avfilter/vf_bwdif: Add a filter_line3 method for optimisation

Add an optional filter_line3 to the available optimisations.

filter_line3 is equivalent to filter_line, memcpy, filter_line

filter_line shares quite a number of loads and some calculations in
common with its next iteration and testing shows that using aarch64
neon filter_line3s performance is 30% better than two filter_lines
and a memcpy.

Adds a test for vf_bwdif filter_line3 to checkasm

Rounds job start lines down to a multiple of 4. This means that if
filter_line3 exists then filter_line will not sometimes be called
once at the end of a slice depending on thread count. The final slice
may do up to 3 extra lines but filter_edge is faster than filter_line
so it is unlikely to create any noticable thread load variation.
Signed-off-by: John Cox <jc@kynesim.co.uk>
Signed-off-by: Martin Storsjö <martin@martin.st>

libavfilter/bwdif.h

@@ -35,6 +35,9 @@ typedef struct BWDIFContext {
     void (*filter_edge)(void *dst, void *prev, void *cur, void *next,
                         int w, int prefs, int mrefs, int prefs2, int mrefs2,
                         int parity, int clip_max, int spat);
+    void (*filter_line3)(void *dst, int dstride,
+                         const void *prev, const void *cur, const void *next, int prefs,
+                         int w, int parity, int clip_max);
 } BWDIFContext;
 
 void ff_bwdif_init_filter_line(BWDIFContext *bwdif, int bit_depth);
@@ -53,4 +56,8 @@ void ff_bwdif_filter_line_c(void *dst1, void *prev1, void *cur1, void *next1,
                             int prefs3, int mrefs3, int prefs4, int mrefs4,
                             int parity, int clip_max);
 
+void ff_bwdif_filter_line3_c(void * dst1, int d_stride,
+                             const void * prev1, const void * cur1, const void * next1, int s_stride,
+                             int w, int parity, int clip_max);
+
 #endif /* AVFILTER_BWDIF_H */

libavfilter/vf_bwdif.c

@@ -150,6 +150,31 @@ void ff_bwdif_filter_line_c(void *dst1, void *prev1, void *cur1, void *next1,
     FILTER2()
 }
 
+#define NEXT_LINE()\
+    dst += d_stride; \
+    prev += prefs; \
+    cur  += prefs; \
+    next += prefs;
+
+void ff_bwdif_filter_line3_c(void * dst1, int d_stride,
+                             const void * prev1, const void * cur1, const void * next1, int s_stride,
+                             int w, int parity, int clip_max)
+{
+    const int prefs = s_stride;
+    uint8_t * dst  = dst1;
+    const uint8_t * prev = prev1;
+    const uint8_t * cur  = cur1;
+    const uint8_t * next = next1;
+
+    ff_bwdif_filter_line_c(dst, (void*)prev, (void*)cur, (void*)next, w,
+                           prefs, -prefs, prefs * 2, - prefs * 2, prefs * 3, -prefs * 3, prefs * 4, -prefs * 4, parity, clip_max);
+    NEXT_LINE();
+    memcpy(dst, cur, w);
+    NEXT_LINE();
+    ff_bwdif_filter_line_c(dst, (void*)prev, (void*)cur, (void*)next, w,
+                           prefs, -prefs, prefs * 2, - prefs * 2, prefs * 3, -prefs * 3, prefs * 4, -prefs * 4, parity, clip_max);
+}
+
 void ff_bwdif_filter_edge_c(void *dst1, void *prev1, void *cur1, void *next1,
                             int w, int prefs, int mrefs, int prefs2, int mrefs2,
                             int parity, int clip_max, int spat)
@@ -212,6 +237,13 @@ static void filter_edge_16bit(void *dst1, void *prev1, void *cur1, void *next1,
     FILTER2()
 }
 
+// Round job start line down to multiple of 4 so that if filter_line3 exists
+// and the frame is a multiple of 4 high then filter_line will never be called
+static inline int job_start(const int jobnr, const int nb_jobs, const int h)
+{
+    return jobnr >= nb_jobs ? h : ((h * jobnr) / nb_jobs) & ~3;
+}
+
 static int filter_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
     BWDIFContext *s = ctx->priv;
@@ -221,8 +253,8 @@ static int filter_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
     int clip_max = (1 << (yadif->csp->comp[td->plane].depth)) - 1;
     int df = (yadif->csp->comp[td->plane].depth + 7) / 8;
     int refs = linesize / df;
-    int slice_start = (td->h *  jobnr   ) / nb_jobs;
-    int slice_end   = (td->h * (jobnr+1)) / nb_jobs;
+    int slice_start = job_start(jobnr, nb_jobs, td->h);
+    int slice_end   = job_start(jobnr + 1, nb_jobs, td->h);
     int y;
 
     for (y = slice_start; y < slice_end; y++) {
@@ -244,6 +276,11 @@ static int filter_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
                                refs << 1, -(refs << 1),
                                td->parity ^ td->tff, clip_max,
                                (y < 2) || ((y + 3) > td->h) ? 0 : 1);
+            } else if (s->filter_line3 && y + 2 < slice_end && y + 6 < td->h) {
+                s->filter_line3(dst, td->frame->linesize[td->plane],
+                                prev, cur, next, linesize, td->w,
+                                td->parity ^ td->tff, clip_max);
+                y += 2;
             } else {
                 s->filter_line(dst, prev, cur, next, td->w,
                                refs, -refs, refs << 1, -(refs << 1),
@@ -280,7 +317,7 @@ static void filter(AVFilterContext *ctx, AVFrame *dstpic,
         td.plane = i;
 
