avfilter/vf_removegrain: add x86 and x86_64 SSE2 functions

Speed of all modes increased by a factor between 7.4 and 19.8 largely depending
on whether bytes are unpacked into words.  Modes 2, 3, and 4 have been sped-up
by a factor of 43 (thanks quick sort!)

All modes are available on x86_64 but only modes 1, 10, 11, 12, 13, 14, 19, 20,
21, and 22 are available on x86 due to the number of SIMD registers used.

With a contribution from James Almer <jamrial@gmail.com>

LICENSE.md

@@ -16,6 +16,7 @@ Specifically, the GPL parts of FFmpeg are:
 - optional x86 optimizations in the files
   - `libavcodec/x86/flac_dsp_gpl.asm`
   - `libavcodec/x86/idct_mmx.c`
+  - `libavfilter/x86/vf_removegrain.asm`
 - libutvideo encoding/decoding wrappers in
   `libavcodec/libutvideo*.cpp`
 - the X11 grabber in `libavdevice/x11grab.c`

libavfilter/removegrain.h

+/*
+ * Copyright (c) 2015 Paul B Mahol
+ * Copyright (c) 2015 James Darnley
+ *
+ * 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"
+
+typedef struct RemoveGrainContext {
+    const AVClass *class;
+
+    int mode[4];
+
+    int nb_planes;
+    int planewidth[4];
+    int planeheight[4];
+    int skip_even;
+    int skip_odd;
+
+    int (*rg[4])(int c, int a1, int a2, int a3, int a4, int a5, int a6, int a7, int a8);
+
+    void (*fl[4])(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+} RemoveGrainContext;
+
+void ff_removegrain_init_x86(RemoveGrainContext *rg);

libavfilter/vf_removegrain.c

@@ -2,6 +2,7 @@
  * Copyright (c) 2012 Laurent de Soras
  * Copyright (c) 2013 Fredrik Mellbin
  * Copyright (c) 2015 Paul B Mahol
+ * Copyright (c) 2015 James Darnley
  *
  * This file is part of FFmpeg.
  *
@@ -20,32 +21,15 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
-/*
- * TODO: add SIMD
- */
-
 #include "libavutil/imgutils.h"
 #include "libavutil/opt.h"
 #include "libavutil/pixdesc.h"
 #include "avfilter.h"
 #include "formats.h"
 #include "internal.h"
+#include "removegrain.h"
 #include "video.h"
 
-typedef struct RemoveGrainContext {
-    const AVClass *class;
-
-    int mode[4];
-
-    int nb_planes;
-    int planewidth[4];
-    int planeheight[4];
-    int skip_even;
-    int skip_odd;
-
-    int (*rg[4])(int c, int a1, int a2, int a3, int a4, int a5, int a6, int a7, int a8);
-} RemoveGrainContext;
-
 #define OFFSET(x) offsetof(RemoveGrainContext, x)
 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
 
@@ -142,6 +126,7 @@ static int mode05(int c, int a1, int a2, int a3, int a4, int a5, int a6, int a7,
 
     const int mindiff = FFMIN(FFMIN(c1, c2), FFMIN(c3, c4));
 
+    /* When adding SIMD notice the return order here: 4, 2, 3, 1. */
     if (mindiff == c4) {
         return av_clip(c, mi4, ma4);
     } else if (mindiff == c2) {
@@ -524,6 +509,9 @@ static int config_input(AVFilterLink *inlink)
         }
     }
 
+    if (ARCH_X86)
+        ff_removegrain_init_x86(s);
+
     return 0;
 }
 
@@ -566,7 +554,19 @@ static int filter_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
         }
 
         *dst++ = *src++;
-        for (x = 1; x < s->planewidth[i] - 1; x++) {
+
+        if (s->fl[i]) {
+            int w_asm = (s->planewidth[i] - 2) & ~15;
+
+            s->fl[i](dst, src, in->linesize[i], w_asm);
+
+            x = 1 + w_asm;
+            dst += w_asm;
+            src += w_asm;
+        } else
+            x = 1;
+
+        for (; x < s->planewidth[i] - 1; x++) {
             const int a1 = src[-op];
             const int a2 = src[-o0];
             const int a3 = src[-om];

