[PATCH 1/2] wined3d: Support stretching compressed format surfaces in surface_cpu_blt().

[Henri Verbeet]

On Fri, 29 Nov 2019 at 17:58, Paul Gofman <gofmanp at gmail.com> wrote:
>  release:
> -    wined3d_context_unmap_bo_address(context, &dst_data, 0, 1, &dst_range);
> +    if (upload)
> +    {
> +        struct wined3d_bo_address data;
> +
> +        data.buffer_object = 0;
> +        data.addr = dst_map.data;
> +
> +        texture_level = dst_sub_resource_idx % dst_texture->level_count;
> +
> +        wined3d_texture_prepare_location(dst_texture, texture_level, context, WINED3D_LOCATION_TEXTURE_RGB);
> +        dst_texture->texture_ops->texture_upload_data(context, wined3d_const_bo_address(&data), dst_format,
> +                dst_box, dst_map.row_pitch, dst_map.slice_pitch, dst_texture, texture_level,
> +                WINED3D_LOCATION_TEXTURE_RGB, dst_box->left, dst_box->top, 0);
> +
> +        wined3d_texture_validate_location(dst_texture, texture_level, WINED3D_LOCATION_TEXTURE_RGB);
> +        wined3d_texture_invalidate_location(dst_texture, texture_level, ~WINED3D_LOCATION_TEXTURE_RGB);
> +
> +        heap_free(dst_map.data);
> +    }
> +    else
> +    {
> +        wined3d_context_unmap_bo_address(context, &dst_data, 0, 1, &dst_range);
> +    }
> +
>
It probably works out in practice because we only jump to "release:"
from error paths when "upload" is FALSE, but it doesn't seem right to
depend on that. I.e., we should probably explicitly check for success
before uploading here.