st/vdpau: Provide YV12 to NV12 putBits conversion v2

mplayer likes putting YV12 data, and if there is a buffer format mismatch,
the vdpau state tracker would try to reallocate the video surface as an
YV12 surface. A virtual driver doesn't like reallocating and doesn't like YV12
surfaces, so if we can't support YV12, try an YV12 to NV12 conversion
instead.

Also advertize that we actually can do the getBits and putBits conversion.

v2: A previous version of this patch prioritized conversion before
reallocating. This has been changed to prioritize reallocating in this version.

Cc: Christian König <[email protected]>
Acked-by: Christian König <[email protected]>
Signed-off-by: Thomas Hellstrom <[email protected]>

3 files changed, 130 insertions, 19 deletions

diff --git a/src/gallium/auxiliary/util/u_video.h b/src/gallium/auxiliary/util/u_video.h
index 99a8fd6e4f3..7cd6268d1fc 100644
--- a/src/gallium/auxiliary/util/u_video.h
+++ b/src/gallium/auxiliary/util/u_video.h
@@ -107,6 +107,48 @@ u_copy_nv12_to_yv12(void *const *destination_data,
    }
 }
 
+/**
+ * \brief  Copy YV12 chroma data while converting it NV12
+ *
+ * Given a set of YV12 source pointers and -pitches, copy the data to a
+ * layout typical for NV12 video buffers.
+ *
+ * \param source data[in]  The plane data pointers. Array of 3.
+ * \param source_pitches[in]  The plane pitches. Array of 3.
+ * \param dst_plane[in]  The destination plane to copy to. For NV12 always 1.
+ * \param dst_field[in]  The destination field if interlaced.
+ * \param dst_stride[in]  The destination stride for this plane.
+ * \param num_fields[in]  The number of fields in the video buffer.
+ * \param dst[in]  The destination plane pointer.
+ * \param width[in]  The source plane width.
+ * \param height[in]  The source plane height.
+ */
+static inline void
+u_copy_nv12_from_yv12(const void *const *source_data,
+                      uint32_t const *source_pitches,
+                      int dst_plane, int dst_field,
+                      int dst_stride, int num_fields,
+                      uint8_t *dst,
+                      int width, int height)
+{
+   int x, y;
+   unsigned u_stride = source_pitches[2] * num_fields;
+   unsigned v_stride = source_pitches[1] * num_fields;
+   uint8_t *u_src = (uint8_t *)source_data[2] + source_pitches[2] * dst_field;
+   uint8_t *v_src = (uint8_t *)source_data[1] + source_pitches[1] * dst_field;
+
+   /* TODO: SIMD */
+   for (y = 0; y < height; y++) {
+      for (x = 0; x < width; x++) {
+         dst[2*x] = u_src[x];
+         dst[2*x+1] = v_src[x];
+      }
+      u_src += u_stride;
+      v_src += v_stride;
+      dst += dst_stride;
+   }
+}
+
 static inline void
 u_copy_yv12_to_nv12(void *const *destination_data,
                     uint32_t const *destination_pitches,
diff --git a/src/gallium/state_trackers/vdpau/query.c b/src/gallium/state_trackers/vdpau/query.c
index e69c9b11f55..435cafd4041 100644
--- a/src/gallium/state_trackers/vdpau/query.c
+++ b/src/gallium/state_trackers/vdpau/query.c
@@ -123,8 +123,21 @@ vlVdpVideoSurfaceQueryGetPutBitsYCbCrCapabilities(VdpDevice device, VdpChromaTyp
 
    switch(bits_ycbcr_format) {
    case VDP_YCBCR_FORMAT_NV12:
+      *is_supported = surface_chroma_type == VDP_CHROMA_TYPE_420;
+      break;
+
    case VDP_YCBCR_FORMAT_YV12:
       *is_supported = surface_chroma_type == VDP_CHROMA_TYPE_420;
+
+      /* We can convert YV12 to NV12 on the fly! */
+      if (*is_supported &&
+          pscreen->is_video_format_supported(pscreen,
+                                             PIPE_FORMAT_NV12,
+                                             PIPE_VIDEO_PROFILE_UNKNOWN,
+                                             PIPE_VIDEO_ENTRYPOINT_BITSTREAM)) {
+         pipe_mutex_unlock(dev->mutex);
+         return VDP_STATUS_OK;
+      }
       break;
 
    case VDP_YCBCR_FORMAT_UYVY:
diff --git a/src/gallium/state_trackers/vdpau/surface.c b/src/gallium/state_trackers/vdpau/surface.c
index 9a80605e24c..e0dff4e90cb 100644
--- a/src/gallium/state_trackers/vdpau/surface.c
+++ b/src/gallium/state_trackers/vdpau/surface.c
@@ -304,9 +304,11 @@ vlVdpVideoSurfacePutBitsYCbCr(VdpVideoSurface surface,
                               uint32_t const *source_pitches)
 {
    enum pipe_format pformat = FormatYCBCRToPipe(source_ycbcr_format);
+   enum getbits_conversion conversion = CONVERSION_NONE;
    struct pipe_context *pipe;
    struct pipe_sampler_view **sampler_views;
    unsigned i, j;
+   unsigned usage = PIPE_TRANSFER_WRITE;
 
    vlVdpSurface *p_surf = vlGetDataHTAB(surface);
    if (!p_surf)
@@ -320,24 +322,53 @@ vlVdpVideoSurfacePutBitsYCbCr(VdpVideoSurface surface,
        return VDP_STATUS_INVALID_POINTER;
 
    pipe_mutex_lock(p_surf->device->mutex);
-   if (p_surf->video_buffer == NULL || pformat != p_surf->video_buffer->buffer_format) {
 