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-$(CONFIG_PSNR_FILTER)                   += x86/vf_psnr_init.o
 OBJS-$(CONFIG_PULLUP_FILTER)                 += x86/vf_pullup_init.o
+OBJS-$(CONFIG_REMOVEGRAIN_FILTER)            += x86/vf_removegrain_init.o
 OBJS-$(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
@@ -22,6 +23,9 @@ YASM-OBJS-$(CONFIG_INTERLACE_FILTER)         += x86/vf_interlace.o
 YASM-OBJS-$(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
+ifdef CONFIG_GPL
+YASM-OBJS-$(CONFIG_REMOVEGRAIN_FILTER)       += x86/vf_removegrain.o
+endif
 YASM-OBJS-$(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

libavfilter/x86/vf_removegrain_init.c

+/*
+ * Copyright (c) 2015 James Darnley
+ *
+ * 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/removegrain.h"
+
+void ff_rg_fl_mode_1_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_10_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_11_12_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_13_14_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_19_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_20_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_21_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_22_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+#if ARCH_X86_64
+void ff_rg_fl_mode_2_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_3_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_4_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_5_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_6_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_7_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_8_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_9_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_15_16_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_17_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_18_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_23_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+void ff_rg_fl_mode_24_sse2(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int pixels);
+#endif
+
+av_cold void ff_removegrain_init_x86(RemoveGrainContext *rg)
+{
+#if CONFIG_GPL
+    int cpu_flags = av_get_cpu_flags();
+    int i;
+
+    for (i = 0; i < rg->nb_planes; i++) {
+        if (EXTERNAL_SSE2(cpu_flags))
+            switch (rg->mode[i]) {
+                case 1: rg->fl[i] = ff_rg_fl_mode_1_sse2; break;
+                case 10: rg->fl[i] = ff_rg_fl_mode_10_sse2; break;
+                case 11: /* fall through */
+                case 12: rg->fl[i] = ff_rg_fl_mode_11_12_sse2; break;
+                case 13: /* fall through */
+                case 14: rg->fl[i] = ff_rg_fl_mode_13_14_sse2; break;
+                case 19: rg->fl[i] = ff_rg_fl_mode_19_sse2; break;
+                case 20: rg->fl[i] = ff_rg_fl_mode_20_sse2; break;
+                case 21: rg->fl[i] = ff_rg_fl_mode_21_sse2; break;
+                case 22: rg->fl[i] = ff_rg_fl_mode_22_sse2; break;
+#if ARCH_X86_64
+                case 2: rg->fl[i] = ff_rg_fl_mode_2_sse2; break;
+                case 3: rg->fl[i] = ff_rg_fl_mode_3_sse2; break;
+                case 4: rg->fl[i] = ff_rg_fl_mode_4_sse2; break;
+                case 5: rg->fl[i] = ff_rg_fl_mode_5_sse2; break;
+                case 6: rg->fl[i] = ff_rg_fl_mode_6_sse2; break;
+                case 7: rg->fl[i] = ff_rg_fl_mode_7_sse2; break;
+                case 8: rg->fl[i] = ff_rg_fl_mode_8_sse2; break;
+                case 9: rg->fl[i] = ff_rg_fl_mode_9_sse2; break;
+                case 15: /* fall through */
+                case 16: rg->fl[i] = ff_rg_fl_mode_15_16_sse2; break;
+                case 17: rg->fl[i] = ff_rg_fl_mode_17_sse2; break;
+                case 18: rg->fl[i] = ff_rg_fl_mode_18_sse2; break;
+                case 23: rg->fl[i] = ff_rg_fl_mode_23_sse2; break;
+                case 24: rg->fl[i] = ff_rg_fl_mode_24_sse2; break;
+#endif /* ARCH_x86_64 */
+            }
+    }
+#endif /* CONFIG_GPL */
+}