avfilter/vf_overlay: exclude nv12/nv21 formats from x86 asm check

They are yet to be supported,
Signed-off-by: Paul B Mahol <onemda@gmail.com>

libavfilter/vf_overlay.c

@@ -921,7 +921,8 @@ static int config_input_main(AVFilterLink *inlink)
 
 end:
     if (ARCH_X86)
-        ff_overlay_init_x86(s, s->format, s->alpha_format, s->main_has_alpha);
+        ff_overlay_init_x86(s, s->format, inlink->format,
+                            s->alpha_format, s->main_has_alpha);
 
     return 0;
 }

libavfilter/vf_overlay.h

@@ -80,6 +80,7 @@ typedef struct OverlayContext {
     int (*blend_slice)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
 } OverlayContext;
 
-void ff_overlay_init_x86(OverlayContext *s, int format, int alpha_format, int main_has_alpha);
+void ff_overlay_init_x86(OverlayContext *s, int format, int pix_format,
+                         int alpha_format, int main_has_alpha);
 
 #endif /* AVFILTER_OVERLAY_H */

libavfilter/x86/vf_overlay_init.c

@@ -32,7 +32,8 @@ int ff_overlay_row_20_sse4(uint8_t *d, uint8_t *da, uint8_t *s, uint8_t *a,
 int ff_overlay_row_22_sse4(uint8_t *d, uint8_t *da, uint8_t *s, uint8_t *a,
                            int w, ptrdiff_t alinesize);
 
-av_cold void ff_overlay_init_x86(OverlayContext *s, int format, int alpha_format, int main_has_alpha)
+av_cold void ff_overlay_init_x86(OverlayContext *s, int format, int pix_format,
+                                 int alpha_format, int main_has_alpha)
 {
     int cpu_flags = av_get_cpu_flags();
 
@@ -46,6 +47,7 @@ av_cold void ff_overlay_init_x86(OverlayContext *s, int format, int alpha_format
     }
 
     if (EXTERNAL_SSE4(cpu_flags) &&
+        (pix_format == AV_PIX_FMT_YUV420P) &&
         (format == OVERLAY_FORMAT_YUV420) &&
         alpha_format == 0 && main_has_alpha == 0) {
         s->blend_row[0] = ff_overlay_row_44_sse4;