lavf/qsv: Support multiple inputs overlay composition

Based on patch by Liu, ChaoX.
    lavfi/vf_overlay_qsv: support mulit-inputs overlay.

    Description:
    1. layout(grid/manual/overlay) default is the preset 'overlay' mode
    2. x_expr,y_expr,w_expr,z_expr,a_expr
        If layout is set to overlay, these exprs will be used to calculate
    the [overlay] channel's output rect while the [main] channle will be
    forced the same as the 1st input.
        If layout is set to manual, these expressions will be used to
    construct the layout. Some const keys and funcs can be used.
        If keyword 'array' is used, the expressions must have enough
    comma-seprated statements to cover each input or it will cause an error.
    Otherwise, the single statement will be applied to each window.

    3. Const values
        @array(\,) x_expr=array(expr1\,expr2\,...\,exprn) an array of
                   expressions indicating each windows x/y/w/h
        @x(n),y(n),w(n),h(n),alpha(n),iw(n),ih(n) a group of functions that can get
        each windows position attributions, where n indicates the index
        counting from 0.
        @n,n Const values, N indicates the number of input windows while n
        indicates the index of current window.

    Examples:
        1. ffmpeg -hwaccel qsv -c:v h264_qsv -i test.mp4 -filter_complex
           "split=4,overlay_qsv=4:manual:x_expr=w(n)*mod(n\,ceil(sqrt(N))):y_expr=h(n)*floor(n/ceil(sqrt(N))):w_expr=W/ceil(sqrt(N)):h_expr=H/ceil(sqrt(N))"
           -preset veryfast -c:v h264_qsv output.mp4
           When calculating the layout, 'n' will be replaced by the index of current window and this is the way how 'grid' mode calculate.
        2. ffmpeg -hwaccel qsv -c:v h264_qsv -i test.mp4 -filter_complex
           "split=4,overlay_qsv=4:manual:x_expr=array(0\,960\,0,960):y_expr=array(0\,0\,540\,540):w_expr=array(960,960,960,960):h_expr=array(540,540,540,540):a_expr=255"
           -preset veryfast -c:v h264_qsv output.mp4
           This command is equal to the former one when test.mp4 is of FHD.
        3. to overlay a video to area [(0,0), (256,256)] on another video, you can use:
           ffmpeg -hwaccel qsv -c:v h264_qsv -i test1.mp4 -hwaccel qsv -c:v h264_qsv -i test2.mp4 -filter_complex
           "overlay_qsv=x_expr=0:y_expr=0:w_expr=256:h_expr=256:a_expr=128" -preset veryfast -c:v h264_qsv output.mp4

Signed-off-by: Ruiling Song <ruiling.song@intel.com>

Author: ruiling-song

libavfilter/qsvvpp.h

@@ -31,6 +31,19 @@
 #define FF_INLINK_IDX(link)  ((int)((link)->dstpad - (link)->dst->input_pads))
 #define FF_OUTLINK_IDX(link) ((int)((link)->srcpad - (link)->src->output_pads))
 
+/* w or h = -n will be treated as factor n */
+#define eval_factor(ow, oh, link) do {\
+    int factor_w = 1, factor_h = 1; \
+    int iw = link->w, ih = link->h; \
+    if (ow < -1)           factor_w = -ow; \
+    if (oh < -1)           factor_h = -oh; \
+    if (ow < 0 && oh < 0)  ow = oh = 0; \
+    if (!ow)               ow = iw; \
+    if (!oh)               oh = ih; \
+    if (ow < 0)            ow = av_rescale(oh, iw, ih * factor_w) * factor_w; \
+    if (oh < 0)            oh = av_rescale(ow, ih, iw * factor_h) * factor_h; \
+} while (0);
+
 typedef struct QSVVPPContext QSVVPPContext;
 
 typedef struct QSVVPPCrop {