ref_stride3q]   ; b4
+    punpcklbw         m2, m7                    ; s1 [word]
+    punpcklbw         m3, m7                    ; s2 [word]
+    punpcklbw         m5, m7                    ; s1 [word]
+    punpcklbw         m8, m7                    ; s2 [word]
+    pmaddwd           m9, m2, m2                ; a3 * a3
+    pmaddwd          m10, m3, m3                ; b3 * b3
+    pmaddwd          m12, m5, m5                ; a4 * a4
+    pmaddwd          m13, m8, m8                ; b4 * b4
+    pmaddwd          m11, m2, m3                ; a3 * b3 = ss12
+    pmaddwd          m14, m5, m8                ; a4 * b4 = ss12
+    paddd             m9, m10
+    paddd            m12, m13
+    paddw             m0, m2
+    paddw             m1, m3
+    paddw             m0, m5
+    paddw             m1, m8
+    paddd             m6, m11
+    paddd             m4, m9
+    paddd             m6, m14
+    paddd             m4, m12
+
+    ; m0 = [word] s1 a,a,a,a,b,b,b,b
+    ; m1 = [word] s2 a,a,a,a,b,b,b,b
+    ; m4 = [dword] ss a,a,b,b
+    ; m6 = [dword] s12 a,a,b,b
+
+    pmaddwd           m0, m15                   ; [dword] s1 a,a,b,b
+    pmaddwd           m1, m15                   ; [dword] s2 a,a,b,b
+    phaddd            m0, m4                    ; [dword] s1 a, b, ss a, b
+    phaddd            m1, m6                    ; [dword] s2 a, b, s12 a, b
+    punpckhdq     m2, m0, m1                    ; [dword] ss a, s12 a, ss b, s12 b
+    punpckldq         m0, m1                    ; [dword] s1 a, s2 a, s1 b, s2 b
+    punpckhqdq    m1, m0, m2                    ; [dword] a s1, s2, ss, s12
+    punpcklqdq        m0, m2                    ; [dword] b s1, s2, ss, s12
+
+    mova  [sumsq+     0], m0
+    mova  [sumsq+mmsize], m1
+
+    add             bufq, mmsize/2
+    add             refq, mmsize/2
+    add            sumsq, mmsize*2
+    sub               wd, mmsize/8
+    jg .loop
+    RET
+
+%endif
+
+INIT_XMM sse4
+cglobal ssim_end_line, 3, 3, 6, sum0, sum1, w
+    pxor              m0, m0
+.loop:
+    mova              m1, [sum0q+mmsize*0]
+    mova              m2, [sum0q+mmsize*1]
+    mova              m3, [sum0q+mmsize*2]
+    mova              m4, [sum0q+mmsize*3]
+    paddd             m1, [sum1q+mmsize*0]
+    paddd             m2, [sum1q+mmsize*1]
+    paddd             m3, [sum1q+mmsize*2]
+    paddd             m4, [sum1q+mmsize*3]
+    paddd             m1, m2
+    paddd             m2, m3
+    paddd             m3, m4
+    paddd             m4, [sum0q+mmsize*4]
+    paddd             m4, [sum1q+mmsize*4]
+    TRANSPOSE4x4D      1, 2, 3, 4, 5
+
+    ; m1 = fs1, m2 = fs2, m3 = fss, m4 = fs12
+    pslld             m3, 6
+    pslld             m4, 6
+    pmulld            m5, m1, m2                ; fs1 * fs2
+    pmulld            m1, m1                    ; fs1 * fs1
+    pmulld            m2, m2                    ; fs2 * fs2
+    psubd             m3, m1
+    psubd             m4, m5                    ; covariance
+    psubd             m3, m2                    ; variance
+
+    ; m1 = fs1 * fs1, m2 = fs2 * fs2, m3 = variance, m4 = covariance, m5 = fs1 * fs2
+    paddd             m4, m4                    ; 2 * covariance
+    paddd             m5, m5                    ; 2 * fs1 * fs2
+    paddd             m1, m2                    ; fs1 * fs1 + fs2 * fs2
+    paddd             m3, [ssim_c2]             ; variance + ssim_c2
+    paddd             m4, [ssim_c2]             ; 2 * covariance + ssim_c2
+    paddd             m5, [ssim_c1]             ; 2 * fs1 * fs2 + ssim_c1
+    paddd             m1, [ssim_c1]             ; fs1 * fs1 + fs2 * fs2 + ssim_c1
+
+    ; convert to float
+    cvtdq2ps          m3, m3
+    cvtdq2ps          m4, m4
+    cvtdq2ps          m5, m5
+    cvtdq2ps          m1, m1
+    mulps             m4, m5
+    mulps             m3, m1
+    divps             m4, m3                    ; ssim_endl
+    addps             m0, m4                    ; ssim
+    add            sum0q, mmsize*4
+    add            sum1q, mmsize*4
+    sub               wd, 4
+    jg .loop
+
+    ; subps the ones we added too much
+    test              wd, wd
+    jz .end
+    add               wd, 4
+    test              wd, 2
+    jz .skip2
+    psrldq            m4, 8
+.skip2:
+    test              wd, 1
+    jz .skip1
+    psrldq            m4, 4
+.skip1:
+    subps             m0, m4
+
+.end:
+    movhlps           m4, m0
+    addps             m0, m4
+    movss             m4, m0
+    shufps            m0, m0, 1
+    addss             m0, m4
+%if ARCH_X86_32
+    movss            r0m, m0
+    fld             r0mp
+%endif
+    RET
diff --git a/libavfilter/x86/vf_ssim_init.c b/libavfilter/x86/vf_ssim_init.c
new file mode 100644
index 0000000..9514b25
--- /dev/null
+++ b/libavfilter/x86/vf_ssim_init.c
@@ -0,0 +1,38 @@
+/*
+ * Copyright (c) 2015 Ronald S. Bultje <rsbultje at gmail.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 "libavutil/x86/cpu.h"
+
+#include "libavfilter/ssim.h"
+
+void ff_ssim_4x4_line_ssse3(const uint8_t *buf, ptrdiff_t buf_stride,
+                            const uint8_t *ref, ptrdiff_t ref_stride,
+                            int (*sums)[4], int w);
+float ff_ssim_end_line_sse4(const int (*sum0)[4], const int (*sum1)[4], int w);
+
+void ff_ssim_init_x86(SSIMDSPContext *dsp)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (ARCH_X86_64 && EXTERNAL_SSSE3(cpu_flags))
+        dsp->ssim_4x4_line = ff_ssim_4x4_line_ssse3;
+    if (EXTERNAL_SSE4(cpu_flags))
+        dsp->ssim_end_line = ff_ssim_end_line_sse4;
+}



More information about the ffmpeg-cvslog mailing list