-      /* destroy the old one */
-      if (p_surf->video_buffer)
-         p_surf->video_buffer->destroy(p_surf->video_buffer);
+   if (p_surf->video_buffer == NULL ||
+       ((pformat != p_surf->video_buffer->buffer_format))) {
+      enum pipe_format nformat = pformat;
+      struct pipe_screen *screen = pipe->screen;
+
+      /* Determine the most suitable format for the new surface */
+      if (!screen->is_video_format_supported(screen, nformat,
+                                             PIPE_VIDEO_PROFILE_UNKNOWN,
+                                             PIPE_VIDEO_ENTRYPOINT_BITSTREAM)) {
+         nformat = screen->get_video_param(screen,
+                                           PIPE_VIDEO_PROFILE_UNKNOWN,
+                                           PIPE_VIDEO_ENTRYPOINT_BITSTREAM,
+                                           PIPE_VIDEO_CAP_PREFERED_FORMAT);
+         if (nformat == PIPE_FORMAT_NONE) {
+            pipe_mutex_unlock(p_surf->device->mutex);
+            return VDP_STATUS_NO_IMPLEMENTATION;
+         }
+      }
 
-      /* adjust the template parameters */
-      p_surf->templat.buffer_format = pformat;
+      if (p_surf->video_buffer == NULL  ||
+          nformat != p_surf->video_buffer->buffer_format) {
+         /* destroy the old one */
+         if (p_surf->video_buffer)
+            p_surf->video_buffer->destroy(p_surf->video_buffer);
 
-      /* and try to create the video buffer with the new format */
-      p_surf->video_buffer = pipe->create_video_buffer(pipe, &p_surf->templat);
+         /* adjust the template parameters */
+         p_surf->templat.buffer_format = nformat;
 
-      /* stil no luck? ok forget it we don't support it */
-      if (!p_surf->video_buffer) {
-         pipe_mutex_unlock(p_surf->device->mutex);
-         return VDP_STATUS_NO_IMPLEMENTATION;
+         /* and try to create the video buffer with the new format */
+         p_surf->video_buffer = pipe->create_video_buffer(pipe, &p_surf->templat);
+
+         /* stil no luck? ok forget it we don't support it */
+         if (!p_surf->video_buffer) {
+            pipe_mutex_unlock(p_surf->device->mutex);
+            return VDP_STATUS_NO_IMPLEMENTATION;
+         }
+         vlVdpVideoSurfaceClear(p_surf);
       }
-      vlVdpVideoSurfaceClear(p_surf);
+   }
+
+   if (pformat != p_surf->video_buffer->buffer_format) {
+      if (pformat == PIPE_FORMAT_YV12 &&
+          p_surf->video_buffer->buffer_format == PIPE_FORMAT_NV12)
+         conversion = CONVERSION_YV12_TO_NV12;
+      else
+         return VDP_STATUS_NO_IMPLEMENTATION;
    }
 
    sampler_views = p_surf->video_buffer->get_sampler_view_planes(p_surf->video_buffer);
@@ -349,21 +380,46 @@ vlVdpVideoSurfacePutBitsYCbCr(VdpVideoSurface surface,
    for (i = 0; i < 3; ++i) {
       unsigned width, height;
       struct pipe_sampler_view *sv = sampler_views[i];
+      struct pipe_resource *tex;
       if (!sv || !source_pitches[i]) continue;
 
+      tex = sv->texture;
       vlVdpVideoSurfaceSize(p_surf, i, &width, &height);
 
-      for (j = 0; j < sv->texture->array_size; ++j) {
+      for (j = 0; j < tex->array_size; ++j) {
          struct pipe_box dst_box = {
             0, 0, j,
             width, height, 1
          };
 
-         pipe->texture_subdata(pipe, sv->texture, 0,
-                               PIPE_TRANSFER_WRITE, &dst_box,
-                               source_data[i] + source_pitches[i] * j,
-                               source_pitches[i] * sv->texture->array_size,
-                               0);
+         if (conversion == CONVERSION_YV12_TO_NV12 && i == 1) {
+            struct pipe_transfer *transfer;
+            uint8_t *map;
+
+            map = pipe->transfer_map(pipe, tex, 0, usage,
+                                     &dst_box, &transfer);
+            if (!map) {
+               pipe_mutex_unlock(p_surf->device->mutex);
+               return VDP_STATUS_RESOURCES;
+            }
+
+            u_copy_nv12_from_yv12(source_data, source_pitches,
+                                  i, j, transfer->stride, tex->array_size,
+                                  map, dst_box.width, dst_box.height);
+
+            pipe_transfer_unmap(pipe, transfer);
+         } else {
+            pipe->texture_subdata(pipe, tex, 0,
+                                  PIPE_TRANSFER_WRITE, &dst_box,
+                                  source_data[i] + source_pitches[i] * j,
+                                  source_pitches[i] * tex->array_size,
+                                  0);
+         }
+         /*
+          * This surface has already been synced
+          * by the first map.
+          */
+         usage |= PIPE_TRANSFER_UNSYNCHRONIZED;
       }
    }
    pipe_mutex_unlock(p_surf->device->mutex);