         ff_filter_execute(ctx, filter_slice, &td, NULL,
-                          FFMIN(h, ff_filter_get_nb_threads(ctx)));
+                          FFMIN((h+3)/4, ff_filter_get_nb_threads(ctx)));
     }
     if (yadif->current_field == YADIF_FIELD_END) {
         yadif->current_field = YADIF_FIELD_NORMAL;
@@ -357,6 +394,7 @@ static int config_props(AVFilterLink *link)
 
 av_cold void ff_bwdif_init_filter_line(BWDIFContext *s, int bit_depth)
 {
+    s->filter_line3 = 0;
     if (bit_depth > 8) {
         s->filter_intra = filter_intra_16bit;
         s->filter_line  = filter_line_c_16bit;

tests/checkasm/vf_bwdif.c

@@ -28,6 +28,10 @@
     for (size_t i = 0; i < count; i++) \
         buf0[i] = buf1[i] = rnd() & mask
 
+#define randomize_overflow_check(buf0, buf1, mask, count) \
+    for (size_t i = 0; i < count; i++) \
+        buf0[i] = buf1[i] = (rnd() & 1) != 0 ? mask : 0;
+
 #define BODY(type, depth)                                                      \
     do {                                                                       \
         type prev0[9*WIDTH], prev1[9*WIDTH];                                   \
@@ -83,6 +87,83 @@ void checkasm_check_vf_bwdif(void)
         report("bwdif10");
     }
 
+    if (!ctx_8.filter_line3)
+        ctx_8.filter_line3 = ff_bwdif_filter_line3_c;
+
+    {
+        LOCAL_ALIGNED_16(uint8_t, prev0, [11*WIDTH]);
+        LOCAL_ALIGNED_16(uint8_t, prev1, [11*WIDTH]);
+        LOCAL_ALIGNED_16(uint8_t, next0, [11*WIDTH]);
+        LOCAL_ALIGNED_16(uint8_t, next1, [11*WIDTH]);
+        LOCAL_ALIGNED_16(uint8_t, cur0,  [11*WIDTH]);
+        LOCAL_ALIGNED_16(uint8_t, cur1,  [11*WIDTH]);
+        LOCAL_ALIGNED_16(uint8_t, dst0,  [WIDTH*3]);
+        LOCAL_ALIGNED_16(uint8_t, dst1,  [WIDTH*3]);
+        const int stride = WIDTH;
+        const int mask = (1<<8)-1;
+        int parity;
+
+        for (parity = 0; parity != 2; ++parity) {
+            if (check_func(ctx_8.filter_line3, "bwdif8.line3.rnd.p%d", parity)) {
+
+                declare_func(void, void * dst1, int d_stride,
+                                          const void * prev1, const void * cur1, const void * next1, int prefs,
+                                          int w, int parity, int clip_max);
+
+                randomize_buffers(prev0, prev1, mask, 11*WIDTH);
+                randomize_buffers(next0, next1, mask, 11*WIDTH);
+                randomize_buffers( cur0,  cur1, mask, 11*WIDTH);
+
+                call_ref(dst0, stride,
+                         prev0 + stride * 4, cur0 + stride * 4, next0 + stride * 4, stride,
+                         WIDTH, parity, mask);
+                call_new(dst1, stride,
+                         prev1 + stride * 4, cur1 + stride * 4, next1 + stride * 4, stride,
+                         WIDTH, parity, mask);
+
+                if (memcmp(dst0, dst1, WIDTH*3)
+                        || memcmp(prev0, prev1, WIDTH*11)
+                        || memcmp(next0, next1, WIDTH*11)
+                        || memcmp( cur0,  cur1, WIDTH*11))
+                    fail();
+
+                bench_new(dst1, stride,
+                         prev1 + stride * 4, cur1 + stride * 4, next1 + stride * 4, stride,
+                         WIDTH, parity, mask);
+            }
+        }
+
+        // Use just 0s and ~0s to try to provoke bad cropping or overflow
+        // Parity makes no difference to this test so just test 0
+        if (check_func(ctx_8.filter_line3, "bwdif8.line3.overflow")) {
+
+            declare_func(void, void * dst1, int d_stride,
+                                      const void * prev1, const void * cur1, const void * next1, int prefs,
+                                      int w, int parity, int clip_max);
+
+            randomize_overflow_check(prev0, prev1, mask, 11*WIDTH);
+            randomize_overflow_check(next0, next1, mask, 11*WIDTH);
+            randomize_overflow_check( cur0,  cur1, mask, 11*WIDTH);
+
+            call_ref(dst0, stride,
+                     prev0 + stride * 4, cur0 + stride * 4, next0 + stride * 4, stride,
+                     WIDTH, 0, mask);
+            call_new(dst1, stride,
+                     prev1 + stride * 4, cur1 + stride * 4, next1 + stride * 4, stride,
+                     WIDTH, 0, mask);
+
+            if (memcmp(dst0, dst1, WIDTH*3)
+                    || memcmp(prev0, prev1, WIDTH*11)
+                    || memcmp(next0, next1, WIDTH*11)
+                    || memcmp( cur0,  cur1, WIDTH*11))
+                fail();
+
+            // No point to benching
+        }
+
+        report("bwdif8.line3");
+    }
+
     {
         LOCAL_ALIGNED_16(uint8_t, prev0, [11*WIDTH]);
         LOCAL_ALIGNED_16(uint8_t, prev1, [11*WIDTH]);